CN116360342A - Machine tool thermal error prediction modeling method - Google Patents
Machine tool thermal error prediction modeling method Download PDFInfo
- Publication number
- CN116360342A CN116360342A CN202310106588.8A CN202310106588A CN116360342A CN 116360342 A CN116360342 A CN 116360342A CN 202310106588 A CN202310106588 A CN 202310106588A CN 116360342 A CN116360342 A CN 116360342A
- Authority
- CN
- China
- Prior art keywords
- data
- thermal error
- machine tool
- domain
- source domain
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 55
- 230000006978 adaptation Effects 0.000 claims abstract description 31
- 238000012545 processing Methods 0.000 claims abstract description 11
- 238000010606 normalization Methods 0.000 claims abstract description 6
- 238000009826 distribution Methods 0.000 claims description 31
- 238000012549 training Methods 0.000 claims description 27
- 238000013528 artificial neural network Methods 0.000 claims description 22
- 230000003044 adaptive effect Effects 0.000 claims description 15
- 238000006073 displacement reaction Methods 0.000 claims description 5
- 230000006870 function Effects 0.000 description 43
- 238000013508 migration Methods 0.000 description 7
- 230000005012 migration Effects 0.000 description 7
- 238000005516 engineering process Methods 0.000 description 5
- 238000003754 machining Methods 0.000 description 5
- 238000013135 deep learning Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 238000004519 manufacturing process Methods 0.000 description 3
- 230000008929 regeneration Effects 0.000 description 3
- 238000011069 regeneration method Methods 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- 238000004364 calculation method Methods 0.000 description 2
- 230000005484 gravity Effects 0.000 description 2
- 239000011159 matrix material Substances 0.000 description 2
- 238000003801 milling Methods 0.000 description 2
- 239000000203 mixture Substances 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 238000013526 transfer learning Methods 0.000 description 2
- 230000009466 transformation Effects 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000005520 cutting process Methods 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 238000013480 data collection Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000000691 measurement method Methods 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B19/00—Programme-control systems
- G05B19/02—Programme-control systems electric
- G05B19/18—Numerical control [NC], i.e. automatically operating machines, in particular machine tools, e.g. in a manufacturing environment, so as to execute positioning, movement or co-ordinated operations by means of programme data in numerical form
- G05B19/406—Numerical control [NC], i.e. automatically operating machines, in particular machine tools, e.g. in a manufacturing environment, so as to execute positioning, movement or co-ordinated operations by means of programme data in numerical form characterised by monitoring or safety
- G05B19/4069—Simulating machining process on screen
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B2219/00—Program-control systems
- G05B2219/30—Nc systems
- G05B2219/32—Operator till task planning
- G05B2219/32342—Real time simulation
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/30—Computing systems specially adapted for manufacturing
Landscapes
- Engineering & Computer Science (AREA)
- Human Computer Interaction (AREA)
- Manufacturing & Machinery (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Automation & Control Theory (AREA)
- Numerical Control (AREA)
Abstract
The invention provides a machine tool thermal error prediction modeling method, which comprises the following steps: step S1, acquiring experimental data corresponding to a thermal error task preset in the machine tool, and generating a data set with characteristic data and a data set with thermal error data in the experimental data together as source domain data; the source domain data comprises a plurality of source domain characteristic data in a spindle of the machine tool and source domain thermal error values in a plurality of directions; s2, generating a thermal error prediction task missing in the thermal error data as a target domain; the target domain comprises a plurality of target domain feature data corresponding to the source domain feature data; and step S3, carrying out normalization processing on the source domain characteristic data and the target domain characteristic data together to obtain a deep multi-core joint domain adaptation network model for thermal error prediction. Compared with the related art, the prediction accuracy of the machine tool thermal error prediction modeling method is high.
Description
Technical Field
The invention relates to the technical field of machine tool design, in particular to a machine tool thermal error prediction modeling method.
Background
At present, a machine tool becomes important equipment for precisely manufacturing a vertical numerical control machining center. In the machining process of the main shaft of the machine tool, the thermal error generated by thermal deformation accounts for 40% -70% of the total machining error of the machine tool, and the higher the precision of the machine tool is, the larger the thermal error accounts for. Thermal errors are a major factor affecting precision manufacturing of a vertical numerical control machining center, and therefore, prediction and compensation techniques for machine tool thermal errors are an important part of machine tool thermal error prediction modeling.
The thermal error prediction compensation technology of the related art is modeled by deep learning, and specifically comprises the following steps: and arranging a multi-source sensor for a specific machine tool, and acquiring data to establish a regression model. However, in the deep learning modeling method of the related art, in the prediction task of single equipment, the same working condition and specific thermal errors, higher prediction precision can be obtained. However, in the situations that thermal error data cannot be acquired during variable working conditions, equipment crossing, multiple types and spindle band-cutting processing, the technical means of the thermal error prediction compensation technology in the related art are proved to be the best for all occasions, and a model under a specific working condition and equipment is difficult to be suitable for all occasions. Therefore, the mobility of the model on the problems of variable working conditions and equipment crossing is discussed, and the method has practical significance on modeling of the thermal error of the machine tool. In addition, the numerical control machine tools are more in variety, and main thermal errors of different machine tools are different due to the arrangement mode of the machine tool main shaft; the spindle is usually switched to rotate speed under different working conditions due to processing limitation; the deep learning modeling means of the related art is difficult to adapt to all situations, and can only obtain higher prediction precision under specific machine tools and working conditions. Therefore, the thermal error of the machine tool has the characteristics of variable working conditions, equipment crossing and multiple types, and the thermal error data is difficult to collect during the spindle cutter-carrying processing, so that the method of building the model by depending on the characteristic data and the labels through deep learning in the related technology is difficult to implement.
In addition, the new technology is increasingly widely applied to precision manufacturing of vertical numerical control machining centers. Among the new techniques are transfer learning and joint adaptation networks. The transfer learning is used for learning experience in the knowledge source field, and the experience is applied to a target task in similar problems, so that the requirements of a machine tool thermal error modeling scene are met. The joint adaptive network is a field adaptive learning model, and reduces the data joint distribution difference between a source field and a target field through iterative learning, so that the model built in the source field can be adapted to a target field thermal error prediction task. The characteristics that the data are related but different in the variable working condition, the cross-equipment and the multi-type prediction tasks of the thermal errors are considered, and a new research direction is provided for the machine tool thermal error prediction modeling technology. Therefore, how to apply the migration learning and the joint adaptation network to machine tool thermal error prediction modeling to improve prediction accuracy is a solved technical problem.
Disclosure of Invention
Aiming at the defects of the prior art, the invention provides a machine tool thermal error prediction modeling method with high prediction precision.
In order to solve the technical problems, an embodiment of the present invention provides a machine tool thermal error prediction modeling method, which is applied to a machine tool, and includes the following steps:
step S1, acquiring experimental data corresponding to a thermal error task preset in the machine tool, and generating a data set with characteristic data and a data set with thermal error data in the experimental data together as source domain data; the source domain data comprises a plurality of source domain characteristic data in a spindle of the machine tool and source domain thermal error values in a plurality of directions;
s2, generating a thermal error prediction task missing in the thermal error data as a target domain; the target domain comprises a plurality of target domain feature data corresponding to the source domain feature data;
and step S3, carrying out normalization processing on the source domain characteristic data and the target domain characteristic data together to obtain a deep multi-core joint domain adaptation network model for thermal error prediction.
Preferably, in step S1, the machine tool is provided with a data acquisition platform matched with the thermal error task, and the experimental data are obtained by using multi-source sensors in the data acquisition platform.
Preferably, the step S3 includes:
step S31, in a preset deep neural network, the source domain characteristic data and the target domain characteristic data are used as input data, the source domain thermal error value is used as output data, the source domain thermal error value is processed in a forward propagation function of the deep neural network, and then a first layer characteristic value and a second layer characteristic value of a full-connection layer of the deep neural network are returned to obtain a source domain first layer characteristic value, a source domain second layer characteristic value, a target domain first layer characteristic value and a target domain second layer characteristic value;
step S32, calculating the first layer characteristic value of the source domain, the second layer characteristic value of the source domain, the first layer characteristic value of the target domain and the second layer characteristic value of the target domain through a kernel joint distribution maximum mean difference loss function;
and step S33, calculating the average absolute error loss function and the total loss function of the predicted value and the measured value output by the source domain data through the deep neural network respectively by the source domain first layer characteristic value, the source domain second layer characteristic value, the target domain first layer characteristic value and the target domain second layer characteristic value.
Preferably, in the step S33, the total loss function is a sum of the kernel joint distribution maximum mean difference loss function and the average absolute error loss function.
Preferably, after the step S3, the machine tool thermal error prediction modeling method further includes:
and S4, training the deep multi-core joint field adaptive network model until the deep multi-core joint field adaptive network model converges.
Preferably, in the step S4, the training is performed by back-propagating the total loss function calculated in the step S33 a plurality of times.
Preferably, in the step S4, the training direction includes: constraining the deep neural network to train towards a gradient direction which makes the difference between the source domain feature data and the target domain feature data small by using the kernel joint distribution maximum mean difference loss function; and constraining the gradient direction training with small difference between the predicted value and the measured value error of the source domain data through the average absolute error loss function.
Preferably, after the step S4, the machine tool thermal error prediction modeling method further includes:
and S5, inputting experimental data acquired in real time in the machine tool into the depth multi-core joint field adaptation network model after training is completed, performing thermal error prediction, and obtaining data output by the depth multi-core joint field adaptation network model according to the thermal error prediction so as to be used for the machining center displacement of the machine tool.
Compared with the related art, the machine tool thermal error prediction modeling method of the invention comprises the following steps of implementing step S1 to step S5: step S1, acquiring experimental data corresponding to a thermal error task preset in the machine tool, and generating a data set with characteristic data and a data set with thermal error data in the experimental data together as source domain data; s2, generating a thermal error prediction task missing in the thermal error data as a target domain; step S3, carrying out normalization processing on the source domain feature data and the target domain feature data together to obtain a depth multi-core joint field adaptation network model for thermal error prediction; s4, training the depth multi-core joint field adaptive network model until the depth multi-core joint field adaptive network model converges; and S5, inputting experimental data acquired in real time in the machine tool into the depth multi-core joint field adaptation network model after training is completed, performing thermal error prediction, and obtaining data output by the depth multi-core joint field adaptation network model according to the thermal error prediction so as to be used for the machining center displacement of the machine tool. According to the operations from step S1 to step S5, the goal of reducing the joint distribution difference between the source domain and the target domain data is achieved through the domain adaptive learning, so that the deep multi-core joint domain adaptive network model established in the source domain can be adapted and migrated to be applied to the target domain prediction task, and higher prediction precision is ensured.
Drawings
The present invention will be described in detail with reference to the accompanying drawings. The foregoing and other aspects of the invention will become more apparent and more readily appreciated from the following detailed description taken in conjunction with the accompanying drawings. In the accompanying drawings:
FIG. 1 is a flow chart diagram of a machine tool thermal error prediction modeling method of the present invention;
FIG. 2 is a block flow diagram of step S3 in the machine tool thermal error prediction modeling method of the present invention;
FIG. 3 is a graph of total loss function versus training times for an embodiment of a machine tool thermal error predictive modeling method of the present invention;
FIG. 4 is a graph of prediction accuracy of a DNN model of an embodiment of a machine tool thermal error prediction modeling method according to the present invention on a source domain thermal error prediction task at a rotational speed of 3000 rpm;
FIG. 5 is a graph of prediction accuracy of a DNN model of an embodiment of a machine tool thermal error prediction modeling method according to the present invention on a source domain thermal error prediction task at a rotational speed of 6000 rpm;
FIG. 6 is a graph of prediction accuracy of a DMK-JAN model of an embodiment of a machine tool thermal error prediction modeling method according to the present invention on a source domain thermal error prediction task at a rotational speed of 3000 rpm;
FIG. 7 is a graph showing prediction accuracy of a DMK-JAN model in an embodiment of a machine tool thermal error prediction modeling method according to the present invention on a source domain thermal error prediction task under a rotation speed of 6000 rpm.
Detailed Description
The following describes in detail the embodiments of the present invention with reference to the drawings.
The detailed description/examples set forth herein are specific embodiments of the invention and are intended to be illustrative and exemplary of the concepts of the invention and are not to be construed as limiting the scope of the invention. In addition to the embodiments described herein, those skilled in the art will be able to adopt other obvious solutions based on the disclosure of the claims and specification of the present application, including those adopting any obvious substitutions and modifications to the embodiments described herein, all within the scope of the present invention.
The invention provides a machine tool thermal error prediction modeling method. The machine tool thermal error prediction modeling method is applied to a machine tool.
Referring to fig. 1, fig. 1 is a flow chart of a machine tool thermal error prediction modeling method according to the present invention.
The machine tool thermal error prediction modeling method comprises the following steps:
step S1, acquiring experimental data corresponding to a thermal error task preset in the machine tool, and generating a data set with characteristic data and a data set with thermal error data in the experimental data together as source domain data D s 。
The source domain data D s Comprising a plurality of source domain feature data X in a spindle of the machine tool i And source domain thermal error values { y over multiple directions i |e x ,e y ,e z }. The source domain data D s The source domain in (a) refers to the domain of knowledge sources.
In the step S1, the machine tool is provided with a data acquisition platform matched with the thermal error task, and the experimental data are obtained by a multi-source sensor in the data acquisition platform.
S2, generating a thermal error prediction task missing in the thermal error data as a target domain D t 。
The target domain D t Including and said source domain feature data X i Corresponding multiple target domain feature data X j . Wherein the target domain D t Refers to the field of applying knowledge to solve problems.
Step S3, the source domain characteristic data X i And the target domain feature data X j Normalization processing is carried out jointly to obtain a Deep Multi-core joint field adaptation network model (Deep Multi-Kernel Joint Domain Adaptation Network, DMK-JAN for short) for thermal error prediction.
The deep multi-core joint field adaptation network model DMK-JAN blends a joint adaptation network (Joint Adaptation Network, JAN) in deep migration learning (DeepTransfer Learning) into a deep neural network (Deep Neural Network, DNN) and is used for machine tool spindle thermal error prediction scenes. The characteristics of the source domain and the target domain in the full-connection layer of the deep neural network are input into a Multi-core maximum mean difference (Multi-KernelJointMaximum Mean Discrepancy, MK-JMMD) loss function of joint distribution, and the Multi-core maximum mean difference is used for iteratively updating a training model, so that the problem that the source domain and the target domain are different in distribution is solved. The deep migration learning (DeepTransfer Learning) is a model of knowledge source domain (source domain) applied to other different but related problems (target domain). In this embodiment, the deep multi-core joint field adaptive network model DMK-JAN may also be referred to as DMK-JAN model.
The thermal error prediction is in this embodiment a thermal error prediction of a spindle of the machine tool, and during the operation of the spindle, an increase in temperature is generated due to the influence of gravity, frictional heat, and the like, which affects the machining accuracy. And (3) collecting other characteristic data of the spindle machining process by arranging a temperature sensor at the key position of the spindle, measuring the actual thermal error of the spindle, and establishing a regression prediction model of the characteristic data and the thermal error of the spindle. The characteristic data of the spindle is used to characterize data affecting thermal errors, such as: temperature characteristics, power characteristics, current characteristics and the like during machine tool processing. The regression prediction model is a predictive modeling technique that studies the relationship between dependent variables (targets) and independent variables (predictors).
