WO2022161000A1

WO2022161000A1 - Training method and training apparatus for machine learning model, and evaluation system

Info

Publication number: WO2022161000A1
Application number: PCT/CN2021/138638
Authority: WO
Inventors: 哈特曼大卫·西蒙; 罗小帆; 赵则昂
Original assignee: 苏州奇流信息科技有限公司
Priority date: 2021-01-29
Filing date: 2021-12-16
Publication date: 2022-08-04
Also published as: CN114818401A

Abstract

A training method and training apparatus for a machine learning model for evaluating the performance of a 3D-printed member, an evaluation system for evaluating the performance of a 3D-printed member, a computer device, and a computer-readable storage medium. In the training method for a machine learning model, variables that affect the printing quality of a printed member are pre-determined, different printing parameters are set for different printed member models to perform actual printing and simulated printing so as to obtain training data, measured values of the performance of the printed member in an actual printing environment in a training data set are taken as output, and equivalent relaxation time data and/or residual stress data are/is taken as input, such that after training is completed, the machine learning model can evaluate the performance of the printed member on the basis of the equivalent relaxation time data and/or the residual stress data of the printed member, and thus printing parameters of printed members of different types can be associated with printing quality.

Description

Training methods, training devices, and evaluation systems for machine learning models

technical field

The present application relates to the field of computer data processing, and in particular to a method for training a machine learning model for evaluating the performance of 3D printing components, a training device for a machine learning model, and an evaluation system, computer equipment, and computer software for evaluating the performance of 3D printing components. Read the storage medium.

Background technique

In the current 3D printing technology, such as in FDM (Fused Deposition Modeling) printing, the bonding strength of the interlayer bonding between different printing layers and the thermal coupling relaxation behavior of the printed polymer material have an impact on the printed object. The mechanical properties have a great influence; the temperature of the wire contact interface and the diffusion time of the polymer molecules of the printing material affect the bonding quality of the wire, and the heating time and cooling rate during printing affect the residual stress and microscopic molecular chain conformation of the wire Therefore, the settings of printing parameters such as extrusion speed, printing speed, layer height, material heating temperature, printing chamber temperature, etc. affect the performance indicators such as the accuracy and bonding strength of the final printed object. In the existing fused deposition printing slicing software, the printing path and printing parameters are determined in a purely geometric way, which cannot effectively guarantee the printing speed and the performance of the final printed product, such as forming ability, shape distortion, interlayer bonding, elastic modulus and strength. Wait.

The physical field information in printing (such as residual stress, local area cooling assessment) obtained by simulating printing with simulation software or simulation system currently exists, but it cannot be compared with the performance parameters of actual printed parts (such as shape distortion, interlayer bonding) , elastic modulus and strength, etc.) are effectively correlated, and it is difficult to predetermine the optimized printing parameter information. Based on the existing slicing technology to analyze the printing process and evaluate the performance of the obtained objects to adjust the printing path and printing parameters, it is necessary to Bear high economic and time costs.

SUMMARY OF THE INVENTION

In view of the shortcomings of the above-mentioned related technologies, the purpose of this application is to provide a training method for a machine learning model for evaluating the performance of a 3D printing component, a training device for the machine learning model, and an evaluation system for evaluating the performance of the 3D printing component, Computer equipment and computer-readable storage media are used to solve the problem that it is difficult to evaluate the performance of 3D printing components in the existing slicing technology and is not conducive to controlling the quality and efficiency of subsequent printing.

In order to achieve the above object and other related objects, the present application discloses, in a first aspect, a training method for a machine learning model for evaluating the performance of 3D printing components, the training method for the machine learning model includes the following steps: obtaining the 3D printing Multiple sets of residual stress data and/or multiple sets of equivalent relaxation time data of the printed component; and multiple sets of measured values of at least one performance evaluation parameter of the 3D printed component in an actual printing environment; and the multiple sets of residual stress data And/or the multiple sets of equivalent relaxation time data are used as input data and the measured values are used as output data to perform associated training to obtain the machine learning model.

In a second aspect, the present application also discloses a training device for a machine learning model, where the machine learning model is used to evaluate the performance of a 3D printing component, and the training device includes: a training sample acquisition module for acquiring multiple data of the 3D printing component. a set of residual stress data and/or multiple sets of equivalent relaxation time data; and multiple sets of measurement values for obtaining at least one performance evaluation parameter of the 3D printing component in an actual printing environment; a training module for applying the multiple sets of residual stress data And/or the multiple sets of equivalent relaxation time data are used as input data and the multiple sets of measured values are used as output data for associated training to obtain the machine learning model.

The present application also discloses, in a third aspect, an evaluation system for evaluating the performance of a 3D printed component, comprising an input module for receiving residual stress data and/or equivalent relaxation time data of the 3D printed component; and a prediction module is used to call the machine learning model generated by the training method of the machine learning model according to any one of the embodiments disclosed in the first aspect of this application to predict the residual stress data and/or the equivalent relaxation time data, and output the An evaluation value of at least one performance evaluation parameter of the 3D printed component.

The present application further discloses a computer device in a fourth aspect, comprising: a storage device for storing at least one program, and preset input data and output data; and a processing device, connected to the storage device, for executing The at least one program is used to call the at least one program in the storage device to execute and implement the method for training a machine learning model according to any implementation manner disclosed in the first aspect of this application.

A fifth aspect of the present application further discloses a computer device, comprising: a storage device for storing at least one program, and the training method for training a machine learning model according to any one of the embodiments disclosed in the first aspect of the present application. the machine learning model; and a processing device connected to the storage device for executing the at least one program to invoke the at least one program execution and the machine learning model in the storage device to analyze the residual stress The data and/or the equivalent relaxation time data are predicted, and an evaluation value of at least one performance evaluation parameter of the 3D printed component is output.

The present application further discloses, in a sixth aspect, a computer-readable storage medium storing at least one program, and when the at least one program is executed by a processor, implements any one of the implementation manners disclosed in the first aspect of the present application. Training methods for machine learning models.

In summary, the present application provides a training method for a machine learning model for evaluating the performance of a 3D printing component, a training device for a machine learning model, and an evaluation system, computer equipment, and computer-readable storage medium for evaluating the performance of a 3D printing component. , has the following beneficial effects: by pre-determining variables that affect the printing quality of the printing member, different printing parameters are set for different printing member models to perform actual printing to obtain the printing obtained when the variables are set to different values or different conditions and states. components, the thus obtained sets of printed components are measured to obtain measurements of at least one performance evaluation parameter; and, temperature fields and residuals of the printed components in actual printing or in a simulated environment simulated by the actual printing environment are obtained The stress field is equivalent to the equivalent relaxation time that can be represented by a single parameter, and a labeled training data set is formed by multiple sets of measurement values, multiple sets of equivalent relaxation time data, and multiple sets of residual stress data. Assuming that the neural network structure or machine learning model can be associated with training based on the training data set, the measured value in the training data set is used as the output, and the equivalent relaxation time data and/or the residual stress data are used as the input, so that after the training is completed, The neural network or machine learning model can evaluate the performance of the printing member based on the equivalent relaxation time data and/or residual stress data of the printing member, so as to realize the correlation of the printing parameters with the printing quality for different types of printing members .

Furthermore, the machine learning model obtained by the training method of the machine learning model provided by the present application can help to predetermine the optimized printing parameter information, for example, set different printing parameters for the printing component model to simulate printing, and convert the simulation results into equivalent relaxation. Temporal data and/or residual stress data are input into the neural network or machine learning model to obtain a prediction result of printing quality, repeat the simulation printing and prediction, and predetermine the optimal printing of the printing component without actual printing Parameter information; or, for the determined printing parameters and printing component models, through simulation printing and prediction, it can be verified whether the printing components under this setting meet the quality requirements, and the qualification rate of actual printing can be improved.

Description of drawings

The invention to which this application relates is set forth with particularity characteristic of the appended claims. The features and advantages of the inventions involved in this application can be better understood by reference to the exemplary embodiments described in detail hereinafter and the accompanying drawings. A brief description of the drawings is as follows:

FIG. 1 shows a schematic flowchart of a training method of a machine learning model of the present application in an embodiment.

FIG. 2 shows a schematic flowchart of a method for determining initial residual stress in an embodiment of the training method of the machine learning model of the present application.

FIG. 3 shows a simplified schematic diagram of an apparatus for training a machine learning model of the present application in an embodiment.

FIG. 4 shows a simplified schematic diagram of the evaluation system of the present application in one embodiment.

FIG. 5 shows a simplified schematic diagram of the computer apparatus of the present application in one embodiment.

FIG. 6 shows a simplified schematic diagram of the computer apparatus of the present application in one embodiment.

Detailed ways

The embodiments of the present application are described below by specific specific examples, and those skilled in the art can easily understand other advantages and effects of the present application from the contents disclosed in this specification.

In the following description, reference is made to the accompanying drawings, which describe several embodiments of the present application. It is to be understood that other embodiments may be utilized and mechanical, structural, as well as operational changes may be made without departing from the spirit and scope of the present disclosure. The following detailed description should not be considered limiting, and the scope of embodiments of the present application is limited only by the claims of the issued patent. The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the application. Spatially related terms, such as "upper," "lower," "left," "right," "below," "below," "lower," "above," "upper," etc., may be used in the text for ease of description The relationship of one element or feature shown in the figures to another element or feature.

Also, as used herein, the singular forms "a," "an," and "the" are intended to include the plural forms as well, unless the context dictates otherwise. It should be further understood that the terms "comprising", "comprising" indicate the presence of stated features, steps, operations, elements, components, items, kinds, and/or groups, but do not exclude one or more other features, steps, operations, The existence, appearance or addition of elements, assemblies, items, categories, and/or groups. The terms "or" and "and/or" as used herein are to be construed to be inclusive or to mean any one or any combination. Thus, "A, B or C" or "A, B and/or C" means "any of the following: A; B; C; A and B; A and C; B and C; A, B and C" . Exceptions to this definition arise only when combinations of elements, functions, steps, or operations are inherently mutually exclusive in some way.

As mentioned in the background art, in the existing 3D printing technology, the setting of printing parameters usually has a decisive effect on the quality (or performance) of the printing component. For example, in FDM printing, the printing component is formed based on the accumulation and stacking of printing filaments. The interlayer bonding quality between the printed layers is related to the overall strength and local strength of the printed component. At the same time, the molding accuracy of the printed component may also show different results based on different printing parameters.

There are usually multiple printing parameters that need to be set in actual printing, so the adjustment of the value of any parameter may affect the performance of the final printed component. For a model of a component to be printed, such as a preset mold model, a medical fixture model, and a customized product model, the optimized printing parameters are determined based on the model shape, which can effectively improve the qualification rate of the printed component and reduce printing failures such as printing component distortion. The resulting time consumption and cost loss, but in the existing slicing technology, it is difficult to correlate the printing parameters with the performance of the printing member, so as to predetermine the optimized printing parameters.

Please refer to FIG. 1 to FIG. 2 , which show schematic flowcharts of performing training steps to generate the machine learning model in the method for training a machine learning model for evaluating the performance of a 3D printing component provided by the present application. Here, based on the training method of the machine learning model, a machine learning model that can be used to evaluate the performance of the printing component is obtained, and the printing parameters can be associated with the performance of the printing component, for example, inputting the input of the machine learning model Data that can characterize printing parameters can be predicted through machine learning models to obtain evaluation values that characterize the performance of printing components.

In order to facilitate the description of the training method of the machine learning model provided by this application, the coordinate system adopted in the embodiments provided by this application is a rectangular three-dimensional coordinate, and the directions of the three-dimensional coordinates are X, Y, and Z directions respectively. Or in the simulation printing, the Z direction is the normal direction of the horizontal plane, that is, the general direction perpendicular to the printing surface, and (x, y, z) can refer to a coordinate point in the defined three-dimensional space.

Please refer to FIG. 1 , which shows a schematic flowchart of an embodiment of the training method of the machine learning model for evaluating the performance of 3D printing components of the present application.

In step S10, multiple sets of residual stress data and/or multiple sets of equivalent relaxation time data of the 3D printing component are acquired; and multiple sets of measurement values of at least one performance evaluation parameter of the 3D printing component in an actual printing environment are acquired.

Here, the 3D printing components are preset printing components, such as common printing components used in 3D printing manufacturing, such as molds, medical fixtures, customized products such as shoe soles, jewelry models, or dental molds. In the embodiments provided in this application, the 3D printing component may refer to the actual printing component entity obtained by printing, or a three-dimensional model of the 3D printing component.

The multiple sets of residual stress data and/or multiple sets of equivalent relaxation time data of the 3D printed component may be residual stress data and/or equivalent relaxation time data obtained by testing the printing process during actual printing; The 3D component model corresponding to the 3D printed component is simulated and printed, such as residual stress data and/or equivalent relaxation time data obtained by simulating printing in a finite element simulation environment.

The relaxation time refers to the time required for the material to return to a normal state after the object is deformed by force and the external force is released. In the embodiments provided in this application, the equivalent relaxation time data can be used to characterize each spatial coordinate point or part of the printing member. The stress relaxation experienced by a region from print initiation to stable formation. Here, the printing material is usually a polymer material, and the mechanical relaxation behavior of the polymer polymer material is the result of the summation of various relaxation behaviors throughout its history, and the mechanical relaxation behavior is related to the material temperature. The sum of mechanical relaxation behaviors generated over time can be equivalent to the sum of mechanical relaxation behaviors generated at low temperature for a long time. The equivalent relaxation time data is equivalent to the sum of the relaxation time under the same preset temperature value from the temperature history experienced by each spatial coordinate point or local area in the printing member during printing. The effective relaxation time data is obtained by using the same preset temperature value, that is, based on the values of the multiple sets of equivalent relaxation times, the mechanical relaxation behaviors of materials corresponding to multiple sets of printing experiments or simulated printing can be compared; or, for a set of equivalent relaxation times Temporal data, which can compare the mechanical relaxation behavior of materials experienced by different spatial coordinate points or regions in the printed component based on its data distribution.

The residual stress data is the residual stress value of the entire printing member after the printing member is cooled and formed, such as the residual stress value of each spatial coordinate point or a local area, and may also be the residual stress value obtained by calculation based on simulated printing. The residual stress data can be used to characterize the performance of the molded component, for example, after the component is molded, its internal tissue has a tendency to deform to eliminate internal stress, thereby changing the accuracy of the printed component.

In the training method of the machine learning model provided in the present application, the multiple sets of residual stress data and/or multiple sets of equivalent relaxation time data obtained in step S10, and at least one of the 3D printing components in the actual printing environment The measured values of the performance evaluation parameters are used to form a training data set for the machine learning model, the training data set includes input data and output data for the training of the machine learning model, and the training is performed based on the training data set to generate a usable The machine learning model for evaluating the performance of 3D printed components. Here, each set of training data includes a determined residual stress data and/or an equivalent relaxation time data, and a measurement value of at least one performance evaluation parameter of a determined 3D printing component in an actual printing environment, each Residual stress data, equivalent relaxation time data, and measured values of at least one performance evaluation parameter in a set of training data have a correspondence to form groups.

It should be understood that the machine learning model obtained via the training method of the machine learning model is intended to correlate the printing parameters of the printing member with the properties of the print. The slice graphics and printing parameter information of the printing components are used as variables to determine the measurement values of the performance evaluation parameters of the printing components. The basis for grouping is the printing component model shape, that is, the slicing graphics and printing parameter information. Based on the same printing component model and printing The residual stress data, the equivalent relaxation time data and the measured value of at least one performance evaluation parameter obtained from the parameter information can be used as a set of training data.

In some examples, the labels or tags in the training dataset are formed based on the correspondences.

To illustrate the corresponding relationship, for example, for a certain 3D printing component model A ₁ , a set of printing parameters B ₁₁ (B _i1 , B _j1 , B _k1 , B _l1 , B _m1 , B _n1 , . . . ) are assigned to the model. Perform a printing experiment or finite element simulation printing to obtain the residual stress data S ₁₁ and the equivalent relaxation time data E ₁₁ , and perform a performance test on the printed component obtained by the printing experiment to obtain a measured value P ₁₁ ( P _c1 , P _d1 , P _e1 , P _f1 , ...), where the measured value P ₁₁ is marked with the residual stress data S ₁₁ and the equivalent relaxation time data E ₁₁ to correspond to each other to form a set of training data Q ₁₁ (S ₁₁ , E ₁₁ , P ₁₁ ), the measured value P ₁₁ can be identified as the expected data of the residual stress data S ₁₁ and/or the equivalent relaxation time data E ₁₁ based on the corresponding relationship during the training process . Among them, B _i1 , B _j1 , B _k1 , B _l1 , B _m1 , B _n1 can refer to different printing parameters such as the printing path of the printing head, the layer height, the moving speed of the printing head, the output speed of the printing material, the heating of the printing material The values of temperature, extrusion temperature, etc., P _c1 , P _d1 , P _e1 , P _f1 can refer to different performance evaluation parameters such as component printability, shape distortion, structural stiffness, interlayer bonding strength, geometric accuracy, etc. Measurements. Here, for the same 3D printing component model A ₁ , configure another set of printing parameters B ₁₂ (B _i2 , B _j2 , B _k2 , B _l2 , B _m2 , B _n2 , ...) for the model to perform printing experiments or finite element simulations The residual stress data S ₁₂ and the equivalent relaxation time data E ₁₂ are obtained by printing, and a performance test is performed on the printed member obtained by the printing experiment to obtain a measurement value P ₁₂ (P _c2 , P _d2 , P ) of at least one performance evaluation parameter _e2 _, _P _f2 _, _. , E ₁₂ , P ₁₂ ), the measured value P ₁₂ can be identified as the expected data of the residual stress data S ₁₂ and/or the equivalent relaxation time data E ₁₂ based on the correspondence during the training process.

Multiple sets of training data can be obtained by changing the slice graphics or printing parameters for printing experiments, which will not be repeated here. The layered (sliced) graphics are different.

It should be understood that in the same set of training data, when the residual stress data and the equivalent relaxation time data are obtained through finite element simulation printing of the printed component model, the training data is used to obtain the at least one performance evaluation parameter. The printing experiment of the measured value is a control experiment of finite element simulation printing. Here, the environment of the printing experiment and finite element simulation printing is: printing the same layered (slice) graphic configuration of the same 3D printing component with the same printing Parameter information is formed. Of course, it should be understood that after determining the printing parameters in the actual printing experiment, the actual printing parameters can be used to construct a comparative finite element simulation printing environment, but the same printing parameters described here are not limited to each printing The parameter values are exactly the same, or the printing parameter information corresponding to the printing experiment and the finite element simulation can be the same within the allowable error range.

The layered (sliced) graphics are obtained in advance based on cross-sectioning of the 3D printed component model in the Z-axis direction. Wherein, a slice pattern delineated by the outline of the 3D printing component model is formed on the cross-sectional layer formed by each adjacent cross-sectional division. In the case where the cross-sectional layer is sufficiently thin, it can be determined that the cross-sectional layer is on the cross-sectional layer. The contour lines of the cross-sectional surface and the lower cross-sectional surface are the same. For 3D printing equipment based on surface projection, each sliced image is also called layered image (pattern) and sliced image (pattern).

In the embodiments provided in this application, the layered (slice) graphics are all layered graphics that constitute the 3D printing component model, and the corresponding slice data includes each layered (slice) graphics and its configured layer height , and the print path for layered graphics.

The layered (sliced) pattern of the 3D printed component is related to the geometry of the 3D printed component, and in some embodiments, the 3D printed component is a regular-shaped geometric structure, including vertical thin walls, horizontal thin plates, inclined Sheets, cubes, cylinders, cylinders, square cylinders, cones, inclined columns, solid blocks, grid-filled structures, etc.

The grid-filled structure is, for example, a structure that is filled by grids with gaps in a limited printing area (or a three-dimensional contour area of a 3D printing component), and the grids are, for example, triangles, quadrilaterals, hexagons, etc.; In one example, the grid fill structure may be determined based on the area area to determine the percentage of fill and the selected grid shape.

Alternatively, in some embodiments, the 3D printing component is a complex structure, and the complex structure is, for example, a complex structure formed by assembling a plurality of the geometric structures with regular shapes, for example, corresponding to some actual customized goods, medical Complex structures of fixtures, organ models, etc.

In one embodiment, based on a preset 3D printing component model and printing parameters, printing data that can be read by a 3D printing device is formed, thereby performing actual printing or finite element simulation printing. The data of the component model can be in any known format, including but not limited to Standard Tessellation Language (STL) or Stereo Lithography Contour (SLC) format, Virtual Reality Modeling Language (Virtual Reality) Modeling Language (VRML), Additive Manufacturing File (AMF) format, Drawing Exchange Format (DXF), Polygon File Format (PLY) or suitable for computer-aided design (Computer-aided design) -Aided Design, CAD) in any other format.

In actual printing, taking the fused deposition process of 3D printing as an example, the fused deposition process is also called FDM printing. During printing, the material is extruded and cooled from the print head in a high-temperature fluid state. The material is gradually added in a molten state, or is brought into a molten state by a moving heat source (such as a heating element), followed by cooling on a continuously evolving surface, it being understood that based on printing convective radiation, ambient cooling, and heat source movement, printing The temperature of the material also evolves over time.

In some instances, the sets of residual stress data and/or the sets of equivalent relaxation time data are obtained through printing experiments.

In one example, during the printing process, a measurement device such as an infrared thermal imager detects temperature field distribution data inside the printing cavity that changes with time during printing, and the equivalent relaxation time data can be obtained based on the temperature field distribution data. In an example, the thermal imaging camera frame is set at the same height as the printing member during the curing process, and the projection surface observed or recorded by the thermal imaging camera is recorded as the height of the printing member increases as the printing material accumulates layer by layer. The temperature change history of the material is also the temperature field distribution data; the recorded temperature change history is in the form of, for example, visual images (photos) converted from temperature distributions or temperature field videos, dynamic graphs, etc. at different printing times. In another example, when the printing structure is a non-axisymmetric structure, thermal imaging cameras can be set up in different directions of the printing member, or multiple printings can be performed by changing the relative directions of the printing member and the thermal imaging camera, such as making The print member is rotated at a certain angle for printing and the process can be repeated to obtain a history of temperature changes in different areas of the print member.

In some cases, for example, during actual printing by FDM equipment, the infrared thermal imager is usually arranged outside the printing cavity, in order to weaken the infrared rays collected by the infrared thermal imager from the cavity structure such as a light-transmitting glass plate. For the attenuation of radiation, the temperature field can also be corrected. Here, the correction process can, for example, obtain temperature field distribution data based on multiple printing experiments, and calculate the attenuation value brought by the cavity structure through data processing.

Here, the device for recording the temperature evolution of the printing material during actual printing may also be a thermocouple, a radiation thermometer, an electronic temperature sensor and other instruments or devices that can be used for temperature detection, which are not limited in this application. It should be understood that the relative positions of the temperature detector and the printing member are determined based on the different types of temperature detectors provided, such as a contact detector or a non-contact detector, so as to obtain the temperature change of the printing member from the start of printing to the cooling and forming of the component Historically, for example, when the temperature detector is a thermocouple, the thermocouple can be arranged at different key positions in the cavity to record the change of the temperature value during printing. By processing the temperature change history, the equivalent relaxation time data in S10 can be obtained.

The temperature change history and the specific shape of the printing component model, such as geometric structure type, size or layered (sliced) graphics, and the printing parameters set for the printing component model, such as printing material properties, printing head moving speed, printing equipment component board heating temperature In an implementation scenario, by configuring different printing parameters for each printing member in different forms of printing members, multiple sets of temperature change histories can be obtained based on printing experiments (that is, actual printing). The multiple sets of equivalent relaxation time data are obtained through the process, and the training method of the machine learning model can be executed based on the multiple sets of equivalent relaxation time data, so as to generate a comparison between the printing parameters and performance of the printing component models of different shapes. Associated machine learning models.

In some embodiments, the residual strain data is obtained by measuring the formed printed member, such as by mechanical methods such as partial separation, segmentation, drilling, grooving, etc., to release the residual stress on the printed member, based on the The resulting deformation of the component is calculated to calculate the residual stress; or the residual stress distribution state of the printed component is measured by non-destructive physical detection methods such as X-ray diffraction, neutron diffraction, magnetic method, ultrasonic method, and indentation strain method. The plurality of sets of residual stresses can be obtained by performing printing experiments for printing component models with different printing parameters, and measuring the residual stress of the formed printing component.

In certain embodiments, the sets of residual stress data or the sets of equivalent relaxation time data are obtained in different finite element simulation environments; wherein, the different finite element simulation environments are obtained by analysing 3D printed components. Model setups are formed with different print parameter information. Here, the 3D printing component model is set to perform simulated printing with different printing parameter information, and the dynamic temperature field distribution data and residual stress data of the printing component during the simulated printing process are output, wherein based on the dynamic temperature field distribution data can be calculated The sets of equivalent relaxation time data are obtained.

Exemplarily, the finite element simulation environment can be built in a device with processing functions, for example, the processor uses any type of grid with an optional preset density to discretize the representation of real-world objects into a plurality of finite elements, where the finite element is a description of the geometric part of a real-world object. The device can be any computing device with mathematical and logical operation and data processing capabilities, including but not limited to: personal computer device, single server, server cluster, distributed server, cloud server, etc.

The finite element simulation environment uses a mathematical approximation method to simulate real physical systems (such as geometry and load conditions). In the embodiment of the present application, a finite element simulation environment close to the real printing environment is constructed based on the construction, that is, The variables that determine (or affect) the temperature history data and the residual stress data during the printing process are made to be close to the same in the actual printing environment and the finite element simulation environment, so as to improve the reliability of the simulated printing.

Here, the residual stress data and equivalent relaxation time data obtained by simulating printing in the finite element simulation environment can represent the residual stress data and equivalent relaxation time data in the real printing environment. The residual stress data or/and the equivalent relaxation time data formed in the environment can form a set of training data with the measured values of at least one performance evaluation parameter obtained in the printing experiment as the finite element simulation environment control experiment. The finite element simulation environment is determined by the layered graphics and printing parameters of the 3D printing component model, and multiple sets of training data can be formed by changing at least one printing parameter information or the layered graphics of the model.

It should be understood that the finite element simulation environment is used to simulate a real printing environment, and in some embodiments provided in the present application, the printing parameter information in the finite element simulation environment is based on the parameters of the actual printing environment in the printing experiment of the comparison The information is determined, for example, the printing equipment information in actual printing, such as the cross section of the print head, is a known quantity. In the comparative finite element simulation, the parameters of the actual printing environment, such as the shape of the cross section of the print head, are obtained to construct a finite element simulation environment.

In some embodiments, such as in FDM printing or finite element simulation of FDM printing, the printing component model can be processed to form G-Code data through slicing, and printing parameter information is configured for the G-Code data, that is, available For actual printing, a finite element simulation environment can also be formed. In practice, the G-Code data includes a series of spatial coordinate points with sequential execution order, and the G-Code data of the 3D printing component model is the printing path expressed in coordinates and time series or sequential order. The G-Code data is input into the computing device of the processing function, and the print head moves in accordance with the path formed by the sequence of each spatial coordinate point of the G-Code data, which constitutes a 3D printing component. The printhead may be a virtual printhead such as a printhead virtually modeled in a finite element simulation environment or a functionally equivalent printhead.

In some embodiments, the printing parameter information includes one or more of the printing path of the printing head, the layer height, the moving speed of the printing head, the output speed of the printing material, the heating temperature of the printing material, and the extrusion temperature. .

Usually in FDM printing, the filamentous hot-melt material is fed into the hot-melt nozzle through a wire feeding mechanism (usually a roller), and the filamentary material is heated and melted in the nozzle, and the nozzle moves along the contour of the part and the filling track. , and extrude the molten material, deposit it at the designated position, solidify and form, bond with the previous layer of already formed material, and accumulate layer by layer to finally form a product model.

It should be understood that the printing head is a heating head (also known as a heating nozzle, a nozzle or a nozzle) in an FDM type 3D printing device, which is used to heat and melt the filament as a 3D printing raw material into a liquid material and apply it to the component plate. on; the build plate (also known as build platform or print platform) is a platform for attaching the target 3D component, which is movable along a vertical Z-axis according to signals provided by a computer-operated controller. The layer height is the slice layer thickness of the printed component model, and in some examples is the distance between adjacent layer printing paths in the G-Code data in the vertical direction (Z-axis direction), and the layer height can be used to indicate the Z-axis The vertical height of the movement.

The parameters defined by the print head, such as the printing path, the moving speed of the print head, the heating temperature with the print head as the moving heat source, etc., can be set based on the print head data in actual printing, or can be set manually. The parameter information of the print head It can be read by 3D printing equipment, or used for finite element simulation.

In some embodiments in which residual stress data and/or equivalent relaxation time data are obtained based on finite element simulations, the print head may serve as a starting point for conveying printing material, shown as having its position determined by the print travel speed and print path section. Wherein, the functionally equivalent print head is achieved by giving the print material the influence of the print head, such as giving the print material the process of increasing the predetermined path of the print head, the effect of the print head on the temperature of the extruded print material, etc. In practice, there is no need to build a model of the printhead.

The output speed of the printing material is the output speed at which the printing material moves relative to the print head, i.e. the speed at which the material accumulates in the printing environment such as on the build platform.

The heating temperature of the printing material is the temperature value or temperature range at which the printing material is heated to a molten state inside the printing device; the extrusion temperature is the temperature of the printing material when the printing head is extruded. In one example, the The extrusion temperature is determined by the heating temperature of the printing material and the temperature set at the print head.

In some embodiments, the printing parameter information further includes printing equipment parameter information, the printing equipment parameter information includes printing initial temperature field information, printing equipment component board heating temperature, printing chamber temperature, and print head shape in one or more types of information.

The printing temperature field information is the temperature field information in the printing device or in the printing chamber at the initial time of printing. In an example, the method for obtaining the initial printing temperature field information includes: measuring the temperature field before printing based on a thermal imager or a thermocouple. The temperature distribution of the printing cavity obtains the printing initial temperature field information of the finite element simulation environment.

Here, the temperature distribution data in the printing chamber before the printing starts is obtained based on the thermal imager or thermocouple, the temperature distribution data is, for example, a temperature visualization obtained by converting the thermal imager, and the printing at the initial printing moment is obtained based on the temperature distribution data. The temperature values at different positions in the chamber form the printing initial temperature field information that can be used for the finite element simulation environment of the component. It should be understood that the temperature detection instrument or device used to obtain the temperature field distribution at the initial printing moment may also be a radiation thermometer, an electronic temperature sensor, etc., which is not limited in this application.

In some examples, the initial print temperature field is obtained based on multiple measurements of the printing device to correct the measured temperature field via data processing.

The initial printing moment is the moment when printing starts or the moment before printing starts, and the temperature field in the printing environment is a function of time. Here, the moment is usually a small time interval such as 0.5s, 1s , 1.5s, etc. In this time interval, the temperature field that changes with time can be equivalent to a constant temperature field.

In the actual printing environment, the heating temperature of the component board of the printing device can be directly set on the component board. In the finite element simulation environment, the heating temperature of the component board of the printing device can be a constant temperature contact surface or a temperature that increases or dissipates heat according to a preset rule. The contact surface is the receiving surface of the first layer of printing material, that is, the bottom surface of the model. Usually, the temperature of the printing substrate is a preset constant value. In the finite element simulation of some examples, the boundary condition at the bottom of the model is a constant temperature with a preset fixed value, which is used to simulate the actual printing environment. or, based on the temperature variation function of the printing substrate in the actual printing environment, the temperature at the bottom of the model is set to the same temperature variation function in the simulation.

The temperature of the printing chamber is the temperature in the molding chamber, which can correspond to the temperature in the preset chamber volume in the finite element simulation environment, and the chamber volume can be determined based on the structure of the printing device in the actual printing experiment, And apply or define heat exchange conditions and material types equivalent to the actual chamber to simulate the actual printing environment. In some embodiments, the printing chamber temperature may be defined in finite element simulations using an equivalent heating or cooling rate of the printing environment, which may be obtained based on boundary conditions of the actual printing environment Heat transfer formula or defined by empirical formula.

The print head usually defaults to a circular cross-section or a polygonal cross-section (such as a square, rectangular, rhombic, pentagon or hexagonal cross-section), and the print material can be transported by the shape (or cross-section) of the print head. Cross section to the printing chamber for cooling and forming. Of course, the cross-sectional shape of the print head is not limited to the above examples, and the geometric parameters may be defined according to the preset print head shape in the simulation.

In some examples, the printing parameter information further includes material property information of the printing member, where the material property information includes: wire type, wire diameter, wire cross-sectional shape, material maximum heating temperature, material thermal parameters, material One or more of dynamic mechanical parameters, and initial residual stress of the material.

The property information of the printing material is the material type and its related characteristic information, which can be properties such as physical properties such as specific heat capacity or thermal conductivity, density, melting point, glass transition temperature, mechanical properties, chemical properties, etc. that are usually related to the material type, Generally speaking, different materials have different properties, corresponding to the parameter information of different values or characteristics.

The filament type is also the type of printing material. In some examples, the printing material includes PLA (Polylactic Acid, polylactic acid), ABS (Acrylonitrile Butadiene Styrene acrylonitrile-butadiene-styrene copolymer), polyurethane TPU , TPE material (thermoplastic elastomer), thermoplastic elastomer, nylon, carbon fiber material (such as Carbon Fiber), semi-crystalline thermoplastic, Metal PLA/Metal ABS metal texture PLA/ABS material, PEEK material, FDM conductive wire, Glow -in-the-Dark luminous material (such as adding different colors of fluorescent agents in PLA or ABS), Wood wood-feeling material (by mixing quantitative wood fibers in PLA), etc.

The diameter of the filament is the diameter of the printing material extruded from the print head. In some embodiments, the filament has a circular cross-section, and the extrusion of the filament can be obtained by defining the moving speed of the print head and the diameter of the filament. out speed.

The cross-sectional shape of the filament is the cross-sectional shape of the forming structure used to stack the forming member during cooling on the continuously evolving surface after the filament has been extruded from the print head. Usually, in an ideal situation, the cross-sectional shape of the wire material is the same shape at the position of the print head, and then the formed wire cross-section may evolve into different shapes based on the heat transfer state and mechanical state of different regions during printing. In one example, the cross-sectional shape of the wire material in the finite element simulation environment is defined as the cross-sectional shape of the wire material at the position just extruded at the print head.

The maximum heating temperature of the printing material is the upper limit of the temperature allowed for maintaining the properties of the printed material during the process of melting, extruding, cooling and forming the specific printing material, for example, the warpage and shrinkage of the printing material caused by excessive temperature. Based on the printing information as a temperature limit condition, the results obtained by the simulation analysis of the printing process by the finite element simulation method can be prevented from deteriorating the printing quality due to temperature in actual printing.

The material thermal parameters include specific heat, thermal conductivity, convective heat transfer coefficient, thermal emissivity (emissivity), etc. of the printing material or the printing environment; in some examples, the material thermal parameters are obtained based on the actual printing environment, such as Parameter reading, measurement or calculation acquisition is performed on the actual printing environment to construct a comparative finite element simulation environment.

In some examples, the convective heat transfer coefficient can be obtained by calculating the convective heat transfer coefficient formula, that is, Newton's law of cooling. For example, based on a certain type of material, the printing filament is uniformly heated to different temperatures and then placed in the printing cavity. , using temperature detection equipment such as infrared thermal imager or thermocouple to record the change law of wire temperature with time, based on the temperature of different areas in the printing chamber, that is, the function of the temperature field with respect to time, and the actual printing environment determined such as convection conversion occurs The thermal heat transfer area is calculated to obtain the convective heat transfer coefficient of the printing material.

In another embodiment, a printing experiment is performed on a geometric structure model with regular shape, the temperature field distribution of the printing material changes with time during the printing process is recorded, and the geometric structure model is set to perform finite element analysis with different convective heat transfer coefficients Simulate, output the simulated temperature field that simulates the printing process, and use the convective heat transfer coefficient corresponding to the simulated temperature field that coincides with the temperature field of the printing experiment as the equivalent convective heat transfer coefficient.

Here, the geometric model with regular shape is, for example, a simple axisymmetric structure such as a horizontal thin plate, a thin-walled cylinder, etc. In the printing experiment, the variation law of the surface temperature of the geometric structure is recorded by a temperature detection device such as an infrared thermal imager; For the same geometric structure model, multiple groups of control variables are simulated and printed for the printing parameter information, wherein the variable is the convection heat transfer coefficient set in the finite element simulation environment, which will affect the temperature change of the printing material during printing in the finite element environment. The rest of the variables are set to be consistent with the printing experiment, and the simulation printing is repeated until the temperature change law of the geometric structure surface output by the simulation printing (that is, the temperature field of the simulation printing) coincides with the temperature field of the printing experiment, and the actual printing temperature will be obtained. The convective heat transfer coefficient set in the finite element environment of the simulated printing temperature field of the field coincidence is regarded as the equivalent convective heat transfer coefficient, and the convective heat transfer coefficient of the corresponding material type can be obtained.

The thermal emissivity can be measured based on a thermal emissivity tester. In an example, the thermal emissivity is obtained by performing a thermal emissivity test on a printed member during actual printing, and the thermal emissivity is used as the input information of the finite element simulation.

The material dynamic mechanical parameters are the material mechanical properties (or mechanical properties) such as: storage modulus (elastic modulus), loss modulus (viscous modulus) and the relationship between time, temperature, frequency, the material dynamic mechanical parameters It can be determined based on DMA (Dynamic thermomechanical analysis).

In some embodiments, the method for determining the dynamic mechanical parameters of the material includes: measuring the response of the printed member, that is, the sample used for testing, after being applied with alternating strain, constant strain, or fixed load at different temperatures, to obtain The curve of the storage modulus and loss modulus of the material with temperature and time, fit the curve to obtain the dynamic mechanical parameter information of the material input in the finite element simulation, such as damping characteristics, creep, stress relaxation, glass transition, etc.

In a specific example, at a determined temperature, an alternating displacement signal is applied to a printing member or a sample formed of the printing material, and the load response of the corresponding printing member is measured, thereby obtaining the energy storage of the printing material at this temperature. Modulus and loss modulus: By changing and repeating the aforementioned measurement process, the function or variation law of the storage modulus and loss modulus of the material with respect to time is obtained, thereby determining the dynamic mechanical parameters of the material for constructing the finite element simulation environment. Here, the sample can be prepared based on the requirements of the DMA test, for example, the length and width of the sample are limited to 2-20 mm, the thickness is 0-5 mm, and the upper and lower surfaces are parallel.

In some embodiments, the alternating strain, constant strain or fixed load causes different deformations of the printed member according to different loading directions, and the types of deformation include tension, compression, bending, 3-point bending, and shearing. at least one. For example, a 3-point bending test can be used to determine the elastic modulus for composite materials, and in a practical scenario, the maximum displacement, maximum amplitude, and deflection force can be set for the printed sample.

Here, the loading direction of the strain or load relative to the specimen can be changed, so that the printed member produces different types of deformation, and different parameters of the material can be obtained from the type of strain applied to the printed member and the loading direction of the load; In an example, the printing member is a sample obtained by printing or a sample obtained by dividing the printing member. In some examples, samples of the same specification can also be tested in different loading directions, and the dynamic mechanical parameter information of the material can be determined by comparison to reduce measurement errors.

In certain embodiments, the alternating strain is a harmonic stress or load, such as a sinusoidal stress applied to the print member.

Usually in the actual printing environment, the initial residual stress will be formed in the initial stage of printing, which is affected by the speed and temperature of the printing material extruded from the printing head. In the finite element simulation, the initial residual stress can be determined by the printing experiment to form the input parameters of the finite element simulation environment, so that the finite element simulation environment is close to the real printing environment.

Please refer to FIG. 2 , which is a schematic flowchart of a method for determining the initial residual stress in an embodiment.

In step S101, multiple sets of monofilament printing experiments are performed on the printing material with different printing parameter information.

Usually, the initial residual stress in printing is related to printing parameters, such as extrusion temperature, type of printing material (filament type), etc. Here, for different types of printing materials, printing experiments with different printing parameters are set to obtain more Sets of monofilament structures available for initial residual stress measurements.

In step S102, the residual stress of the monofilament component obtained by the multi-group monofilament printing experiments is calculated or measured.

In some examples, in step S102, the monofilament structure may be processed to release residual strain, the deformation of the monofilament may be measured to calculate the residual stress; or the residual stress of the monofilament component may be determined based on a physical detection method.

In some specific examples, the monofilament structure obtained by printing is heated to above the glass transition temperature, for example, heated to 10°C above the glass transition temperature, and the lengths of the monofilament structure before and after heating are recorded as L ₀ and L, respectively, at During this process, the residual stress in the monofilament is released, and the deformation produced by the monofilament, namely (L ₀ -L)/L, can determine its residual stress.

Here, the residual stress of the monofilament structure can also be released by mechanical methods such as partial separation, division, drilling, grooving, etc., and the residual stress is calculated based on the resulting monofilament deformation; or, for example, X-ray diffraction method is used. , neutron diffraction method, magnetic method, ultrasonic method and indentation strain method and other non-destructive physical detection methods to measure the residual stress distribution state of the monofilament structure. The residual stress of the wire can be obtained, and the multiple sets of residual stress can be obtained.

In actual printing, the initial residual stress is formed in the initial stage of printing, and the structure extruded during this time is usually a monofilament structure. Here, it should be understood that the residual stress of the monofilament can be used to characterize the initial residual stress of the printed component. stress.

In step S103, a residual stress database including the different printing parameter information and the residual stress of the monofilament component obtained by multiple sets of printing experiments is obtained; wherein, the residual stress of the monofilament component obtained by the printing experiment and the printing parameters of the monofilament component are obtained. Information has a corresponding relationship.

The residual stress obtained by the measurement of the monofilament and the printing parameters of the monofilament are marked, so that each residual stress data corresponds to a printing parameter. Residual strain database. In an example, the residual strain database is input into the finite element simulation system, and the simulation system in the finite element simulation can match the corresponding residual stress, ie the initial residual stress, in the residual strain database based on the received printing parameter information.

Here, after the layered graphics and printing parameter information of the 3D printing component are determined, a finite element simulation environment can be constructed. In some examples, the 3D printing component is subjected to coupled simulation calculation in the finite element simulation environment. To obtain the multiple sets of residual stress data and/or multiple sets of equivalent relaxation time data, the coupling simulation calculation is a coupling simulation calculation under preset boundary conditions, and the boundary conditions include thermal convection boundary conditions and/or Thermal radiation boundary condition.

In one example, the boundary conditions include boundary conditions for heat transfer and mechanical contact. In some embodiments, after setting the 3D printing component model in the finite element simulation environment, for example, the size specification is preset. The regular hexahedron is used as the basic unit, and the 3D printing component model is composed of multiple basic units that are regular hexahedrons. The thermal boundary condition of each basic unit is the heat exchange between the basic unit and the external environment and adjacent basic units, including : Convective heat exchange with the constant temperature component plate at the bottom of the model, internal negative heat source brought by ambient cooling, convection between different printing layers and cooling caused by radiation in the printing chamber.

In some examples, the printing process is described with dynamic boundary conditions by re-identifying the surface of the model to assign thermal convection boundary conditions and thermal radiation boundary conditions on the surface of the model at different times in the simulation printing, such as every certain time interval. influence of ambient temperature.

In some embodiments, a coupled simulation calculation is performed on the 3D printed component in the finite element simulation environment to obtain the sets of residual stress data and/or sets of equivalent relaxation time data, the coupled simulation calculation The model includes a linear viscoelastic model for describing the mechanical deformation of the printing material, or/and a transverse isotropic heat conduction model or an orthotropic heat conduction model for describing the thermal conduction behavior of the printing material.

In one example, the linear elastic-viscous model is a multi-branch thermo-viscoelastic model that takes into account temperature-dependent relaxation behavior and flow-shear phenomena of the printed material. Among them, the mechanical behavior of the material is related to temperature, and the total strain in printing consists of thermal strain and elastic strain. The thermal-mechanical-chemical coupling model is used to describe the change of the material over time. The obtained simulation results also consider the material properties, The interaction between temperature and mechanical state, as well as temperature and stress-strain, the final deformation data is more in line with the actual printing state, which can reduce the simulation error.

In some embodiments, in the process of simulation calculation, one of the stress contour corresponding to the residual stress field, the deformed displacement contour or the temperature contour corresponding to the temperature field can be selected. It should be understood that in the calculation, all The temperature field and the stress field are completely coupled, and the transient temperature field and the residual stress field can be obtained at the same time in the calculation at each moment.

Here, in step S10, the calculation result of the simulated printing is output through finite element simulation, or the printed residual stress data and dynamic temperature field data can be obtained by observing the actual printing process. Usually, the residual stress data and the dynamic temperature field data in the calculation result of the simulated printing are the residual stress data and the temperature field data at each spatial coordinate in the printing member. In some instances, the density of the spatial coordinate points is based on a finite The mesh density set in the meta environment is determined. For example, the higher the mesh density, the smaller each mesh, and the greater the density of spatial coordinate points in the calculated residual stress data and temperature field data.

In some embodiments, the equivalent relaxation time data is based on the WLF equation (Williams-Landel-Ferry equation) or/and a The Arrhenius equation is obtained when the temperature is equivalent to the same preset temperature value.

Based on the relationship between the relaxation behavior of polymer materials and temperature, the relaxation behavior at low temperature for a long time can be equivalent to the relaxation behavior at high temperature for a short time, and the following formula (1) is expressed:

Among them, τ(T ₁ ) and τ(T ₂ ) are the characteristic relaxation times of the material at different temperatures T ₁ and T ₂ respectively, and the characteristic relaxation times can represent the stress relaxation ability of the material, a(T ₁ ), a (T ₂ ) is the temperature-dependent transformation factor (also known as the shift factor), that is, the transformation factor a is a function of temperature. Here, when the temperature in the material is generally higher than the glass transition temperature, the transformation factor a follows the WLF equation , as shown in the following formula (2):

Among them, T is the actual temperature in printing, _TM is the reference temperature, C ₁ and C ₂ are empirical parameters, which are determined by the value of the reference temperature _TM . The actual temperature T in the area, when the actual temperature T is higher than the glass transition temperature of the printing material, after the preset reference temperature _TM is determined, the current actual temperature corresponding to the set reference temperature can be obtained from the formula (2). The conversion factor a of _TM can be equivalent to the mechanical relaxation experienced by the material at the reference temperature _TM within the time t/a for the mechanical relaxation experienced at the current temperature T within the time t.

Furthermore, when the temperature of the material in the printed component is lower than the glass transition temperature, the conversion factor a follows the Arrhenius equation in the low temperature state, as shown in the following equation (3):

Among them, T is the actual temperature during printing, T _g is the reference temperature, A is the material constant, F _C is the configurational energy, and k _B is the Boltzmann constant. Usually, the reaction occurs when the temperature variation range is not large. The activation energy can be regarded as a constant. For the actual temperature T at any spatial coordinate in the printing member or in the local area, when the actual temperature T is lower than the glass transition temperature of the printing material, after determining the preset reference temperature T _g , the formula (3) can be used to obtain The current actual temperature corresponds to the conversion factor a of the set reference temperature T _g . For the mechanical relaxation experienced at the current temperature T within the time t, it can be equivalent to the material at the reference temperature T _g within the time t/a experienced mechanical relaxation.

It should be understood that in the actual printing or simulated printing process, the temperature field in the printing member changes with time, and at the same time, the temperature histories at different positions or spatial coordinate points in the printing member are different. Here, the printing process is divided into a plurality of tiny The time period dt is composed of, here, the temperature of the printing material in dt can be considered to be a constant value, and the stress relaxation behavior at the dt time can be converted into dt/a at the reference temperature by formula (2) and formula (3). The stress relaxation behavior in time can integrate the relaxation behavior experienced by the printing materials at different positions during the entire printing process, that is, from the start of printing, such as time 0 to cooling and forming, such as time t ₁ , as shown in the following formula (4). Show:

Among them, (x, y, z) is the spatial coordinate point, and t _r (x, y, z) is the equivalent relaxation time of the point during the printing process from time 0 to time t ₁ .

For any spatial coordinate point or a region in the actual printing environment or the simulated printing environment, the temperature history of the entire printing process can be equivalent to the equivalent relaxation time at the preset reference temperature, based on the equivalent relaxation of the printing component Time distribution data can also evaluate stress relaxation behavior at different locations.

With the methods disclosed in the above embodiments, the multiple sets of residual stress data and multiple sets of equivalent relaxation time data in step S10 can be obtained. Here, step S10 also includes obtaining multiple sets of corresponding 3D printing components in the step S10. The measured value of at least one performance evaluation parameter in an actual printing environment to form the multiple sets of training data, that is, a training data set.

In certain embodiments, the performance evaluation parameters include component printability, shape distortion, structural stiffness, interlayer bond strength, geometric accuracy, minimum print gap, resolution, bridging performance, drape performance, surface waviness, At least one of minimum print layer thickness and squareness. For example, the shape distortions include local shape distortions of the printed member, and the corresponding measurement values may be represented by a data array such as a matrix.

The performance evaluation result corresponding to the performance evaluation parameter, that is, the measurement value, may be, for example, a unit and a numerical value corresponding to the performance. Here, the form of the numerical value includes a singular number, a data array, or an identifier used to characterize the evaluation result under preset rules. Symbols such as text symbols, mathematical symbols, letters, etc.

Here, it should be understood that the measured values of at least one performance evaluation parameter of the multiple sets of 3D printing components in the actual printing environment correspond to the multiple sets of residual stress data and multiple sets of equivalent relaxation time data; therefore, in practice In the printing environment, experiments are carried out for each combination of 3D printing components with different geometric structures, different slicing methods, and slicing graphics and printing parameters set with different printing parameters, so as to obtain enough training data.

The printability of the component is an evaluation of whether printing can be performed to obtain a printed component that meets quality specifications for the determined layered graphics and printing parameters. In some embodiments, the method for determining the printability of the component includes: setting the three-dimensional model of the component with different printing parameters, and determining the combination of the three-dimensional model of the component with local collapse or distortion of the 3D printed component in an actual printing environment and the printing parameters as Not printable. In the actual scene, the printing process is observed and tracked to determine whether the printing component has local collapse or shape distortion from the beginning to the end of printing. If so, it is considered that the printing component under the current printing parameter settings is not printable.

Generally, the local collapse or shape distortion is related to the temperature state of the local material in printing. For example, when the material in the region is in a high temperature state for a long time, local collapse or shape distortion is prone to occur; it should be understood that under different printing parameter settings , the temperature field of each coordinate position in the printing member may change accordingly. In order to determine the relationship between the printability of the printing member and the temperature field of the printing process, printing experiments are carried out on members of different geometric configurations. Type of printing component model, print experiments with different slicing methods and printing parameters, mark the combination of slicing method and printing parameter that successfully prints as 1, and mark the combination of slicing method and printing parameter that fails to print by another mark, such as The mark is 0, from which an evaluation result characterizing the printability can be formed. Of course, the way of marking the printability is not limited to this.

The shape distortion parameters include overall shape distortion and local shape distortion, which can be used to evaluate the geometric error between the printed component entity and the expected outline of the printed model. In some examples, determining the measure of shape distortion includes: comparing a curvature error between a surface mesh of the 3D printed component in an actual printing environment and a surface mesh of a model of the 3D printed component to calculate the surface mesh of the component. Distortion of the shape of a local area or/and of the component as a whole. Here, the surface mesh of the 3D printing component model is a 3D printing component model that is preset before printing, for example, in the 3D printing preprocessing.

In one example, the surface mesh of the 3D printed member in the actual printing environment is obtained by scanning the 3D printed member with a three-dimensional optical scanner. The surface mesh is, for example, a triangular mesh, and a plurality of triangular meshes are used for piecewise linear fitting to the surface of the object. Usually, the 3D printed component model in the preprocessing can generate or display the triangular mesh in the preprocessing equipment. Of course, the mesh obtained by the optical scanning or the surface mesh of the 3D printing component model may also be quadrilateral, polygon, etc., which is not limited in this application.

In some examples, a measure of the shape distortion parameter of the printed member is obtained by calculating the curvature error between the triangular mesh obtained from the optical scan and the surface mesh of the printed member model, where the curvature may be Gaussian, for example Curvature, principal curvature, average curvature, etc. are used as equivalent curvatures. In a specific example, the mesh obtained by optical scanning of the printed component does not coincide with the 3D printing component model such as the CAD model mesh. When calculating the mesh curvature error of the component surface before and after printing, the triangular mesh obtained by each actual scan is used. The three closest CAD model meshes can be determined based on the centroid position of the mesh, and the three curvature errors can be obtained by comparing the CAD model mesh with the actual scanning mesh, for example, expressed as Δκ _i1 , Δκ _i2 , Δκ _i3 , taking the average of the 3 curvature errors

As the curvature error value between the actual scanned triangular mesh and the CAD model mesh, the curvature error value can be used as the measurement value of the shape distortion of each actual scanned mesh. For a local area of the model surface or the model as a whole, its shape Distortion κ=ΣS _i _Δκ _i /ΣS _i , where Si is the area of each grid in the region evaluated by the shape distortion parameter k.

It should be understood that the degree to which the mesh plane is spatially curved in different directions can be indicated by the equivalent curvature of the mesh. When the component surface is replaced by a number of tiny mesh fittings, for a certain type of curvature such as the principal curvature, The principal curvature between each mesh and its adjacent meshes is similar. Here, the CAD model meshes with close distances used for evaluating the actual scanned triangular meshes can also be 2, 4, 5, etc., This application is not limited.

Of course, the measurement method of the shape distortion parameter can also be based on binocular structured light to measure the overall three-dimensional shape of the printing member, and obtain the three-dimensional point cloud of the printing member based on three-dimensional laser scanning. The surface morphology of the printed component in the actual printing environment obtained in different ways can be compared with the contour of the preset model before printing to realize the shape distortion analysis of the overall and local areas of the component.

In an example, the structural stiffness of the printed member is an overall stiffness, and the overall stiffness can be used to evaluate the performance that the printed member needs to achieve in practical use. For different printed members, it can be based on the geometrical features of the printed member. Determine the loading direction of the force in the stiffness test to obtain a measurement for the stiffness evaluation of the printed component. For example, if the printed member has a cuboid structure, it can be defined as the Z direction according to the length direction, for example, in three-dimensional space, the tensile stiffness is measured according to the Z direction, and the bending stiffness is measured in the X-Y direction by using a three-point bending experiment; here , Different types of printing structures need to measure the stiffness in the Z direction and the stiffness in the X-Y direction, and determine the corresponding measurement method based on the geometric structure of the printing component. In some implementations, the overall stiffness can be represented by a stiffness matrix of the entire printed member.

In some embodiments, the printed member is tested for structural stiffness according to different loading directions of forces, the structural stiffness comprising at least one of bending stiffness, tensile stiffness, compression stiffness, shear stiffness, and torsional stiffness By. The type of structural stiffness to be measured can be determined from the practical scenario of the printed component, for example, a printed component that needs to transmit torque in practice, where its torsional stiffness can be measured.

In some examples, the structural stiffness is represented by the local stiffness of the component, and the method of measuring the local structure of the component includes: cutting the 3D printed component into a preset regular geometric structure, and measuring the structural stiffness of each geometric structure. Here, based on the spatial position of each geometric structure in the printing member, the local structural stiffness distribution of the different regions of the printing member, ie, the regions of the divided geometric structures, can be formed from the structural stiffness of each geometrical structure. For the structural stiffness of each local area, the measurement method of its structural stiffness is similar to the above-mentioned test method of overall stiffness, but the structural stiffness of each local structure is the stiffness with its spatial position information in the 3D printed component, such as the The local stiffness measurements of the 3D printed component are integrated to obtain a matrix characterized by the local stiffness measurements. The matrix or other types of data arrays formed by the local stiffness measurements can characterize the distribution law of the stiffness of the printed member in different regions.

In some examples, when the structural stiffness of the printing member is represented by the local stiffness, the printing member is divided by a preset number of divisions or the size of the local member, for example, the printing member is divided into equal-sized filling blocks, and the Each infill block is measured for its structural stiffness to obtain a stiffness distribution for the local structure of the printed member.

The interlayer bonding strength can represent the consolidation strength between the printing layers of the printing member, and the direction in which each layered pattern is stacked during printing is defined as the Z direction, and the interlayer bonding strength is used to evaluate the printing member. Z-direction performance.

In some examples, the interlayer bonding strength includes the overall interlayer bonding strength and local interlayer bonding strength of the 3D printed component. For example, in one example, a tensile test is performed on the printing member along the Z direction, and the stress value when the printing member is damaged is recorded, which can be used as a measurement value of the interlayer bonding strength of the printing member as a whole. For another example, in another example, the printing member is divided to obtain a plurality of geometric structures (also referred to as samples), a tensile test is performed on each sample along the Z direction, and the damage of the sample when the sample is destroyed is recorded. The stress value is a measure of the interlaminar bond strength of the sample, ie the local interlaminar bond strength of the printed member. The measured value of the interlayer bonding strength of different local components obtained by dividing the printed component can form the distribution data of the interlayer bonding strength of different regions of the 3D printed component. formed matrix.

The geometric accuracy includes, for example, surface roughness, dimensional error, shape error, etc. The surface roughness is, for example, the microscopically uneven traces formed on the surface of the printing member with small distances and peaks and valleys; the dimensional error For example, it includes the diameter error, length error, etc. of the printing member; the shape error includes, for example, the position error of the geometric features of the surface of the printing member, such as points, lines, and surfaces.

The surface waviness can also be used as an evaluation content of geometric accuracy, which is a surface shape error between macroscopic and microscopic geometrical errors. Usually, the length of peaks and valleys and spacings on the component surface in the surface waviness error are orders of magnitude greater than the order of magnitude of the surface roughness, and the component surface changes periodically. By testing the surface waviness of the printing member, it can be determined whether the printing member reaches a preset quality specification, for example, whether there is surface waviness on the surface of the printing member, and if so, the printing is considered unqualified. In some examples, the surface waviness test and the surface roughness test can be performed using the same test instrument.

The minimum printing gap can be used to evaluate the performance of the joint in the printing structure formed by the splicing of different composite parts, and the printing gap refers to the distance between any two parts, thin walls or columns. Usually, the minimum printing gap is related to the geometric structure of the printing component, printing parameters such as the property information of the printing material, printing equipment, and the control of the printing molding accuracy. Therefore, for printing experiments using different slice graphics and different printing parameter information, the The physical measurement of the printed component obtains multiple sets of evaluation values for the minimum printing gap.

The resolution is characterized by, for example, the wire diameter of the printing member, slice layer thickness before printing (ie, Z-axis layer thickness), dots per inch (DPI), pixel size, beam spot size, nozzle diameter, etc. Generally, the printing resolution is The higher the rate, the higher the printing accuracy, and the printing resolution can be used to evaluate the accuracy of the printed member.

The bridging performance or sagging performance may be determined based on the type of the printed member. For example, for a member with a bridging portion in the printing member, the number of filaments hanging or sagging at the bridging portion is measured as a measure of the bridging performance; There are overhanging structures in the printing member, such as an arch bridge-shaped central overhanging structure and an inverted concave-shaped structure, which can be used as a measure of the overhang performance by measuring the sagging of the edge of the wire and the overflow of the wire.

The perpendicularity can be used to evaluate the vertical state between straight lines, between planes, or between straight lines and planes, wherein straight lines or planes are the evaluation criteria, and here, the straight lines can be the straight line part or the linear motion trajectory of the tested printing member, and the plane It can be a plane formed by a plane part of the sample to be tested or a motion track; for example, when the printing member is a cylinder, the plane can be a plane formed by the rotation and rolling of the cylinder along the axis.

Here, for each performance evaluation parameter, the measured value includes the results of multiple sets of printing experiments, and corresponds to multiple sets of residual stress data and/or multiple sets of equivalent relaxation time data; it should be understood that the multiple sets of printing experiments It is a printing experiment carried out on different types of printing components or different layered graphics configurations with different printing parameter information. In step S10, by forming multiple sets of training data, the training data set used for neural network training can be expanded, so that the machine learning model obtained by training is suitable for predicting different printing components configured with different printing parameter information.

It should be noted that, in the embodiments provided in this application, in step S10, multiple sets of residual stress data and/or multiple sets of equivalent relaxation time data of the 3D printing component are acquired, and multiple sets of 3D printing components are acquired during actual printing. The measured values of at least one performance evaluation parameter in the environment together form training data Q(S, E, P). Here, residual stress data S, equivalent relaxation time data E, and performance evaluation parameters are obtained for each set of training data. The order of the measured values P is not limited; at the same time, for multiple sets of training data, the measured values of at least one performance evaluation parameter of the multiple sets of 3D printing components in the multiple sets of training data in the actual printing environment can be obtained first, or the 3D printing components can be obtained first. Multiple sets of residual stress data and/or multiple sets of equivalent relaxation time data of the component, or may first obtain a set of training data Q ₁₁ (S ₁₁ , E ₁₁ , P ₁₁ ), and then obtain another set of training data Q ₁₂ ( S ₁₂ , E ₁₂ , P ₁₂ ), repeat to obtain multiple sets of training data.

In an actual scenario, the training method of the machine learning model can be performed by, for example, a processing device, and the processing device can be any computing device with mathematical and logical operations and data processing capabilities, including but not limited to: personal computer equipment, single Servers, server clusters, distributed servers, cloud servers, etc. Here, multiple sets of training data are acquired by the device. It should be understood that each set of training data in the training data is determined by unique hierarchical graphics and printing parameter information. In some examples, the processing device may Obtain all data in the training data set, and identify the corresponding hierarchical graphics and printing parameter information to divide the training data set into multiple sets of training data.

In step S11, the multiple sets of residual stress data and/or the multiple sets of equivalent relaxation time data are used as input data and the measured values are used as output data to perform associated training to obtain the machine learning model.

In some embodiments, the measured value of the at least one performance evaluation parameter is a data array, for example, when the stiffness of the printed member is characterized by local stiffness, the measured value is a data matrix formed by the stiffness values of the local structure of the printed member , here, the data matrix may be a two-dimensional matrix, a three-dimensional matrix, or a multi-dimensional matrix, which is not limited in this application. Here, it should be understood that the stiffness value of each local structure can also be represented as a stiffness matrix. For another example, the performance evaluation parameter of the printing member is shape distortion, and the shape distortion is represented by the local distortion on the surface of the printing member, and the corresponding measurement value can be represented by a two-dimensional matrix, and each element in the matrix corresponds to a local area. The average value or root mean square of the shape distortion of each grid represents the equivalent value of the shape distortion in the local area. For another example, the interlayer bonding strength of the performance evaluation parameter is represented by the local interlayer bonding strength, and the interlayer bonding strength value distribution of the printing member corresponding to the measured value in different regions can be represented by a data matrix.

It should be understood that the data array is only selected as one form of expression, and here, the performance measurement values of local areas of the 3D printing component can also be expressed in other forms, such as corresponding measurement values in multiple areas, each The measurement value includes corresponding area information such as coordinate information.

Here, when the performance evaluation parameter of the printing member is characterized by the performance evaluation result of the local area, the distribution law of the measured value of the performance evaluation parameter in the printing member can be obtained. Therefore, for a printing member with a complex structure, the printing member is divided into a plurality of parts for performance evaluation. In some examples, the distribution of the measurement value of the performance evaluation between different parts has continuity, and the measurement value is used as a function. In the output data of the machine learning model training, the machine learning model can be based on the shape of the component, the distribution law of the equivalent relaxation time data and the residual stress data in different geometric positions, and the performance distribution law at different positions in the printed component such as distribution For example, after the trained machine learning model receives the residual stress data or/and equivalent relaxation time data of the printed component, the predicted value (i.e. the evaluation value) of the performance evaluation parameters in different areas of the component ) can be performed based on the distribution law of the internal performance parameters of the printing member obtained in the training, whereby the machine learning model can predict the performance evaluation parameters of each position in the printing member.

In certain embodiments, the multiple sets of residual stress data and/or multiple sets of equivalent relaxation time data are residual stress data or/or equivalent relaxation time data of each spatial coordinate point in the 3D printing component. For example, the residual stress data input to the neural network is the equivalent relaxation time data obtained from the thermal history at each spatial coordinate point in the printed component.

In certain embodiments, the sets of residual stress data or the sets of equivalent relaxation times are data arrays. Wherein, the data array includes a multi-dimensional matrix, such as a two-dimensional matrix, a three-dimensional matrix, and the like. In one example, each value in the data array may be an average value of residual stress data or equivalent relaxation time data at a plurality of spatial coordinate points in the print member.

In some embodiments, the simulation domain of the 3D printing component model in the finite element simulation environment is divided into multiple sub-regions according to a preset size or a preset number of partitions, the multiple sets of residual stress data and/or multiple sets of The equivalent relaxation time data is the mean value of residual stress or/and the mean value of equivalent relaxation time in each sub-region in the simulation domain. Wherein, the sub-region can be obtained by dividing the simulation domain based on the preset size of the sub-region, or obtained based on the preset number of divisions of the simulation domain, for example, dividing the simulation domain into equal parts in the length direction l, width direction m, etc. divided into n equal parts in the height direction, where l, m, and n are all natural numbers greater than or equal to 1; for another example, in a certain simulation domain, set the preset sub-region size to a×b×c, and simulate The threshold is divided into multiple sub-regions of size a×b×c.

The simulation domain may be a computational domain for thermal-mechanical-chemical coupling calculation of simulated printing in a finite element environment. Generally, the larger the range of the simulation domain, the closer to the printing state in actual production. In the finite element simulation, in order to reduce the consumption of computing resources and computing time, the scope of the simulation domain includes the printing component model and the equivalent printing chamber boundary and component plate, and the indoor environment where the printing equipment is located is equivalent to the simulation domain boundaries.

Here, the residual stress data and equivalent relaxation time data at each spatial coordinate point in each sub-region can be equivalent to the equivalent values in the sub-region, such as average value, root mean square, median, etc., which can be simplified The output results of the finite element simulation calculation reduce the data volume of residual stress data and equivalent relaxation time data from the mesh density to the number of sub-regions, and use the reduced results as the output data for training the machine learning model.

In some embodiments, the average value of residual stress or/and the average value of equivalent relaxation time in each of the sub-regions is two-dimensional data obtained by transforming three-dimensional data. For example, the simulation domain in the finite element simulation is divided into l×m×n sub-regions according to the length, width, and height of l, m, and n, respectively, and the corresponding equivalent relaxation time data and residual stress data are l×m ×n three-dimensional data, each data value corresponds to a sub-region; here, the three-dimensional data can be processed to form two-dimensional data, and the two-dimensional data can be used as training data to facilitate training calculations.

In some examples, the multiple sets of residual stress data and/or multiple sets of equivalent relaxation time data are the average value of residual stress or/and the average value of equivalent relaxation time in each sub-region in the 3D printing component model; wherein, The sub-regions are obtained by dividing the 3D printing component model based on a preset size or a preset number of partitions. For example, when the multiple sets of residual stress data and/or multiple sets of equivalent relaxation time data are obtained by measuring the printing process in an actual printing environment, the sub-regions may be obtained by dividing the three-dimensional contour of the printing member, for example, by dividing the The printing component model is divided into multiple parts based on a preset number. For example, the printing component model is divided into l, m, and n equal parts according to the length, width, and height, respectively, and is divided into l×m×n sub-areas. The size divides the printed component model into multiple sub-regions; correspondingly, the residual stress data and temperature field data at different positions in each sub-region are measured and converted into equivalent values representing the whole sub-region, such as average value, root mean square, median In order to reduce the data amount of residual stress data and equivalent relaxation time data in the training data, the calculation amount of training the machine learning model can be reduced. In some examples, three-dimensional data of equivalent relaxation time data can also be transformed into two-dimensional data for residual stress data characterized by sub-regions.

Here, based on the determined residual stress data and equivalent relaxation time data of the printing member as input data, and the measured value of at least one performance evaluation parameter of the printing member in the actual printing environment as output data, correlation training can be performed to Obtain the machine learning model.

The association training is the process of associating the input data with the output data in the initial structure of the preset machine learning model to form a stable evaluation network. For example, after the association training is completed, the machine learning model receives the residual stress of the printing member. The data and/or the equivalent relaxation time data can predict and output the predicted value (evaluation value) of the corresponding printing component performance parameter. The association training can also be understood as a process of supervised learning based on the input data and output data.

In the embodiments provided in this application, the machine learning model is an algorithm model constructed based on supervised machine learning, which can be based on the residual stress data, equivalent relaxation time data of the corresponding printing components and the actual printing environment of the printing components. The measurement value of at least one performance evaluation parameter in the training sample (ie the input data and output data of the training sample) is trained, so that the generation can be predicted when the residual stress data or/and the equivalent relaxation time data of the printing member are received. Performance evaluation results of printed components. Wherein, the functions in the algorithm model are, for example: SVM (support vector machines, support vector machines), AdaBoost (Adaptive Boosting, adaptive enhancement), linear regression algorithm, linear discriminant analysis (LDA), decision tree algorithm (such as Classification and regression tree), random forest algorithm, naive Bayes, K-nearest neighbor algorithm, Learning Vector Quantization (LVQ), etc.

In some embodiments, the machine learning model is a neural network model.

It should be understood that in the association training, a preset neural network structure can be selected, such as a feedforward neural network, a feedback neural network, a deep neural network, a convolutional neural network, a self-organizing neural network, etc. The training is performed to associate the input data with the output data. During the association process, the preset neural network structure is adjusted under the corresponding algorithm, so as to gradually form the expected neural network during the training.

Here, multiple sets of training data formed based on the equivalent relaxation time data, multiple sets of residual stress data, and multiple sets of measurement values of at least one performance evaluation parameter of the printing component in the actual printing environment can be used to train the preset neural network structure or A machine learning model, so that the connection weights between different neurons in the neural network are adjusted or the data weights in the machine learning model are adjusted. The neural network or machine learning model obtained after training can be based on the received equivalent relaxation time data or/and residual stress data. The performance of the printed matter is predicted, and the predicted value, that is, the evaluation value, is output.

In some examples, step S11 includes a step of adjusting the neural network by performing error evaluation based on the predicted data and the output data; wherein the predicted data is a performance evaluation result of the prediction made by the machine learning model based on the input data. For example, the preset model of the machine learning model is a neural network structure using a supervised learning algorithm, such as a BP neural network. During the training process, a sample data, that is, a set of training data, is selected from the training data set, and the preset neural network Calculate the evaluation value of the corresponding performance evaluation parameter based on the input data in the training data, that is, the equivalent relaxation time data or/and the stress relaxation time; compare the evaluation value with the output data in the sample data, that is, the performance in the actual printing environment The measured values of the evaluation parameters are compared to obtain a deviation value, and the connection weights between different neurons in the neural network are adjusted according to the deviation value; this process is repeated until the deviation value meets the specified error range.

In some examples, the preset neural network structure is, for example, a deep neural network, wherein a corresponding loss function (Loss Function), such as a mean square error loss function, can be set to measure the output loss of the training data to determine whether the prediction result is accurate , where the loss function is used to indicate the degree of inconsistency between the prediction results of the machine learning model and the actual results of the output data, and is a non-negative real-valued function. The smaller the loss function, the better the robustness of the machine learning model. Its types include logarithmic loss (Logistic Loss) function, square loss function and exponential loss function. In the training process of the neural network, based on multiple sets of training data, the parameters of the prediction algorithm in the neural network, that is, the connection weights of neurons between different layers, are adjusted iteratively until the optimization extremum of the loss function is reached, that is, it is determined that the neural network is based on training. The loss between the result obtained by the prediction of the input data in the data and the output data reaches the set threshold steadily.

It should be noted that the above-mentioned preset neural network structure or machine learning model is only an example and not a limitation of the present application, and the specific neural network structure or machine learning model that can be used to execute the training method of the applied machine learning model in practice is not limited. here. The training data of the present application is obtained by configuring different printing parameters for the slice graphics of different printing components, and the training data formed has a corresponding relationship between the input data and the output data, that is, the training samples with labels. Usually, it can be based on supervised learning. The neural network trained by the algorithm or other machine learning models can be obtained through the training data provided by the training method of the present application to obtain the required neural network for evaluating the performance of the printing component. In some examples, the machine learning model or the preset algorithm model of the neural network can also use an unsupervised learning algorithm, such as using a self-organizing neural network to establish the corresponding relationship between the input data and the output data in the training data; When there is a corresponding relationship between the data and the output data, there is usually a supervised learning algorithm for training, which is easy to implement the method of associating the input data with the output data.

Based on the data types in the training data set in the training method of the machine learning model, the machine learning models obtained by performing the training are different.

For example, when the measurement value in the training data set, that is, the output data is a data array, such as a matrix represented by local stiffness or a matrix of local interlayer bonding strength, the trained machine learning model is based on the input data. It is predicted that the output evaluation result is also a data array, that is, the structural stiffness evaluation value represented by the local stiffness of the printed component or the interlayer bonding strength evaluation value represented by the local interlayer bonding strength; here, during the training process, machine learning The model automatically learns the material performance parameters at different positions, their residual stress data, equivalent relaxation time data, and the inherent continuity or variation law of the geometric position. It can predict the performance parameters of each position in the printed component and output it as data. Array of predicted results.

Or if the structural stiffness of the printed structure used as training data is the overall stiffness, the machine learning model predicts based on the input data, and outputs the stiffness evaluation value of the overall structure of the printed component; another example, in the printability experiment of the printed component, the The combination of component model and printing parameters that are successfully printed is marked as 1, otherwise, the failure is marked as 0. The machine learning model after the training is predicted based on the output data, and the printability of the printed component is output as 0 or 1. Printability evaluation. value.

In some examples, when the measured value of the performance evaluation parameter of the printing member in the training data set is a single evaluation parameter or several evaluation parameters, such as the overall shape distortion parameter and the overall structural stiffness, the machine learning model obtained by training is After receiving the equivalent relaxation time data of the printing member, the residual stress data is output, and the predicted value of a single evaluation parameter or several evaluation parameters (the types of which correspond to the evaluation parameters in the training data) are output.

In some examples, the input data of the training data set is equivalent relaxation time data, the output data is the printability measurement of the printed member, and the machine learning model obtained by training is based on a single input, that is, the equivalent relaxation of the printed member The time data predicts the printability of the printed member, and outputs a single predicted value, that is, an evaluation value for the printability of the member.

In some examples, when the input data in the training data set are equivalent relaxation time data and residual stress data, and the output data are measured values of various performance evaluation parameters of the printed component, the machine learning model obtained by training is based on multiple The inputs are equivalent relaxation time data and residual stress data to predict various performance evaluation parameters of the printed component, and output a plurality of corresponding evaluation values.

In some embodiments, the training method of the machine learning model further includes the step of encapsulating multiple sub-models as the machine learning model; wherein, the multiple sets of residual stress data and/or the multiple sets, etc. The effective relaxation time data is used as input data and a measured value of the performance evaluation parameter is used as output data to perform associated training to obtain a sub-model.

Here, the sub-model takes the equivalent relaxation time data or/and the residual stress data as input in the training data set received in the associated training, and uses the measured value of a performance evaluation parameter of the printing component as the output, and executes the completed sub-model. When the model receives the equivalent relaxation time data or/and residual stress data of the printed component, it can predict a performance evaluation parameter of the printed component (the type of which is the performance evaluation parameter in the training data when training the sub-network) and predict it. Output evaluation value. It should be understood that there are multiple sub-models, and different sub-models respectively use different types of performance evaluation parameter measurement values as training data during training, and correspondingly obtain sub-models for predicting different performance evaluation parameters.

Here, the initial structure of the preset machine learning model for training the sub-model is similar to that described above, which is not repeated here. It should be noted that when each sub-model only predicts one performance evaluation parameter, the The output value of the output layer of the sub-model is one type. In some examples, each sub-model can also predict several performance evaluation parameters. The output value of the sub-model is several. A submodel for predicting structural stiffness and local structural stiffness.

After training to obtain different sub-models, the sub-models can be encapsulated to form a machine learning model that can output an evaluation result of at least one performance parameter of the printing member, that is, the machine learning model receives the equivalent relaxation of the printing member After the time data or/and residual stress data are obtained, the sub-models in it respectively predict different performance evaluation parameters, and the machine learning model outputs the evaluation results of each sub-model. In some examples, a part of the sub-models in the machine learning model can be called for prediction based on the type of performance evaluation parameters of the printing member to be evaluated. For example, when only the printability evaluation of the printing member needs to be obtained, only the machine learning model can be called The sub-model used to evaluate the printability in the model predicts and outputs the predicted value, which can improve the calculation efficiency in the prediction and reduce the calculation amount.

Here, based on the training method of the machine learning model provided by the first aspect of the present application, by pre-determining variables that affect the printing quality of the printing component, different printing parameters are set for different printing component models to perform actual printing to obtain the variable settings For the printing members obtained under different numerical values or different conditions and states, the obtained multiple groups of printing members are measured to obtain measurement values of at least one performance evaluation parameter; and, obtained in actual printing or simulated by the actual printing environment The temperature field and residual stress field of the printed component in the simulated environment, and the temperature field is equivalent to the equivalent relaxation time that can be characterized by a single parameter, which is composed of multiple sets of measured values and multiple sets of equivalent relaxation time data, multiple sets of The residual stress data forms a training data set with labels, and the preset neural network structure or machine learning model can be used for association training based on the training data set, and the measured values in the training data set are used as output, equivalent relaxation time data and/or Residual stress data is used as input, so that after the training is completed, the neural network or machine learning model can evaluate the performance of the printed member based on the equivalent relaxation time data and/or residual stress data of the printed member, which can be achieved for different types of The print component of , associates its print parameters with print quality.

Furthermore, the machine learning model obtained by the training method of the machine learning model provided by the present application can help to predetermine the optimized printing parameter information, for example, set different printing parameters for the printing component model to simulate printing, and convert the simulation results into equivalent relaxation. Temporal data and/or residual stress data are input into the neural network or machine learning model to obtain a prediction result of printing quality, repeat the simulation printing and prediction, and predetermine the optimal printing of the printing component without actual printing Parameter information; or, for the determined printing parameters and printing component models, by simulating printing and predicting by the machine learning model, you can verify whether the printing components under this setting meet the quality requirements and improve the actual printing pass rate.

The present application also discloses a machine learning model training apparatus in a second aspect. Please refer to FIG. 3 , which is a simplified schematic diagram of an embodiment of the machine learning model training apparatus of the present application.

The machine learning model is used to evaluate the performance of the 3D printing component, and the training device includes: a training sample acquisition module 21 for acquiring multiple sets of residual stress data and/or multiple sets of equivalent relaxation time data of the 3D printed component; and obtaining Multiple sets of measurement values of at least one performance evaluation parameter of the 3D printing component in an actual printing environment; a training module 22 for using the multiple sets of residual stress data and/or the multiple sets of equivalent relaxation time data as input data and Correlation training is performed on the sets of measurements as output data to obtain the machine learning model.

In some embodiments, each module in the training device of the machine learning model may be a software module, and the software module may also be configured in a software system based on a programming language. The software module can be provided by a system of an electronic device. In some embodiments, the electronic device is, for example, an electronic device loaded with an APP application or an electronic device with webpage/website access capability, the electronic device includes a memory, a memory controller , one or more processing units (CPUs), peripheral interfaces, RF circuits, audio circuits, speakers, microphones, input/output (I/O) subsystems, displays, other output or control devices, and components such as external ports , these components communicate over one or more communication buses or signal lines. The electronic devices include, but are not limited to, personal computers such as desktop computers, notebook computers, tablet computers, smart phones, and smart TVs. The electronic device may also be an electronic device composed of a host with multiple virtual machines and a human-computer interaction device (such as a touch display screen, a keyboard and a mouse) corresponding to each virtual machine.

In some embodiments, the training device obtains training data, performs training based on the training data (including generating intermediate results), and obtains the functional modules of the target machine learning model, which can be implemented by various types of devices (such as terminal devices, servers, etc.). , server cluster, or cloud server system), or computing resources such as processors, communication resources (such as for supporting communication in various ways such as optical cables, cellular, etc.) are collaboratively implemented; the training data is trained by the training device. Any one of the obtained machine learning model and prediction results can be stored in the electronic device configuring the training device, and can also be transmitted to other terminal devices, servers, server clusters, cloud servers that communicate with the electronic device network system, etc.

In some embodiments of the present application, the cloud server system may be arranged on one or more physical servers according to various factors such as function and load. Wherein, when distributed in multiple physical servers, the server can be composed of servers based on cloud architecture. For example, a server based on a cloud architecture includes a public cloud (Public Cloud) server and a private cloud (Private Cloud) server, wherein the public or private cloud server includes a Software-as-a-Service (Software as a Service, SaaS) ), Platform-as-a-Service (Platform as a Service, PaaS) and Infrastructure-as-a-Service (Infrastructure as a Service, IaaS), etc. The private cloud service end is, for example, Alibaba cloud computing service platform, Amazon (Amazon) cloud computing service platform, Baidu cloud computing platform, Tencent cloud computing platform, and the like. The server can also be composed of distributed or centralized server clusters. For example, the server cluster consists of at least one physical server. Each physical server is configured with a plurality of virtual servers, each virtual server runs at least one functional module in the catering business information management server, and the virtual servers communicate with each other through a network.

The network may be the Internet, a mobile network, a local area network (LAN), a wide area network (WLAN), a storage area network (SAN), or one or more intranets, or an appropriate combination thereof. The type, or the type or protocol of the communication network between the publisher terminal and the server or between the responder terminal and the server, etc. are not limited in this application.

Regarding the presentation form of the training device of the machine learning model in the actual application scenario, the present application also provides the following exemplary descriptions:

In a scenario A, each module in the training device can be embedded in the APP of the electronic device, and the APP of the electronic device can obtain the above-mentioned information from the storage medium of the electronic device or other devices, servers, etc. that communicate with the electronic device network. training data to complete the training process of the target machine learning model. During the training process performed by the training module, in some examples, the training device of the machine learning model may be implemented by computing resources provided by an electronic device used to configure the training device, and in other examples, The computing resources required to perform the training process can also be allocated to terminal devices, servers, cloud server systems, or processors, etc. that communicate with the device network; at the same time, the prediction results generated by the training module during the training process can be The local storage of the device can also be transmitted to the terminal device, server, cloud server system, or processor for network communication of the device, and can also be provided to other applications or modules for use.

In a scenario B, the training device is a software module running on the server side, and the server side can also be a distributed and parallel computing platform composed of multiple servers. In the usage scenario, the required training data can be uploaded. platform to perform the training process. The server side can perform the training process of the machine learning model based on the stored data in the storage medium of the server side or the data from other devices in communication with the server side, whereby the machine learning model generated by training is based on the temperature history data of the printing member or The performance evaluation results obtained by predicting the stress history data can be stored on the server side, and can also be provided to other applications or modules for use.

In a scenario C, each module of the training device can be an API (Application Programming Interface), a plug-in or a software development kit (SDK, Software) provided to a server (including a cloud server) or an electronic device. Development Toolkit), etc., the API, plug-in or SDK, etc. can realize the functions of each module in the training device, such as the acquisition of training data, and the function of performing training based on the training data to generate a target machine learning model; in some implementations In this manner, the training device presented in this form can be invoked by other servers or electronic equipment to be embedded in various application programs.

The present application also provides an evaluation system for evaluating the performance of a 3D printing component in a third aspect. Please refer to FIG. 4 , which shows a simplified structure of the evaluation system for evaluating the performance of a 3D printing component of the present application in an embodiment. Schematic. As shown in the figure, the evaluation system includes an input module 31 and a prediction module 32 .

The input module 31 is configured to receive residual stress data and/or equivalent relaxation time data of the 3D printed component.

The prediction module 32 is configured to invoke the training method of the machine learning model described in any one of the embodiments provided in the first aspect of the present application to train the generated machine learning model on the residual stress data and/or the residual stress data received by the input module. or equivalent relaxation time data to predict, and output the evaluation value of at least one performance evaluation parameter of the 3D printing component.

The present application also provides a computer device in a fourth aspect. Please refer to FIG. 5 , which shows a simplified schematic block diagram of the structure of an embodiment of the computer device provided in the fourth aspect of the present application.

As shown in the figure, the computer equipment includes a storage device 41 and a processing device 42 .

Wherein, the storage device 41 is used to store at least one program, as well as preset input data and output data.

In some embodiments, the storage device 41 is, for example, a network attached storage device accessed via an RF circuit or an external port and a communication network, which may be the Internet, one or more intranets, a local area network, a wide area network, a storage LAN, etc., or a suitable combination thereof. The storage device controller may control access to the storage device by other components of the device, such as the CPU and peripheral interfaces. The storage device 41 optionally includes high-speed random access memory, and optionally also includes non-volatile memory, such as one or more magnetic disk storage devices, flash memory devices, or other non-volatile solid-state memory devices. Access to memory is optionally controlled by a memory controller by other components of the device such as the CPU and peripheral interfaces.

In some embodiments, the storage device 41 may further include volatile memory, such as random access memory; the memory may also include non-volatile memory, such as read-only memory, flash memory, hard disk or solid-state disk.

The processing device 42 is connected to the storage device 41, and is configured to execute the at least one program, so as to call the at least one program in the storage device to execute and implement any one of the embodiments provided in the first aspect of the present application The training method of the machine learning model described in the embodiment. Here, the preset input data and output data stored in the storage device 41 correspond to the input data and output data in the training data set in the training method of the machine learning model.

In some embodiments, the processing device 42 includes an integrated circuit chip with signal processing capability; or includes a general-purpose processor, which may be a microprocessor or any conventional processor, such as a central processing unit. For example, it may be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a discrete gate or transistor logic device, or a discrete hardware component, which can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of the present application, For example, based on the preset input data and output data stored in the storage device 41, the training method of the machine learning model described in any one of the embodiments provided in the first aspect of the present application is executed.

The present application also provides a computer device in a fifth aspect. Please refer to FIG. 6 , which shows a simplified schematic block diagram of the structure of the computer device provided in the fifth aspect of the present application in an embodiment. As shown in the figure, the computer equipment includes a storage device 51 and a processing device 52 .

The storage device 51 is configured to store at least one program and a machine learning model generated by the method for training a machine learning model in any one of the embodiments provided in the first aspect of the present application.

In some embodiments, the storage device 51 is, for example, a network attached storage device accessed via an RF circuit or an external port and a communication network, which may be the Internet, one or more intranets, a local area network, a wide area network, a storage LAN, etc., or a suitable combination thereof. The storage device controller may control access to the storage device by other components of the device, such as the CPU and peripheral interfaces. The storage device 51 optionally includes high-speed random access memory, and optionally also includes non-volatile memory, such as one or more magnetic disk storage devices, flash memory devices, or other non-volatile solid-state memory devices. Access to memory is optionally controlled by a memory controller by other components of the device such as the CPU and peripheral interfaces.

In some embodiments, the storage device 51 may further include volatile memory, such as random access memory; the memory may also include non-volatile memory, such as read-only memory, flash memory, hard disk or solid-state disk.

The processing device 52 is connected to the storage device 51, and is configured to execute the at least one program, so as to call the storage device 51 to execute and implement any one of the implementation manners in the embodiments provided in the first aspect of this application. The machine learning model generated by the training method of the machine learning model, so that the machine learning model predicts the residual stress data and/or the equivalent relaxation time data, and outputs at least one performance evaluation parameter of the 3D printing component. Evaluation value.

In some embodiments, the processing device 52 includes an integrated circuit chip with signal processing capabilities; or includes a general-purpose processor, which may be a microprocessor or any conventional processor, such as a central processing unit. For example, it may be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a discrete gate or transistor logic device, or a discrete hardware component, which can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of the present application, For example, the program is invoked to execute the machine learning model generated by the machine learning model training method according to any one of the embodiments provided in the first aspect of this application to perform prediction.

In a sixth aspect, the present application further provides a computer-readable storage medium, which stores at least one program, and when the at least one program is executed by a processor, implements any one of the embodiments provided in the first aspect of the present application The training method of the machine learning model.

The functions, if implemented in the form of software functional units and sold or used as independent products, may be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application can be embodied in the form of a software product in essence, or the part that contributes to the prior art or the part of the technical solution. The computer software product is stored in a storage medium, including Several instructions are used to cause a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present application.

In the embodiments provided in this application, the computer readable and writable storage medium may include read-only memory, random access memory, EEPROM, CD-ROM or other optical disk storage devices, magnetic disk storage devices or other magnetic storage devices, flash memory, A USB stick, a removable hard disk, or any other medium that can be used to store the desired program code in the form of instructions or data structures and that can be accessed by a computer. Also, any connection is properly termed a computer-readable medium. For example, if the instructions are sent from a website, server, or other remote source using coaxial cable, fiber optic cable, twisted pair, digital subscriber line (DSL), or wireless technologies such as infrared, radio, and microwave Coaxial cable, fiber optic cable, twisted pair, DSL, or wireless technologies such as infrared, radio, and microwave are included in the definition of the medium. It should be understood, however, that computer-readable storage media and data storage media do not include connections, carrier waves, signals, or other transitory media, but are instead intended to be non-transitory, tangible storage media. Disk and disc, as used in the application, includes compact disc (CD), laser disc, optical disc, digital versatile disc (DVD), floppy disk, and blu-ray disc, where disks usually reproduce data magnetically, while discs use lasers to optically reproduce data replicate the data.

In one or more exemplary aspects, the functions described in the computer program for executing the machine learning model training method described herein may be implemented in hardware, software, firmware, or any combination thereof. When implemented in software, the functions may be stored on or transmitted over as one or more instructions or code on a computer-readable medium. The steps of the methods or algorithms disclosed herein may be embodied in processor-executable software modules, where the processor-executable software modules may reside on a tangible, non-transitory computer readable and writable storage medium. Tangible, non-transitory computer-readable storage media can be any available media that can be accessed by a computer.

The flowcharts and block diagrams in the above-described figures of the present application illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various embodiments of the present application. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of code that contains one or more logical functions for implementing the specified functions executable instructions. It should also be noted that, in some alternative implementations, the functions noted in the blocks may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It is also noted that each block of the block diagrams and/or flowchart illustrations, and combinations of blocks in the block diagrams and/or flowchart illustrations, can be implemented by dedicated hardware-based systems that perform the specified functions or operations , or can be implemented by a combination of dedicated hardware and computer instructions.

The above-mentioned embodiments merely illustrate the principles and effects of the present application, but are not intended to limit the present application. Anyone skilled in the art can make modifications or changes to the above embodiments without departing from the spirit and scope of the present application. Therefore, all equivalent modifications or changes made by those with ordinary knowledge in the technical field without departing from the spirit and technical ideas disclosed in this application should still be covered by the claims of this application.

Claims

A training method for a machine learning model for evaluating the performance of 3D printing components, characterized in that the training method for the machine learning model comprises the following steps:

obtaining multiple sets of residual stress data and/or multiple sets of equivalent relaxation time data of the 3D printed component; and obtaining multiple sets of measurement values of at least one performance evaluation parameter of the 3D printed component in an actual printing environment; and

Correlation training is performed using the sets of residual stress data and/or the sets of equivalent relaxation time data as input data and the sets of measurement values as output data to obtain the machine learning model.
The method for training a machine learning model according to claim 1, further comprising the step of adjusting the machine learning model by performing error evaluation based on predicted data and the output data; wherein the predicted data is machine learning The performance evaluation result of the model making predictions based on the input data.
The method for training a machine learning model according to claim 1, wherein the multiple sets of residual stress data or the multiple sets of equivalent relaxation time data are obtained in different finite element simulation environments; wherein, the Different finite element simulation environments are formed by setting the 3D printing component model with different printing parameter information.
The method for training a machine learning model according to claim 3, wherein the printing parameter information includes the printing path of the printing head, the layer height, the moving speed of the printing head, the output speed of the printing material, the heating temperature of the printing material, and One or more information on extrusion temperature.
The method for training a machine learning model according to claim 3, wherein the printing parameter information further comprises printing equipment parameter information, and the printing equipment parameter information comprises printing initial temperature field information, printing equipment component board heating temperature, One or more of print chamber temperature, and print head shape.
The method for training a machine learning model according to claim 5, wherein the method for obtaining the initial printing temperature field information comprises: obtaining a finite element simulation based on a thermal imager or a thermocouple measuring the temperature distribution of the printing cavity before printing The printed initial temperature field information of the environment.
The method for training a machine learning model according to claim 3, wherein the printing parameter information further comprises material property information of the printing member, and the material property information comprises: wire type, wire diameter, wire cross section One or more information of shape, material maximum heating temperature, material thermal parameters, material dynamic mechanical parameters, and material initial residual stress.
The method for training a machine learning model according to claim 7, wherein the method for determining the thermal parameters of the material comprises at least one of the following:

Measure the thermal emissivity of printing materials based on the thermal emissivity tester;

Based on the wire printing experiment, record the temperature change law of the printing wire after being heated and placed in the printing chamber, thereby calculating the equivalent convective heat transfer coefficient; or

The printing experiment is carried out on the geometric structure model with regular shape, and the temperature field distribution of the printing material changes with time during the printing process is recorded, and the geometric structure model is set to perform finite element simulation with different convective heat transfer coefficients, and output the simulated printing process. Simulate the temperature field, and take the convective heat transfer coefficient corresponding to the simulated temperature field that coincides with the temperature field of the printing experiment as the equivalent convective heat transfer coefficient.
The method for training a machine learning model according to claim 7, wherein the method for determining the dynamic mechanical parameters of the material comprises: measuring the alternating strain, constant strain, or fixed strain applied to the printed member or the sample at different temperatures The response after loading, the change curves of the storage modulus and loss modulus of the material with temperature are obtained, and the curve is fitted to obtain the dynamic mechanical parameter information of the material input in the finite element simulation.
The method for training a machine learning model according to claim 9, wherein the alternating strain, constant strain or fixed load cause different deformations of the printing member according to different loading directions, and the deformation types include tension, compression At least one of , bending, 3-point bending, and shearing.
The method for training a machine learning model according to claim 9, wherein the alternating strain is a harmonic stress or a harmonic load.
The method for training a machine learning model according to claim 7, wherein the method for determining the initial residual stress of the material comprises the following steps:

Perform multiple sets of monofilament printing experiments with different printing parameter information for printing material settings;

Calculate or measure the residual stress of monofilament components obtained from multiple sets of monofilament printing experiments; and

Obtain a residual strain database including the different printing parameter information and the residual stress of the monofilament component obtained by multiple sets of printing experiments; wherein, the residual stress of the monofilament component obtained by the printing experiment has a corresponding relationship with the printing parameter information of the monofilament component.
The method for training a machine learning model according to claim 12, characterized in that, in the step of calculating or measuring the residual stress of monofilament components obtained from multiple sets of monofilament printing experiments, the monofilament structures are processed to The residual strain is released, and the deformation of the monofilament is measured to calculate the residual stress; or, the residual stress of the monofilament member is determined based on a physical detection method.
The method for training a machine learning model according to claim 3, wherein a coupled simulation calculation is performed on the 3D printed component in the finite element simulation environment to obtain the multiple sets of residual stress data and/or multiple sets of residual stress data. Equivalent relaxation time data, the coupling simulation calculation is a coupling simulation calculation under preset boundary conditions, and the boundary conditions include thermal convection boundary conditions and/or thermal radiation boundary conditions.
The method for training a machine learning model according to claim 3, wherein a coupled simulation calculation is performed on the 3D printed component in the finite element simulation environment to obtain the multiple sets of residual stress data and/or multiple sets of residual stress data. Equivalent relaxation time data, the model calculated by the coupled simulation includes a linear viscoelastic model for describing the mechanical deformation of the printing material, or/and a transverse isotropic heat conduction model for describing the thermal conduction behavior of the printing material, or Orthotropic heat conduction model.
The method for training a machine learning model according to claim 3, wherein the simulation domain of the 3D printing component model in the finite element simulation environment is divided into a plurality of sub-regions according to a preset size or a preset number of partitions, and the The multiple sets of residual stress data and/or multiple sets of equivalent relaxation time data are the average value of residual stress or/and the average value of equivalent relaxation time in each sub-region in the simulation domain.
The method for training a machine learning model according to claim 15 or 16, wherein the average value of residual stress or/and the average value of equivalent relaxation time in each sub-region is a two-dimensional data obtained by transforming three-dimensional data. data.
The method for training a machine learning model according to claim 1, wherein the performance evaluation parameters include component printability, shape distortion, structural stiffness, interlayer bonding strength, geometric accuracy, minimum printing gap, resolution , at least one of bridging performance, drape performance, surface waviness, minimum print layer thickness, squareness.
The method for training a machine learning model according to claim 18, wherein the method of determining the measurement value of the shape distortion comprises: comparing the surface mesh of the 3D printing component and the surface mesh of the 3D printing component model in an actual printing environment The curvature error between the cells can be calculated to obtain the shape distortion of the local area of the component surface or/and the overall component.
The method for training a machine learning model according to claim 19, wherein the surface mesh of the 3D printing component in the actual printing environment is obtained by scanning the 3D printing component with a three-dimensional optical scanner.
The method for training a machine learning model according to claim 18, wherein the method for determining the printability of the component comprises: setting the three-dimensional model of the component with different printing parameters, and setting the local collapse of the 3D-printed component in an actual printing environment Or the 3D model of the distorted component in combination with the printing parameters is determined not to be printable.
The method for training a machine learning model according to claim 18, wherein the structural stiffness comprises at least one of bending stiffness, tensile stiffness, compression stiffness, shear stiffness, and torsional stiffness.
The method for training a machine learning model according to claim 18, wherein the structural stiffness is represented by the local stiffness of the component, and the method for measuring the local structure of the component comprises: cutting the 3D printed component partition into a preset regular geometry Structure, measures the structural stiffness of each geometry.
The method for training a machine learning model according to claim 18, wherein the interlayer bonding strength includes the overall interlayer bonding strength and local interlayer bonding strength of the 3D printing component.
The method for training a machine learning model according to claim 1, wherein the equivalent relaxation time data is based on the dynamic temperature field data of the printing material changing with time during the actual printing process or the simulated printing process based on the WLF equation or / and Arrhenius equation (Arrhenius equation) when the temperature is equivalent to the preset temperature value obtained.
The method for training a machine learning model according to claim 1, further comprising the step of encapsulating multiple sub-models as the machine learning model; wherein the multiple sets of residual stress data and/or the multiple sets of residual stress data and/or the multiple A set of equivalent relaxation time data is used as input data and a measured value of one of the performance evaluation parameters is used as output data for associated training to obtain a sub-model.
The method for training a machine learning model according to claim 1, wherein the multiple sets of residual stress data and/or multiple sets of equivalent relaxation time data are the residual stress data of each spatial coordinate point in the 3D printing component and/or or equivalent relaxation time data.
The method for training a machine learning model according to claim 1, wherein the multiple sets of residual stress data and/or multiple sets of equivalent relaxation time data are the average value of residual stress in each sub-region in the 3D printing component model and/or the average value of the equivalent relaxation time; wherein, the sub-regions are obtained by dividing the 3D printing component model based on a preset size or a preset number of partitions.
The method for training a machine learning model according to claim 1, wherein the measurement value of the at least one performance evaluation parameter is a data array.
The method for training a machine learning model according to claim 1 or 3, wherein the 3D printing component is a geometric structure with regular shape, including vertical thin walls, horizontal thin plates, inclined thin plates, cubes, cylinders, Cylinder, square cylinder, cone, inclined column, solid block, grid filling structure.
The method for training a machine learning model according to claim 1, wherein the machine learning model is a neural network model.
A training device for a machine learning model, wherein the machine learning model is used to evaluate the performance of a 3D printing component, and the training device includes:

A training sample acquisition module for acquiring multiple sets of residual stress data and/or multiple sets of equivalent relaxation time data of the 3D printing component; and acquiring multiple sets of measurement values of at least one performance evaluation parameter of the 3D printing component in an actual printing environment;

A training module, configured to use the multiple sets of residual stress data and/or the multiple sets of equivalent relaxation time data as input data and the multiple sets of measurement values as output data to perform associated training to obtain the machine learning model.
An evaluation system for evaluating the performance of 3D printing components, characterized in that it includes:

an input module for receiving residual stress data and/or equivalent relaxation time data for the 3D printed component; and

A prediction module, used to call the machine learning model generated by the training method of the machine learning model according to any one of claims 1-31 to predict the residual stress data and/or the equivalent relaxation time data, and output the 3D An evaluation value of at least one performance evaluation parameter of the printed member.
A computer equipment, characterized in that, comprising:

a storage device for storing at least one program, and preset input data and output data; and

A processing device, connected to the storage device, for executing the at least one program, so as to call the at least one program in the storage device to execute and implement the machine learning model according to any one of claims 1-31. training method.
A computer equipment, characterized in that, comprising:

A storage device for storing at least one program and a machine learning model generated by the training method for a machine learning model according to any one of claims 1-31; and

A processing device, connected to the storage device, for executing the at least one program to invoke the at least one program execution and the machine learning model in the storage device to perform an analysis of the residual stress data and/or equivalent The relaxation time data is used for prediction, and an evaluation value of at least one performance evaluation parameter of the 3D printed component is output.
A computer-readable storage medium, characterized by storing at least one program, which implements the method for training a machine learning model according to any one of claims 1-31 when the at least one program is executed by a processor.