CN112801091A - Additive manufacturing part forming quality monitoring and predicting method based on deep learning - Google Patents
Additive manufacturing part forming quality monitoring and predicting method based on deep learning Download PDFInfo
- Publication number
- CN112801091A CN112801091A CN202110104828.1A CN202110104828A CN112801091A CN 112801091 A CN112801091 A CN 112801091A CN 202110104828 A CN202110104828 A CN 202110104828A CN 112801091 A CN112801091 A CN 112801091A
- Authority
- CN
- China
- Prior art keywords
- forming quality
- geometric
- characteristic parameters
- temperature characteristic
- neural network
- 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.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
- G06Q10/06395—Quality analysis or management
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
- G06V10/267—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion by performing operations on regions, e.g. growing, shrinking or watersheds
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Business, Economics & Management (AREA)
- Human Resources & Organizations (AREA)
- Software Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Biophysics (AREA)
- Mathematical Physics (AREA)
- Biomedical Technology (AREA)
- Artificial Intelligence (AREA)
- Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Strategic Management (AREA)
- Multimedia (AREA)
- Educational Administration (AREA)
- Economics (AREA)
- Entrepreneurship & Innovation (AREA)
- Development Economics (AREA)
- Game Theory and Decision Science (AREA)
- Marketing (AREA)
- Tourism & Hospitality (AREA)
- Quality & Reliability (AREA)
- Operations Research (AREA)
- General Business, Economics & Management (AREA)
- General Factory Administration (AREA)
- Image Analysis (AREA)
Abstract
The embodiment of the invention discloses an additive manufacturing part forming quality monitoring and predicting method based on deep learning, which comprises the following steps: the method comprises the steps of obtaining a geometric characteristic image and a temperature characteristic image of a molten pool in the additive manufacturing process by using an image obtaining module, obtaining geometric characteristic parameters and temperature characteristic parameters, using the geometric characteristic parameters and the temperature characteristic parameters as input of a double-current convolution neural network, learning hidden corresponding relations among the geometric characteristic parameters, the temperature characteristic parameters and the forming quality of a workpiece, and corresponding the hidden corresponding relations among the geometric characteristic parameters, the temperature characteristic parameters and the forming quality of the workpiece. By adopting the method, the geometric and temperature characteristic parameters of three areas of the molten pool, the plume and the splash are extracted, the dynamic monitoring process of the additive manufacturing technology and the forming quality of the workpiece are more comprehensively reflected, the classification of the forming quality of the workpiece is realized by using the double-current convolution neural network, the advance prediction of the forming quality of the workpiece and the timely correction of the process parameters are realized by using the long-short term memory neural network, the processing time is reduced, and the production quality is improved.
Description
Technical Field
The invention relates to the technical field of material manufacturing monitoring and prediction, in particular to a method for monitoring and predicting the forming quality of an additive manufacturing part based on deep learning.
Background
Additive manufacturing is an advanced manufacturing technology integrating multiple technologies such as an information technology, a new material technology, a manufacturing technology and the like, is known as a representative technology expected to generate the third industrial revolution, and is a leading technology for developing a large-batch manufacturing mode to a personalized manufacturing mode. In recent 20 years, the technology is rapidly developed and widely applied to the fields of aviation, aerospace, nuclear power, weapons, ships and the like. However, the existing additive manufacturing technology has the problems that defects such as cracks, pores, spheroidization, unfused holes and the like are easy to generate in the processing process, so that the forming quality of a finished piece is unstable and the process repeatability is difficult to ensure. Therefore, the real-time monitoring of the state of the additive manufacturing process and the prediction of the forming quality of the workpiece have important significance for improving the production quality and reducing the production cost.
With the advent of the big data era and the rise of intelligent manufacturing technology, the application of artificial intelligence to the field of additive manufacturing becomes a hotspot of academic and industrial research. Deep learning belongs to a branch of machine learning, the deep learning completes target tasks by learning the characteristics of data, and works such as image classification and equipment life prediction are completed by extracting a mapping relation hidden in the data by using a neural network algorithm. However, in the field of additive manufacturing, there are few systems that combine features of two parts, namely, monitoring and predicting of the forming quality of a workpiece, so that the forming quality of the workpiece cannot be predicted only by monitoring the process features in the additive manufacturing process, the forming quality of the workpiece cannot be guaranteed, and a reference cannot be provided for subsequent process compensation, so that the repeatability of the additive manufacturing process becomes unstable.
Disclosure of Invention
The technical problem to be solved by the embodiment of the invention is to provide a method for monitoring and predicting the forming quality of an additive manufacturing part based on deep learning. The forming quality of the finished piece can be monitored and predicted.
In order to solve the technical problem, an embodiment of the present invention provides a method for monitoring and predicting the forming quality of an additive manufacturing product based on deep learning, including the following steps:
s1: acquiring a geometric characteristic image and a temperature characteristic image of a molten pool in an additive manufacturing process by using an image acquisition module, dividing the geometric characteristic image into three characteristic distribution areas of the molten pool, a plume and a splash, extracting geometric characteristic parameters of the molten pool, the plume and the splash, dividing the temperature characteristic image into three temperature characteristic distribution areas of the molten pool, the plume and the splash corresponding to the geometric characteristic image, and acquiring the temperature characteristic parameters;
s2: the geometric characteristic parameters and the temperature characteristic parameters are used as the input of a double-current convolution neural network, the geometric characteristic parameters and the temperature characteristic parameters are combined to learn the hidden corresponding relation among the geometric characteristic parameters, the temperature characteristic parameters and the forming quality of the workpiece, the characteristics of a molten pool, a plume and splashes in different states are corresponding to the forming quality of the corresponding workpiece, and the type of the forming quality is classified and monitored;
s3: and inputting the geometric and temperature characteristic parameters of the molten pool, the plume and the splash corresponding to the quality type classification result into a trained forming quality prediction model, and predicting the forming quality grade of the nth second by the model by combining different quality classification results.
The step S3 is that the equipment parameter correction unit compares the predicted forming quality grade with the standard forming quality grade, and if the difference value is within the alarm value, the forming quality is judged to be qualified; if the difference value is between the alarm value and the stop value, judging that the forming quality has a problem and needing to modify process parameters; if the difference value exceeds the shutdown value, judging that the molding quality cannot be repaired by modifying the process parameters, giving an alarm and stopping machining.
And S3, when the forming quality is judged to be in problem and the process parameters need to be modified, adjusting the process parameters by the equipment parameter correction module according to the difference value between the predicted forming quality grade and the preset forming quality grade.
The three temperature characteristic distribution areas of the molten pool, the plume and the splash are used for positioning the central position of the molten pool by using a target tracking method of Kalman filtering, and the geometric characteristic image is segmented based on the center of the molten pool.
The double-current convolutional neural network training method comprises the following steps:
extracting the geometric and temperature characteristic images under different extracted experimental conditions by adopting an opencv image processing algorithm, extracting the interested regions of the geometric and temperature characteristic images, performing image filtering and noise reduction operation on the interested region images to remove interference factors, and then performing normalization processing; the quality grade is calibrated by adopting a single-hot coding mode, three types of labels with qualified quality, problematic quality and irreparable quality are added to the forming quality grade pictures corresponding to different geometric and temperature characteristics, and the data acquired in the previous three experiments are used as a training set after being processed.
The method for establishing the double-current convolutional neural network comprises the following steps:
the double-flow convolutional neural network comprises two parallel convolutional neural networks, wherein the first flow convolutional neural network comprises a convolutional layer, a pooling layer and a full-connection layer, the convolutional layer adopts a 5 x 5 convolutional kernel, the convolutional layer is used for extracting geometric features hidden in a molten pool image, the second flow convolutional neural network is used for extracting temperature characteristic parameters, the first flow convolutional neural network and the second flow convolutional neural network are combined and connected at the full-connection layer, the full-connection layer is used for classifying the forming quality by combining the geometric characteristic parameters and the temperature characteristic parameters, and the used classification function is a softmax function.
The method also comprises the steps of constructing y layers of long and short term memory neural networks as a backbone network in the stage of a finished piece forming quality prediction model, wherein each layer of long and short term memory neural network comprises an input gate, a forgetting gate, a state updating gate and an output gate; in order to prevent overfitting, random inactivation operation is added after the long-term and short-term memory neural network; and finally outputting and adding a Dense layer to aggregate the dimensionality into 1, namely outputting the forming quality grade.
The embodiment of the invention has the following beneficial effects:
(1) compared with the traditional additive manufacturing forming quality monitoring system, the system extracts the geometric and temperature characteristic parameters of a molten pool, a plume and a splash, and more comprehensively reflects the dynamic monitoring process of the additive manufacturing technology and the forming quality of a workpiece;
(2) the Two-stream CNN network is used for classifying and monitoring the forming quality categories, so that the repeatability of the additive manufacturing process is greatly improved. The long-short term memory neural network is combined with different forming quality results to predict the forming quality of the finished piece n seconds later, so that the forming quality of the finished piece is predicted in advance and technological parameters are corrected in time, the processing time is reduced, and the production quality is improved; when the predicted forming quality parameter of the workpiece exceeds the alarm value, adjusting the process parameter in time and continuing the subsequent processing process; when the predicted forming quality parameter of the workpiece exceeds the shutdown value, an alarm is sent out in time, and the workpiece is shut down, so that the quality of the material increase manufacturing workpiece is improved, and the material loss is reduced.
(3) The method combines the characteristics of the forming quality monitoring and the prediction of the workpiece, applies deep learning to the classification and the quality prediction of the forming quality category of the workpiece, and expands the application field of the additive manufacturing technology.
Drawings
Fig. 1 is a block diagram of a method for monitoring and predicting the forming quality of an additive manufacturing part based on deep learning according to the present invention;
FIG. 2 is a schematic view of an additive manufacturing process monitoring platform according to the present invention;
FIG. 3 is a flowchart of a method for monitoring and predicting additive manufacturing part forming quality based on deep learning according to the present invention
FIG. 4 is a flow chart of neural network classification model and part forming quality prediction model training.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention will be described in further detail with reference to the accompanying drawings.
As shown in fig. 1 and fig. 2, the carrier structure of the present invention includes a laser emitter 1, an optical fiber interface 2, a high-speed camera and Infrared (Infrared) camera combination module 3, a laser cladding head 4, a powder feeding channel 5, a molten pool 6, a powder feeder 7, a formed part 8, and a substrate 9, wherein the high-speed camera and Infrared camera combination module 3 captures a real-time image of a dynamic feature in a region of the molten pool 6 during an additive manufacturing process. Geometric and temperature characteristic images of the molten pool, plume and spatter are acquired.
Referring to fig. 3, in an additive manufacturing part forming quality monitoring and predicting method based on deep learning according to an embodiment of the present invention,
the method comprises the following steps:
t1, the additive manufacturing process monitoring platform carries out additive manufacturing process monitoring according to the current process parameters.
T2, the forming quality monitoring unit, acquires images of the area containing the molten pool, plume and spatter using the high speed camera in the image acquisition module. Temperature profiles of the melt pool, plume and spatter during the additive manufacturing process are collected using an Infrared (IR) camera. Positioning the central position of a molten pool by a geometric characteristic image shot by a high-speed camera through an image characteristic extraction and processing module and by using a target tracking method based on Kalman filtering; based on the center of the molten pool, dividing the shot image into three areas of the molten pool, plume and splash through an image feature extraction and processing module; and extracting characteristic parameters of the molten pool, the plume and the splash in the three image areas through an image characteristic extraction and processing module, wherein the geometric characteristic parameters comprise the area of the molten pool, the area of the plume and the area of the splash. And the temperature characteristic image acquired by the infrared camera is divided into three temperature characteristic distribution areas of a molten pool, a plume and a splash corresponding to the geometric characteristics through an image characteristic extraction and processing module based on a temperature image processing algorithm, and the temperature characteristic parameters comprise the molten pool temperature, the plume temperature and the splash temperature.
T3, forming quality classification and monitoring: inputting the geometric characteristics and temperature characteristic parameters of the molten pool, the plume and the splash extracted by the image characteristic extraction and processing module into a Two-stream CNN network, respectively extracting a plurality of characteristics and temperature characteristic parameters of the molten pool, the plume and the splash by a double-stream convolution neural network, learning the hidden corresponding relation between the geometric characteristics and the temperature characteristics of the molten pool, the plume and the splash and the forming quality of a workpiece by combining Two dimensions of the geometric characteristics and the temperature characteristics, corresponding the characteristics of the molten pool, the plume and the splash in different states to the corresponding forming quality of the workpiece, and classifying and monitoring the forming quality types
T4, forming quality prediction: inputting the geometric and temperature characteristic parameters of the molten pool, the plume and the splash corresponding to different forming quality classification results obtained in the step T3 into a trained forming quality prediction model, and predicting the forming quality grade of the nth second by the model according to the different quality classification results; the equipment parameter correction unit compares the predicted forming quality grade with the standard forming quality grade, and if the difference value is within the alarm value, the forming quality is judged to be qualified; if the difference value is between the alarm value and the stop value, judging that the forming quality has a problem and needing to modify process parameters; if the difference value exceeds the shutdown value, judging that the forming quality cannot be repaired, giving an alarm and stopping machining.
T5, equipment parameter modification or equipment alarm shutdown: according to the prediction result of the step T4, when the forming quality is judged to have problems and the process parameters need to be modified, the equipment parameter correction module adjusts the process parameters according to the difference value between the predicted forming quality grade and the preset forming quality grade so as to ensure that the forming quality grade n seconds later is within the preset quality grade range; and when the forming quality is judged to be incapable of being repaired, an alarm is given out to stop processing so as to avoid producing unqualified workpieces and wasting materials.
The training process of the neural network classification model and the part forming quality prediction model is shown in fig. 4, and specifically comprises the following steps:
(1.1) data acquisition: in order to obtain a data set corresponding to a training neural network, a plurality of times of additive manufacturing experiments are carried out, so that model training is convenient, the time of each experiment is the same, and experiment parameters are changed for a plurality of times in the process of each experiment to construct different forming quality grades; according to the method provided in the steps T2 and T3, the change conditions of the geometric characteristic parameters and the temperature characteristic parameters with time under different experimental conditions and the time points corresponding to different product forming quality parameters are recorded to form a product forming quality original data set.
(1.2) data preprocessing: in order to construct the input of the Two-stream CNN and LSTM networks, extracting the interested regions of geometric features and temperature feature images under different extracted experimental conditions by adopting an opencv image processing algorithm, carrying out image filtering and noise reduction on the interested region images to remove interference factors, then carrying out normalization processing, and compressing the numerical values of the geometric and temperature feature images to be between 0 and 1 through the normalization processing; the quality grade is calibrated by adopting a One-Hot (One-Hot) coding mode, the One-Hot coding is also called One-bit effective coding, N states are mainly coded by adopting an N-bit state register, each state has an independent register bit, only One bit is effective at any time, three types of labels with qualified quality, problems in quality and incapability of repairing quality are added to the forming quality grade pictures corresponding to different geometric and temperature characteristics by utilizing the characteristic of the One-Hot coding, each type of state is respectively represented by the One-bit effective state register of the One-Hot coding, and the 3-bit state register is used for calibrating the three types of labels in the forming quality grade calibration; the data collected in the previous three experiments are used as a training set after being processed, wherein the training set consists of 8000 images, and the geometric characteristic image and the temperature characteristic image are 4000 images respectively; the main purpose is to train and fit the model, and the rest experimental data is used as a test set after being processed as above, and the test set data is composed of 2000 images, wherein the geometric characteristic image and the temperature characteristic image are 1000 images respectively. The main objective is to evaluate the model capabilities.
(1.3) establishing and training a network model: firstly, building x layers of double-current convolutional neural network, wherein the double-current convolutional neural network is composed of two parallel convolutional neural networks, the first layer (branch) convolutional neural network comprises a convolutional layer, a pooling layer and a full-connection layer, the convolutional layer adopts a 5 multiplied by 5 convolutional kernel, the convolutional layer is mainly used for extracting geometric features hidden in a molten pool image, and the pooling layer is used for extracting geometric features hidden in a molten pool imageThe maximum pooling mode is adopted, the size of the pooling specification is 2 multiplied by 2, and the method is mainly used for reducing the dimensionality of image data. Therefore, the first flow convolution neural network is used for extracting the geometric characteristic parameter attribute of the molten pool. The structure of the second flow (branch) convolutional neural network is consistent with the structure of the first flow (branch) convolutional neural network in composition, only different from the characteristic identification module, and the convolution kernel of the second flow is designed according to the characteristic attribute of the temperature of the molten pool. Therefore, the second stream convolutional neural network is used for extracting the temperature characteristic parameters, the first stream and the second stream are merged and connected in a full connection layer, the full connection layer combines the geometric characteristic parameters and the temperature characteristic parameters to classify the forming quality, and the classification function is a softmax function. In probability theory, the softmax function formula:
and (3) a value range (0, 1), wherein e is a natural logarithm, t is an input quantity, and sigma (t) is an output probability value, after the full-connection layer convergence is carried out, a softmax classification function is used in the classifier classification process to classify the forming quality grades. (the specific implementation way of the network training is that experimental data in the data preprocessing step is firstly obtained, the experimental data is input into a double-current convolutional neural network after being preprocessed, a forward propagation training network is carried out, a loss function is calculated, network weight parameters are updated by using a back propagation algorithm, a network model is updated until the classification precision of the model on the forming quality reaches a preset target.) then, in the stage of a workpiece forming quality prediction model, y layers of long and short-term memory neural networks are built to serve as a backbone network, and each layer of long and short-term memory neural network comprises an input gate, a forgetting gate, a state updating gate and an output gate; to prevent overfitting, a Dropout (random inactivation) operation was added after the long-short term memory neural network; finally, a Dense layer is added to output, the dimension is aggregated to be 1, namely, the forming quality grade is output, and the loss function used for training the model is an MAE (mean Absolute error) MAE function formula:
where m is the total number of data sets of a sample,
is a sample prediction value, yiThe true value of the sample label. Using an MAE loss function in the process of carrying out forward propagation training model, wherein an Adam optimization algorithm is adopted as an optimization algorithm; the specific implementation mode of the training network comprises the steps of preprocessing the obtained geometrical and temperature characteristic parameters of a molten pool, a plume and a splash corresponding to different forming quality classification results, inputting the preprocessed parameters into a long-short term memory neural network model, flexibly storing and extracting the input data by adopting a threshold structure, establishing the relation between the future moment and the existing historical data, carrying out forward propagation training network, calculating a loss function, updating network parameters by adopting a backward propagation algorithm, optimizing the network by using an Adma algorithm, and directly reaching the preset target value of the quality prediction accuracy of the network model.
(1.4) saving the model: after the training of the double-current convolution neural network and the workpiece forming quality prediction model is finished, the model weight is stored, and the model weight can be used for workpiece forming quality classification and prediction.
While the invention has been described in connection with what is presently considered to be the most practical and preferred embodiment, it is to be understood that the invention is not to be limited to the disclosed embodiment, but on the contrary, is intended to cover various modifications and equivalent arrangements included within the spirit and scope of the appended claims.
Claims (7)
1. A method for monitoring and predicting the forming quality of an additive manufacturing part based on deep learning is characterized by comprising the following steps:
s1: acquiring a geometric characteristic image and a temperature characteristic image of a molten pool in an additive manufacturing process by using an image acquisition module, dividing the geometric characteristic image into three characteristic distribution areas of the molten pool, a plume and a splash, extracting geometric characteristic parameters of the molten pool, the plume and the splash, dividing the temperature characteristic image into three temperature characteristic distribution areas of the molten pool, the plume and the splash corresponding to the geometric characteristic image, and acquiring the temperature characteristic parameters;
s2: the geometric characteristic parameters and the temperature characteristic parameters are used as the input of a double-current convolution neural network, the geometric characteristic parameters and the temperature characteristic parameters are combined to learn the hidden corresponding relation among the geometric characteristic parameters, the temperature characteristic parameters and the forming quality of the workpiece, the characteristics of a molten pool, a plume and splashes in different states are corresponding to the forming quality of the corresponding workpiece, and the type of the forming quality is classified and monitored;
s3: and inputting the geometric and temperature characteristic parameters of the molten pool, the plume and the splash corresponding to the quality type classification result into a trained forming quality prediction model, and predicting the forming quality grade of the nth second by the model by combining different quality classification results.
2. The deep learning based additive manufacturing part forming quality monitoring and predicting method according to claim 1, wherein the S3 further comprises an equipment parameter modification unit comparing the predicted forming quality grade with a standard forming quality grade, and if the difference is within an alarm value, determining that the forming quality is qualified; if the difference value is between the alarm value and the stop value, judging that the forming quality has a problem and needing to modify process parameters; if the difference value exceeds the shutdown value, judging that the molding quality cannot be repaired by modifying the process parameters, giving an alarm and stopping machining.
3. The method for monitoring and predicting the forming quality of the additive manufacturing part based on the deep learning of claim 2, wherein the step S3 further comprises adjusting the process parameters according to the difference between the predicted forming quality level and the preset forming quality level when the process parameters need to be modified when it is determined that the forming quality is in problem.
4. The deep learning based additive manufacturing part forming quality monitoring and predicting method according to claim 1, wherein the three temperature characteristic distribution areas of the molten pool, the plume and the splash are obtained by locating a central position of the molten pool by using a target tracking method of Kalman filtering and segmenting the geometric characteristic image based on the center of the molten pool.
5. The deep learning based additive manufacturing part forming quality monitoring and predicting method according to claim 1, wherein the dual-flow convolutional neural network training method comprises:
extracting the geometric and temperature characteristic images under different extracted experimental conditions by adopting an opencv image processing algorithm, extracting the interested regions of the geometric and temperature characteristic images, performing image filtering and noise reduction operation on the interested region images to remove interference factors, and then performing normalization processing; the quality grade is calibrated by adopting a single-hot coding mode, three types of labels with qualified quality, problematic quality and irreparable quality are added to the forming quality grade pictures corresponding to different geometric and temperature characteristics, and the data acquired in the previous three experiments are used as a training set after being processed.
6. The deep learning based additive manufacturing part forming quality monitoring and predicting method according to claim 1, wherein the method for establishing the dual-flow convolutional neural network comprises the following steps:
the double-flow convolutional neural network comprises two parallel convolutional neural networks, wherein the first flow convolutional neural network comprises a convolutional layer, a pooling layer and a full-connection layer, the convolutional layer adopts a 5 x 5 convolutional kernel, the convolutional layer is used for extracting geometric features hidden in a molten pool image, the second flow convolutional neural network is used for extracting temperature characteristic parameters, the first flow convolutional neural network and the second flow convolutional neural network are combined and connected at the full-connection layer, the full-connection layer is used for classifying the forming quality by combining the geometric characteristic parameters and the temperature characteristic parameters, and the used classification function is a softmax function.
7. The deep learning-based additive manufacturing part forming quality monitoring and predicting method according to claim 6, further comprising the steps of constructing y layers of long-short term memory neural networks as a backbone network in a part forming quality prediction model stage, wherein each layer of long-short term memory neural network comprises an input gate, a forgetting gate, a state updating gate and an output gate; in order to prevent overfitting, random inactivation operation is added after the long-term and short-term memory neural network; and finally outputting and adding a Dense layer to aggregate the dimensionality into 1, namely outputting the forming quality grade.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110104828.1A CN112801091B (en) | 2021-01-26 | 2021-01-26 | Deep learning-based method for monitoring and predicting forming quality of additive manufacturing part |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110104828.1A CN112801091B (en) | 2021-01-26 | 2021-01-26 | Deep learning-based method for monitoring and predicting forming quality of additive manufacturing part |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112801091A true CN112801091A (en) | 2021-05-14 |
| CN112801091B CN112801091B (en) | 2023-06-13 |
Family
ID=75811943
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110104828.1A Active CN112801091B (en) | 2021-01-26 | 2021-01-26 | Deep learning-based method for monitoring and predicting forming quality of additive manufacturing part |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112801091B (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113762366A (en) * | 2021-08-24 | 2021-12-07 | 华中科技大学 | Additive manufacturing forming state prediction control method and system |
| CN114619119A (en) * | 2022-03-29 | 2022-06-14 | 北京理工大学 | Electric arc vibration material disk makes on-line monitoring and real-time control system |
| CN114707181A (en) * | 2022-04-08 | 2022-07-05 | 北京国信网联科技有限公司 | Machine learning-based data security exchange system and method |
| CN116758527A (en) * | 2023-08-16 | 2023-09-15 | 山东巨野盛鑫电器材料有限公司 | Quality identification system of polyimide insulating film with low expansion coefficient |
| CN117340280A (en) * | 2023-12-05 | 2024-01-05 | 成都斐正能达科技有限责任公司 | LPBF additive manufacturing process monitoring method |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20130178952A1 (en) * | 2010-06-28 | 2013-07-11 | Precitec Itm Gmbh | Method for closed-loop controlling a laser processing operation and laser material processing head using the same |
| US20140277735A1 (en) * | 2013-03-15 | 2014-09-18 | JIBO, Inc. | Apparatus and methods for providing a persistent companion device |
| US20170151720A1 (en) * | 2015-11-27 | 2017-06-01 | Metal Industries Research & Development Centre | Additive manufacturing method |
| CN107807568A (en) * | 2017-10-27 | 2018-03-16 | 中国电子产品可靠性与环境试验研究所 | Increasing material manufacturing monitoring system, method, apparatus and increasing material manufacturing equipment |
| US20180341248A1 (en) * | 2017-05-24 | 2018-11-29 | Relativity Space, Inc. | Real-time adaptive control of additive manufacturing processes using machine learning |
| CN110472698A (en) * | 2019-08-22 | 2019-11-19 | 四川大学 | Increase material based on the metal of depth and transfer learning and shapes fusion penetration real-time predicting method |
| CN110490866A (en) * | 2019-08-22 | 2019-11-22 | 四川大学 | Metal based on depth characteristic fusion increases material forming dimension real-time predicting method |
| US20200030880A1 (en) * | 2018-07-25 | 2020-01-30 | Jtekt Corporation | Additive manufacturing, learning model generation apparatus, manufacturing condition determination apparatus for shaped article to be produced by additive manufacturing and status estimation apparatus for shaped article to be produced by additive manufacturing |
| CN110789128A (en) * | 2019-10-24 | 2020-02-14 | 汕头大学 | Additive manufacturing part forming quality prediction and control system and method |
| CN111151751A (en) * | 2020-01-02 | 2020-05-15 | 汕头大学 | Three-laser-beam intelligent material-increasing and material-decreasing composite manufacturing system and method |
-
2021
- 2021-01-26 CN CN202110104828.1A patent/CN112801091B/en active Active
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20130178952A1 (en) * | 2010-06-28 | 2013-07-11 | Precitec Itm Gmbh | Method for closed-loop controlling a laser processing operation and laser material processing head using the same |
| US20140277735A1 (en) * | 2013-03-15 | 2014-09-18 | JIBO, Inc. | Apparatus and methods for providing a persistent companion device |
| US20170151720A1 (en) * | 2015-11-27 | 2017-06-01 | Metal Industries Research & Development Centre | Additive manufacturing method |
| US20180341248A1 (en) * | 2017-05-24 | 2018-11-29 | Relativity Space, Inc. | Real-time adaptive control of additive manufacturing processes using machine learning |
| CN107807568A (en) * | 2017-10-27 | 2018-03-16 | 中国电子产品可靠性与环境试验研究所 | Increasing material manufacturing monitoring system, method, apparatus and increasing material manufacturing equipment |
| US20200030880A1 (en) * | 2018-07-25 | 2020-01-30 | Jtekt Corporation | Additive manufacturing, learning model generation apparatus, manufacturing condition determination apparatus for shaped article to be produced by additive manufacturing and status estimation apparatus for shaped article to be produced by additive manufacturing |
| CN110472698A (en) * | 2019-08-22 | 2019-11-19 | 四川大学 | Increase material based on the metal of depth and transfer learning and shapes fusion penetration real-time predicting method |
| CN110490866A (en) * | 2019-08-22 | 2019-11-22 | 四川大学 | Metal based on depth characteristic fusion increases material forming dimension real-time predicting method |
| CN110789128A (en) * | 2019-10-24 | 2020-02-14 | 汕头大学 | Additive manufacturing part forming quality prediction and control system and method |
| CN111151751A (en) * | 2020-01-02 | 2020-05-15 | 汕头大学 | Three-laser-beam intelligent material-increasing and material-decreasing composite manufacturing system and method |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113762366A (en) * | 2021-08-24 | 2021-12-07 | 华中科技大学 | Additive manufacturing forming state prediction control method and system |
| CN113762366B (en) * | 2021-08-24 | 2024-02-06 | 华中科技大学 | Method and system for predictive control of forming state of additive manufacturing |
| CN114619119A (en) * | 2022-03-29 | 2022-06-14 | 北京理工大学 | Electric arc vibration material disk makes on-line monitoring and real-time control system |
| CN114619119B (en) * | 2022-03-29 | 2023-01-13 | 北京理工大学 | Electric arc vibration material disk makes on-line monitoring and real-time control system |
| CN114707181A (en) * | 2022-04-08 | 2022-07-05 | 北京国信网联科技有限公司 | Machine learning-based data security exchange system and method |
| CN116758527A (en) * | 2023-08-16 | 2023-09-15 | 山东巨野盛鑫电器材料有限公司 | Quality identification system of polyimide insulating film with low expansion coefficient |
| CN116758527B (en) * | 2023-08-16 | 2023-11-10 | 山东巨野盛鑫电器材料有限公司 | Quality identification system of polyimide insulating film with low expansion coefficient |
| CN117340280A (en) * | 2023-12-05 | 2024-01-05 | 成都斐正能达科技有限责任公司 | LPBF additive manufacturing process monitoring method |
| CN117340280B (en) * | 2023-12-05 | 2024-02-13 | 成都斐正能达科技有限责任公司 | LPBF additive manufacturing process monitoring method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112801091B (en) | 2023-06-13 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112801091B (en) | Deep learning-based method for monitoring and predicting forming quality of additive manufacturing part | |
| CN110789128B (en) | Additive manufacturing part forming quality prediction and control system and method | |
| CN109740676B (en) | Object detection and migration method based on similar targets | |
| CN112381788B (en) | Part surface defect increment detection method based on double-branch matching network | |
| CN110909507A (en) | Method for suggesting soldering tin process parameters | |
| CN111815555A (en) | Metal additive manufacturing image detection method and device combining anti-neural network with local binary | |
| CN110648310A (en) | Weak supervision casting defect identification method based on attention mechanism | |
| CN110610482A (en) | Workpiece flaw detection method based on resnet | |
| Kao et al. | Laser cladding quality monitoring using coaxial image based on machine learning | |
| CN115936060B (en) | Substation capacitance temperature early warning method based on depth deterministic strategy gradient | |
| CN112258459A (en) | Lithium battery defect detection method based on integrated deep learning | |
| CN110599459A (en) | Underground pipe network risk assessment cloud system based on deep learning | |
| Wu et al. | Hot-rolled steel strip surface inspection based on transfer learning model | |
| CN117152119A (en) | Profile flaw visual detection method based on image processing | |
| Kumar et al. | Tool wear classification based on machined surface images using convolution neural networks | |
| CN116572535A (en) | Production optimization method and system for high polymer material product | |
| CN112507929A (en) | Vehicle body spot welding slag accurate detection method based on improved YOLOv3 network | |
| Cao et al. | Wafer surface defect detection based on improved YOLOv3 network | |
| Liu et al. | A deep learning approach to defect detection in additive manufacturing of titanium alloys | |
| CN112365014A (en) | GA-BP-CBR-based industrial equipment fault diagnosis system and method | |
| CN112967266B (en) | Laser directional energy deposition area calculation method of full convolution neural network | |
| CN115169617B (en) | Mold maintenance prediction model training method, mold maintenance prediction method and system | |
| CN116525295B (en) | Metallized film for high-frequency pulse capacitor and preparation method thereof | |
| CN114372181B (en) | Equipment production intelligent planning method based on multi-mode data | |
| CN116843615B (en) | Lead frame intelligent total inspection method based on flexible light path |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |