Full-automatic motor production line and method thereof

Abstract

The invention provides a full-automatic motor production line and a method thereof, relates to the field of intelligent production, and adopts an image analysis technology based on artificial intelligence to carry out graying processing and morphological structure feature extraction of a welding area image. And intelligently judging whether the welding quality is qualified or not according to the fusion feature of the welding region image between the structural feature and the morphological feature of the welding region. Therefore, tiny welding defects which are difficult to perceive by human eyes can be accurately recognized, so that potential product faults are avoided, meanwhile, the detection speed is greatly improved, a production line can quickly respond and meet the requirement of large-scale production, and the intelligent level of quality detection in a motor production line is further improved.

Description

Technical Field

[0001] The present application relates to the field of intelligent production, and more specifically, to a fully automatic motor production line and method thereof. Background Art

[0002] With the acceleration of industrialization, the demand for efficient and reliable motors continues to grow, driving the development of motor production lines. This demand growth has prompted manufacturers to continuously optimize production processes and improve product quality and performance to meet the specific requirements of motor performance in different industries. In the motor manufacturing industry, welding quality is one of the key manufacturing processes, and its quality directly affects the performance and life of the motor.

[0003] However, the quality inspection of welding in traditional motor production lines is highly dependent on the eyesight and experience of the inspectors, which not only makes the inspection results susceptible to individual differences, but also makes it difficult to ensure consistency. In particular, the human eye may have difficulty detecting tiny welding defects such as cracks, pores or incomplete penetration. Although these defects are small, they may cause malfunctions during product use or even lead to serious accidents. In addition, manual inspection is slow and prone to errors, and cannot maintain high efficiency and accuracy continuously, which limits its application in large-scale production environments.

[0004] Therefore, an optimized motor production line is desired. Summary of the invention

[0005] In order to solve the above technical problems, the present application is proposed. The embodiment of the present application provides a fully automatic motor production line and method thereof, which adopts an artificial intelligence-based image analysis technology to perform grayscale processing and morphological structural feature extraction of the welding area image, so as to intelligently judge whether the welding quality is qualified according to the fusion characteristics between the structural features and morphological features of the welding area image. In this way, tiny welding defects that are difficult for the human eye to detect can be accurately identified, thereby avoiding potential product failures, and at the same time greatly improving the detection speed, so that the production line can respond quickly and adapt to the needs of large-scale production, thereby improving the level of intelligence of quality inspection in the motor production line.

[0006] According to one aspect of the present application, there is provided a fully automatic motor production line, which includes: a raw material processing and preparation part, a winding and embedding part, an assembly and welding part, a varnishing and drying part, a detection and testing part, and a packaging and warehousing part;

[0007] The raw material processing and preparation part is used to convey and process the raw materials, and the raw materials include copper wire, iron core, housing and bearing;

[0008] The winding and embedding parts are used to wind the copper wire into stator coils and rotor coils, and embed the stator coils and rotor coils into the core slots to obtain the stator and rotor;

[0009] The assembly and welding part is used to automatically assemble and weld the stator, the rotor, the bearing and the housing to obtain an assembled motor;

[0010] The varnish dipping and drying part is used to immerse the assembled motor into insulating varnish and dry the motor after varnish dipping to obtain the motor to be tested;

[0011] The detection and testing part is used to perform key parameter detection and performance test on the motor to be detected to determine whether the motor to be detected is qualified;

[0012] The packaging and warehousing part is used to package the qualified motors to be tested.

[0013] According to another aspect of the present application, a fully automatic motor production method is provided, comprising:

[0014] Acquire the welding area image captured by the camera;

[0015] Performing grayscale processing on the welding area image to obtain a welding area grayscale image;

[0016] Inputting the welding area grayscale image into a welding area morphological feature extractor based on a multi-layer convolutional layer to obtain a welding area morphological representation vector;

[0017] Inputting the welding area grayscale image into a welding area structural feature extractor based on a hole convolutional neural network model to obtain a welding area structural feature map;

[0018] Inputting the welding area structural feature map into a self-attention enhanced cross-channel feature space structure preservation module to obtain a welding area structural enhancement feature map;

[0019] Inputting the welding area structure enhancement feature map and the welding area morphology representation vector into a heterogeneous Transformer cross-domain joint encoder assisted by morphological features to obtain a welding area multi-scale fusion feature map as a welding area multi-scale fusion feature;

[0020] Based on the multi-scale fusion features of the welding area, an evaluation result is obtained, and the evaluation result is used to indicate whether the welding quality is qualified.

[0021] Compared with the prior art, the fully automatic motor production line and method provided by the present application adopts an artificial intelligence-based image analysis technology to grayscale the welding area image and extract the morphological structure features, so as to intelligently judge whether the welding quality is qualified according to the fusion features between the structural features and morphological features of the welding area image. In this way, tiny welding defects that are difficult for the human eye to detect can be accurately identified, thereby avoiding potential product failures, and greatly improving the detection speed, so that the production line can respond quickly and adapt to the needs of large-scale production, thereby improving the intelligent level of quality detection in the motor production line. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the prior art descriptions. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. In the drawings:

[0023] Figure 1 It is a block diagram of a fully automatic motor production line according to an embodiment of the present application.

[0024] Figure 2 It is a block diagram of the assembly and welding parts in the fully automatic motor production line according to an embodiment of the present application.

[0025] Figure 3 Schematic diagram of data flow of the assembly and welding parts in the fully automatic motor production line according to an embodiment of the present application.

[0026] Figure 4 It is a block diagram of a welding area structure strengthening module in a fully automatic motor production line according to an embodiment of the present application.

[0027] Figure 5 It is a block diagram of a multi-scale fusion module of a welding area in a fully automatic motor production line according to an embodiment of the present application.

[0028] Figure 6 Flow chart of a fully automatic motor production method according to an embodiment of the present application. DETAILED DESCRIPTION

[0029] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0030] With the rapid development of industrialization, the demand for efficient and stable motors is growing, which has promoted the continuous improvement of motor manufacturing production lines. In order to meet the specific requirements of motor performance in different industries, manufacturers continue to improve production processes and are committed to improving product quality and performance.

