Production defect identification method and system for metal rubber composite plate

Abstract

The invention relates to the field of production quality detection, and provides a production defect identification method and system for a metal rubber composite board, and the method comprises the steps: carrying out the time domain semantic feature extraction of each ultrasonic echo detection signal in a set of ultrasonic echo detection signals through employing a signal processing and analysis technology based on deep learning; and automatically judging whether the first scanning position data is abnormal or not according to the ultrasonic echo detection signal semantic feature corresponding to the first scanning position data in the scanning position data set and the query comparison between each ultrasonic echo detection signal semantic feature. Therefore, tiny internal defects which cannot be touched by traditional optical or contact detection in the metal rubber composite plate can be detected, and the detection accuracy can be improved by performing defect detection through a deep learning technology.

Description

Technical Field

[0001] The present application relates to the field of production quality inspection, and more specifically, to a production defect identification method and system for metal-rubber composite plates. Background Art

[0002] Metal-rubber composite sheet is a material formed by combining metal and rubber. This material combines the high strength and rigidity of metal with the elasticity and vibration damping properties of rubber. Metal-rubber composite sheet is usually made of one or more layers of metal sheets (such as mild steel) tightly bonded with a rubber layer through a special process. This material is widely used in the automotive, aerospace, construction and other fields, especially in applications that require high stress, vibration absorption, sealing and insulation. There are many defects that may occur during the production process of metal-rubber composite sheets, which can affect the quality and performance of the product. Therefore, identifying and controlling these defects is essential to ensure product quality and extend product life.

[0003] However, traditional mechanical inspection equipment can usually only detect specific types of defects, such as surface cracks or holes, but the detection effect is often poor for internal defects or more complex defect types. Specifically, because traditional mechanical inspection equipment often relies on contact measurement or simple optical imaging technology, it is difficult for these technologies to penetrate the interior of the material to detect hidden defects. For example, for metal-rubber composite panels, internal delamination, poor bonding and other problems are difficult to detect through surface observation. In addition, mechanical inspection equipment also encounters difficulties when dealing with complex geometries or irregular surfaces, resulting in incomplete detection results and possible damage to the sample.

[0004] Therefore, an optimized production defect identification scheme for metal-rubber composite panels is needed. Summary of the invention

[0005] In view of the shortcomings of the prior art, the present application provides a production defect identification method and system for metal-rubber composite plates.

[0006] According to one aspect of the present application, a method for identifying production defects of a metal-rubber composite plate is provided, comprising:

[0007] Scanning the metal-rubber composite plate to be inspected using an ultrasonic probe to obtain a set of scanning position data and a set of ultrasonic echo detection signals;

[0008] Extracting echo signal time domain features from each ultrasonic echo detection signal in the set of ultrasonic echo detection signals to obtain a set of ultrasonic echo detection signal semantic feature vectors;

[0009] Performing local neighborhood gated query processing on the ultrasonic echo detection signal semantic feature vector corresponding to the first scanning position data in the set of scanning position data and the set of ultrasonic echo detection signal semantic feature vectors to obtain an ultrasonic detection single point-global query comparison implicit representation;

[0010] Based on the ultrasonic detection single point-global query comparison implicit representation, a recognition result is obtained, and the recognition result is used to indicate whether the first scanning position data has an abnormality;

[0011] Among them, the ultrasonic echo detection signal semantic feature vector corresponding to the first scanning position data in the set of scanning position data and the set of the ultrasonic echo detection signal semantic feature vectors are subjected to local neighborhood gated query processing to obtain an implicit representation of ultrasonic detection single point-global query comparison, including: using the ultrasonic echo detection signal semantic feature vector corresponding to the first scanning position data in the set of scanning position data as a query feature vector, using the set of ultrasonic echo detection signal semantic feature vectors as a set of analog key feature vectors, and performing attention implicit optimization on the query feature vector and the set of analog key feature vectors to obtain the implicit representation of ultrasonic detection single point-global query comparison.

[0012] In the above-mentioned production defect identification method for metal-rubber composite plates, echo signal time domain feature extraction is performed on each ultrasonic echo detection signal in the set of ultrasonic echo detection signals to obtain a set of ultrasonic echo detection signal semantic feature vectors, including: inputting each ultrasonic echo detection signal in the set of ultrasonic echo detection signals into an echo signal time domain feature extractor based on a 1D-CNN model to obtain a set of ultrasonic echo detection signal semantic feature vectors.

[0013] In the above-mentioned production defect identification method for metal-rubber composite plates, the ultrasonic echo detection signal semantic feature vector corresponding to the first scanning position data in the set of scanning position data is used as a query feature vector, and the set of the ultrasonic echo detection signal semantic feature vectors is used as a sequence of analog key feature vectors. The query feature vector and the sequence of the analog key feature vectors are implicitly optimized by attention to obtain the ultrasonic detection single point-global query comparison implicit representation, including: performing one-dimensional convolution encoding on the set of ultrasonic echo detection signal semantic feature vectors to obtain a set of ultrasonic echo detection signal local neighborhood semantic association feature vectors, wherein the size of the one-dimensional convolution encoding is The length is equal to an integer multiple of the length of the ultrasonic echo detection signal semantic feature vector; the ultrasonic echo detection signal semantic feature vector corresponding to the first scanning position data is embedded and encoded to obtain an ultrasonic echo detection signal semantic value feature vector; the set of the ultrasonic echo detection signal semantic feature vectors and the set of the ultrasonic echo detection signal local neighborhood semantic association feature vectors are respectively used as a set of key vectors, and the ultrasonic echo detection signal semantic value feature vector is used as a value vector, and the query vector, the value vector and the set of the key vector are subjected to local neighborhood implicit optimization and attention-based gating processing to obtain the ultrasonic detection single point-global query comparison implicit representation.

[0014] In the above-mentioned production defect identification method for metal-rubber composite plates, the semantic feature vector of the ultrasonic echo detection signal corresponding to the first scanning position data is embedded and encoded to obtain the semantic value feature vector of the ultrasonic echo detection signal, including: using a value embedding coding matrix to process the semantic feature vector of the ultrasonic echo detection signal corresponding to the first scanning position data to obtain the semantic value feature vector of the ultrasonic echo detection signal.

[0015] In the above-mentioned production defect identification method for metal-rubber composite plates, the semantic feature vector of the ultrasonic echo detection signal corresponding to the first scanning position data is processed using a value embedding coding matrix to obtain the semantic value feature vector of the ultrasonic echo detection signal, including: multiplying the semantic feature vector of the ultrasonic echo detection signal corresponding to the first scanning position data with the value embedding coding matrix, and then adding the obtained ultrasonic echo detection signal embedding vector and the value bias vector by position to obtain the semantic value feature vector of the ultrasonic echo detection signal.

[0016] In the above-mentioned production defect identification method for metal-rubber composite plates, the set of ultrasonic echo detection signal semantic feature vectors and the set of ultrasonic echo detection signal local neighborhood semantic association feature vectors are respectively used as a set of key vectors, and the ultrasonic echo detection signal semantic value feature vector is used as a value vector, and the query vector, the value vector and the set of key vectors are subjected to local neighborhood implicit optimization and attention-based gating processing to obtain the ultrasonic detection single point-global query comparison implicit representation, including: using the ultrasonic echo detection signal semantic feature vector corresponding to the first scanning position data as the query vector, the ultrasonic echo detection signal semantic value feature vector as the value vector and the set of ultrasonic echo detection signal semantic feature vectors as the set of key vectors, inputting the query vector, the value vector and the set of key vectors into a first multi-head attention query module based on a converter structure to obtain a set of ultrasonic echo detection signal key granularity query matching implicit vectors; using the ultrasonic echo detection signal semantic feature vector corresponding to the first scanning position data as the query vector, the ultrasonic echo detection signal semantic value feature vector as the value vector and the set of ultrasonic echo detection signal semantic feature vectors as the set of key vectors; The semantic feature vector of the echo detection signal is used as the query vector, the semantic value feature vector of the ultrasonic echo detection signal is used as the value vector, and the set of the local neighborhood semantic association feature vectors of the ultrasonic echo detection signal is used as the set of key vectors. The query vector, the value vector and the set of the key vectors are input into a second multi-head attention query module based on a converter structure to obtain a set of ultrasonic echo detection signal key neighborhood granularity query matching implicit vectors; the positional mean vector of the set of ultrasonic echo detection signal key granularity query matching implicit vectors and the set of ultrasonic echo detection signal key neighborhood granularity query matching implicit vectors are calculated to obtain an ultrasonic echo detection signal key granularity query matching feature vector and an ultrasonic echo detection signal key neighborhood granularity query matching feature vector; the ultrasonic echo detection signal key granularity query matching feature vector and the ultrasonic echo detection signal key neighborhood granularity query matching feature vector are fused to obtain an ultrasonic detection single point-global query comparison implicit representation vector as the ultrasonic detection single point-global query comparison implicit representation.

[0017] In the above-mentioned production defect identification method for metal-rubber composite plates, the semantic feature vector of the ultrasonic echo detection signal corresponding to the first scanning position data is used as a query vector, the semantic value feature vector of the ultrasonic echo detection signal is used as a value vector, and the set of the ultrasonic echo detection signal semantic feature vectors is used as a set of key vectors. The query vector, the value vector and the set of key vectors are input into a first multi-head attention query module based on a converter structure to obtain a set of ultrasonic echo detection signal key granularity query matching implicit vectors, including: calculating the semantic feature vector of the ultrasonic echo detection signal corresponding to the first scanning position data and the ultrasonic echo detection signal; After obtaining the ultrasonic echo detection signal semantic association matrix by multiplying the transposed vectors of the ultrasonic echo detection signal semantic feature vectors, the ultrasonic echo detection signal semantic association matrix and the square root of the length of the ultrasonic echo detection signal semantic feature vector are divided by position to obtain the ultrasonic echo detection signal semantic association modulation matrix; the ultrasonic echo detection signal semantic association modulation matrix is ​​input into the softmax function to obtain the ultrasonic echo detection signal semantic association activation matrix; the ultrasonic echo detection signal semantic association activation matrix is ​​multiplied by the ultrasonic echo detection signal semantic value feature vector to obtain the ultrasonic echo detection signal key-granularity query matching implicit vector.

[0018] In the above-mentioned production defect identification method for metal-rubber composite plates, the ultrasonic echo detection signal key granularity query matching feature vector and the ultrasonic echo detection signal key neighborhood granularity query matching feature vector are fused to obtain the ultrasonic detection single point-global query comparison implicit representation vector, including: respectively calculating the position-by-position subtraction, position-by-position dot multiplication and position-by-position addition between the ultrasonic echo detection signal key granularity query matching feature vector and the ultrasonic echo detection signal key neighborhood granularity query matching feature vector to obtain the ultrasonic echo detection signal key granularity-key neighborhood granularity difference vector, the ultrasonic echo detection signal key granularity-key neighborhood granularity dot product vector and the ultrasonic echo detection signal key granularity-key neighborhood granularity difference vector. The ultrasonic echo detection signal key granularity-key neighborhood granularity sum vector is cascaded to obtain an ultrasonic echo detection signal key granularity-key neighborhood granularity cascade vector; after performing one-dimensional convolution encoding on the ultrasonic echo detection signal key granularity-key neighborhood granularity cascade vector, the obtained ultrasonic echo detection signal key granularity-key neighborhood granularity convolution encoding vector is subjected to local window-based maximum pooling processing to obtain the ultrasonic detection single point-global query comparison implicit representation vector.

[0019] In the above-mentioned production defect identification method for metal-rubber composite plates, the identification result is obtained based on the ultrasonic detection single point-global query comparison implicit representation, including: inputting the ultrasonic detection single point-global query comparison implicit representation vector into a classifier-based anomaly identifier to obtain the identification result.

[0020] According to another aspect of the present application, a production defect identification system for a metal-rubber composite plate is provided, comprising:

[0021] A data acquisition module related to defect detection of metal-rubber composite plates, used to use an ultrasonic probe to scan the metal-rubber composite plate to be detected to obtain a set of scanning position data and a set of ultrasonic echo detection signals;

[0022] An ultrasonic echo detection signal time domain feature extraction module, used for performing echo signal time domain feature extraction on each ultrasonic echo detection signal in the set of ultrasonic echo detection signals to obtain a set of ultrasonic echo detection signal semantic feature vectors;

[0023] An ultrasonic detection single point-global query comparison implicit representation generation module is used to perform local neighborhood gated query processing on the ultrasonic echo detection signal semantic feature vector corresponding to the first scanning position data in the set of scanning position data and the set of ultrasonic echo detection signal semantic feature vectors to obtain an ultrasonic detection single point-global query comparison implicit representation;

[0024] A metal-rubber composite plate defect recognition result generation module is used to obtain a recognition result based on the ultrasonic detection single point-global query comparison implicit representation, and the recognition result is used to indicate whether the first scanning position data is abnormal;

[0025] Among them, the ultrasonic detection single point-global query comparison implicit representation generation module is used to: use the ultrasonic echo detection signal semantic feature vector corresponding to the first scanning position data in the set of scanning position data as the query feature vector, and use the set of ultrasonic echo detection signal semantic feature vectors as the set of analog key feature vectors, and perform attention implicit optimization on the query feature vector and the set of analog key feature vectors to obtain the ultrasonic detection single point-global query comparison implicit representation.

[0026] This application has significant technical effects due to the adoption of the above technical solutions:

[0027] The production defect identification method and system for metal-rubber composite plates provided in the present application adopts signal processing and analysis technology based on deep learning to extract the time domain semantic features of each ultrasonic echo detection signal in the set of ultrasonic echo detection signals, so as to automatically judge whether the first scanning position data has abnormalities according to the query comparison between the semantic features of the ultrasonic echo detection signal corresponding to the first scanning position data in the set of scanning position data and the semantic features of each ultrasonic echo detection signal. In this way, tiny internal defects in metal-rubber composite plates that cannot be touched by traditional optical or contact detection can be detected, and defect detection through deep learning technology can improve the accuracy of detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other purposes, features and advantages of the present application will become more apparent. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0029] Figure 1 The present invention is a flowchart of a method for identifying production defects of a metal-rubber composite plate according to an embodiment of the present application.

[0030] Figure 2 The figure is a data flow diagram of a method for identifying production defects of a metal-rubber composite plate according to an embodiment of the present application.

[0031] Figure 3 In a production defect identification method for metal-rubber composite plates according to an embodiment of the present application, a semantic feature vector of an ultrasonic echo detection signal corresponding to the first scanning position data in the set of scanning position data is used as a query feature vector, and the set of the semantic feature vectors of the ultrasonic echo detection signal is used as a sequence of analog key feature vectors. Attention implicit optimization is performed on the query feature vector and the sequence of the analog key feature vectors to obtain a flowchart of an implicit representation of the ultrasonic detection single-point-global query comparison.

[0032] Figure 4 In the production defect identification method for metal-rubber composite plates according to an embodiment of the present application, a set of semantic feature vectors of the ultrasonic echo detection signal and a set of local neighborhood semantic association feature vectors of the ultrasonic echo detection signal are used as a set of key vectors, and the semantic value feature vector of the ultrasonic echo detection signal is used as a value vector, and local neighborhood implicit optimization and attention-based gating processing are performed on the query vector, the value vector and the set of key vectors to obtain a flowchart of the implicit representation of the ultrasonic detection single point-global query comparison.

[0033] Figure 5 The system block diagram of the production defect identification system for metal-rubber composite panels according to an embodiment of the present application is shown. DETAILED DESCRIPTION

[0034] 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.

[0035] Metal-rubber composite sheet is a composite material that combines the high strength and rigidity of metal with the elasticity and vibration damping properties of rubber. It is made by tightly combining a metal sheet with a rubber layer through a special process and is widely used in the automotive, aerospace, construction and other fields, especially in applications that require high stress, vibration absorption, sealing and insulation. However, metal-rubber composite sheets may produce a variety of defects during the production process, which can affect the quality and performance of the product. Therefore, identifying and controlling these defects is essential to ensure product quality and extend product life.

[0036] Traditional mechanical inspection equipment has limitations when inspecting metal-rubber composite panels. They can usually only detect surface defects, such as cracks or holes, but have poor detection effects on internal defects or more complex defect types. Specifically, traditional mechanical inspection equipment relies on contact measurement or simple optical imaging techniques, which have difficulty penetrating the interior of the material to detect hidden defects, such as internal delamination or poor bonding. In addition, these devices also have difficulties when dealing with complex geometries or irregular surfaces, which may lead to incomplete inspection results and may even cause damage to the sample.

[0037] In response to the above technical problems, the technical concept of the present application is to obtain a set of scanning position data and a set of ultrasonic echo detection signals by scanning the metal-rubber composite plate to be inspected with an ultrasonic probe, and to extract the time domain semantic features of each ultrasonic echo detection signal by using signal processing and analysis technology based on deep learning, so as to automatically judge whether there is an abnormality in the first scanning position data according to the query comparison between the semantic features of the ultrasonic echo detection signal corresponding to the first scanning position data in the set of scanning position data and the semantic features of each ultrasonic echo detection signal. In this way, through ultrasonic detection, it is possible to go deep into the material in a non-contact manner and detect tiny internal defects that cannot be reached by traditional optical or contact detection, and it will not be limited by the surface characteristics of the material. And extracting key features from complex ultrasonic echo signals through deep learning technology can improve the accuracy of detection and realize automated defect judgment.

[0038] Figure 1The present invention is a flowchart of a method for identifying production defects of a metal-rubber composite plate according to an embodiment of the present application. Figure 2 Schematic diagram of data flow for the method for identifying production defects of metal-rubber composite plates according to an embodiment of the present application. Figure 1 and Figure 2 As shown, according to the embodiment of the present application, the production defect identification method for metal-rubber composite plates includes: S110, using an ultrasonic probe to scan the metal-rubber composite plate to be inspected to obtain a set of scanning position data and a set of ultrasonic echo detection signals; S120, performing echo signal time domain feature extraction on each ultrasonic echo detection signal in the set of ultrasonic echo detection signals to obtain a set of ultrasonic echo detection signal semantic feature vectors; S130, performing local neighborhood gated query processing on the ultrasonic echo detection signal semantic feature vector corresponding to the first scanning position data in the set of scanning position data and the set of ultrasonic echo detection signal semantic feature vectors to obtain an ultrasonic detection single point-global query comparison implicit representation; S140, obtaining a recognition result based on the ultrasonic detection single point-global query comparison implicit representation, and the recognition result is used to indicate whether there is an abnormality in the first scanning position data.

[0039] In step S110, an ultrasonic probe is used to scan the metal-rubber composite plate to be inspected to obtain a set of scanning position data and a set of ultrasonic echo detection signals. It should be understood that an ultrasonic probe is a device that can emit and receive ultrasonic waves. In the process of scanning the metal-rubber composite plate to be inspected, the ultrasonic probe will emit ultrasonic pulses into the material and receive echo signals reflected from different interfaces inside the material (such as the interface between metal and rubber, internal defects, etc.), and these echo signals carry important information about the internal structure and characteristics of the material. Specifically, the ultrasonic probe is scanned point by point along the surface of the metal-rubber composite plate to be inspected, and each scanning position will record the ultrasonic echo signal and the corresponding position information. By sorting the scanning data, a set of scanning position data and a set of ultrasonic echo detection signals can be obtained. This ensures that the state of each position on the surface of the composite plate can be accurately recorded, and then analyzing it can improve the accuracy and reliability of defect detection of the metal-rubber composite plate.

[0040] In step S120, the echo signal time domain feature extraction is performed on each ultrasonic echo detection signal in the set of ultrasonic echo detection signals to obtain a set of ultrasonic echo detection signal semantic feature vectors. Specifically, in an embodiment of the present application, the echo signal time domain feature extraction is performed on each ultrasonic echo detection signal in the set of ultrasonic echo detection signals to obtain a set of ultrasonic echo detection signal semantic feature vectors, including: inputting each ultrasonic echo detection signal in the set of ultrasonic echo detection signals into an echo signal time domain feature extractor based on a 1D-CNN model to obtain a set of ultrasonic echo detection signal semantic feature vectors. It should be understood that ultrasonic echo signals are essentially one-dimensional time series data, which vary with time and contain information about the internal structure of the material, that is, each ultrasonic echo detection signal in the set of ultrasonic echo detection signals contains local semantic features in the time domain. Considering that 1D-CNN can effectively extract local features in time series signals through the one-dimensional convolution kernel of the convolution layer, and reduce the feature dimension through the pooling layer, thereby retaining the most important information, based on this, in the technical solution of the present application, each ultrasonic echo detection signal in the set of ultrasonic echo detection signals is input into an echo signal time domain feature extractor based on the 1D-CNN model to capture and refine the local time domain semantic features in each detection signal, and obtain a set of ultrasonic echo detection signal semantic feature vectors.

[0041] In step S130, the ultrasonic echo detection signal semantic feature vector corresponding to the first scanning position data in the set of scanning position data and the set of ultrasonic echo detection signal semantic feature vectors are subjected to local neighborhood gated query processing to obtain an ultrasonic detection single point-global query comparison implicit representation. Accordingly, considering that each ultrasonic echo detection signal semantic feature in the set of ultrasonic echo detection signal semantic feature vectors represents key semantic information of the ultrasonic echo detection signal at a position, such as reflection intensity, time delay, etc. Based on this, in order to compare the features of a specific position (for example, the first scanning position) with the features of all other positions in a concentrated manner, so as to more carefully evaluate whether significantly different features appear, thereby improving the accuracy of defect recognition, in the technical solution of the present application, the ultrasonic echo detection signal semantic feature vector corresponding to the first scanning position data in the set of scanning position data and the set of ultrasonic echo detection signal semantic feature vectors are subjected to local neighborhood gated query processing to obtain an ultrasonic detection single point-global query comparison implicit representation vector as an ultrasonic detection single point-global query comparison implicit representation. In particular, when processing ultrasonic signals, information from the local neighborhood is crucial to understanding the context of the signal, and through gated query processing, subtle changes in the signal can be captured, which is very useful for detecting subtle material defects.

[0042] Specifically, first, the set of the ultrasonic echo detection signal semantic feature vectors is subjected to one-dimensional convolution coding to obtain a set of ultrasonic echo detection signal local neighborhood semantic association feature vectors. In particular, the scale of the one-dimensional convolution coding here is equal to an integer multiple of the length of the ultrasonic echo detection signal semantic feature vector. It should be understood that the local neighborhood semantic features in the ultrasonic echo detection signal can be captured and mined by performing one-dimensional convolution coding. By setting the size of the convolution kernel to an integer multiple of the length of the ultrasonic echo detection signal semantic feature vector, it can be ensured that the convolution kernel can identify significant patterns that last for a certain period of time, thereby more effectively capturing the semantic connection between local neighborhoods, which helps to more accurately perform query matching in subsequent steps. Then, the ultrasonic echo detection signal semantic feature vector (query feature) corresponding to the first scanning position data is processed using the value embedding coding matrix to obtain the ultrasonic echo detection signal semantic value feature vector. That is, using the embedding technology, the query feature vector is mapped to a higher dimensional space, which enables these feature vectors to more accurately reflect the query intent in the subsequent attention module processing process and to achieve more efficient interaction with other feature vectors. Subsequently, the processed query vector, value vector and key vector are input into the multi-head attention query module based on the converter structure to obtain the set of ultrasonic echo detection signal key granularity query matching implicit vectors and the set of ultrasonic echo detection signal key neighborhood granularity query matching implicit vectors respectively. The multi-head attention query module can focus on different dimensions of features to identify multiple and complex connections between features, thereby enhancing the strength of feature expression. Specifically, when the first multi-head attention query module generates a set of ultrasonic echo detection signal key granularity query matching implicit vectors, it mainly focuses on the attributes of the key vector itself; while when the second multi-head attention query module generates a set of ultrasonic echo detection signal key neighborhood granularity query matching implicit vectors, it considers the characteristics of the key vector and its neighborhood at the same time, so as to comprehensively consider the local and overall information of the neighbors and improve the accuracy of query matching. Afterwards, the mean vector of the set of ultrasonic echo detection signal key granularity query matching implicit vectors and the set of ultrasonic echo detection signal key neighborhood granularity query matching implicit vectors is calculated to synthesize the global features of ultrasonic echo detection signals of different granularities to obtain the ultrasonic echo detection signal key granularity query matching feature vector and the ultrasonic echo detection signal key neighborhood granularity query matching feature vector. Finally, the matching features of the two are fused to synthesize the information of two different granularities, ensuring that the final ultrasonic detection single point-global query comparison implicit representation vector can fully capture the diverse interactive correlations between the query and the key vector.

[0043] Specifically, in an embodiment of the present application, a local neighborhood gated query processing is performed on the ultrasonic echo detection signal semantic feature vector corresponding to the first scanning position data in the set of scanning position data and the set of the ultrasonic echo detection signal semantic feature vectors to obtain an implicit representation of ultrasonic detection single point-global query comparison, including: S131, using the ultrasonic echo detection signal semantic feature vector corresponding to the first scanning position data in the set of scanning position data as a query feature vector, using the set of the ultrasonic echo detection signal semantic feature vectors as a set of analog key feature vectors, and performing attention implicit optimization on the query feature vector and the set of the analog key feature vectors to obtain an implicit representation of the ultrasonic detection single point-global query comparison.

[0044] More specifically, Figure 3 In the production defect identification method for metal-rubber composite plates according to the embodiment of the present application, the ultrasonic echo detection signal semantic feature vector corresponding to the first scanning position data in the set of scanning position data is used as the query feature vector, and the set of ultrasonic echo detection signal semantic feature vectors is used as the sequence of analog key feature vectors, and the query feature vector and the sequence of analog key feature vectors are implicitly optimized to obtain the flowchart of the ultrasonic detection single point-global query comparison implicit representation. Figure 3 As shown, the ultrasonic echo detection signal semantic feature vector corresponding to the first scanning position data in the set of scanning position data is used as the query feature vector, and the set of the ultrasonic echo detection signal semantic feature vectors is used as the sequence of analog key feature vectors, and the query feature vector and the sequence of the analog key feature vectors are implicitly optimized by attention to obtain the ultrasonic detection single point-global query comparison implicit representation, including: S1311, one-dimensional convolution encoding is performed on the set of the ultrasonic echo detection signal semantic feature vectors to obtain a set of ultrasonic echo detection signal local neighborhood semantic association feature vectors, wherein the scale of the one-dimensional convolution encoding is equal to the ultrasonic echo detection signal An integer multiple of the length of the semantic feature vector; S1312, embedding the ultrasonic echo detection signal semantic feature vector corresponding to the first scanning position data to obtain the ultrasonic echo detection signal semantic value feature vector; S1313, respectively using the set of the ultrasonic echo detection signal semantic feature vectors and the set of the ultrasonic echo detection signal local neighborhood semantic association feature vectors as the set of key vectors, and using the ultrasonic echo detection signal semantic value feature vector as the value vector, performing local neighborhood implicit optimization and attention-based gating processing on the query vector, the value vector and the set of the key vector to obtain the ultrasonic detection single point-global query comparison implicit representation.

[0045] More specifically, in an embodiment of the present application, the semantic feature vector of the ultrasonic echo detection signal corresponding to the first scanning position data is embedded and encoded to obtain a semantic value feature vector of the ultrasonic echo detection signal, including: using a value embedding coding matrix to process the semantic feature vector of the ultrasonic echo detection signal corresponding to the first scanning position data to obtain the semantic value feature vector of the ultrasonic echo detection signal.

[0046] More specifically, in an embodiment of the present application, a value embedding coding matrix is ​​used to process the semantic feature vector of the ultrasonic echo detection signal corresponding to the first scanning position data to obtain the semantic value feature vector of the ultrasonic echo detection signal, including: multiplying the semantic feature vector of the ultrasonic echo detection signal corresponding to the first scanning position data with the value embedding coding matrix, and then adding the obtained ultrasonic echo detection signal embedding vector and the value bias vector by position to obtain the semantic value feature vector of the ultrasonic echo detection signal.

[0047] More specifically, Figure 4 In the production defect identification method for metal-rubber composite plates according to the embodiment of the present application, the set of ultrasonic echo detection signal semantic feature vectors and the set of ultrasonic echo detection signal local neighborhood semantic association feature vectors are used as the set of key vectors, and the ultrasonic echo detection signal semantic value feature vector is used as the value vector, and the query vector, the value vector and the set of key vectors are subjected to local neighborhood implicit optimization and attention-based gating processing to obtain the flowchart of the ultrasonic detection single point-global query comparison implicit representation. Figure 4As shown, the set of the ultrasonic echo detection signal semantic feature vectors and the set of the ultrasonic echo detection signal local neighborhood semantic association feature vectors are respectively used as the set of key vectors, and the ultrasonic echo detection signal semantic value feature vector is used as the value vector, and the query vector, the value vector and the set of the key vectors are implicitly optimized in the local neighborhood and gated based on attention to obtain the ultrasonic detection single point-global query comparison implicit representation, including: S13131, the ultrasonic echo detection signal semantic feature vector corresponding to the first scanning position data is used as the query vector, the ultrasonic echo detection signal semantic value feature vector is used as the value vector and the set of the ultrasonic echo detection signal semantic feature vectors is used as the set of key vectors, and the query vector, the value vector and the set of the key vector are input into the first multi-head attention query module based on the converter structure to obtain the set of ultrasonic echo detection signal key granularity query matching implicit vectors; S13132, the ultrasonic echo detection signal semantic feature vector corresponding to the first scanning position data is used as the query vector, the ultrasonic echo detection signal semantic value feature vector is used as the value vector and the set of the ultrasonic echo detection signal semantic feature vectors is used as the key vector. The vector is used as a query vector, the ultrasonic echo detection signal semantic value feature vector is used as a value vector, and the set of the ultrasonic echo detection signal local neighborhood semantic association feature vectors is used as a set of key vectors, and the query vector, the value vector and the set of the key vectors are input into a second multi-head attention query module based on a converter structure to obtain a set of ultrasonic echo detection signal key neighborhood granularity query matching implicit vectors; S13133, calculating the positional mean vector of the set of ultrasonic echo detection signal key granularity query matching implicit vectors and the set of ultrasonic echo detection signal key neighborhood granularity query matching implicit vectors to obtain the ultrasonic echo detection signal key granularity query matching feature vector and the ultrasonic echo detection signal key neighborhood granularity query matching feature vector; S13134, fusing the ultrasonic echo detection signal key granularity query matching feature vector and the ultrasonic echo detection signal key neighborhood granularity query matching feature vector to obtain the ultrasonic detection single point-global query comparison implicit representation vector as the ultrasonic detection single point-global query comparison implicit representation.

[0048] More specifically, in an embodiment of the present application, the ultrasonic echo detection signal semantic feature vector corresponding to the first scanning position data is used as a query vector, the ultrasonic echo detection signal semantic value feature vector is used as a value vector, and the set of ultrasonic echo detection signal semantic feature vectors is used as a set of key vectors. The query vector, the value vector, and the set of key vectors are input into a first multi-head attention query module based on a converter structure to obtain a set of ultrasonic echo detection signal key-granularity query matching implicit vectors, including: calculating the ultrasonic echo detection signal semantic feature vector corresponding to the first scanning position data and the ultrasonic echo detection signal semantic feature vector. After obtaining the ultrasonic echo detection signal semantic association matrix by multiplying the transposed vectors of the eigenvectors, the ultrasonic echo detection signal semantic association matrix and the square root of the length of the ultrasonic echo detection signal semantic feature vector are divided by position to obtain the ultrasonic echo detection signal semantic association modulation matrix; the ultrasonic echo detection signal semantic association modulation matrix is ​​input into the softmax function to obtain the ultrasonic echo detection signal semantic association activation matrix; the ultrasonic echo detection signal semantic association activation matrix is ​​multiplied by the ultrasonic echo detection signal semantic value eigenvector to obtain the ultrasonic echo detection signal key-granularity query matching implicit vector.

[0049] More specifically, in an embodiment of the present application, the ultrasonic echo detection signal key granularity query matching feature vector and the ultrasonic echo detection signal key neighborhood granularity query matching feature vector are fused to obtain an ultrasonic detection single point-global query comparison implicit representation vector, including: respectively calculating the positional subtraction, positional dot multiplication and positional addition between the ultrasonic echo detection signal key granularity query matching feature vector and the ultrasonic echo detection signal key neighborhood granularity query matching feature vector to obtain an ultrasonic echo detection signal key granularity-key neighborhood granularity difference vector, an ultrasonic echo detection signal key granularity-key neighborhood granularity dot product vector and an ultrasonic echo detection signal key granularity -key neighborhood granularity sum vector; cascade the ultrasonic echo detection signal key granularity-key neighborhood granularity difference vector, the ultrasonic echo detection signal key granularity-key neighborhood granularity dot product vector and the ultrasonic echo detection signal key granularity-key neighborhood granularity sum vector to obtain an ultrasonic echo detection signal key granularity-key neighborhood granularity cascade vector; after performing one-dimensional convolution encoding on the ultrasonic echo detection signal key granularity-key neighborhood granularity cascade vector, perform local window-based maximum pooling processing on the obtained ultrasonic echo detection signal key granularity-key neighborhood granularity convolution encoding vector to obtain the ultrasonic detection single point-global query comparison implicit representation vector.

[0050] In the embodiment of the present application, specifically, the local neighborhood gated query processing is performed on the ultrasonic echo detection signal semantic feature vector corresponding to the first scanning position data in the set of scanning position data and the set of the ultrasonic echo detection signal semantic feature vectors, which can be expressed as:

[0051] K={k1,k2,...,k i ,...,k n}

[0052] K′=Conv 1×l ({k1,k2,...,k i ,...,k n})

[0053] K′={k1′,k2′,...,k i ′,...,k m ′}

[0054] v v =v q W q +b q

[0055]

[0056] Wherein, K represents the set of semantic feature vectors of the ultrasonic echo detection signal, k1, k2, ..., k i ,...,k n is the first, second, ..., i-th, ..., n-th ultrasonic echo detection signal semantic feature vector in the set of ultrasonic echo detection signal semantic feature vectors, Conv 1×l (·) is a one-dimensional convolutional code, l is an integer multiple of the length of the semantic feature vector of the ultrasonic echo detection signal, K′ is a set of local neighborhood semantic association feature vectors of the ultrasonic echo detection signal, k1′, k2′, ..., k i ′,...,k m ′ is the first, second, ..., i-th, ..., m-th ultrasonic echo detection signal local neighborhood semantic association feature vector in the set of ultrasonic echo detection signal local neighborhood semantic association feature vectors, v q is the semantic feature vector of the ultrasonic echo detection signal corresponding to the first scanning position data, W q is the value embedding encoding matrix, b q is the value bias vector, v v is the semantic value feature vector of the ultrasonic echo detection signal, k i T and k i ' T ki and k i ′, d1 and d2 are the lengths of the ultrasonic echo detection signal semantic feature vector and the ultrasonic echo detection signal local neighborhood semantic association feature vector, softmax(·) is the softmax function, v ki is the ith ultrasonic echo detection signal key granularity query matching implicit vector in the set of ultrasonic echo detection signal key granularity query matching implicit vectors, n is the number of feature vectors in the set of ultrasonic echo detection signal key granularity query matching implicit vectors, v ni is the ith ultrasonic echo detection signal key neighborhood granularity query matching implicit vector in the set of ultrasonic echo detection signal key neighborhood granularity query matching implicit vectors, m is the number of feature vectors in the set of ultrasonic echo detection signal key neighborhood granularity query matching implicit vectors, v k is the ultrasonic echo detection signal key granularity query matching feature vector, v n is the ultrasonic echo detection signal key neighborhood granularity query matching feature vector, ⊙ and are position-wise subtraction, position-wise multiplication, and position-wise addition, respectively. [·,·,·] is a cascade process. conv1D(·) is a one-dimensional convolutional encoding operation. MaxPool(·) is a maximum pooling operation. p It is the implicit representation vector of the ultrasonic detection single point-global query comparison.

[0057] In step S140, based on the ultrasonic detection single point-global query comparison implicit representation, a recognition result is obtained, and the recognition result is used to indicate whether the first scanning position data has an abnormality. Specifically, in an embodiment of the present application, based on the ultrasonic detection single point-global query comparison implicit representation, a recognition result is obtained, including: inputting the ultrasonic detection single point-global query comparison implicit representation vector into a classifier-based abnormality identifier to obtain the recognition result. That is, the ultrasonic detection single point-global query comparison implicit representation obtained by performing a global query comparison using the set of semantic feature vectors of the ultrasonic echo detection signal is classified and processed, so as to automatically determine whether the first scanning position data has an abnormality. In this way, through ultrasonic detection, it is possible to penetrate into the material in a non-contact manner and detect tiny internal defects that cannot be reached by traditional optical or contact detection, and it will not be limited by the surface characteristics of the material. And extracting key features from complex ultrasonic echo signals through deep learning technology can improve the accuracy of detection and realize automated defect judgment.

[0058] In particular, in the technical solution of the present application, each ultrasonic echo detection signal semantic feature vector in the set of ultrasonic echo detection signal semantic feature vectors respectively represents the signal waveform semantic features of each ultrasonic echo detection signal at each position. Therefore, when the ultrasonic echo detection signal semantic feature vector corresponding to the first scanning position data in the set of scanning position data is used as the query feature vector and the set of ultrasonic echo detection signal semantic feature vectors is used as the sequence of analog key feature vectors, and the query feature vector and the sequence of analog key feature vectors are subjected to local neighborhood gated query processing, the signal noise in each ultrasonic echo detection signal in the set of ultrasonic echo detection signals will affect the accuracy of extracting the association information between the query feature vector and the analog key feature vector and the accuracy of extracting the association information between the analog key feature vectors, so that the ultrasonic detection single point-global query comparison implicit representation vector has a fine-grained query response aggregation difference caused by the difference in fine-grained query encoding response weights, so it is expected to improve the detail semantic aggregation expression effect of the fine-grained query response aggregation difference of the ultrasonic detection single point-global query comparison implicit representation vector.

[0059] Preferably, inputting the ultrasonic detection single point-global query comparison implicit representation vector into a classifier-based anomaly identifier to obtain an identification result includes: calculating the sum of the absolute values ​​of each eigenvalue of the ultrasonic detection single point-global query comparison implicit representation vector to obtain a first ultrasonic detection single point-global query comparison implicit and modulation value, and calculating the square root of the sum of the squares of each eigenvalue of the ultrasonic detection single point-global query comparison implicit representation vector to obtain a second ultrasonic detection single point-global query comparison implicit and modulation value; performing dot-wise subtraction on the ultrasonic detection single point-global query comparison implicit representation vector and the second ultrasonic detection single point-global query comparison implicit and modulation value, respectively performing dot-wise multiplication with the number of eigenvalues ​​of the ultrasonic detection single point-global query comparison implicit representation vector and the reciprocal of the first ultrasonic detection single point-global query comparison implicit and modulation value, and taking the reciprocal of each eigenvalue to obtain the first ultrasonic detection single point-global query comparison implicit and modulation value. domain query comparison implicit phase conversion vector; after dot-wise subtraction of the ultrasonic detection single point-global query comparison implicit representation vector and the first ultrasonic detection single point-global query comparison implicit and modulation value, respectively perform dot multiplication with the square root of the number of eigenvalues ​​of the ultrasonic detection single point-global query comparison implicit representation vector and the reciprocal of the second ultrasonic detection single point-global query comparison implicit and modulation value, and take the reciprocal of each eigenvalue to obtain the second ultrasonic detection single point-global query comparison implicit phase conversion vector; subtract the dot product vector of the second ultrasonic detection single point-global query comparison implicit phase conversion vector and the weighted hyperparameter from the first ultrasonic detection single point-global query comparison implicit phase conversion vector to obtain an optimized ultrasonic detection single point-global query comparison implicit representation vector; input the optimized ultrasonic detection single point-global query comparison implicit representation vector into the classifier-based anomaly identifier to obtain the recognition result.

[0060] Here, the optimization expression of the implicit representation vector of the ultrasonic detection single point-global query comparison is:

[0061]

[0062]

[0063] v i ∈V∈R n

[0064] Wherein, V represents the implicit representation vector of the ultrasonic detection single point-global query comparison, v irepresents the eigenvalue of the ith position of the ultrasonic detection single point-global query comparison implicit representation vector, n represents the number of eigenvalues ​​of the ultrasonic detection single point-global query comparison implicit representation vector, α represents the first ultrasonic detection single point-global query comparison implicit and modulation value, β represents the second ultrasonic detection single point-global query comparison implicit and modulation value, ⊙ is the point product by position, Indicates point-by-point subtraction, [·] ⊙-1 Take the inverse of each eigenvalue in the vector, V1 represents the implicit phase transition vector of the first ultrasonic detection single point-global query comparison, V2 represents the implicit phase transition vector of the second ultrasonic detection single point-global query comparison, ω represents the weighted hyperparameter, and V′ represents the optimized ultrasonic detection single point-global query comparison implicit representation vector.

[0065] In this preferred embodiment, the difference of the characteristic value of the ultrasonic detection single point-global query comparison implicit representation vector relative to the vector feature set of the ultrasonic detection single point-global query comparison implicit representation vector as a whole and the difference of the modulation representation is used as the semantic change intensity information, and the phase-like conversion corresponding to the position-based intensity modulation is performed through the different and modulation representation forms, so that the aggregation enhancement of the semantic change phase perception can improve the axial aggregation receptive field along the feature aggregation direction by performing the space translation operation based on alternating stacking under the vector set scale balance of the ultrasonic detection single point-global query comparison implicit representation vector, thereby improving the perception effect of the aggregation semantics of the ultrasonic detection single point-global query comparison implicit representation vector for the detail semantic change, so as to improve the expression effect of the ultrasonic detection single point-global query comparison implicit representation vector, and improve the accuracy of the recognition result obtained by the abnormal identifier based on the classifier. In this way, through ultrasonic detection, it is possible to go deep into the material in a non-contact manner, detect tiny internal defects that cannot be touched by traditional optical or contact detection, and will not be limited by the surface characteristics of the material. In addition, deep learning technology can be used to extract key features from complex ultrasonic echo signals to improve detection accuracy and achieve automated defect judgment.

[0066] In summary, the production defect identification method for metal-rubber composite plates based on the embodiment of the present application is explained, which uses signal processing and analysis technology based on deep learning to extract the time domain semantic features of each ultrasonic echo detection signal in the set of ultrasonic echo detection signals, so as to automatically determine whether the first scanning position data is abnormal based on the query comparison between the semantic features of the ultrasonic echo detection signal corresponding to the first scanning position data in the set of scanning position data and the semantic features of each ultrasonic echo detection signal. In this way, tiny internal defects in metal-rubber composite plates that cannot be reached by traditional optical or contact detection can be detected, and defect detection through deep learning technology can improve the accuracy of detection.

[0067] Figure 5 FIG. 1 is a system block diagram of a production defect identification system for a metal-rubber composite plate according to an embodiment of the present application. Figure 5 As shown, according to an embodiment of the present application, a production defect identification system 100 for metal-rubber composite plates includes: a metal-rubber composite plate defect detection related data acquisition module 110, which is used to use an ultrasonic probe to scan the metal-rubber composite plate to be detected to obtain a set of scanning position data and a set of ultrasonic echo detection signals; an ultrasonic echo detection signal time domain feature extraction module 120, which is used to extract echo signal time domain features of each ultrasonic echo detection signal in the set of ultrasonic echo detection signals to obtain a set of ultrasonic echo detection signal semantic feature vectors; an ultrasonic detection single point-global query comparison implicit representation generation module 130, which is used to extract the ultrasonic echo detection signal semantic feature vector corresponding to the first scanning position data in the set of scanning position data and the ultrasonic echo detection signal semantic feature vector. The set is subjected to local neighborhood gated query processing to obtain an ultrasonic detection single point-global query comparison implicit representation; a metal-rubber composite plate defect recognition result generation module 140 is used to obtain a recognition result based on the ultrasonic detection single point-global query comparison implicit representation, and the recognition result is used to indicate whether there is an abnormality in the first scanning position data; wherein the ultrasonic detection single point-global query comparison implicit representation generation module is used to: use the ultrasonic echo detection signal semantic feature vector corresponding to the first scanning position data in the set of scanning position data as the query feature vector, use the set of the ultrasonic echo detection signal semantic feature vectors as the set of analog key feature vectors, and perform attention implicit optimization on the query feature vector and the set of the analog key feature vectors to obtain the ultrasonic detection single point-global query comparison implicit representation.

[0068] Here, those skilled in the art will appreciate that the specific functions and operations of the various units and modules in the above-mentioned production defect identification system 100 for metal-rubber composite panels have been described in detail above. Figures 1 to 4 The description of the production defect identification method for the metal-rubber composite plate has been described in detail, and therefore, its repeated description will be omitted.

[0069] In summary, the production defect identification system 100 for metal-rubber composite plates based on the embodiment of the present application is explained, which uses signal processing and analysis technology based on deep learning to extract time domain semantic features of each ultrasonic echo detection signal in a set of ultrasonic echo detection signals, so as to automatically determine whether the first scanning position data is abnormal based on the query comparison between the semantic features of the ultrasonic echo detection signal corresponding to the first scanning position data in the set of scanning position data and the semantic features of each ultrasonic echo detection signal. In this way, tiny internal defects in metal-rubber composite plates that cannot be reached by traditional optical or contact detection can be detected, and defect detection through deep learning technology can improve the accuracy of detection.

[0070] As described above, the production defect identification system 100 for metal-rubber composite plates according to the embodiment of the present application can be implemented in various wireless terminals, such as a server for production defect identification of metal-rubber composite plates. In one example, the production defect identification system 100 for metal-rubber composite plates according to the embodiment of the present application can be integrated into the wireless terminal as a software module and / or a hardware module. For example, the production defect identification system 100 for metal-rubber composite plates can be a software module in the operating system of the wireless terminal, or can be an application developed for the wireless terminal; of course, the production defect identification system 100 for metal-rubber composite plates can also be one of the many hardware modules of the wireless terminal.

[0071] Alternatively, in another example, the production defect identification system 100 for metal-rubber composite panels and the wireless terminal may also be separate devices, and the production defect identification system 100 for metal-rubber composite panels may be connected to the wireless terminal via a wired and / or wireless network and transmit interactive information in accordance with an agreed data format.

[0072] The above is only a preferred embodiment of the present application and does not constitute any form of limitation to the present application. Although the present application has been disclosed as a preferred embodiment as above, it is not intended to limit the present application. Any technician familiar with the profession can make some changes or modifications to equivalent embodiments of the technical contents disclosed above without departing from the scope of the technical solution of the present application. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present application without departing from the content of the technical solution of the present application still falls within the scope of the technical solution of the present application.

Claims

1. A method for identifying production defects of metal rubber composite panels, characterized in that: include: Scanning the metal-rubber composite plate to be inspected using an ultrasonic probe to obtain a set of scanning position data and a set of ultrasonic echo detection signals; Extracting echo signal time domain features from each ultrasonic echo detection signal in the set of ultrasonic echo detection signals to obtain a set of ultrasonic echo detection signal semantic feature vectors; Performing local neighborhood gated query processing on the ultrasonic echo detection signal semantic feature vector corresponding to the first scanning position data in the set of scanning position data and the set of ultrasonic echo detection signal semantic feature vectors to obtain an ultrasonic detection single point-global query comparison implicit representation; Based on the ultrasonic detection single point-global query comparison implicit representation, a recognition result is obtained, and the recognition result is used to indicate whether the first scanning position data has an abnormality; Among them, the ultrasonic echo detection signal semantic feature vector corresponding to the first scanning position data in the set of scanning position data and the set of the ultrasonic echo detection signal semantic feature vectors are subjected to local neighborhood gated query processing to obtain an implicit representation of ultrasonic detection single point-global query comparison, including: using the ultrasonic echo detection signal semantic feature vector corresponding to the first scanning position data in the set of scanning position data as a query feature vector, using the set of ultrasonic echo detection signal semantic feature vectors as a set of analog key feature vectors, and performing attention implicit optimization on the query feature vector and the set of analog key feature vectors to obtain the implicit representation of ultrasonic detection single point-global query comparison.

2. The method for identifying production defects of metal-rubber composite panels according to claim 1, characterized in that: Performing echo signal time domain feature extraction on each ultrasonic echo detection signal in the set of ultrasonic echo detection signals to obtain a set of ultrasonic echo detection signal semantic feature vectors, including: inputting each ultrasonic echo detection signal in the set of ultrasonic echo detection signals into an echo signal time domain feature extractor based on a 1D-CNN model to obtain a set of ultrasonic echo detection signal semantic feature vectors.

3. The method for identifying production defects of metal-rubber composite panels according to claim 2, characterized in that: The ultrasonic echo detection signal semantic feature vector corresponding to the first scanning position data in the set of scanning position data is used as a query feature vector, the set of the ultrasonic echo detection signal semantic feature vectors is used as a sequence of analog key feature vectors, and the query feature vector and the sequence of analog key feature vectors are implicitly optimized by attention to obtain the ultrasonic detection single point-global query comparison implicit representation, including: Performing one-dimensional convolution coding on the set of ultrasonic echo detection signal semantic feature vectors to obtain a set of ultrasonic echo detection signal local neighborhood semantic association feature vectors, wherein the scale of the one-dimensional convolution coding is equal to an integer multiple of the length of the ultrasonic echo detection signal semantic feature vector; Embedding the semantic feature vector of the ultrasonic echo detection signal corresponding to the first scanning position data to obtain a semantic value feature vector of the ultrasonic echo detection signal; The set of the ultrasonic echo detection signal semantic feature vectors and the set of the ultrasonic echo detection signal local neighborhood semantic association feature vectors are respectively used as the set of key vectors, and the ultrasonic echo detection signal semantic value feature vector is used as the value vector. Local neighborhood implicit optimization and attention-based gating processing are performed on the query vector, the value vector and the set of key vectors to obtain the ultrasonic detection single point-global query comparison implicit representation.

4. The method for identifying production defects of metal-rubber composite panels according to claim 3, characterized in that: The ultrasonic echo detection signal semantic feature vector corresponding to the first scanning position data is embedded and encoded to obtain the ultrasonic echo detection signal semantic value feature vector, including: using a value embedding coding matrix to process the ultrasonic echo detection signal semantic feature vector corresponding to the first scanning position data to obtain the ultrasonic echo detection signal semantic value feature vector.

5. The method for identifying production defects of metal-rubber composite panels according to claim 4, characterized in that: The ultrasonic echo detection signal semantic feature vector corresponding to the first scanning position data is processed using a value embedding coding matrix to obtain the ultrasonic echo detection signal semantic value feature vector, including: multiplying the ultrasonic echo detection signal semantic feature vector corresponding to the first scanning position data with the value embedding coding matrix, and then adding the obtained ultrasonic echo detection signal embedding vector and the value bias vector by position to obtain the ultrasonic echo detection signal semantic value feature vector.

6. The method for identifying production defects of metal-rubber composite panels according to claim 5, characterized in that: The set of ultrasonic echo detection signal semantic feature vectors and the set of ultrasonic echo detection signal local neighborhood semantic association feature vectors are respectively used as a set of key vectors, and the ultrasonic echo detection signal semantic value feature vector is used as a value vector, and the query vector, the value vector and the set of key vectors are subjected to local neighborhood implicit optimization and attention-based gating processing to obtain the ultrasonic detection single point-global query comparison implicit representation, including: Taking the ultrasonic echo detection signal semantic feature vector corresponding to the first scanning position data as a query vector, the ultrasonic echo detection signal semantic value feature vector as a value vector and the set of ultrasonic echo detection signal semantic feature vectors as a set of key vectors, the query vector, the value vector and the set of key vectors are input into a first multi-head attention query module based on a converter structure to obtain a set of ultrasonic echo detection signal key granularity query matching implicit vectors; Taking the ultrasonic echo detection signal semantic feature vector corresponding to the first scanning position data as a query vector, the ultrasonic echo detection signal semantic value feature vector as a value vector and the set of ultrasonic echo detection signal local neighborhood semantic association feature vectors as a set of key vectors, the query vector, the value vector and the set of key vectors are input into a second multi-head attention query module based on a converter structure to obtain a set of ultrasonic echo detection signal key neighborhood granularity query matching implicit vectors; Calculate the position mean vector of the set of ultrasonic echo detection signal key granularity query matching implicit vectors and the set of ultrasonic echo detection signal key neighborhood granularity query matching implicit vectors to obtain ultrasonic echo detection signal key granularity query matching feature vectors and ultrasonic echo detection signal key neighborhood granularity query matching feature vectors; The ultrasonic echo detection signal key granularity query matching feature vector and the ultrasonic echo detection signal key neighborhood granularity query matching feature vector are fused to obtain an ultrasonic detection single point-global query comparison implicit representation vector as the ultrasonic detection single point-global query comparison implicit representation.

7. The method for identifying production defects of metal-rubber composite panels according to claim 6, characterized in that: Taking the ultrasonic echo detection signal semantic feature vector corresponding to the first scanning position data as a query vector, the ultrasonic echo detection signal semantic value feature vector as a value vector and the set of ultrasonic echo detection signal semantic feature vectors as a set of key vectors, the query vector, the value vector and the set of key vectors are input into a first multi-head attention query module based on a converter structure to obtain a set of ultrasonic echo detection signal key granularity query matching implicit vectors, including: After calculating the product between the ultrasonic echo detection signal semantic feature vector corresponding to the first scanning position data and the transposed vector of the ultrasonic echo detection signal semantic feature vector to obtain an ultrasonic echo detection signal semantic association matrix, the ultrasonic echo detection signal semantic association matrix is ​​divided by the square root of the length of the ultrasonic echo detection signal semantic feature vector by position to obtain an ultrasonic echo detection signal semantic association modulation matrix; Inputting the ultrasonic echo detection signal semantic association modulation matrix into a softmax function to obtain an ultrasonic echo detection signal semantic association activation matrix; The ultrasonic echo detection signal semantic association activation matrix is ​​multiplied by the ultrasonic echo detection signal semantic value feature vector to obtain the ultrasonic echo detection signal key granularity query matching implicit vector.

8. The method for identifying production defects of metal-rubber composite panels according to claim 7, characterized in that: The ultrasonic echo detection signal key granularity query matching feature vector and the ultrasonic echo detection signal key neighborhood granularity query matching feature vector are integrated to obtain an ultrasonic detection single point-global query comparison implicit representation vector, including: Respectively calculating the positional subtraction, positional dot product and positional addition between the ultrasonic echo detection signal key granularity query matching feature vector and the ultrasonic echo detection signal key neighborhood granularity query matching feature vector to obtain an ultrasonic echo detection signal key granularity-key neighborhood granularity difference vector, an ultrasonic echo detection signal key granularity-key neighborhood granularity dot product vector and an ultrasonic echo detection signal key granularity-key neighborhood granularity sum vector; Cascade processing is performed on the ultrasonic echo detection signal key granularity-key neighborhood granularity difference vector, the ultrasonic echo detection signal key granularity-key neighborhood granularity dot product vector and the ultrasonic echo detection signal key granularity-key neighborhood granularity sum vector to obtain an ultrasonic echo detection signal key granularity-key neighborhood granularity cascade vector; After performing one-dimensional convolution encoding on the ultrasonic echo detection signal key granularity-key neighborhood granularity cascade vector, the obtained ultrasonic echo detection signal key granularity-key neighborhood granularity convolution encoding vector is subjected to local window-based maximum pooling processing to obtain the ultrasonic detection single point-global query comparison implicit representation vector.

9. The method for identifying production defects of metal-rubber composite panels according to claim 8, characterized in that: Based on the ultrasonic detection single point-global query comparison implicit representation, obtaining a recognition result includes: inputting the ultrasonic detection single point-global query comparison implicit representation vector into a classifier-based anomaly identifier to obtain the recognition result.

10. A production defect identification system for metal rubber composite panels, characterized in that: include: A data acquisition module related to defect detection of metal-rubber composite plates, used to use an ultrasonic probe to scan the metal-rubber composite plate to be detected to obtain a set of scanning position data and a set of ultrasonic echo detection signals; An ultrasonic echo detection signal time domain feature extraction module, used for performing echo signal time domain feature extraction on each ultrasonic echo detection signal in the set of ultrasonic echo detection signals to obtain a set of ultrasonic echo detection signal semantic feature vectors; An ultrasonic detection single point-global query comparison implicit representation generation module is used to perform local neighborhood gated query processing on the ultrasonic echo detection signal semantic feature vector corresponding to the first scanning position data in the set of scanning position data and the set of ultrasonic echo detection signal semantic feature vectors to obtain an ultrasonic detection single point-global query comparison implicit representation; A metal-rubber composite plate defect recognition result generation module is used to obtain a recognition result based on the ultrasonic detection single point-global query comparison implicit representation, and the recognition result is used to indicate whether the first scanning position data is abnormal; Among them, the ultrasonic detection single point-global query comparison implicit representation generation module is used to: use the ultrasonic echo detection signal semantic feature vector corresponding to the first scanning position data in the set of scanning position data as the query feature vector, and use the set of ultrasonic echo detection signal semantic feature vectors as the set of analog key feature vectors, and perform attention implicit optimization on the query feature vector and the set of analog key feature vectors to obtain the ultrasonic detection single point-global query comparison implicit representation.

Images