A method for assembly process reuse based on deep learning

Abstract

The application discloses a kind of based on deep learning's assembly process reuse method, steps are as follows: according to enterprise investigation, establish historical assembly process instance database;Through the secondary development of SolidWorks, obtain initial data set, and the initial data set is expanded, divided, and with MobileNetV3 model as base network model, fusion attention mechanism, the assembly object to be identified is sampled in multiple perspectives and similarity retrieval is carried out;The highest similarity historical assembly object is obtained, and the assembly process of the historical assembly object is the assembly process for reuse;Through artificial interactive modification, make the reused process more suitable for the assembly object to be identified, and store the modified process in historical assembly process instance database, realize the expansion of historical assembly process instance database.The application realizes the quick reuse of assembly process by deep learning, not only reduces the labor cost of enterprise, but also improves assembly process generation efficiency and automation level.

Description

TECHNICAL FIELD

[0001] The present application relates to an assembly process, in particular a method for assembly process reuse based on deep learning. BACKGROUND

[0002] Assembly process is a process of combining a group of scattered parts into a product through reasonable process flow and various necessary ways according to certain precision standards and technical requirements. Assembly process is the last link in the whole product manufacturing process, and the efficiency and quality of assembly work have a great influence on the manufacturing cost of products and the final quality of products. In the product production process, about 1 / 3 of the human resources are engaged in activities related to assembly, and more than 40% of the production cost in industrial countries is used for product assembly. In particular, with the restructuring and optimization of enterprises, assembly-type enterprises have emerged in large numbers, and this enterprise organization mode will dominate the future development of manufacturing industry. Traditional assembly process documents are mostly based on two-dimensional process cards, and with the accumulation of time, most process documents are in disarray. For the assembly object to be identified, it is difficult to quickly find reusable assembly process, and only the assembly process can be redesigned. And the rapid reuse of assembly process in some enterprises mostly relies on the experience of process personnel, which not only increases labor cost, but also is time-consuming and laborious, low in efficiency, not conducive to inventory management, and seriously affects the design efficiency of assembly process and the development of enterprises.

[0003] Under the background of Industry 4.0, more and more enterprises have realized intelligentization, and the intelligentization of assembly process is one of the key links of enterprises. Convolutional neural network is a popular recognition method in recent years, which avoids the complex steps of manual feature extraction and can learn relevant features by itself, thereby improving the accuracy and efficiency of image classification. The similarity of shape features of assembly objects largely determines the similarity of assembly process. Based on the similarity of shape features, the rapid reuse of assembly process gradually develops towards digitization, informatization and high-speed intelligentization. In the field of mechanical manufacturing, to realize the rapid design of product assembly process, a new technology is needed to identify the assembly object and complete the automatic assembly process reuse. Deep neural network is the key direction of research on assembly object type identification and classification technology, which has the advantages of high recognition accuracy, high efficiency and no contact with mechanical assembly objects, so it is the key technology of current manufacturing industry. Therefore, based on deep learning to solve the reuse of assembly process is a technical problem that needs to be solved urgently.

[0004] The closest prior art to the present application is the patent "Assembly process knowledge mining method and rapid construction method of assembly case ontology", which automatically extracts application concepts and instances and their relationships from assembly process text through CRF to realize assembly process knowledge mining. In the mining process, the crawler technology is used to enrich the assembly field knowledge corpus, improve the knowledge mining accuracy, ensure the accuracy of the generated assembly process, and improve the efficiency of the generated assembly process. And through the assembly process knowledge mining method, the assembly case database is generated, and the conversion rule is established to realize the conversion of the assembly case database to the assembly case ontology. It solves the technical defects of low efficiency and poor consistency of manual construction of assembly case ontology, and improves the efficiency of assembly process generation. However, this method only mines existing assembly process knowledge, and when a new assembly process is generated, the assembly process knowledge needs to be mined again to achieve a truly intelligent effect.

[0005] Another prior art close to the present application is the patent "Assembly process procedure generation system and method based on three-dimensional CAD model", which fully utilizes the existing various information in the three-dimensional CAD assembly model to push similar assembly processes, realizes the pushing of assembly process knowledge, and thus can significantly reduce the manual operation in the assembly process design process and improve the efficiency and quality of process design. However, this method only compares each extracted assembly part with the assembly part data under each assembly process procedure in the assembly process knowledge base to determine the assembly parts in each assembly process procedure in the assembly part knowledge base that have the same name as the model to be processed and the assembly quantity of each same name assembly part. And in the implementation process, standard parts in the product need to be filtered out, so the applicable range is relatively narrow and the retrieval process is relatively cumbersome.

[0006] The existing technology has the following problems:

[0007] (1) Most of the traditional assembly process files are mainly two-dimensional process cards, which lead to chaos in most process files over time. It is difficult to quickly find the reusable assembly process of the assembly object to be identified, and the assembly process needs to be redesigned, which seriously affects the design efficiency of the assembly process and the development of the enterprise.

[0008] (2) At present, most enterprises use manual classification to divide the similarity of assembly objects, and the similarity of shape features of assembly objects determines the similarity of assembly process to a large extent. Since the classification standard is determined by the experience of process engineers, it cannot achieve standardization. And with the increase of assembly objects, there may be misclassification, which is time-consuming and laborious, and seriously affects the development and automation level of the enterprise.

[0009] (3) Throughout the country and abroad, it is difficult to find a general and fast method for the rapid reuse of assembly process, so there is still a lack of a systematic and effective generation method for the rapid design of assembly process in enterprises. SUMMARY

[0010] The purpose of the application is to provide an assembly process reuse method based on deep learning, so as to realize the rapid reuse of assembly process through deep learning, reduce the labor cost of enterprises, and improve the generation efficiency and automation level of assembly process.

[0011] Technical scheme: The assembly process reuse method based on deep learning comprises the following steps:

[0012] (1) By investigating the products produced by an enterprise, the historical assembly process instances of the products are established and stored in a SQL Sever database to establish a historical assembly process instance database;

[0013] The historical assembly process instance database of step (1) is used to store various assembly process procedure information, which includes historical assembly process basic information and process information, the historical assembly process basic information includes historical assembly process number, historical assembly process name, historical assembly object number, historical assembly object name, historical assembly resource information, historical assembly method information, etc., and the process information includes process number, process name, process content, etc.

[0014] (2) Based on the secondary development of SolidWorks, API functions are called to obtain initial data sets of various historical assembly objects;

[0015] Based on VB.NET programming language, SolidWorks is secondarily developed, API functions in SolidWorks are used to rotate historical assembly objects 360°, and a two-dimensional screenshot is captured every 1° to obtain two-dimensional view data sets of various historical assembly objects; OpenDoc function in SolidWorks is used to open SolidWorks files, CloseDoc function is used to close SolidWorks files, SaveAs3 function is used to save files as specified files, ViewZoomtofit2 function is used to scale the current view, ShowNamedView2 function is used to display the specified view, and RotateAboutCenter function is used to rotate model view around the X-axis and Y-axis of the screen, mainly including XAngle function for rotating around the X-axis and YAngle function for rotating around the Y-axis.

[0016] (3) The initial data sets are expanded and divided;

[0017] Based on the convolutional neural network to identify images, the role of the data set is to train the network model, so that the model extracts the relevant features of the picture, so that the model has better learning ability. Of course, the more diverse the data is, the stronger the generalization ability of the model is, and the higher the accuracy of identifying pictures is. In order to improve the learning ability of the model, various image processing methods are needed to expand the original data set, so that the data volume and the size of the model are matched, so that the model can extract more detailed feature information, and the accuracy is also improved. When training the model, it is necessary to ensure that these expanded pictures do not change the important features of the original pictures, so as to avoid certain disturbances, otherwise it will be counterproductive. The randomly generated pictures belong to a certain class of original pictures, but they are different, so when training the model, the diversity of the data can be ensured, and the computer will not think that these are completely the same pictures, which improves the generalization ability of the model as a whole.

[0018] (3.1) Expand the initial data set, and the expansion method includes mirror change, random cropping, Gaussian noise, whitening processing, and contrast change;

[0019] (3.2) Divide the initial data set, divide the data set into training set, validation set and test set according to a certain proportion, so as to better test.

[0020] (4) Train the data set based on the improved MobolenetV3 network model to obtain the trained model;

[0021] The step (4) is specifically: first, use a 1x1 convolution kernel to increase the dimension of the input features, increase the image feature information, then use a 3x3 convolution kernel to perform depth separable convolution on the features, reduce the parameter operation amount, reduce the running cost, and increase the running speed. Enter the SE attention mechanism module, after global pooling and two fully connected layers, automatically allocate weights to the feature map, increase the weight of valuable feature channels, and suppress useless feature channels, and finally perform dimension reduction processing through a 1x1 convolution kernel.

[0022] (5) Multi-view sampling of the to-be-identified assembly object to obtain a two-dimensional view set of the to-be-identified object;

[0023] Open the three-dimensional model of the to-be-identified assembly object in SolidWorks, perform multi-view sampling, save the sampled two-dimensional views to a specified folder for extraction.

[0024] Similarity retrieval is performed in the model, i=1,2,3,……,n.

[0025] (6) Insert the obtained two-dimensional view of the to-be-identified object into the trained network model for similarity retrieval;

[0026] (7) According to the similarity retrieval, the assembly process to be reused is obtained and stored in the temporary database;

[0027] The similarity retrieval is performed, and if the similarity rate of the to-be-identified assembly object i and the historical assembly object is less than 70%, the similarity of the two cannot reach the degree of reuse of the assembly process, and thus the reuse of the assembly process cannot be performed; otherwise, the reuse of the assembly process is performed, and the assembly process name information corresponding to the reused process is saved to the temporary database for convenient query and use.

[0028] (8) After the two-dimensional view of all to-be-identified assembly objects is cycled, the similarity rates of various to-be-reused assembly processes are compared, and the assembly process that can be reused by the to-be-identified assembly object is output according to the name of the historical assembly object;

[0029] (8.1) After the two-dimensional view of all to-be-identified assembly objects is cycled, the similarity rates of various to-be-reused assembly processes are compared, and if the i-th to-be-identified assembly object is n, i = 1, 2, 3, …, n, all the sampled two-dimensional views are identified, the similarity rate data obtained from the database is compared, and after comparison, the assembly process of the historical assembly object of the type with the largest similarity rate stored in the temporary database is output, otherwise, the process returns to step (6) to continue the cycle until the cycle ends;

[0030] (8.2) After all the two-dimensional views of the to-be-identified objects are cycled, if none of them has a similarity rate greater than or equal to 70% with the historical assembly object, the to-be-identified assembly object has no reusable process.

[0031] (9) Through artificial interactive modification, the reused process is more suitable for the to-be-identified assembly object;

[0032] After the to-be-identified object is obtained, it may need to be modified in some aspects, and after interactive modification by the process engineer, the reused process corresponding to the assembly process name can be directly found in the historical assembly process instance database by searching the assembly process name (in the historical assembly process instance database, each historical assembly process corresponds to a unique process name).

[0033] (10) The reused assembly process is given a new assembly process name and stored in the historical assembly process database for next use.

[0034] On the basis of the obtained reused process, the to-be-identified assembly object is given a new assembly process name and stored in the historical assembly process instance database for next use.

[0035] Finally, the classic CNN model, MobileNetV3 network model and improved MobileNetV3 network model are compared.

[0036] The classic CNN model takes 50s for each training, and the accuracy curve is not more than 0.95, and the verification set fluctuates greatly; the MobileNetV3 network model takes about 35s for each training, compared with the classic CNN model, the training time is greatly reduced each time, and the accuracy is greatly improved, but the accuracy curve of the verification set still has a large fluctuation, and the fitting effect is poor; the improved MobileNetV3 network model takes about 39s for each training, because the CBAM attention mechanism is fused, so the training time of each time is slightly higher than that of the MobileNetV3 network model, but the training set accuracy and the verification set accuracy are more stable. Therefore, in comparison, the improved MobileNetV3 network model is more stable, there is no large fluctuation, and the verification accuracy is higher.

[0037] Advantages: Compared with the prior art, the present application has the following advantages:

[0038] 1. Reduce production cost and improve work efficiency: The reuse of assembly process can realize the standardization, standardization and automation of assembly process, avoid the process of repeated design and formulation, reduce the time and cost of process design, and improve the work efficiency;

[0039] 2. Improve product quality: Assembly process reuse can avoid human error and improve product stability and reliability, thereby improving product quality;

[0040] 3. Improve production flexibility: Assembly process reuse can automatically call the corresponding process based on the shape and size of the product, realize the rapid conversion of the production line and the flexible adjustment of the production organization, and better meet the market demand. BRIEF DESCRIPTION OF DRAWINGS

[0041] Figure 1 The step flow chart of the method described in the application;

[0042] Figure 2 The principle diagram of the method described in the application;

[0043] Figure 3 The MobileNetV3 basic unit;

[0044] Figure 4 The CBMA structure diagram;

[0045] Figure 5 The CNN accuracy curve;

[0046] Figure 6 MobileNetV3 accuracy curve;

[0047] Figure 7 Improved MobileNetV3 accuracy curve. DETAILED DESCRIPTION

[0048] The technical solutions of the present application will be further described below with reference to the accompanying drawings.

[0049] As shown in Figure 1 and 2 A deep learning-based assembly process reuse method, comprising the following steps:

[0050] (1) By investigating the products produced by a certain enterprise, a historical assembly process instance of the product is established and stored in a SQL Sever database, and a historical assembly instance database is established;

[0051] The historical assembly process instance database of step (1) is used to store various assembly process procedure information, and the information includes historical assembly process basic information and process information, the historical assembly process basic information includes historical assembly process number, historical assembly process name, historical assembly object number, historical assembly object name, historical assembly resource information, and historical assembly method information, and the process information includes process number, process name, and process content.

[0052] (2) Based on the secondary development of SolidWorks, API functions are called to obtain initial data sets of various historical assembly objects;

[0053] Based on VB.NET programming language, SolidWorks is secondarily developed, API functions in SolidWorks are used, historical assembly objects are rotated by 360°, a two-dimensional screenshot is captured every 1°, and two-dimensional view data sets of various historical assembly objects are obtained; OpenDoc functions in SolidWorks are used to open SolidWorks files, CloseDoc functions are used to close SolidWorks files, SaveAs3 functions are used to save files as specified files, ViewZoomtofit2 functions are used to scale the current view, ShowNamedView2 functions are used to display the specified view, and RotateAboutCenter functions are used to rotate the model view around the X-axis and Y-axis of the screen, mainly including using XAngle functions to rotate around the X-axis, and using YAngle functions to rotate around the Y-axis.

[0054] (3) The initial data set is expanded and divided;

[0055] Based on the convolutional neural network, the data set is used to train the network model, so that the model extracts the relevant features of the picture, and the model has better learning ability. Of course, the more diverse the data is, the stronger the generalization ability of the model is, and the higher the accuracy of identifying the picture is. In order to improve the learning ability of the model, various image processing methods are used to expand the original data set, so that the data quantity and the size of the model are matched, so that the model can extract more detailed feature information, and the accuracy is also improved. When training the model, it is necessary to ensure that the expanded pictures do not change the important features of the original pictures, so as to avoid certain disturbance, otherwise it will be counterproductive. The randomly generated pictures belong to a certain class of original pictures, but they are different, so when training the model, the diversity of the data can be ensured, and the computer will not think that these are completely the same pictures, which improves the generalization ability of the model as a whole.

[0056] (3.1) Expand the initial data set, and the expansion method includes mirror change, random cropping, Gaussian noise, whitening processing, and contrast change;

[0057] (3.2) Divide the initial data set, divide the data set into training set, validation set and test set in the ratio of 8:1:1, so as to better test.

[0058] (4) Based on the improved MobolenetV3 network model, the data set is trained to obtain a training model;

[0059] As shown in Figure 3 MobileNetV3 first introduces depth separable convolution to effectively replace the traditional convolution layer. And the depth separable convolution separates the spatial filter from the feature generation mechanism, and effectively decomposes the traditional convolution. The basic unit first uses a 1x1 convolution kernel to upgrade the input feature, increases the image feature information, then uses a 3x3 convolution kernel to perform depth separable convolution on the feature, reduces the number of parameter operations, reduces the running cost, increases the running speed, and then enters the attention mechanism (SE) module. After global pooling and two fully connected layers, the feature maps of various assembled objects are automatically assigned weights, and finally the features are processed by a 1x1 convolution kernel. MobileNetV3 network can be used for classification, detection and segmentation. The network can improve the model accuracy and running speed on the basis of ensuring the lightweight of the model. Compared with the traditional CNN network and MobileNetV1 network and MobileNetV2 network, there is a great improvement, so the MobileNetV3 network is used as the basic network of the compartment car compartment assembly object classification in the present application, and the features of various assembly objects are extracted.

[0060] As shown in Figure 4As shown, CBAM (Convolutional Block Attention Module) is a lightweight module that combines channel attention (Channel Attention Module, CAM) and spatial attention (Spartial Attention Module, SAM).

[0061] During the identification process of the assembly object, many similar features will appear. By adding CBAM to the MobilenetV3 feature extraction network, the main features of the assembly object are enhanced, and the useless features are suppressed, thereby improving the feature recognition accuracy of the main assembly object. Based on the improved MobolenetV3 network model, the data set is trained, and the obtained training model is used for similarity retrieval.

[0062] (5) Multi-view sampling of the assembly object to be identified is performed to obtain a two-dimensional view set of the object to be identified;

[0063] The three-dimensional model of the assembly object to be identified is opened in SolidWorks, multi-view sampling is performed, and the sampled two-dimensional views are saved to a designated folder for extraction.

[0064] Model similarity retrieval, i = 1, 2, 3, …, n.

[0065] (6) The two-dimensional view of the object to be identified is inserted into the trained network model for similarity retrieval;

[0066] (7) According to the similarity retrieval, the assembly process to be reused is obtained and stored in a temporary database;

[0067] The similarity retrieval is performed, and if the similarity rate of the output assembly object i to be identified and the historical assembly object is less than 70%, the similarity of the two cannot reach the degree of assembly process reuse, so the assembly process cannot be reused; otherwise, the assembly process is reused, and the assembly process name information corresponding to the reused process is saved to the temporary database for query.

[0068] (8) After all the two-dimensional views of the assembly objects to be identified are cycled, the similarity rates of various assembly processes to be reused are compared, and the assembly processes that can be reused by the assembly object to be identified are output according to the name of the historical assembly object;

[0069] (8.1) all the two-dimensional views of the to-be-identified assembly objects are cycled, the similarity rates of various to-be-reused assembly processes are compared, if the i-th to-be-identified assembly object is n, i = 1, 2, 3, …, n, then all the sampled two-dimensional views are identified, the similarity rate data obtained in the database are compared, after comparison, the assembly process of the historical assembly object with the largest similarity rate stored in the temporary database is output, otherwise, it is returned to step (6) to continue the cycle until the cycle is completed;

[0070] (8.2) all the two-dimensional views of the to-be-identified objects are cycled, if none of them has a similarity rate greater than or equal to 70% with the historical assembly object, then the to-be-identified assembly object has no reusable process.

[0071] (9) through artificial interactive modification, the reused process is more suitable for the to-be-identified assembly object;

[0072] After obtaining the reusable process of the to-be-identified object, it may need to be modified in some aspects, after interactive modification by the process engineer, the reused process corresponding to the assembly process name can be directly found in the historical assembly process instance database (in the historical assembly process instance database, each historical assembly process corresponds to a unique process name) by searching the assembly process name.

[0073] (10) the reused assembly process is given a new assembly process name and stored in the historical assembly process database for next use.

[0074] On the basis of the obtained reusable process, the to-be-identified assembly object is given a new assembly process name and stored in the historical assembly process instance database for next use.

[0075] Finally, the classical CNN model, the MobileNetV3 network model and the improved MobileNetV3 network model are compared, respectively as shown in Figure 5 、 Figure 6 、 Figure 7

[0076] ​The training time of the classic CNN model is 50 s each time, and the accuracy curve is not more than 0.95, and the validation set fluctuates greatly; the training time of the MobileNetV3 network model is about 35 s each time, which is greatly reduced compared with the classic CNN model, and the accuracy is greatly improved, but the accuracy curve of the validation set still fluctuates greatly, and the fitting effect is relatively poor; the training time of the improved MobileNetV3 network model is about 39 s each time, and since the CBAM attention mechanism is fused, the training time each time is slightly higher than that of the MobileNetV3 network model, but the training set accuracy and the validation set accuracy are more stable. Therefore, in comparison, the improved MobileNetV3 network model is more stable, there is no large fluctuation, and the validation accuracy is higher.

Claims

1. A method for deep learning based assembly process reuse, the method comprising: The method comprises the following steps: (1) establishing a historical assembly process instance of a product by investigating the product produced by an enterprise, and storing the historical assembly process instance in an SQL Sever database to establish a historical assembly process instance database; (2) calling API functions based on secondary development of SolidWorks to obtain initial data sets of various historical assembly objects; (3) expanding and dividing the initial data sets; (4) training the data sets based on an improved MobolenetV3 network model to obtain a training model; (5) sampling a to-be-recognized assembly object from multiple perspectives to obtain a two-dimensional view set of the to-be-recognized object; (6) inserting the obtained two-dimensional view of the to-be-recognized object into the trained network model to perform similarity retrieval; (7) obtaining an assembly process to be reused according to the similarity retrieval, and storing the assembly process to be reused in a temporary database; (8) completing the two-dimensional view of all to-be-recognized assembly objects, comparing the similarity rates of various assembly processes to be reused, and outputting an assembly process that can be reused by the to-be-recognized assembly object according to the name of the historical assembly object; (9) modifying the reused process through artificial interactive modification to make the reused process more suitable for the to-be-recognized assembly object; (10) assigning a new assembly process name to the reused assembly process, and storing the reused assembly process in the historical assembly process database for next time use.

2. The method of claim 1, wherein, The historical assembly process instance database in step (1) is used for storing various assembly process procedure information, and the information comprises historical assembly process basic information and process information, the historical assembly process basic information comprises a historical assembly process number, a historical assembly process name, a historical assembly object number, a historical assembly object name, historical assembly resource information and historical assembly method information, and the process information comprises a process number, a process name and process content.

3. The method of claim 1, wherein, The step (2) is specifically as follows: The SolidWorks is secondarily developed based on a VB.NET programming language, API functions in the SolidWorks are used to rotate a historical assembly object at 360°, a two-dimensional screenshot is captured every 1° to obtain a two-dimensional view data set of each historical assembly object, OpenDoc functions in the SolidWorks are used to open a SolidWorks file, CloseDoc functions are used to close the SolidWorks file, SaveAs3 functions are used to save the file as a specified file, ViewZoomtofit2 functions are used to scale a current view, ShowNamedView2 functions are used to display a specified view, RotateAboutCenter functions are used to rotate a model view around an X-axis and a Y-axis, XAngle functions are used to rotate around the X-axis, and YAngle functions are used to rotate around the Y-axis; and API functions in the SolidWorks are used to obtain data sets of various historical assembly objects.

4. The method of claim 1, wherein, The step (3) is specifically as follows: (3.1) expanding the initial data sets by using expansion methods including mirror change, random cropping, Gaussian noise, whitening processing and contrast change; (3.2) the initial data set is divided, and the data set is divided into a training set, a validation set and a test set according to a certain proportion, so that the test can be better.

5. The method of claim 1, wherein, The step (4) is specifically: first, the input features are dimensioned with a 1*1 convolution kernel, the image feature information is increased, then a 3*3 convolution kernel is used for deep separable convolution of the features, the parameter operation amount is reduced, the running cost is reduced, and the running speed is increased; enter the SE attention mechanism module, after global pooling and two fully connected layers, the feature map is automatically allocated weight, the weight of the valuable feature channel is increased, and the useless feature channel is suppressed, and finally the dimension is reduced by a 1*1 convolution kernel; The data set is trained, and finally the corresponding model is obtained.

6. The method of claim 1, wherein, The step (5) is specifically: opening the three-dimensional model of the to-be-identified assembly object in SolidWorks, performing multi-view sampling, saving the sampled two-dimensional views to a specified folder for extraction.

7. The method of claim 1, wherein, The step (6) is specifically: inserting the obtained two-dimensional view of the i-th to-be-identified assembly object into the trained network model for similarity retrieval, i=1,2,3,...,n.

8. The method of claim 1, wherein, The step (7) is specifically: similarity retrieval is performed, and if the similarity rate of the to-be-identified assembly object i and the historical assembly object is less than 70%, the similarity of the two cannot reach the degree of assembly process reuse, so the assembly process cannot be reused; Otherwise, the assembly process is reused, and the assembly process name information corresponding to the reused process is saved to the temporary database for query.

9. The method of claim 1, wherein, The step (8) is specifically: (8.1) all to-be-identified assembly object two-dimensional views are cycled, the similarity rates of various to-be-reused assembly processes are compared, if the i-th to-be-identified assembly object is n, i=1,2,3,...,n, all the sampled two-dimensional views are identified, the similarity rate data obtained in the database is compared, after comparison, the assembly process of the historical assembly object with the largest similarity rate stored in the temporary database is output, otherwise, return to step (6) to continue the cycle until the cycle is completed; (8.2) all the sampled two-dimensional views of the to-be-identified object are cycled, if none of them has a similarity rate greater than or equal to 70% with the historical assembly object, the to-be-identified assembly object has no reusable process.

Images