[0031] In the field of motor manufacturing, the quality of the welding process is a crucial link, which is directly related to the performance and service life of the motor. However, the current motor production line relies heavily on the eyesight and experience of inspectors in welding quality inspection. This dependence makes the inspection results susceptible to individual differences and difficult to achieve consistency in the inspection results. In particular, it may be difficult for the human eye to detect tiny welding defects such as cracks, pores or lack of penetration. These seemingly minor defects may cause malfunctions during the use of the motor, and may even cause serious accidents. In addition, manual inspection is slow and prone to errors, which limits its efficiency and accuracy in large-scale production.

[0032] Based on this, the present application proposes a fully automatic motor production line, which includes: a raw material processing and preparation part, a winding and wire embedding part, an assembly and welding part, a varnishing and drying part, a detection and testing part, and a packaging and warehousing part; the raw material processing and preparation part is used to convey and process the raw materials, and the raw materials include copper wire, iron core, shell and bearing; the winding and wire embedding part is used to wind the copper wire into a stator coil and a rotor coil, and embed the stator coil and the rotor coil into the iron core slot to obtain a stator and a rotor; the assembly and welding part is used to automatically assemble and weld the stator, the rotor, the bearing and the shell to obtain an assembled motor; the varnishing and drying part is used to immerse the assembled motor in insulating varnish and dry the motor after varnishing to obtain a motor to be tested; the detection and testing part is used to perform key parameter detection and performance test on the motor to be tested to determine whether the motor to be tested is qualified; the packaging and warehousing part is used to package the qualified motor to be tested.

[0033] Accordingly, in the above-mentioned process of assembling and welding the parts to obtain the assembled motor, it is necessary to detect the welding quality to ensure the structural stability of the motor during operation. Therefore, in the technical solution of the present application, the welding area image is captured by a camera, and the image recognition and analysis technology based on artificial intelligence is used to grayscale the welding area image and extract the morphological structure features, so as to intelligently judge whether the welding quality is qualified according to the fusion characteristics between the structural features and morphological features of the welding area image. In this way, the use of artificial intelligence technology to conduct in-depth analysis of the welding image can accurately identify tiny welding defects that are difficult for the human eye to detect, such as cracks, pores or incomplete penetration, thereby avoiding potential product failures. In addition, automated image acquisition and analysis greatly improves the detection speed, allowing the production line to respond quickly and adapt to the needs of large-scale production, thereby improving the level of intelligence of quality inspection in the motor production line.

[0034] Figure 1 FIG. 1 is a block diagram of a fully automatic motor production line according to an embodiment of the present application. Figure 1 As shown, in the fully automatic motor production line 100, it includes: a raw material processing and preparation part 110, a winding and wire embedding part 120, an assembly and welding part 130, a varnishing and drying part 140, a detection and testing part 150 and a packaging and warehousing part 160; the raw material processing and preparation part 110 is used to transport and process the raw materials, and the raw materials include copper wire, iron core, housing and bearing; the winding and wire embedding part 120 is used to wind the copper wire into a stator coil and a rotor coil, and embed the stator coil and the rotor coil into the iron core slots to obtain the stator and the rotor; the assembly and welding part 130 is used to automatically assemble and weld the stator, the rotor, the bearing and the housing to obtain an assembled motor; the varnishing and drying part 140 is used to immerse the assembled motor in insulating varnish and dry the motor after varnishing to obtain the motor to be tested; the detection and testing part 150 is used to perform key parameter detection and performance testing on the motor to be tested to determine whether the motor to be tested is qualified; the packaging and warehousing part 160 is used to package the qualified motor to be tested.

[0035] Figure 2 It is a block diagram of the assembly and welding parts in the fully automatic motor production line according to an embodiment of the present application. Figure 3 Schematic diagram of data flow of the assembly and welding parts in the fully automatic motor production line according to the embodiment of the present application. Figure 2 and Figure 3As shown, the assembly and welding part 130 includes: a welding area image acquisition module 131, which is used to obtain a welding area image acquired by a camera; a welding area image grayscale processing module 132, which is used to perform grayscale processing on the welding area image to obtain a welding area grayscale image; a welding area morphological feature extraction module 133, which is used to input the welding area grayscale image into a welding area morphological feature extractor based on a multi-layer convolutional layer to obtain a welding area morphological representation vector; a welding area structural feature extraction module 134, which is used to input the welding area grayscale image into a welding area structural feature extractor based on a hole convolutional neural network model to obtain a welding area structural feature. Figure; a welding area structure enhancement module 135, which is used to input the welding area structure feature map into a self-attention enhanced cross-channel feature space structure preservation module to obtain a welding area structure enhancement feature map; a welding area multi-scale fusion module 136, which is used to input the welding area structure enhancement feature map and the welding area morphological representation vector into a heterogeneous Transformer cross-domain joint encoder assisted by morphological features to obtain a welding area multi-scale fusion feature map as a welding area multi-scale fusion feature; a welding quality assessment module 137, which is used to obtain an assessment result based on the welding area multi-scale fusion feature, and the assessment result is used to indicate whether the welding quality is qualified.

[0036] In the embodiment of the present application, the welding area image acquisition module 131 is used to obtain the welding area image acquired by the camera. It should be understood that the welding area image contains geometric feature information such as the shape of the weld and the welding position. In addition, it also contains possible surface defect information. By processing and analyzing the welding area image, efficient and accurate welding quality detection can be achieved, thereby ensuring the reliability and safety of the motor product, thereby effectively improving the overall quality of motor production.

[0037] In the embodiment of the present application, the welding area image grayscale processing module 132 is used to grayscale the welding area image to obtain the welding area grayscale image. Accordingly, considering that the welding area image contains welding features in the welding area, such as pores, cracks or incomplete penetration and other welding features. In order to make these welding-related features more prominent and obvious, in the technical solution of the present application, the welding area image is grayscale processed to obtain the welding area grayscale image. That is, grayscale processing of the welding area image can enhance the structural features of the welding area related to welding in the image, thereby improving subsequent feature analysis. In particular, in a specific implementation of the embodiment of the present application, the image grayscale conversion function in Opencv can be called to grayscale process the welding area image to obtain the welding area grayscale image.

[0038] In the embodiment of the present application, the welding area morphological feature extraction module 133 is used to input the welding area grayscale image into a welding area morphological feature extractor based on a multi-layer convolution layer to obtain a welding area morphological representation vector. It should be understood that the welding area grayscale image reflects the complex morphological features of different levels of the welding area, and the multi-layer convolution layer extracts the feature representation from shallow to deep at different levels, and can capture and mine the morphological feature information of different scales and depths of the welding area from the welding area grayscale image. Therefore, in the technical solution of the present application, the welding area grayscale image is input into a welding area morphological feature extractor based on a multi-layer convolution layer to obtain a welding area morphological representation vector. In particular, the welding area morphological feature extractor based on a multi-layer convolution layer here is a multi-layer convolution neural network model. Through the processing of the welding area morphological feature extractor based on a multi-layer convolution layer, the morphological features of the welding area can be effectively extracted, which provides a solid data analysis basis for subsequent welding quality evaluation.

[0039] In an embodiment of the present application, the welding area structural feature extraction module 134 is used to input the welding area grayscale image into a welding area structural feature extractor based on a hole convolutional neural network model to obtain a welding area structural feature map. It should be understood that the hole convolutional neural network model uses a hole convolution kernel to extract features from the input data. It introduces holes between the convolution kernels to expand the receptive field without increasing the amount of calculation, which allows the network to capture a wider range of contextual information when processing input data. Considering that the structural features contained in the welding area grayscale image play an important role in the evaluation of welding quality, in order to effectively extract the structural features contained in the welding area grayscale image, in the technical solution of the present application, the welding area grayscale image is input into a welding area structural feature extractor based on a hole convolutional neural network model to obtain a welding area structural feature map.

[0040] In an embodiment of the present application, the welding region structure enhancement module 135 is used to input the welding region structure feature map into the self-attention enhanced cross-channel feature space structure maintenance module to obtain the welding region structure enhancement feature map. Accordingly, considering that there is a mutual correlation and interaction relationship between different welding region structure features in the welding region structure feature map. Therefore, in order to more carefully understand and analyze the interaction relationship between different channels of the welding region structure feature map, while retaining the spatial structural information of each feature, in the technical solution of the present application, the welding region structure feature map is input into the self-attention enhanced cross-channel feature space structure maintenance module to obtain the welding region structure enhancement feature map. Specifically, firstly, the activation outputs of different layers in the neural network are adjusted and scaled by performing layer normalization processing on the welding region structure feature map, so that the processing process is stably distributed. Then, the context association between each channel of the normalized feature map is extracted by point convolution processing, and the spatial structure information of the welding region structure feature map is captured by convolution coding. After that, the backup spatial feature information obtained by copying the convolution coding is retained. The spatial context information of the original welding region structure. Subsequently, the channel, space, and replicated feature maps are converted into matrix form through feature reshaping, and the self-attention mechanism is used to perform cross-channel global correlation interaction of welding region structural features to quantify the correlation between welding region structural features of different channels. The covariance matrix calculated by activating channel features and spatial features is then activated using the Softmax function to effectively highlight the key information of the welding region structure in the data and generate a global interaction attention matrix. Finally, the attention-enhanced feature representation matrix obtained by multiplying the backup spatial features with the global interaction attention matrix is ​​reshaped and restored to the form of a feature map to obtain a welding region structure-enhanced feature map.

[0041] Specifically, Figure 4 FIG. 1 is a block diagram of a welding area structure strengthening module in a fully automatic motor production line according to an embodiment of the present application. Figure 4As shown, the welding area structure strengthening module 135 includes: a welding area structure feature map normalization processing unit 1351, which is used to perform layer normalization on the welding area structure feature map to obtain a normalized welding area structure feature map; a welding area structure channel context association representation feature generation unit 1352, which is used to perform point convolution processing on the normalized welding area structure feature map to obtain a welding area structure channel context association representation feature map; a welding area structure channel context association representation feature convolution encoding unit 1353, which is used to perform convolution encoding on the welding area structure channel context association representation feature map to obtain a welding area structure space context association representation feature map; a welding area structure space context association representation feature backup unit 1354, which is used to copy the welding area structure space context association representation feature map to obtain a welding area structure backup space context association representation feature map; a welding area structure context association representation feature shape reshaping unit 1355, which is used to perform feature processing on the welding area structure channel context association representation feature map, the welding area structure space context association representation feature map and the welding area structure backup space context association representation feature map respectively. The shape is reshaped to obtain the welding area structure channel context association representation feature matrix, the welding area structure space context association representation feature matrix and the welding area structure backup space context association representation feature matrix; the welding area structure covariance matrix calculation unit 1356 is used to calculate the welding area structure cross-channel cross covariance matrix between the welding area structure channel context association representation feature matrix and the welding area structure space context association representation feature matrix; the welding area structure cross-channel cross covariance matrix activation unit 1357 is used to use the Softmax function to activate the welding area structure cross-channel cross covariance matrix to obtain the welding area structure feature global interactive attention matrix; the welding area structure attention enhancement feature representation generation unit 1358 is used to calculate the product between the welding area structure backup space context association representation feature matrix and the welding area structure feature global interactive attention matrix to obtain the welding area structure attention enhancement feature representation matrix; the welding area structure enhancement feature generation unit 1359 is used to perform feature shape reshaping on the welding area structure attention enhancement feature representation matrix to obtain the welding area structure enhancement feature map.

[0042] More specifically, in an embodiment of the present application, the welding area structure covariance matrix calculation unit is used to: calculate the matrix product between the welding area structure channel context association representation feature matrix and the welding area structure space context association representation feature matrix to obtain the welding area structure channel-space context association representation feature matrix; calculate the positional division of the welding area structure channel-space context association representation feature matrix and a predetermined hyperparameter to obtain the welding area structure cross-channel cross covariance matrix.

[0043] In the embodiment of the present application, specifically, the welding area structure strengthening module is used to: input the welding area structure feature map into the self-attention enhanced cross-channel feature space structure maintaining module, and process it with the following self-attention enhancement formula to obtain the welding area structure strengthening feature map; wherein the self-attention enhancement formula is:

[0044] F ln =Layer Normalization(F i )

[0045] F Q =Conv 1×1 (F ln )

[0046] F K =Conv 3×3 (F Q )

[0047] F V =Copy(F K )

[0048] M Q =Reshape(F Q )

[0049] M K =Reshape(F K )

[0050] M V =Reshape(F V )

[0051]

[0052]

[0053] Among them, F i is the structural feature map of the welding area, Layer Normalization(·) represents the layer normalization operation on the feature map, and F ln is the normalized welding area structural feature map, Conv 1×1(·) is the point convolution operation, F Q The welding area structure channel context association represents the feature map, Conv 3×3 (·) is the convolution operation with a convolution kernel of 3×3, F K is the spatial contextual representation feature map of the welding area structure, Copy(·) is the copy operation, and F V is the spatial context association representation feature map of the welding area structure backup space, Reshape(·) is the shape reshaping operation, M Q 、M K and M V are the welding region structure channel context association representation feature matrix, the welding region structure space context association representation feature matrix and the welding region structure backup space context association representation feature matrix, θ is a predetermined hyperparameter, softmax(·) is a softmax function, M a is the global interactive attention matrix of the welding area structural features, is the matrix multiplication, F a It is a structural strengthening feature map of the welding area.

[0054] In an embodiment of the present application, the welding region multi-scale fusion module 136 is used to input the welding region structure enhancement feature map and the welding region morphological representation vector into a heterogeneous Transformer cross-domain joint encoder assisted by morphological features to obtain a welding region multi-scale fusion feature map as a welding region multi-scale fusion feature. Accordingly, considering that the welding region structure enhancement feature map and the welding region morphological representation vector respectively express the structural information such as the geometric and texture features of the welding region and the morphological features such as the local shape and size of the welding region. In order to effectively fuse the features of different scales and descriptions of the two together to obtain a more comprehensive welding region feature representation, in the technical solution of the present application, by using fine feature processing and the self-attention mechanism of the Transformer structure, the effective fusion and optimization of different modal features are realized by using a heterogeneous Transformer cross-domain joint encoder assisted by morphological features to jointly encode the welding region structure enhancement feature map and the welding region morphological representation vector to obtain a welding region multi-scale fusion feature map.

[0055] Specifically, the welding region morphology representation modulation matrix is ​​first generated by linearly transforming and reshaping the welding region morphology representation vector. And the welding region structure enhancement feature map is linearly transformed and reshaped twice to obtain two sets of linear transformation feature matrices. Then, the two sets of linear transformation feature matrices are used as key matrices and value matrices, and the welding region morphology representation modulation matrix is ​​used as the query matrix. The welding region morphology attention query welding region structure matrix is ​​obtained by using the self-attention mechanism of Transformer to capture the mutual correlation and dependency between the welding region structure and morphology, and reshape and point convolution encoding are performed to generate the welding region morphology attention query welding region structure feature map. Finally, the multi-scale fusion feature map of the welding region is generated by calculating the position point multiplication between the welding region morphology attention query welding region structure feature map and the welding region structure enhancement feature map, which integrates the cross-modal features of the welding region structure and morphology and the information of the original structure feature map.

[0056] Specifically, Figure 5 FIG. 1 is a block diagram of a multi-scale fusion module for welding areas in a fully automatic motor production line according to an embodiment of the present application. Figure 5As shown, the welding area multi-scale fusion module 136 includes: a welding area morphological feature linear transformation shape reshaping unit 1361, which is used to perform linear transformation and feature shape reshaping on the welding area morphological representation vector to obtain a welding area morphological representation modulation matrix; a welding area structural enhancement feature linear transformation unit 1362, which is used to perform a first linear transformation and a second linear transformation on the welding area structural enhancement feature map to obtain a first linear transformation welding area structural enhancement feature map and a second linear transformation welding area structural enhancement feature map; a linear transformation welding area structural enhancement feature shape reshaping unit 1363, which is used to perform feature shape reshaping on the first linear transformation welding area structural enhancement feature map and the second linear transformation welding area structural enhancement feature map to obtain a first linear transformation welding area structural enhancement feature matrix and a second linear transformation welding area structural enhancement feature matrix; welding area morphological query attention calculation Unit 1364 is used to use the first linear transformation welding area structure enhancement feature matrix and the second linear transformation welding area structure enhancement feature matrix as the key matrix and the value matrix and use the welding area morphology representation modulation matrix as the query matrix, and input the query matrix, the key matrix and the value matrix into the Transformer structure to obtain the welding area morphology attention query welding area structure matrix; the welding area morphology attention query welding area structure feature generation unit 1365 is used to reshape and point convolutionally encode the welding area morphology attention query welding area structure matrix to obtain the welding area morphology attention query welding area structure feature map; the welding area multi-scale fusion feature generation unit 1366 is used to perform position point multiplication of the welding area morphology attention query welding area structure feature map and the welding area structure enhancement feature map to obtain the welding area multi-scale fusion feature map.

[0057] More specifically, in an embodiment of the present application, the welding area morphological feature linear transformation shape reshaping unit is used to: after matrix multiplying the welding area morphological representation vector and the welding area morphological modulation matrix to obtain the welding area morphological modulation vector, the welding area morphological modulation vector and the welding area morphological bias vector are added according to position to obtain the welding area morphological modulation bias vector; the welding area morphological modulation bias vector is input into the Sigmoid function for activation processing to obtain the welding area morphological modulation activation vector; and the welding area morphological modulation activation vector is reshaped to obtain the welding area morphological representation modulation matrix.

[0058] More specifically, in an embodiment of the present application, the welding area structure strengthening feature linear transformation unit is used to: multiply each feature matrix along the channel dimension in the welding area structure strengthening feature map by position with the first welding area structure modulation matrix, and then add them by position with the first welding area structure bias matrix to obtain the first welding area structure bias feature map; input the first welding area structure bias feature map into the Sigmoid function for activation processing to obtain the first linear transformation welding area structure strengthening feature map; multiply each feature matrix along the channel dimension in the welding area structure strengthening feature map by position with the second welding area structure modulation matrix, and then add them by position with the second welding area structure bias matrix to obtain the second welding area structure bias feature map; input the second welding area structure bias feature map into the Sigmoid function for activation processing to obtain the second linear transformation welding area structure strengthening feature map.

[0059] More specifically, in an embodiment of the present application, the welding area morphology query attention calculation unit is used to: calculate the matrix product between the transposed matrix of the welding area morphology representation modulation matrix and the first linear transformation welding area structure enhancement feature matrix and then input the matrix product into the softmax function to obtain the welding area morphology-structure feature association matrix; use the welding area morphology-structure feature association matrix as a weight matrix, calculate the matrix product between the weight matrix and the second linear transformation welding area structure enhancement feature matrix to obtain the welding area morphology attention query welding area structure matrix.

[0060] In the embodiment of the present application, specifically, the welding area multi-scale fusion module is used to: input the welding area structure enhancement feature map and the welding area morphology representation vector into a heterogeneous Transformer cross-domain joint encoder assisted by morphological features, and process them with the following cross-domain joint encoding formula to obtain the welding area multi-scale fusion feature map; wherein the cross-domain joint encoding formula is:

[0061] V l =σ(W l V s +B l )

[0062] M l = reshape(V l )

[0063] F1=σ(W1F a +B1)

[0064] F2=σ(W2F a +B2)

[0065] M1=reshape(F1)

[0066] M2=reshape(F2)

[0067]

[0068]

[0069] F w =Conv 1×1 [reshape(M t )]

[0070] F s =F w ⊙F a

[0071] Among them, V s is the welding area morphology representation vector, W l is the welding area morphology modulation matrix, B l is the morphological bias vector of the welding area, σ(·) is the Sigmoid function, V l is the activation vector for the weld region morphology modulation, reshape(·) is the shape reshaping operation, M l The modulation matrix, F, is the morphology representation of the welding area. a is the welding region structural strengthening characteristic diagram, W1 and W2 are the first welding region structural modulation matrix and the second welding region structural modulation matrix, B1 and B2 are the first welding region structural bias matrix and the second welding region structural bias matrix, F1 and F2 are the first linear transformation welding region structural strengthening characteristic diagram and the second linear transformation welding region structural strengthening characteristic diagram, M1 and M2 are the first linear transformation welding region structural strengthening characteristic matrix and the second linear transformation welding region structural strengthening characteristic matrix, M l T is the transposed matrix of the welding area morphology modulation matrix, softmax(·) is the softmax function, M b is the welding area morphology-structure feature association matrix, is the matrix product, M t Query the welding area structure matrix for the welding area morphology attention, Conv 1×1 (·) is the point convolution operation, F w is the welding area morphology attention query welding area structure feature map, ⊙ is the position point multiplication, F s ′ is the multi-scale fusion feature map of the welding area.

[0072] In an embodiment of the present application, the welding quality assessment module 137 is used to obtain an assessment result based on the multi-scale fusion feature of the welding area, and the assessment result is used to indicate whether the welding quality is qualified. Specifically, in an embodiment of the present application, the welding quality assessment module is used to: input the multi-scale fusion feature map of the welding area into a welding quality evaluator based on a classifier to obtain the assessment result, and the assessment result is used to indicate whether the welding quality is qualified. That is, the multi-scale fusion feature of the welding area obtained by cross-domain joint encoding of the welding area structure enhancement feature map and the welding area morphology representation vector is used for classification processing to intelligently judge whether the welding quality is qualified. In this way, by using artificial intelligence technology to conduct in-depth analysis of welding images, tiny welding defects that are difficult for the human eye to detect, such as cracks, pores or incomplete penetration, can be accurately identified, thereby avoiding potential product failures. In addition, automated image acquisition and analysis greatly improve the detection speed, allowing the production line to respond quickly and adapt to the needs of large-scale production, thereby promoting the intelligent level of quality inspection in the motor production line.

[0073] In particular, in a preferred example, when the welding area structure enhancement feature map and the welding area morphology representation vector respectively express image semantic features of different dimensions of the welding area grayscale image, and perform morphology-assisted heterogeneous Transformer cross-domain joint encoding on them, the obtained welding area multi-scale fusion feature map will also cause the probability convergence and divergence of the image semantic feature distribution regression mode due to the difference in cross-domain encoding constraints of the heterogeneous image semantic feature channel dimensions, that is, the accuracy of the evaluation result obtained by inputting the welding area multi-scale fusion feature map into the classifier-based welding quality evaluator is reduced.

[0074] Therefore, when the multi-scale fusion feature map of the welding area is classified by a welding quality evaluator based on a classifier, the multi-scale fusion feature map of the welding area is first optimized, including the following steps:

[0075] Determine the median eigenvalue, the maximum eigenvalue and the minimum eigenvalue in the feature set of the multi-scale fusion feature map of the welding area, and divide the median eigenvalue by the difference between the maximum eigenvalue and the minimum eigenvalue to obtain the multi-scale fusion distribution probability value of the welding area, that is, p = f mid / (f max -f min ), where f mid 、f max 、f min They respectively represent the median eigenvalue, the maximum eigenvalue and the minimum eigenvalue in the feature set of the multi-scale fusion feature map of the welding area, and p represents the multi-scale fusion distribution probability value of the welding area;

[0076] Each eigenvalue in the multi-scale fusion feature map of the welding area is normalized to the maximum value to obtain the multi-scale fusion probability eigenvalue f of the welding area. i ;

[0077] The multi-scale fusion probability characteristic value f of each welding area is calculated by subtracting the multi-scale fusion distribution probability value p of the welding area as an exponent. i The power function is used to obtain the convergence characteristic value f of the multi-scale fusion probability of the first welding area. 1i =f i 1 -p , where p represents the probability value of multi-scale fusion distribution of welding area, f i represents the multi-scale fusion probability eigenvalue of the welding area, f 1i represents the convergence eigenvalue of multi-scale fusion probability in the first welding area;

[0078] The multi-scale fusion distribution probability value p of the welding area is used as an index to calculate a multi-scale fusion probability characteristic value f of each welding area. i The power function of the difference between the two is used to obtain the multi-scale fusion probability convergence eigenvalue f of the second welding area. 2i =(1-f i ) p , where f 2i It represents the convergence eigenvalue of multi-scale fusion probability in the second welding area;

[0079] The multi-scale fusion probability feature value f of each welding area i The difference between the probability value p of the multi-scale fusion distribution of the welding area and one is multiplied to obtain the first welding area multi-scale fusion probability defined characteristic value f 3i =f i ×(1-p), where f 3i represents the multi-scale fusion probability limiting characteristic value of the first welding area;

[0080] The multi-scale fusion probability feature value f of each welding area is i The difference between the values ​​of and is multiplied by the multi-scale fusion distribution probability value p of the welding area to obtain the second welding area multi-scale fusion probability limiting characteristic value f 4i =(1-f i )×p, where f 4i represents the multi-scale fusion probability limiting eigenvalue of the second welding area;

[0081] After multiplying the first welding area multiscale fusion probability convergence eigenvalue and the second welding area multiscale fusion probability convergence eigenvalue, further add the first welding area multiscale fusion probability limit eigenvalue and the second welding area multiscale fusion probability limit eigenvalue to obtain the eigenvalue f′ of the optimized welding area multiscale fusion feature map i =f 1i ×f 2i +f 3i +f 4i , where f′ i The characteristic value of the multi-scale fusion feature map of the optimized welding area is represented. Finally, the multi-scale fusion feature map of the optimized welding area is input into the welding quality evaluator based on the classifier to obtain the evaluation result.

[0082] Based on this, by using the cross entropy form power series of the regression probability of the multi-scale fusion feature map of the welding area as the probability distribution convergence limit, the regression probability convergence approximation is performed on the basis of the combination of the feature set distribution and the probability density distribution of the multi-scale fusion feature map of the welding area, so as to avoid the conservative convergence control of the fatal potential divergence of the multi-scale fusion feature map of the welding area, and further use the probability distribution cross entropy as the objective function to guide the limited recovery strategy, so as to achieve the common agility of the convergence of the probability convergence and divergence of the feature set of the multi-scale fusion feature map of the welding area to the probability density distribution space, and improve the accuracy of the evaluation results obtained by the welding quality evaluator based on the classifier. In this way, the use of artificial intelligence technology to conduct in-depth analysis of welding images can accurately identify tiny welding defects that are difficult for the human eye to detect, such as cracks, pores or incomplete penetration, thereby avoiding potential product failures. In addition, automated image acquisition and analysis greatly improve the detection speed, so that the production line can respond quickly and adapt to the needs of large-scale production, thereby promoting the intelligent level of quality inspection in the motor production line.

[0083] In summary, the fully automatic motor production line 100 based on the embodiment of the present application is explained, which uses an artificial intelligence-based image analysis technology to perform grayscale processing and morphological structural feature extraction of the welding area image, so as to intelligently judge whether the welding quality is qualified according to the fusion characteristics between the structural features and morphological features of the welding area image. In this way, tiny welding defects that are difficult for the human eye to detect can be accurately identified, thereby avoiding potential product failures, and at the same time greatly improving the detection speed, so that the production line can respond quickly and adapt to the needs of large-scale production, thereby improving the level of intelligence of quality inspection in the motor production line.

[0084] As described above, the fully automatic motor production line 100 according to the embodiment of the present application can be implemented in various terminal devices, such as a server for fully automatic motor production. In one example, the fully automatic motor production line 100 according to the embodiment of the present application can be integrated into the terminal device as a software module and / or a hardware module. For example, the fully automatic motor production line 100 can be a software module in the operating system of the terminal device, or can be an application developed for the terminal device; of course, the fully automatic motor production line 100 can also be one of the many hardware modules of the terminal device.

[0085] Alternatively, in another example, the fully automatic motor production line 100 and the terminal device may also be separate devices, and the fully automatic motor production line 100 may be connected to the terminal device via a wired and / or wireless network and transmit interactive information in accordance with an agreed data format.

[0086] In other embodiments of the present application, a fully automatic motor production line is also provided, which includes a winding production part, a paper making production part, a wire entry production part, a coil production part, an inner winding and outer winding production part, a paint dipping and baking production part, a motor housing production part, a motor assembly production part and a quality inspection part. Specifically, the winding production part is responsible for the traditional coil enameled wire winding; the paper making production part is responsible for inserting the insulating paper into the stator core; the wire entry production part is responsible for inserting the wound coil into the stator core; the wire entry production part is mainly responsible for completing the shaping, binding, measuring, wiring, soldering and other processes of the stator coil with the wire entered; the inner winding and outer winding production part is responsible for directly winding the enameled wire on the stator wire frame or directly winding the enameled wire on the stator wire frame and then embedding it into the stator core. In the inner winding and outer winding production, a fully automatic detection equipment that can automatically identify defects is used to improve Production quality; the varnishing and baking production section uses the most advanced fully automatic assembly line varnishing and baking equipment to automatically varnish and automatically bake the inspected stator coil products at a constant temperature; the motor shell production section uses equipment such as batch removal machines and automatic tapping and drilling machines to produce motor shells; the motor assembly production section is responsible for the assembly of front and rear aluminum shells, axial rotors, coil products, gear boxes, capacitors and other components, as well as motor inspection and finished motor packaging; the quality inspection section is responsible for randomly sampling some products for relevant inspections on incoming materials and all aspects of the entire production.

[0087] Figure 6 Flow chart of the fully automatic motor production method according to an embodiment of the present application. Figure 6As shown, in the fully automatic motor production method, it includes: S1, acquiring a welding area image collected by a camera; S2, grayscale processing the welding area image to obtain a welding area grayscale image; S3, inputting the welding area grayscale image into a welding area morphological feature extractor based on a multi-layer convolutional layer to obtain a welding area morphological representation vector; S4, inputting the welding area grayscale image into a welding area structural feature extractor based on a hole convolutional neural network model to obtain a welding area structural feature map; S5, inputting the welding area structural feature map into a self-attention enhanced cross-channel feature space structure preservation module to obtain a welding area structural enhancement feature map; S6, inputting the welding area structural enhancement feature map and the welding area morphological representation vector into a heterogeneous Transformer cross-domain joint encoder assisted by morphological features to obtain a welding area multi-scale fusion feature map as a welding area multi-scale fusion feature; S7, based on the welding area multi-scale fusion feature, obtaining an evaluation result, and the evaluation result is used to indicate whether the welding quality is qualified.

[0088] Here, those skilled in the art will appreciate that the specific operations of each step in the above-mentioned fully automatic motor production method have been described in detail above. Figures 1 to 5 The fully automatic motor production line has been described in detail, and therefore, its repeated description will be omitted.

[0089] In summary, the fully automatic motor production method based on the embodiment of the present application is explained, which uses an artificial intelligence-based image analysis technology to grayscale the welding area image and extract the morphological structure features, so as to intelligently judge whether the welding quality is qualified according to the fusion characteristics between the structural features and morphological features of the welding area image. In this way, tiny welding defects that are difficult for the human eye to detect can be accurately identified, thereby avoiding potential product failures, and at the same time greatly improving the detection speed, so that the production line can respond quickly and adapt to the needs of large-scale production, thereby improving the level of intelligence of quality inspection in the motor production line.

[0090] The basic principles of the present application are described above in conjunction with specific embodiments. However, it should be noted that the advantages, strengths, effects, etc. mentioned in the present application are only examples and not limitations, and it cannot be considered that these advantages, strengths, effects, etc. are required by each embodiment of the present application. In addition, the specific details of the above application are only for the purpose of illustration and ease of understanding, rather than limitation, and the above details do not limit the present application to the use of the above specific details to be implemented.

Claims

1. A fully automatic motor production line, characterized in that: include: Raw material processing and preparation, winding and embedding, assembly and welding, varnishing and drying, inspection and testing, and packaging and warehousing; The raw material processing and preparation part is used to convey and process the raw materials, and the raw materials include copper wire, iron core, housing and bearing; The winding and embedding parts are used to wind the copper wire into stator coils and rotor coils, and embed the stator coils and rotor coils into the core slots to obtain the stator and rotor; The assembly and welding part is used to automatically assemble and weld the stator, the rotor, the bearing and the housing to obtain an assembled motor; The varnish dipping and drying part is used to immerse the assembled motor into insulating varnish and dry the motor after varnish dipping to obtain the motor to be tested; The detection and testing part is used to perform key parameter detection and performance test on the motor to be detected to determine whether the motor to be detected is qualified; The packaging and warehousing part is used to package the qualified motors to be tested.

2. The fully automatic motor production line according to claim 1 is characterized in that: The assembly and welding part includes: A welding area image acquisition module, used to acquire the welding area image acquired by the camera; A welding area image grayscale processing module, used for performing grayscale processing on the welding area image to obtain a welding area grayscale image; A welding area morphological feature extraction module, used for inputting the welding area grayscale image into a welding area morphological feature extractor based on a multi-layer convolution layer to obtain a welding area morphological representation vector; A welding area structural feature extraction module, used for inputting the welding area grayscale image into a welding area structural feature extractor based on a hole convolutional neural network model to obtain a welding area structural feature map; A welding region structure strengthening module, used for inputting the welding region structure feature map into a self-attention enhanced cross-channel feature space structure maintaining module to obtain a welding region structure strengthening feature map; A welding area multi-scale fusion module, used for inputting the welding area structure enhancement feature map and the welding area morphology representation vector into a heterogeneous Transformer cross-domain joint encoder assisted by morphological features to obtain a welding area multi-scale fusion feature map as a welding area multi-scale fusion feature; The welding quality assessment module is used to obtain an assessment result based on the multi-scale fusion features of the welding area, and the assessment result is used to indicate whether the welding quality is qualified.

3. The fully automatic motor production line according to claim 2 is characterized in that: The welding area structure strengthening module comprises: A welding region structural feature map normalization processing unit, used for performing layer normalization on the welding region structural feature map to obtain a normalized welding region structural feature map; A welding region structure channel context association representation feature generation unit, used for performing point convolution processing on the normalized welding region structure feature map to obtain a welding region structure channel context association representation feature map; A welding region structure channel context association representation feature convolution encoding unit, used for performing convolution encoding on the welding region structure channel context association representation feature map to obtain a welding region structure space context association representation feature map; A welding region structure space context association representation feature backup unit, used to copy the welding region structure space context association representation feature map to obtain a welding region structure backup space context association representation feature map; A welding region structure context association representation feature shape reshaping unit is used to reshape the feature shape of the welding region structure channel context association representation feature map, the welding region structure space context association representation feature map and the welding region structure backup space context association representation feature map to obtain a welding region structure channel context association representation feature matrix, a welding region structure space context association representation feature matrix and a welding region structure backup space context association representation feature matrix; A welding region structure covariance matrix calculation unit, used to calculate a welding region structure cross-channel cross covariance matrix between the welding region structure channel context association representation feature matrix and the welding region structure space context association representation feature matrix; A welding region structure cross-channel cross-covariance matrix activation unit is used to activate the welding region structure cross-channel cross-covariance matrix using a Softmax function to obtain a welding region structure feature global interactive attention matrix; A welding area structure attention enhancement feature representation generation unit is used to calculate the product between the welding area structure backup space context association representation feature matrix and the welding area structure feature global interaction attention matrix to obtain the welding area structure attention enhancement feature representation matrix; The welding area structure enhancement feature generation unit is used to reshape the welding area structure attention enhancement feature representation matrix to obtain the welding area structure enhancement feature map.

4. The fully automatic motor production line according to claim 3 is characterized in that: The welding area structure covariance matrix calculation unit is used for: Calculating the matrix product between the welding region structure channel context association representation feature matrix and the welding region structure space context association representation feature matrix to obtain the welding region structure channel-space context association representation feature matrix; The position-wise division of the welding region structure channel-spatial context association representation feature matrix and a predetermined hyperparameter is calculated to obtain the welding region structure cross-channel cross covariance matrix.

5. The fully automatic motor production line according to claim 4 is characterized in that: The welding area multi-scale fusion module includes: A welding region morphology characteristic linear transformation shape reshaping unit is used to perform linear transformation and characteristic shape reshaping on the welding region morphology representation vector to obtain a welding region morphology representation modulation matrix; A welding region structural strengthening characteristic linear transformation unit, used for performing a first linear transformation and a second linear transformation on the welding region structural strengthening characteristic map to obtain a first linear transformation welding region structural strengthening characteristic map and a second linear transformation welding region structural strengthening characteristic map; A linear transformation welding region structural strengthening characteristic shape reshaping unit is used to reshape the first linear transformation welding region structural strengthening characteristic map and the second linear transformation welding region structural strengthening characteristic map to obtain a first linear transformation welding region structural strengthening characteristic matrix and a second linear transformation welding region structural strengthening characteristic matrix; A welding region morphology query attention calculation unit is used to use the first linear transformation welding region structure enhancement feature matrix and the second linear transformation welding region structure enhancement feature matrix as a key matrix and a value matrix and use the welding region morphology representation modulation matrix as a query matrix, and input the query matrix, the key matrix and the value matrix into a Transformer structure to obtain a welding region morphology attention query welding region structure matrix; A welding region morphology attention query welding region structure feature generation unit is used to reshape and dot convolutional encode the welding region morphology attention query welding region structure matrix to obtain a welding region morphology attention query welding region structure feature map; The welding area multi-scale fusion feature generation unit is used to multiply the welding area morphological attention query welding area structure feature map and the welding area structure enhancement feature map by position points to obtain the welding area multi-scale fusion feature map.

6. The fully automatic motor production line according to claim 5, characterized in that: The welding area morphological feature linear transformation shape reshaping unit is used to: After matrix multiplication of the welding area morphology representation vector and the welding area morphology modulation matrix to obtain the welding area morphology modulation vector, the welding area morphology modulation vector and the welding area morphology bias vector are added according to position to obtain the welding area morphology modulation bias vector; Inputting the welding area morphology modulation bias vector into a Sigmoid function for activation processing to obtain a welding area morphology modulation activation vector; The welding area morphology modulation activation vector is reshaped to obtain the welding area morphology representation modulation matrix.

7. The fully automatic motor production line according to claim 6, characterized in that: The welding area structural strengthening characteristic linear transformation unit is used for: Each characteristic matrix along the channel dimension in the welding region structural strengthening characteristic map is multiplied by position with the first welding region structural modulation matrix, and then is added by position with the first welding region structural bias matrix to obtain the first welding region structural bias characteristic map; Inputting the first welding region structural bias feature map into the Sigmoid function for activation processing to obtain the first linear transformation welding region structural enhancement feature map; Each characteristic matrix along the channel dimension in the welding region structural strengthening characteristic map is multiplied by position with the second welding region structural modulation matrix, and then is added by position with the second welding region structural bias matrix to obtain the second welding region structural bias characteristic map; The second welding region structural bias feature map is input into the Sigmoid function for activation processing to obtain the second linear transformation welding region structural enhancement feature map.

8. The fully automatic motor production line according to claim 7, characterized in that: The welding area morphology query attention calculation unit is used to: Calculating the matrix product between the transposed matrix of the welding region morphology representation modulation matrix and the first linear transformation welding region structure enhancement feature matrix and inputting the matrix into a softmax function to obtain a welding region morphology-structure feature association matrix; The welding area morphology-structure feature association matrix is ​​used as a weight matrix, and the matrix product between the weight matrix and the second linear transformation welding area structure enhancement feature matrix is ​​calculated to obtain the welding area morphology attention query welding area structure matrix.

9. The fully automatic motor production line according to claim 8, characterized in that: The welding quality assessment module is used to: input the multi-scale fusion feature map of the welding area into a welding quality assessor based on a classifier to obtain the assessment result, and the assessment result is used to indicate whether the welding quality is qualified.

10. A fully automatic motor production method, characterized in that: include: Acquire the welding area image captured by the camera; Performing grayscale processing on the welding area image to obtain a welding area grayscale image; Inputting the welding area grayscale image into a welding area morphological feature extractor based on a multi-layer convolutional layer to obtain a welding area morphological representation vector; Inputting the welding area grayscale image into a welding area structural feature extractor based on a hole convolutional neural network model to obtain a welding area structural feature map; Inputting the welding area structural feature map into a self-attention enhanced cross-channel feature space structure preservation module to obtain a welding area structural enhancement feature map; Inputting the welding area structure enhancement feature map and the welding area morphology representation vector into a heterogeneous Transformer cross-domain joint encoder assisted by morphological features to obtain a welding area multi-scale fusion feature map as a welding area multi-scale fusion feature; Based on the multi-scale fusion features of the welding area, an evaluation result is obtained, and the evaluation result is used to indicate whether the welding quality is qualified.

Images