Referring to fig. 2, fig. 2 is a block flow diagram of step S3 in the machine tool thermal error prediction modeling method of the present invention. Specifically, the step S3 includes:
step S31, in a preset deep neural network (Deep Neural Network, DNN) the source domain feature data X is obtained i And the target domain feature data X j As input data and to take the source domain thermal error value { y }, as i |e x ,e y ,e z Processing in the forward propagation function of the deep neural network DNN as output data, and returning the first layer characteristic value and the second layer characteristic value of the fully connected layer of the deep neural network DNN to obtain a source domain first layer characteristic value
Source field second layer eigenvalue +.>
Target domain first layer feature values
Second layer characteristic value of target domain +.>
In this embodiment, the deep neural network DNN may also be simply referred to as a DNN model.
Step S32, the source domain first layer characteristic value is obtained
The source domain second layer eigenvalue
The target domain first layer characteristic value +.>
The target domain second layer characteristic value +.>
Loss function Loss through kernel joint distribution maximum mean difference (MK-JMMD) MK-JMMD And (5) performing calculation.
Step S33, the source domain first layer characteristic value is obtained
The source domain second layer eigenvalue
The target domain first layer characteristic value +.>
The target domain second layer characteristic value +.>
By the source domain data D s Predicted value +.>
And the measured value y i Is a Loss function Loss of average absolute error (Mean Absolute Error, MAE for short) MAE And the total Loss function Loss are calculated separately.
In the step S33, the total Loss function Loss is the kernel-combined distribution maximum mean difference Loss function Loss MK-JMMD And the average absolute error Loss function MAE . Namely, the following conditions are satisfied: loss=loss MK-JMMD +Loss MAE 。
In this embodiment, after the step S3, the machine tool thermal error prediction modeling method further includes:
and S4, training the depth multi-core joint field adaptation network model DMK-JAN until the depth multi-core joint field adaptation network model DMK-JAN converges.
In the step S4, the training is performed by back-propagating the total Loss function Loss calculated in the step S33 a plurality of times.
In the step S4, the training direction includes:
loss function Loss through the kernel joint distribution maximum mean difference (MK-JMMD) MK-JMMD Constraining the deep neural network DNN to orient the source domain feature data X i And the target domain feature data X j Training the gradient direction with small difference;
loss function Loss by the mean absolute error (Mean Absolute Error, MAE for short) MAE Constraining the source domain data D s Predicted value of (2)
And the measured value y i Gradient direction training with small difference between the two errors.
In this embodiment, after the step S4, the machine tool thermal error prediction modeling method further includes:
and S5, inputting experimental data acquired in real time in the machine tool into the depth multi-core joint field adaptation network model DMK-JAN after training is completed, performing thermal error prediction, and obtaining data output by the depth multi-core joint field adaptation network model DMK-JAN according to the thermal error prediction so as to be used for the displacement of a machining center of the machine tool.
The working principle of the machine tool thermal error prediction modeling method is described in detail below:
the joint distribution probability is represented by an edge probability distribution P (X s ,Y s ) And conditional probability distribution Q (X t ,Y t ) Composition, wherein source domain D s ＝{X s ,Y s Target domain D t ＝{X t The method proposes two hypotheses:
assuming that the first, source and target domain edge probability distributions are different, i.e., P s (X s )≠P t (X t )；
Assuming that the conditional probability distributions of the two, source and target domains are different, i.e. Q s (y s |x s )≠Q t (y t |x t ). The core idea of joint distribution self-adaption is to find a transformation A so that the edge probability distribution P after transformation s (A T x a ) And P t (A T x t ) Conditional probability distribution P S (y s |A T x s ) And P S (y t |A T x t ) While approaching.
X and Y represent characteristic data and thermal error value respectively, the data is input into a deep neural network to extract deep characteristics, and the joint distribution P (Z) between network layers is utilized s1 ,Z s2 ,…Z sL ) And Q (Z) t1 ,Z t2 ,…Z tL ) Replacement P (X) s ,Y s ) And Q (X) t ,Y t ) Wherein Z is 1 ,Z 2 ,…Z L Representing the fully connected layers in different positions.
First, source domain and destination are reducedEdge probability distribution difference of the mark domain, and the edge probability distribution difference of the two distributions is measured by using the distance of the maximum average difference (Maximum Mean Discrepancy, MMD) so that P s (A T x s ) And P t (A T x t ) As close as possible, as shown in the following equation (1), where n s And n t The number of samples representing the source domain and the target domain, respectively.
Then the conditional probability distribution of the source domain and the target domain is adapted such that P s (y s |A T x s ) And P t (y t |A T x T ) The distance of (2) is as small as possible, and the total optimization objective is as shown in formula 3:
in formula (3): phi is the Lagrangian multiplier; h is a central matrix; i is an identity matrix.
The Maximum Mean Difference (MMD) loss function maps two different but related data to a high-dimensional regeneration kernel Hilbert space through a kernel function k (&. Cndot.,. Cndot.) and measures the distribution distance of the two data in the space, wherein the regeneration kernel Hilbert space is a complete inner product function space constructed by the sum kernel function k (& cndot. ). According to the traditional distance measurement method, single-core MMD is used for carrying out minimum distance calculation and solving of a source domain and a target domain, and alignment of two-domain data features is achieved.
More preferably, in this embodiment, it is proposed that the source domain and the target domain data are mapped to the regenerated kernel hilbert space through a kernel function, and the joint distribution difference of the network adaptation layer is aligned by using a JMMD algorithm, where the JMMD is defined as shown in formula (4).
Where k is the kernel function and L is the different adaptation layer.
The Multi-core JMMD (Multi-Kernel JMMD, MK JMMD) in this embodiment is obtained by linearly combining a plurality of core functions on the basis of JMMD, and specifically is shown in formula (5):
wherein: k is a formula of a plurality of kernel functions subjected to linear combination; k is a kernel function; beta is the number of kernel functions. Compared with single-core JMMD, MK-JMMD enables source domain and target domain data to have better adaptability in a high-dimensional regeneration core Hilbert space.
The best quality of the DMK-JAN loss function can be defined as shown in equation (6).
The following analysis of the beneficial effects of implementing the machine tool thermal error predictive modeling method is performed by specific embodiments:
in the embodiment, two working conditions of idling (A) and milling (B) of a main shaft of the same vertical machine tool are taken as research objects, and the modeling process of the depth multi-core combined field adaptation network model DMK-JAN under the A and the migration prediction effect of the model from the A to the B are shown. The key thermal error of the machine tool is a main shaft Z-direction thermal error under the influence of gravity and the temperature rise of the main shaft, a data acquisition platform is arranged on the main shaft of the machine tool, a data set of a working condition A is taken as a source domain and comprises characteristic data and thermal error data, a data set of a working condition B is taken as a target domain and comprises characteristic data similar to the source domain, the target domain is provided with thermal error data for verifying the migration effect of the depth multi-core combined domain adaptive network model, but the thermal error value is not required in a real scene, and the data sets of the source domain and the target domain are shown in the table one. The source domain is a data set of the working condition A main shaft idling at 3000rpm, the target domain is a data set of the working condition B main shaft milling with a cutter at 6000rpm, and specific parameters of the two data sets are shown in the following table 1:
table 1, source domain and destination domain data sets.
The steps 1 to 4 are implemented, namely, the following operations: after normalizing the characteristic data of the source domain and the target domain, respectively inputting the two data sets into the depth multi-core joint domain adaptive network model DMK-JAN, and returning the characteristic values of the first layer and the second layer of the full-connection layer in a forward propagation function of the depth multi-core joint domain adaptive network model DMK-JAN to respectively obtain the characteristic values of the first layer of the source domain
The source domain second layer characteristic value +.>
The target domain first layer characteristic value +.>
The target domain second layer characteristic value +.>
Calculating the multi-core joint distribution maximum mean difference (MK-JMMD) Loss function Loss of 4 features MK-JMMD The method comprises the steps of carrying out a first treatment on the surface of the Simultaneously calculating a predicted value of a source domain output through the deep neural network DNN>
And the measured value y i Average absolute error (Mean Absolute Error, MAE) Loss function Loss of (2) MAE . The total Loss function is loss=loss MK-JMMD +Loss MAE . When the model converges, the network reaches the expected target, andthe training is terminated.
And then through the kernel joint distribution maximum mean difference (MK-JMMD) Loss function Loss MK-JMMD Constraining the deep neural network DNN to orient the source domain feature data X i And the target domain feature data X j Training the gradient direction with small difference; at the same time, the Loss function Loss is realized through the average absolute error (Mean Absolute Error, MAE for short) MAE Constraining the source domain data D s Predicted value of (2)
And the measured value y i Gradient direction training with small difference between the two errors.
Wherein the different feature extractors are in different forms, the hidden layers used for regression are as follows: "Dense", "flame" and the like can all be considered as extracting features in different forms.
The results of implementation of this example are as follows:
referring to fig. 3, fig. 3 is a graph showing total loss function and training times according to an embodiment of the machine tool thermal error prediction modeling method of the present invention. The depth multi-core joint field adaptation network model DMK-JAN iterates 200 times, and the total Loss function is Loss and is shown in figure 3. Wherein, A1 is the total Loss function obtained by the source domain training set is Loss; a2 is the total Loss function obtained for the test set is Loss.
In order to compare the prediction precision of the DMK-JAN model, a DNN model and a DMK-JAN model are respectively established by using source domain data, the prediction precision of the two models on a source domain is compared, and the DNN model and the DMK-JAN model are transferred to a target domain to perform prediction precision, so that four groups of prediction values are obtained.
Referring to fig. 4 to 5, fig. 4 is a graph showing prediction accuracy of a DNN model of an embodiment of a machine tool thermal error prediction modeling method according to the present invention on a source domain thermal error prediction task under a rotation speed of 3000 rpm; fig. 5 is a graph of prediction accuracy of a DNN model of an embodiment of a machine tool thermal error prediction modeling method according to the present invention on a source domain thermal error prediction task at a rotation speed of 6000 rpm.
In FIG. 4, B1 is a prediction accuracy curve of an actual thermal error at a rotation speed of 3000 rpm. B2 is a predicted value curve of the DNN model under the condition of 3000rpm of rotating speed. In FIG. 5, C1 is a prediction accuracy curve of the actual thermal error at a rotation speed of 6000 rpm. C2 is a predicted value curve of the DNN model under the condition of 6000rpm of rotation speed.
Referring to fig. 6 to fig. 7, fig. 6 is a graph showing prediction accuracy of a DMK-JAN model according to an embodiment of the machine tool thermal error prediction modeling method according to the present invention on a source domain thermal error prediction task under a rotation speed of 3000 rpm; FIG. 7 is a graph showing prediction accuracy of a DMK-JAN model in an embodiment of a machine tool thermal error prediction modeling method according to the present invention on a source domain thermal error prediction task under a rotation speed of 6000 rpm.
In FIG. 6, D1 is a prediction accuracy curve of an actual thermal error at a rotation speed of 3000 rpm. D2 is a predicted value curve of the DMK-JAN model at a rotating speed of 3000 rpm. In FIG. 7, E1 is a prediction accuracy curve of the actual thermal error at a rotation speed of 6000 rpm. E2 is a predicted value curve of the DMK-JAN model under the condition of 6000rpm of rotation speed.
The data collection from fig. 4 to 7 described above is shown in table 2 below:
| average absolute error of source field test set/μm | Average absolute error of target domain test set/μm | |
| DNN | 1.9263 | 2.2172 |
| DMK-JAN | 1.8526 | 1.3920 |
And a second table and a prediction precision comparison table.
From the data analysis of the above figures 4 to 7 and table 2, it can be seen that: the migration prediction effect of the DMK-JAN model established by the machine tool thermal error prediction modeling method is better than that of a model which singly uses DNN
In this embodiment, the machine tool thermal error prediction modeling method has high prediction accuracy by the implementation of the steps S1 to S5.
Compared with the related art, the machine tool thermal error prediction modeling method of the invention comprises the following steps of implementing step S1 to step S5: step S1, acquiring experimental data corresponding to a thermal error task preset in the machine tool, and generating a data set with characteristic data and a data set with thermal error data in the experimental data together as source domain data D s The method comprises the steps of carrying out a first treatment on the surface of the S2, generating a thermal error prediction task missing in the thermal error data as a target domain D t The method comprises the steps of carrying out a first treatment on the surface of the Step S3, the source domain characteristic data X i And the target domain feature data X j Carrying out normalization processing together to obtain a Deep Multi-core joint field adaptation network model (Deep Multi-Kernel Joint Domain Adaptation Network, DMK-JAN for short) for thermal error prediction; s4, training the depth multi-core joint field adaptation network model DMK-JAN until the depth multi-core joint field adaptation network model DMK-JAN converges; and S5, inputting experimental data acquired in real time in the machine tool into the depth multi-core joint field adaptation network model DMK-JAN after training is completed, performing thermal error prediction, and obtaining data output by the depth multi-core joint field adaptation network model DMK-JAN according to the thermal error prediction so as to be used for the displacement of a machining center of the machine tool. According to the operations from step S1 to step S5, the goal of reducing the joint distribution difference between the source domain and the target domain data is achieved through the domain adaptive learning, so that the deep multi-core joint domain built in the source domain is adapted to the network model DMKThe JAN can adapt to the migration application on the target domain prediction task and ensure higher prediction precision.
The foregoing is merely exemplary of the present invention, and those skilled in the art should not be considered as limiting the invention, since modifications may be made in the specific embodiments and application scope of the invention in light of the teachings of the present invention.
Claims (8)
1. The machine tool thermal error prediction modeling method is applied to a machine tool and is characterized by comprising the following steps of:
step S1, acquiring experimental data corresponding to a thermal error task preset in the machine tool, and generating a data set with characteristic data and a data set with thermal error data in the experimental data together as source domain data; the source domain data comprises a plurality of source domain characteristic data in a spindle of the machine tool and source domain thermal error values in a plurality of directions;
s2, generating a thermal error prediction task missing in the thermal error data as a target domain; the target domain comprises a plurality of target domain feature data corresponding to the source domain feature data;
and step S3, carrying out normalization processing on the source domain characteristic data and the target domain characteristic data together to obtain a deep multi-core joint domain adaptation network model for thermal error prediction.
2. The method according to claim 1, wherein in step S1, the machine tool is provided with a data acquisition platform matched with the thermal error task, and the experimental data are obtained by multi-source sensors in the data acquisition platform.
3. The machine tool thermal error prediction modeling method according to claim 1, wherein the step S3 includes:
step S31, in a preset deep neural network, the source domain characteristic data and the target domain characteristic data are used as input data, the source domain thermal error value is used as output data, the source domain thermal error value is processed in a forward propagation function of the deep neural network, and then a first layer characteristic value and a second layer characteristic value of a full-connection layer of the deep neural network are returned to obtain a source domain first layer characteristic value, a source domain second layer characteristic value, a target domain first layer characteristic value and a target domain second layer characteristic value;
step S32, calculating the first layer characteristic value of the source domain, the second layer characteristic value of the source domain, the first layer characteristic value of the target domain and the second layer characteristic value of the target domain through a kernel joint distribution maximum mean difference loss function;
and step S33, calculating the average absolute error loss function and the total loss function of the predicted value and the measured value output by the source domain data through the deep neural network respectively by the source domain first layer characteristic value, the source domain second layer characteristic value, the target domain first layer characteristic value and the target domain second layer characteristic value.
4. A machine tool thermal error prediction modeling method according to claim 3, wherein in the step S33, the total loss function is a sum of the kernel-combined distribution maximum mean difference loss function and the average absolute error loss function.
5. The machine tool thermal error prediction modeling method according to claim 4, characterized in that after the step S3, the machine tool thermal error prediction modeling method further comprises:
and S4, training the deep multi-core joint field adaptive network model until the deep multi-core joint field adaptive network model converges.
6. The machine tool thermal error prediction modeling method according to claim 5, wherein in the step S4, the training is achieved by back-propagating the total loss function calculated in the step S33 a plurality of times.
7. The machine tool thermal error prediction modeling method according to claim 5, wherein in the step S4, the training direction includes:
constraining the deep neural network to train towards a gradient direction which makes the difference between the source domain feature data and the target domain feature data small by using the kernel joint distribution maximum mean difference loss function;
and constraining the gradient direction training with small difference between the predicted value and the measured value error of the source domain data through the average absolute error loss function.
8. The machine tool thermal error prediction modeling method according to claim 5, characterized in that after the step S4, the machine tool thermal error prediction modeling method further comprises:
and S5, inputting experimental data acquired in real time in the machine tool into the depth multi-core joint field adaptation network model after training is completed, performing thermal error prediction, and obtaining data output by the depth multi-core joint field adaptation network model according to the thermal error prediction so as to be used for the machining center displacement of the machine tool.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310106588.8A CN116360342A (en) | 2023-02-13 | 2023-02-13 | Machine tool thermal error prediction modeling method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310106588.8A CN116360342A (en) | 2023-02-13 | 2023-02-13 | Machine tool thermal error prediction modeling method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116360342A true CN116360342A (en) | 2023-06-30 |
Family
ID=86910580
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310106588.8A Pending CN116360342A (en) | 2023-02-13 | 2023-02-13 | Machine tool thermal error prediction modeling method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116360342A (en) |
Citations (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104070406A (en) * | 2014-06-23 | 2014-10-01 | 广东工业大学 | Device and method for compensating composition error of main shaft in real time |
| CN110083125A (en) * | 2019-05-19 | 2019-08-02 | 重庆理工大学 | A kind of machine tool thermal error modeling method based on deep learning |
| CN112192318A (en) * | 2020-09-28 | 2021-01-08 | 北京天泽智云科技有限公司 | Machining tool state monitoring method and system |
| CN112801139A (en) * | 2021-01-06 | 2021-05-14 | 南京航空航天大学 | Intelligent tool wear state identification method based on heterogeneous domain adaptive migration learning |
| US20210191359A1 (en) * | 2019-12-20 | 2021-06-24 | Hexagon Technology Center Gmbh | Advanced thermal compensation of mechanical processes |
| CN113569356A (en) * | 2021-07-27 | 2021-10-29 | 重庆大学 | Modeling method and migration learning method of depth residual error LSTM network and thermal error prediction model |
| CN114241265A (en) * | 2021-12-15 | 2022-03-25 | 北京工业大学 | Multi-source domain multi-level transfer learning method based on sample weight |
| CN114239733A (en) * | 2021-12-21 | 2022-03-25 | 华中科技大学 | Machine tool response modeling method and system based on transfer learning and response prediction method |
| CN114676941A (en) * | 2022-05-30 | 2022-06-28 | 国网天津市电力公司经济技术研究院 | Electric-thermal load combined self-adaptive prediction method and device for park comprehensive energy system |
| WO2022157152A2 (en) * | 2021-01-20 | 2022-07-28 | Thyssenkrupp Industrial Solutions Ag | Aircraft and method for inspecting equipment in coking plants to detect sources of error |
| CN115309106A (en) * | 2022-08-05 | 2022-11-08 | 襄阳华中科技大学先进制造工程研究院 | Machine tool thermal field simulation system based on multi-source heterogeneous data |
| CN115470842A (en) * | 2022-08-24 | 2022-12-13 | 襄阳华中科技大学先进制造工程研究院 | Machine tool thermal error prediction method, device, equipment and storage medium |
| CN115657600A (en) * | 2022-11-02 | 2023-01-31 | 重庆大学 | CNN-BLRNET thermal error prediction model and transfer learning method and intelligent integrated framework thereof |
-
2023
- 2023-02-13 CN CN202310106588.8A patent/CN116360342A/en active Pending
Patent Citations (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104070406A (en) * | 2014-06-23 | 2014-10-01 | 广东工业大学 | Device and method for compensating composition error of main shaft in real time |
| CN110083125A (en) * | 2019-05-19 | 2019-08-02 | 重庆理工大学 | A kind of machine tool thermal error modeling method based on deep learning |
| US20210191359A1 (en) * | 2019-12-20 | 2021-06-24 | Hexagon Technology Center Gmbh | Advanced thermal compensation of mechanical processes |
| CN112192318A (en) * | 2020-09-28 | 2021-01-08 | 北京天泽智云科技有限公司 | Machining tool state monitoring method and system |
| CN112801139A (en) * | 2021-01-06 | 2021-05-14 | 南京航空航天大学 | Intelligent tool wear state identification method based on heterogeneous domain adaptive migration learning |
| WO2022157152A2 (en) * | 2021-01-20 | 2022-07-28 | Thyssenkrupp Industrial Solutions Ag | Aircraft and method for inspecting equipment in coking plants to detect sources of error |
| CN113569356A (en) * | 2021-07-27 | 2021-10-29 | 重庆大学 | Modeling method and migration learning method of depth residual error LSTM network and thermal error prediction model |
| CN114241265A (en) * | 2021-12-15 | 2022-03-25 | 北京工业大学 | Multi-source domain multi-level transfer learning method based on sample weight |
| CN114239733A (en) * | 2021-12-21 | 2022-03-25 | 华中科技大学 | Machine tool response modeling method and system based on transfer learning and response prediction method |
| CN114676941A (en) * | 2022-05-30 | 2022-06-28 | 国网天津市电力公司经济技术研究院 | Electric-thermal load combined self-adaptive prediction method and device for park comprehensive energy system |
| CN115309106A (en) * | 2022-08-05 | 2022-11-08 | 襄阳华中科技大学先进制造工程研究院 | Machine tool thermal field simulation system based on multi-source heterogeneous data |
| CN115470842A (en) * | 2022-08-24 | 2022-12-13 | 襄阳华中科技大学先进制造工程研究院 | Machine tool thermal error prediction method, device, equipment and storage medium |
| CN115657600A (en) * | 2022-11-02 | 2023-01-31 | 重庆大学 | CNN-BLRNET thermal error prediction model and transfer learning method and intelligent integrated framework thereof |
Non-Patent Citations (2)
| Title |
|---|
| 邓柯楠: "基于迁移学习的铣削机器人定位误差补偿方法", 机械工程学报, no. 14, pages 170 - 180 * |
| 阳红;方辉;刘立新;张定金;殷国富;徐德炜;: "基于热误差神经网络预测模型的机床重点热刚度辨识方法研究", 机械工程学报, no. 11 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108803486B (en) | Numerical control machine tool thermal error prediction and compensation method based on parallel deep learning network | |
| WO2022141669A1 (en) | Bearing fault diagnosis method for dynamic joint distribution alignment network under variable working conditions | |
| Ma et al. | Self learning-empowered thermal error control method of precision machine tools based on digital twin | |
| CN104835153B (en) | Non-rigid surface's alignment schemes based on rarefaction representation | |
| CN105022344A (en) | Numerical control machine tool thermal error compensating method | |
| CN104950808B (en) | Machine tool thermal error compensation method based on augmented naive Bayes network | |
| CN106354942B (en) | Assembly contact stress distribution evaluation method based on entropy theory | |
| Wang et al. | A pose estimation system based on deep neural network and ICP registration for robotic spray painting application | |
| Shaik | Optimal selection of operating parameters in end milling of Al-6061 work materials using multi-objective approach | |
| Yang et al. | Applying neural network based on fuzzy cluster pre-processing to thermal error modeling for coordinate boring machine | |
| CN112561211B (en) | Drilling circulation pressure consumption prediction method based on neural network deep learning | |
| Li et al. | Prediction of thermal error for feed system of machine tools based on random radial basis function neural network | |
| Li et al. | Roughness prediction model of milling noise-vibration-surface texture multi-dimensional feature fusion for N6 nickel metal | |
| CN112364444A (en) | Numerical control machine tool temperature measuring point optimization method based on finite element model numerical simulation | |
| Durasov et al. | Debosh: Deep bayesian shape optimization | |
| CN109709895B (en) | Machine tool spindle thermal error modeling training method based on image threshold setting | |
| Mu et al. | A point cloud registration-based calibration algorithm for robot offline programming automatic loading in aero-grinding applications | |
| CN108256688B (en) | Steam flooding development effect prediction method for optimizing GRNN by improving firefly algorithm | |
| CN116360342A (en) | Machine tool thermal error prediction modeling method | |
| CN111192158A (en) | Transformer substation daily load curve similarity matching method based on deep learning | |
| Dai et al. | A prediction model of milling force for aviation 7050 aluminum alloy based on improved RBF neural network | |
| CN115933531B (en) | Machine tool thermal error modeling method and system based on depth domain countermeasure migration | |
| CN108197809B (en) | Real-time scheduling method of weight sharing deep network based on dimension optimal conversion | |
| Zhang et al. | An adaptive grinding method for precision-cast blades with geometric deviation | |
| CN116245944B (en) | Cabin automatic docking method and system based on measured data |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |