Infrared-based airborne electronic system circuit board operational amplifier chip fault diagnosis method

Abstract

The invention relates to the technical field of circuit board fault diagnosis, in particular to an infrared-based airborne electronic system circuit board operational amplifier chip fault diagnosis method, which comprises the following steps of: acquiring infrared sequence images II and corresponding sequence temperature matrixes TM of an airborne electronic system circuit board in different working modes; and S2, preprocessing the infrared sequence image II in the step S1, and sending the preprocessed infrared sequence image II into a semantic segmentation model to obtain an infrared segmentation image III of the key electronic component-operational amplifier chip in the region of interest. According to the invention, by considering the temperature change trend and fault category information of the temperature sequence and utilizing the infrared image and the deep learning algorithm to carry out fault classification, the problem of inaccurate fault detection caused by factors such as external environment change, unobvious fault category temperature difference and component aging in the prior art is effectively solved; and the accuracy and generalization of fault detection are improved.

Description

Technical Field

[0001] The invention relates to the technical field of circuit board fault diagnosis, and in particular to an infrared-based method for diagnosing operational amplifier chip faults on an airborne electronic system circuit board. Background Art

[0002] With the rapid development of avionics technology, the miniaturization and integration of airborne electronic system circuit boards are constantly improving, which brings challenges to the data extraction of components on circuit boards and circuit fault diagnosis. The traditional contact circuit board fault diagnosis method can no longer meet the detection and maintenance needs of increasingly complex airborne electronic system circuit boards. Infrared thermal imaging detection technology has been widely used due to its advantages such as non-contact, fast, and intuitive detection results.

[0003] For example, a method for diagnosing circuit board faults based on visible light images and infrared thermal image sequences provided by publication number CN116429783A collects infrared images and visible light images of a normal circuit board and a circuit board to be tested, respectively, and calculates the thermal feature vectors of each device on the two circuit boards after heterogeneous registration of the infrared images and visible light images of the two circuit boards; then, the thermal similarity of each device between the thermal feature vector of the normal circuit board and the thermal feature vector of the circuit board to be tested is calculated, and it is determined whether the thermal similarity exceeds the performance failure threshold of the device, thereby determining the fault state of the circuit board to be tested.

[0004] However, when the above technology is actually used, the temperature value of the normal circuit board under different external environments is not a constant and may even vary greatly; the temperature difference of some fault categories is only different from the normal state in a certain time segment, and after reaching a steady state, the heating difference between the two is not much. In addition, the same type of circuit boards may also have large temperature differences under normal working conditions due to different aging degrees of components. This makes the threshold setting method inaccurate and easily affected by the environment, resulting in low fault diagnosis efficiency and poor accuracy, which limits its effectiveness in practical applications. Summary of the invention

[0005] The purpose of the present invention is to provide an infrared-based method for diagnosing faults of operational amplifier chips on circuit boards of airborne electronic systems, so as to solve the problem that the above-mentioned threshold setting method is not accurate enough and is easily affected by the environment, resulting in low fault diagnosis efficiency and poor accuracy, which limits its effect in practical applications.

[0006] To achieve the above object, the present invention provides the following technical solution: an infrared-based method for diagnosing faults of operational amplifier chips on circuit boards of airborne electronic systems, comprising the following steps:

[0007] S1. Data collection: Collect infrared sequence images of the airborne electrical system circuit board in different working modes I I And the corresponding sequence temperature matrix TM ;

[0008] S2, data preprocessing: the infrared sequence image I of step S1 is I After preprocessing, it is sent to the semantic segmentation model to obtain the infrared segmentation image I of the key electronic components in the area of ​​interest - op amp chips II ;

[0009] S3, temperature sequence matrix acquisition: Combine the sequence temperature matrix T in step S1 M and the infrared segmentation image I of the operational amplifier chip in step S2 II , get the temperature sequence matrix T of the operational amplifier chip I ;

[0010] S4. Temperature state area division and data set extraction: Based on the working time, the temperature state of the electronic components in the area of ​​interest of the airborne electronic system circuit board - the operational amplifier chip is divided into a dynamic temperature rise area and a steady-state constant temperature area, and the temperature sequence matrix T is extracted. I The data of the part where the temperature continues to rise is taken as the dynamic temperature rising zone data set and is recorded as T ID , the data set T ID Further processing into multi-dimensional temperature sequence data T; extracting infrared segmentation image I of operational amplifier chip II The image of the stable temperature part is taken as the image data set of the steady-state constant temperature area and is denoted as I IS ;

[0011] S5, data set processing and model training: The multidimensional temperature series data in step S4 is processed. and image dataset of steady-state constant temperature region I IS Take 70% of them as training set and name them as T train and I train ; The remaining 30% of the data are set as validation sets and named T val and I val , and multidimensional temperature series data The training set is input into the long short-term memory time series feature extraction model with fusion attention mechanism to extract classification features. At the same time, the image dataset I of the steady-state constant temperature area IS The training set is input into the infrared thermal image classification feature extraction model to extract classification features;

[0012] S6. Feature fusion and model classification training: Fusion of the multidimensional temperature series data T obtained in S5 and the image data set I of the steady-state constant temperature zone IS The two types of feature information are input into the multi-layer perceptron in the model to complete the classification training of the model;

[0013] S7. Fault classification and diagnosis: Input the known category data set into the trained model to obtain the classification of component faults in the area of ​​interest of the airborne electronic system circuit board. Based on the classification results, diagnose the faults of operational amplifier chips on the airborne electronic system circuit board.

[0014] Preferably, in step S1, the infrared sequence image I I And the corresponding sequence temperature matrix T M The acquisition is achieved by using multiple infrared thermal imagers that are perpendicular to the circuit board of the airborne electronic system to be tested, wherein the circuit board of the airborne electronic system to be tested is connected to the excitation source through a programmable relay switch, and the host computer is connected to the infrared camera and the programmable relay switch respectively.

[0015] Preferably, in step S2, the infrared sequence image I I The preprocessing includes but is not limited to translation, rotation and mirror flipping. The semantic segmentation model in step S2 is a DeepLab v3+ semantic segmentation model using a Mobilenet network.

[0016] Preferably, the serial temperature matrix T of the operational amplifier chip in step S4 I Determine the start time t0 of the dynamic temperature rise zone, the start time t1 of the stable thermal equilibrium state, and the stop working time t2, define t0-t1 as the dynamic temperature rise zone, and define t1-t2 as the steady-state constant temperature zone.

[0017] Preferably, in step S5, T train The specific steps of extracting classification features from the long short-term memory time series feature extraction model that integrates the attention mechanism are as follows:

[0018] For the training set T train Perform data preprocessing and obtain data set T train The expectation μ and variance σ of each dimension in , convert each data point into a standard score, as shown in the formula:

[0019]

[0020] Among them, z i,t is the temperature data of each sequence point in the i-th dimension after standardization; x i is the temperature data of each sequence point in the i-th dimension in the training set;

[0021] After that, the processed T train The data is input into the long short-term memory time series classification feature extraction model integrating the attention mechanism to extract classification features.

[0022] Compared with the prior art, the present invention has the following beneficial effects:

[0023] 1. The present invention effectively solves the problem of inaccurate fault detection caused by factors such as external environment changes, unclear temperature differences between fault categories and aging of components in the prior art by considering the temperature change trend and fault category information of the temperature sequence, and using infrared images and deep learning algorithms for fault classification, thereby improving the accuracy and generalization of fault detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 The present invention is a flow chart of the overall structure of the infrared-based airborne electronic system circuit board operational amplifier chip fault diagnosis method. DETAILED DESCRIPTION

[0025] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0026] This embodiment selects the airborne weather radar system circuit board power module as the test target.

[0027] See also Figure 1 The present invention provides a technical solution: an infrared-based method for diagnosing faults of operational amplifier chips on a circuit board of an airborne electronic system, comprising the following steps:

[0028] S1. Data acquisition: First, an infrared image acquisition system is built, which includes multiple infrared thermal imagers. The circuit board of the airborne electronic system to be tested is placed flat on a horizontal plane, and the infrared thermal imager lens is vertical to the top of the circuit board of the airborne electronic system to be tested. The circuit board of the airborne electronic system to be tested is connected to the excitation source through a programmable relay switch. The host computer is respectively connected to the infrared camera and the programmable relay switch. The host computer is used to control the programmable relay and the infrared camera to open and close synchronously to ensure that the working state of the circuit board of the airborne electronic system to be tested is synchronized with the working state of the infrared camera. That is, in the present invention, only the infrared image information of the circuit board of the airborne electronic system in the working state is directionally acquired. When the circuit board of the airborne electronic system works in different working modes, the infrared images of the circuit board of the airborne electronic system in different working modes are obtained through the infrared image acquisition system and named I I , and sent back to the host computer for storage, the collected infrared image information is processed to obtain the temperature information corresponding to each pixel of the infrared image of the airborne electronic system circuit board. The processed temperature information is in the form of a two-dimensional data matrix, that is, the sequence temperature matrix corresponding to each pixel of the infrared sequence image of the airborne electronic system circuit board under different working modes, recorded as T M, where the size of the infrared image of the airborne electrical system circuit board is 320×256, the temperature accuracy of each pixel is ±1%, and the temperature resolution of each pixel is less than or equal to 0.015K.

[0029] S2. Data preprocessing: Use Labelme labeling software to annotate the infrared sequence images I of the airborne electrical system circuit board collected in step S1. I The key electronic components in the area of ​​interest, the op amp chip, are annotated. After the image annotation is completed, the image transformation methods such as translation, rotation, and mirror flipping are used for data preprocessing. The preprocessed infrared sequence image I I The improved DeepLab v3+ semantic segmentation model is used to obtain the infrared segmentation image I of the key electronic components in the area of ​​interest. II , where the improved DeepLab v3+ semantic segmentation model consists of two parts: an encoder and a decoder.

[0030] The encoder part should include the following parts: backbone network Mobilenet, void space pyramid pooling module, convolutional block attention module, upsampling module;

[0031] The decoder part should include the following parts: 1×1 convolution module, feature fusion module and upsampling module.

[0032] The present invention is mainly applied to the field of infrared images of circuit boards. Therefore, the present invention replaces the Xception backbone network of the original DeepLab v3+ semantic segmentation model with a more lightweight Mobilenet network, so as to improve the model running speed and reduce the complexity of the algorithm; the atrous space pyramid pooling module can adjust the sampling ratio and receptive field, and can capture targets and related contextual information at multiple scales; the convolutional block attention module can highlight important features and suppress unimportant features (here mainly refers to the ability to highlight the segmentation of target components and suppress the segmentation of background), thereby improving the segmentation performance of the model.

[0033] S3, combined sequence temperature matrix T M Infrared segmentation image of key electronic components in the area of ​​interest - op amp chips I II , get the temperature sequence matrix T of the key electronic components in the area of ​​interest - op amp chips I , where the temperature sequence matrix T I The specific cargo volume area method is as follows:

[0034] Infrared segmentation image I of the key electronic component in the region of interest in step S2 - the op amp chip IIThe edge coordinates of each electronic component in the image are obtained, and the coordinates of all pixels inside the edge are obtained by traversing the point set in the contour. A matrix M with a size of 320×256 is established, and the above coordinate positions are set to 1 and the rest are set to 0. The matrix M is combined with the sequence temperature matrix T M Multiply by elements to get the temperature sequence matrix T of the key electronic components in the area of ​​interest - the op amp chip I .

[0035] T I (i,j)=M(i,j)*T M (i,j)

[0036] Where i and j represent the rows and columns of the matrix M respectively.

[0037] S4, temperature state area division and data set extraction: According to the working time, the temperature state of the electronic components of the airborne electronic system circuit board area of ​​interest in step S3 - the operational amplifier chip is divided into a dynamic temperature rise area and a steady-state constant temperature area. Extract the temperature sequence matrix T I The data of the part where the temperature continues to rise is taken as the dynamic temperature rising zone data set and is recorded as T ID ; Extract infrared segmentation image I of op amp chips II The image of the stable temperature part is taken as the image data set of the steady-state constant temperature area and is denoted as I IS , the specific steps are as follows:

[0038] By analyzing the serial temperature matrix T of the op amp chip I Determine the start time t0 of the dynamic temperature rise zone, the start time t1 of the stable thermal equilibrium state, and the stop working time t2, define t0-t1 as the dynamic temperature rise zone, and define t1-t2 as the steady-state constant temperature zone.

[0039] Intercept the temperature sequence matrix T of the key electronic components in the area of ​​interest - the operational amplifier chip I The temperature matrix sequence data of the time period t0-t1 (the time interval from the start of the circuit board to the stable thermal equilibrium state) is used as the dynamic temperature rise zone data set and named T ID , represents the change of the temperature matrix of the electronic components-op-amp chips in the area of ​​interest during the time interval from the start of the circuit board to the stable thermal equilibrium state. The temperature information of the dynamic heating area sequence refers to the temperature information of the key electronic components-op-amp chips in the area of ​​interest during the time interval from the start of the circuit board to the stable thermal equilibrium state. Its characteristic is that the sequence temperature information of the key electronic components in the area of ​​interest in this area will change significantly over time. ID Perform the average and maximum value operations to obtain the mean sequence data T of the key electronic components in the area of ​​interest - the op amp chip in the dynamic heating stageD-mean and maximum value sequence data T D-Max ; In order to further expand the feature information, the temperature change rate T′ of the mean sequence and the maximum sequence is extracted respectively D-mean and T′ D-Max . The mean T D-mean , maximum value T D-Max , mean series temperature change rate T′ D-mean , the temperature change rate of the maximum sequence T′ D-Max By merging and splicing, we can obtain the multi-dimensional temperature series data T of the key electronic component in the area of ​​interest - the operational amplifier chip, which characterizes the temperature variation law of the target operational amplifier chip in various working modes (normal mode and different fault modes).

[0040]

[0041] Capture the infrared segmentation image of the key electronic components in the area of ​​interest - the op amp chip I II The infrared image sequence of the time period t1-t2 (the time interval from the circuit board being in a stable working state to stopping working) is named I IS , represents the change of infrared image of key electronic components in the area of ​​interest - op amp chip in the interval from stable thermal equilibrium state to shutdown of circuit board. In this interval, the serial temperature information of key electronic components in the area of ​​interest - op amp chip only changes slightly with time.

[0042] S5. Dataset processing and model training: ID Further processing into multidimensional temperature series data T, multidimensional temperature series data and image dataset of steady-state constant temperature region I IS Set the same index and take with I IS 70% of the data is set as the training set and named T train and I train ; The rest with I IS 30% of the data are set as the validation set and named T val and I val . train Input to the long short-term memory time series feature extraction model with fusion attention mechanism to extract classification features; train Input to the improved Transformer-Encoder network V8 infrared thermal image classification feature extraction model to extract classification features, where the specific steps are as follows:

[0043] First, the training set T train Perform data preprocessing and obtain data set T trainThe expected μ and variance σ of each dimension (different features, such as the maximum value feature and the maximum value change rate feature) in , convert each data point into a standard score, as shown in the formula. The validation set is not processed.

[0044]

[0045] Among them, z i,t is the temperature data of each sequence point in the i-th dimension after standardization; x i is the temperature data of each sequence point in the i-th dimension in the training set.

[0046] After that, the processed T train The input is fed into the long short-term memory time series classification feature extraction model integrated with attention mechanism for classification feature extraction. The long short-term memory time series feature extraction model integrated with attention mechanism (LSTM&Att) consists of two LSTM networks with mixed attention mechanism connected in series, in which a time query and a set of key-value pairs are mapped to an output. Query, key, and value are matrices obtained from the feature extraction of the LSTM network. The output is defined as the weighted sum of values ​​with full representation. In the LSTM&Att classification feature extraction model, the first network is used to extract the basic relationship between the temperature series data, and the second network is used to expand the relationship details to obtain more complex complex relationships hidden in the data. For each LSTM network with mixed attention mechanism, its mathematical expression is as shown in the formula:

[0047] I q =f LSTM-Q (z)

[0048] I k =f LSTM-K (z)

[0049] I v =f LSTM-V (z)

[0050] Among them, f LSTM-Q 、f LSTM-K 、f LSTM-V Respectively represent the LSTM networks used to obtain query, key and value, z represents the input of the training set; I q ,I k and I vRepresents the vector matrix composed of query, key and value respectively.

[0051] Finally, the output of the LSTM network mixed with the attention mechanism is:

[0052]

[0053] Among them, f softmax Represents the Softmax function; Represents the transpose of the vector matrix composed of keys.

[0054] Among them, I train The input is into the infrared thermal image classification feature extraction model based on the Transformer-Encoder network to complete the classification feature extraction. Compared with the recognition algorithm, the Transformer-Encoder network detection algorithm can focus on the objects in the area of ​​interest in the image and ignore the background information, thereby making full use of the heat influence between multiple op amp chips in the area of ​​interest during the operation of the circuit board, and incorporating the features generated by the mutual influence of various components under different fault conditions into the recognition network, thereby improving the accuracy of classification.

[0055] S6. Feature fusion and model classification training: Fusion of the multidimensional temperature series data T obtained in S5 and the image data set I of the steady-state constant temperature zone IS The two types of feature information are input into the multi-layer perceptron in the model to complete the classification training of the model.

[0056] S7. Fault classification and diagnosis: Input the known category data set into the trained model to obtain the classification of component faults in the area of ​​interest of the airborne electronic system circuit board. Based on the classification results, diagnose the faults of operational amplifier chips on the airborne electronic system circuit board.

[0057] The present invention has brought significant beneficial effects after in-depth improvement and optimization against the shortcomings of the prior art, as follows:

[0058] Reduce the risk of secondary damage to faulty aircraft parts: In the prior art, when the traditional injection-type detection method is applied to airborne circuit boards, it may cause secondary damage to faulty aircraft parts due to the complex structure and high cost. However, the non-destructive testing technology used in the present invention, especially infrared technology, can accurately locate the fault without large-scale disassembly of complex airborne electronic systems, effectively avoiding the dangers and secondary damage that may be caused by high voltage and high current working environments.

[0059] Simplify the detection process and reduce the detection difficulty: When detecting circuit board faults, the existing technology often requires a detailed analysis of the circuit board's operating mechanism, which not only increases the difficulty of detection, but also increases the requirements for the professional ability of technicians. However, by collecting infrared temperature data of the circuit board, the present invention can intuitively extract the temperature characteristics of different fault modes and complete fault diagnosis without in-depth analysis of the circuit board's operating mechanism, thus reducing the difficulty of detection.

[0060] Improve the generalization and applicability of the detection method: The traditional circuit detection method in the prior art often requires the purchase of corresponding diagnostic equipment according to different models and systems, which increases the maintenance cost and reduces the generalization of the detection method. The infrared detection technology adopted by the present invention only requires an infrared circuit board data acquisition system composed of a numerical control computer and a thermal imager to complete the fault diagnosis of most circuit board op amp chips, which not only reduces the maintenance cost, but also improves the generalization and applicability of the detection method.

[0061] Improve the accuracy and efficiency of fault detection: The accuracy of fault detection in existing technologies is often challenged when facing complex external environments (such as light and room temperature) and different aging levels of circuit board components. However, the present invention considers the temperature change trend of the temperature series and the fault category information hidden therein, and combines deep learning and other algorithms to more accurately mine the fault information of the circuit board and improve the accuracy and efficiency of fault detection.

[0062] In summary, the present invention overcomes the limitations of traditional methods and improves the diagnostic accuracy by combining visible light images with infrared thermal image sequences. By building an infrared image acquisition system, infrared sequence images and temperature matrices of circuit boards in different working modes are collected, and then classification training is performed using a deep learning algorithm. This method is particularly suitable for fault diagnosis in complex scenarios such as airborne electronic system circuit boards, achieving non-contact, fast and intuitive detection results.

[0063] This paper elaborates on the method of infrared image processing and temperature state analysis of key electronic components. First, the accurate segmentation of infrared images is achieved through an improved model to obtain the temperature information of components. Secondly, the working state is divided into different areas, and the temperature sequence data and infrared image data are extracted respectively. In addition, this paper also proposes a method for processing the temperature information of electronic component sequences, and constructs a rich data set by calculating multi-dimensional temperature data and intercepting infrared image sequences.

[0064] In terms of feature extraction, this paper uses LSTM networks and Transformer-Encoder networks that integrate attention mechanisms. These models can extract the basic relationships between temperature series data and reveal more complex relationships, thereby effectively expanding feature information and improving classification accuracy. Experimental results show that this method can focus on the area of ​​interest, ignore background information, and achieve efficient and accurate diagnosis of faults.

[0065] Compared with traditional detection methods, this method reduces the difficulty of detection, improves generalization, and avoids secondary damage and the dangers brought by high voltage and high current environments. This technology is of great significance for improving the efficiency and accuracy of fault diagnosis of airborne electronic system circuit boards, and is expected to reduce maintenance costs and improve system safety. Therefore, this method has broad application prospects in fault detection of airborne electronic system circuit boards.

[0066] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0067] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An infrared-based method for diagnosing faults of operational amplifier chips on airborne electronic system circuit boards, characterized in that: The following steps are involved: S1. Data collection: Collect infrared sequence images of the airborne electrical system circuit board in different working modes I I And the corresponding sequence temperature matrix T M ; S2, data preprocessing: the infrared sequence image I of step S1 is I After preprocessing, it is fed into the semantic segmentation model to obtain the infrared segmentation image I of the key electronic components in the area of ​​interest - op amp chips II ; S3, temperature sequence matrix acquisition: Combine the sequence temperature matrix T in step S1 M and the infrared segmentation image I of the operational amplifier chip in step S2 II , get the temperature sequence matrix T of the operational amplifier chip I ; S4. Temperature state area division and data set extraction: Based on the working time, the temperature state of the electronic components in the area of ​​interest of the airborne electronic system circuit board - the operational amplifier chip is divided into a dynamic temperature rise area and a steady-state constant temperature area, and the temperature sequence matrix T is extracted. I The data of the part where the temperature continues to rise is taken as the dynamic temperature rising zone data set and is recorded as T ID , the data set T ID Further processing into multi-dimensional temperature sequence data T; extracting infrared segmentation image I of operational amplifier chip II The image of the stable temperature part is taken as the image data set of the steady-state constant temperature area and is denoted as I IS ; S5, data set processing and model training: The multidimensional temperature series data in step S4 is processed. and image dataset of steady-state constant temperature region I IS Take 70% of them as training set and name them as T train and I train ; The remaining 30% of the data are set as validation sets and named T val and I val , and multidimensional temperature series data The training set is input into the long short-term memory time series feature extraction model with fusion attention mechanism to extract classification features. At the same time, the image dataset I of the steady-state constant temperature area IS The training set is input into the infrared thermal image classification feature extraction model to extract classification features; S6. Feature fusion and model classification training: Fusion of the multidimensional temperature series data T obtained in S5 and the image data set I of the steady-state constant temperature zone IS The two types of feature information are input into the multi-layer perceptron in the model to complete the classification training of the model; S7. Fault classification and diagnosis: Input the known category data set into the trained model to obtain the classification of component faults in the area of ​​interest of the airborne electronic system circuit board. Based on the classification results, diagnose the faults of operational amplifier chips on the airborne electronic system circuit board.

2. The infrared-based airborne electronic system circuit board operational amplifier chip fault diagnosis method according to claim 1, characterized in that: In step S1, the infrared sequence image I I And the corresponding sequence temperature matrix T M The acquisition is achieved by using multiple infrared thermal imagers that are perpendicular to the top of the circuit board of the airborne electronic system to be tested, wherein the circuit board of the airborne electronic system to be tested is connected to the excitation source through a programmable relay switch, and the host computer is connected to the infrared camera and the programmable relay switch respectively.

3. The infrared-based airborne electronic system circuit board operational amplifier chip fault diagnosis method according to claim 1, characterized in that: In step S2, the infrared sequence image I I The preprocessing includes but is not limited to translation, rotation and mirror flipping. The semantic segmentation model in step S2 is a DeepLab v3+ semantic segmentation model using a Mobilenet network.

4. The infrared-based airborne electronic system circuit board operational amplifier chip fault diagnosis method according to claim 1, characterized in that: The serial temperature matrix T of the operational amplifier chip in step S4 I Determine the start time t0 of the dynamic temperature rise zone, the start time t1 of the stable thermal equilibrium state, and the stop working time t2, define t0-t1 as the dynamic temperature rise zone, and define t1-t2 as the steady-state constant temperature zone.

5. The infrared-based airborne electronic system circuit board operational amplifier chip fault diagnosis method according to claim 1, characterized in that: In step S5, T train The specific steps of extracting classification features from the long short-term memory time series feature extraction model that integrates the attention mechanism are as follows: For the training set T train Perform data preprocessing and obtain data set T train The expectation μ and variance σ of each dimension in , convert each data point into a standard score, as shown in the formula: Among them, z i,t is the temperature data of each sequence point in the i-th dimension after standardization; x i is the temperature data of each sequence point in the i-th dimension in the training set; After that, the processed T train The data is input into the long short-term memory time series classification feature extraction model integrating the attention mechanism to extract classification features.

Images