CNN-LSTM-based analog circuit fault detection method and apparatus, readable storage device and program product

Abstract

The invention discloses an analog circuit fault detection method and device based on CNN-LSTM, readable storage equipment and a program product, and relates to the technical field of deep learning for circuit fault diagnosis. The problems that an existing fault diagnosis method based on deep learning is difficult to effectively deal with complex characteristics of high-dimensional data and cannot fully capture key characteristic information related to faults, so that the diagnosis effect is poor, and the comprehensiveness and reliability of diagnosis results are poor are solved. By constructing the CNN-LSTM fault detection model, spatial features and time sequence features in circuit fault signals can be processed at the same time, and accurate detection and classification of faults can be realized without depending on an accurate mathematical model. The method is mainly used for fault diagnosis of the analog circuit.

Description

Technical Field

[0001] The present invention relates to the technical field of deep learning for circuit fault diagnosis. Background Art

[0002] With the rapid development of electronic technology, analog circuits have been widely used in industrial control, communication equipment, consumer electronics and other fields. However, the complex structure and diverse failure modes of analog circuits make fault diagnosis an important and challenging technical problem. In recent years, fault diagnosis methods based on deep learning have gradually attracted widespread attention due to their automated feature extraction and powerful classification capabilities. Compared with traditional fault diagnosis methods that rely on circuit models or signal processing, deep learning methods can directly extract features from data, providing a new technical means for fault diagnosis of complex circuits. However, these methods still have obvious defects in practical applications.

[0003] Existing deep learning methods often have difficulty effectively dealing with the complex characteristics of high-dimensional data when processing analog circuit signals. The operating state of analog circuits is often manifested as multi-dimensional dynamic signals, and the feature extraction capabilities of traditional methods are limited and cannot fully capture the key feature information related to faults, resulting in poor fault diagnosis results.

[0004] In addition, existing technologies are weak in dealing with class imbalance. There are many types of faults in analog circuits, but in practice, the distribution of samples of different fault types is often uneven, which makes the existing models prone to bias towards the main categories and ignore the fault characteristics of a few categories, thereby reducing the comprehensiveness and reliability of the diagnosis results. Therefore, the above problems need to be solved urgently. Summary of the invention

[0005] The purpose of the present invention is to solve the problems that the existing fault diagnosis methods based on deep learning are difficult to effectively deal with the complex characteristics of high-dimensional data, cannot fully capture the key feature information related to the fault, resulting in poor diagnostic effect, and the comprehensiveness and reliability of the diagnostic results are poor. The present invention provides an analog circuit fault detection method, device, readable storage device and program product based on CNN-LSTM.

[0006] The analog circuit fault detection method based on CNN-LSTM includes the following steps:

[0007] S1. Data collection and preprocessing:

[0008] The tolerance of each component of the analog circuit to be detected is set, and the voltage value of the circuit output node at each sampling time under the fault type corresponding to each sampling period is collected when the analog circuit to be detected is in operation state, and the voltage value of the circuit output node at each sampling time in the sampling period is used as the input data of the training sample, and the fault type corresponding to the sampling period is used as the label of the training sample;

[0009] After linear interpolation of all training samples, the input data of each training sample is reshaped into a two-dimensional matrix;

[0010] S2. Build a CNN-LSTM fault detection model and train it using training samples;

[0011] S3. Use the trained CNN-LSTM fault detection model to perform fault detection on the analog circuit to be detected.

[0012] Preferably, each fault type is caused by one or more component failures in the analog circuit to be detected.

[0013] Preferably, the CNN-LSTM fault detection model includes a CNN, a flattening layer, a LSTM and a SOFTMAX layer arranged in sequence;

[0014] CNN, used to extract features from the input two-dimensional matrix to obtain a multi-dimensional feature map;

[0015] A flattening layer, used to flatten the multi-dimensional feature map into a one-dimensional vector array;

[0016] LSTM, used to extract time series correlation features from one-dimensional vector arrays to obtain time series feature vectors;

[0017] The SOFTMAX layer is used to reduce the dimension of the time series feature vector output by LSTM.

[0018] Preferably, the CNN includes a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer, a third convolutional layer and a third pooling layer arranged in sequence.

[0019] Preferably, the dimension of the two-dimensional matrix is ​​28×28;

[0020] The convolution kernel size of the first convolution layer is 3×3, the sampling window size of the first pooling layer is 2×2, the convolution kernel size of the second convolution layer is 3×3, the sampling window size of the second pooling layer is 2×2, the convolution kernel size of the third convolution layer is 2×2, and the sampling window size of the third pooling layer is 2×2.

[0021] Preferably, during the training of the CNN-LSTM fault detection model, a back propagation algorithm is used to update the model parameters.

[0022] The analog circuit fault detection device based on CNN-LSTM includes a storage device, a processor, and a computer program stored in the storage device and executable on the processor. The processor executes the computer program to implement the analog circuit fault detection method based on CNN-LSTM.

[0023] A computer-readable storage device stores a computer program, and when the computer program is executed, the analog circuit fault detection method based on CNN-LSTM is implemented.

[0024] A computer program product comprises a computer program, which, when executed by a processor, implements the analog circuit fault detection method based on CNN-LSTM as described above.

[0025] Advantages of the present invention:

[0026] The present invention proposes a CNN-LSTM fault detection model constructed by combining a convolutional neural network (CNN) and a long short-term memory network (LSTM), and uses it to perform analog circuit fault diagnosis. CNN extracts local features of high-dimensional signals to capture the spatial features of fault signals, and then uses LSTM to model the temporal correlation of signals and extract the dynamic evolution law of fault features, thereby making up for the deficiencies of existing methods in dynamic modeling and feature extraction, and improving the accuracy of fault detection. At the same time, the fault categories corresponding to the training samples artificially constructed by the present invention avoid the problem of category imbalance, ensure that the model has good recognition ability in different categories to ensure the comprehensiveness of samples, improve the comprehensiveness and reliability of diagnostic results, and improve the diagnostic accuracy of a few categories of faults. Overall, the present invention significantly improves the accuracy and applicability of analog circuit fault diagnosis through multi-level feature extraction and timing modeling, and provides reliable technical support for analog circuit fault management.

[0027] The CNN-LSTM-based analog circuit fault detection method, device, readable storage device and program product described in the present invention is a data-driven method based on deep learning. Unlike the traditional physical model-based method, the present invention relies on data-driven deep learning technology and can achieve accurate fault detection and classification without relying on precise mathematical models. It is more accurate and robust than traditional methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 It is a schematic diagram of the principle of the analog circuit fault detection method based on CNN-LSTM of the present invention;

[0029] Figure 2 This is a schematic diagram of the principle of CNN;

[0030] Figure 3 This is a schematic diagram of the LSTM structure;

[0031] Figure 4 It is a schematic diagram of the structure of the Shallen-key bandpass filter circuit;

[0032] Figure 5 This is a visual T-SNE plot of the input data of the Shallen-key bandpass filter circuit;

[0033] Figure 6 Yes Figure 5 A T-SNE diagram showing the visual classification effect of the output data after the input data is detected by the CNN-LSTM fault detection model of the present invention;

[0034] Figure 7 This is a schematic diagram of the detection accuracy of each category of the Shallen-key bandpass filter circuit changing with the training rounds. DETAILED DESCRIPTION

[0035] 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.

[0036] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other.

[0037] Specific implementation method 1. Combination Figure 1 The present embodiment is described. The analog circuit fault detection method based on CNN-LSTM described in the present embodiment comprises the following steps:

[0038] S1. Data collection and preprocessing:

[0039] The tolerance of each component of the analog circuit to be detected is set, and the voltage value of the circuit output node at each sampling time under the fault type corresponding to each sampling period is collected when the analog circuit to be detected is in operation state, and the voltage value of the circuit output node at each sampling time in the sampling period is used as the input data of the training sample, and the fault type corresponding to the sampling period is used as the label of the training sample;

[0040] After linear interpolation of all training samples, the input data of each training sample is reshaped into a two-dimensional matrix; wherein, linear interpolation of all training samples is performed to ensure that the sampling time in all training samples is unified;

[0041] S2. Build a CNN-LSTM fault detection model and train it using training samples;

[0042] S3. Use the trained CNN-LSTM fault detection model to perform fault detection on the analog circuit to be detected. Specifically, the voltage values ​​of the circuit output nodes at each sampling time in the current sampling period are reshaped into a two-dimensional matrix and input into the trained CNN-LSTM fault detection model to predict the fault type.

[0043] The present invention uses a convolutional neural network (CNN) to extract high-dimensional features of circuit state data, and combines it with a long short-term memory network (LSTM) for dynamic timing modeling to form a CNN-LSTM fault detection model, which can achieve accurate detection and classification of faults without relying on precise mathematical models.

[0044] Convolutional Neural Network (CNN): It is used to extract the spatial features of time domain signals from the input circuit fault data. Through multi-layer convolution and pooling operations, CNN can automatically extract local features in circuit signals and perform dimensionality reduction.

[0045] Long Short-Term Memory Network (LSTM): It is used to model the timing characteristics of circuit fault data. It can handle the timing dynamic changes of circuit faults, capture long-term dependencies through memory units, and effectively predict the time series patterns of fault occurrence.

[0046] The local features of high-dimensional signals are extracted by CNN to capture the spatial features of fault signals, and the time correlation of LSTM modeling signals is then used to extract the dynamic evolution law of fault features, thereby making up for the deficiencies of existing methods in dynamic modeling and feature extraction and improving the accuracy of fault detection. At the same time, the fault categories corresponding to the training samples artificially constructed by the present invention avoid the problem of category imbalance, ensure that the model has good recognition ability in different categories, ensure the comprehensiveness of samples, and improve the comprehensiveness and reliability of diagnostic results.

[0047] Specifically, each fault type is caused by one or more component faults in the analog circuit to be detected.

[0048] Further, the CNN-LSTM fault detection model includes CNN, flattening layer, LSTM and SOFTMAX layers arranged in sequence;

[0049] CNN, used to extract features from the input two-dimensional matrix to obtain a multi-dimensional feature map;

[0050] A flattening layer, used to flatten the multi-dimensional feature map into a one-dimensional vector array;

[0051] LSTM, used to extract time series correlation features from one-dimensional vector arrays to obtain time series feature vectors;

[0052] The SOFTMAX layer is used to reduce the dimension of the time series feature vector output by LSTM.

[0053] In this preferred embodiment, since the output feature map of CNN is small in size, directly inputting it into the LSTM model will cause data incompatibility. Therefore, a flattening layer is added after the CNN model to "flatten" the multi-dimensional feature map into a one-dimensional vector array so that the data can be smoothly passed to the LSTM. This conversion is crucial in the data flow process and ensures the seamless connection between CNN and LSTM.

[0054] LSTM's timing modeling capability: Long short-term memory networks can learn the temporal dependencies in circuit signals, and combine the static features extracted by CNN with temporal context information, further enhancing the model's ability to identify fault modes. For an existing specific structure of LSTM, see Figure 3 The core of LSTM is the cell state, which is represented by a horizontal line running through the cell. The cell state is equivalent to the path of information transmission, allowing information to be passed down in the sequence. It can be regarded as the "memory" of the network. In theory, the cell state can always pass on relevant information during the sequence processing. Therefore, even information from earlier time steps can be carried to cells in later time steps, which overcomes the influence of short-term memory. The addition and removal of information is achieved through the "gate" structure, which learns what information to save or forget during the training process.

[0055] Combining CNN with LSTM fully utilizes the feature extraction capability of convolutional networks and the temporal learning capability of long short-term memory networks to achieve more accurate fault detection.

[0056] See also Figure 2 , CNN includes a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer, a third convolutional layer and a third pooling layer which are arranged in sequence.

[0057] The dimension of the two-dimensional matrix received by the first convolutional layer is 28×28;

[0058] The convolution kernel size of the first convolutional layer is 3×3, and it outputs 32 feature maps of size 26×26;

[0059] The sampling window size of the first pooling layer is 2×2, and the output is 32 feature maps of size 13×13;

[0060] The convolution kernel size of the second convolution layer is 3×3, and the output is 64 feature maps of size 11×11;

[0061] The sampling window size of the second pooling layer is 2×2, and the output is 64 feature maps of size 5×5;

[0062] The convolution kernel size of the third convolution layer is 2×2, and the output is 128 feature maps of size 5×5;

[0063] The sampling window size of the third pooling layer is 2×2, and the output is 128 feature maps of size 1×1.

[0064] In specific applications, during the training of the CNN-LSTM fault detection model, the back propagation algorithm is used to update the model parameters.

[0065] Specific implementation method 2: The analog circuit fault detection device based on CNN-LSTM described in this implementation method includes a storage device, a processor, and a computer program stored in the storage device and executable on the processor. The processor executes the computer program to implement the analog circuit fault detection method based on CNN-LSTM.

[0066] Specific implementation method three: A computer-readable storage device described in this implementation method stores a computer program, and is characterized in that when the computer program is executed, the analog circuit fault detection method based on CNN-LSTM is implemented.

[0067] Specific implementation method 4: A computer program product described in this implementation method includes a computer program, and when the computer program is executed by a processor, the analog circuit fault detection method based on CNN-LSTM as described above is implemented.

[0068] Verification test:

[0069] The Shallen-key bandpass filter, also known as the finite gain positive feedback filter, is a circuit based on an operational amplifier. Its design is simple and efficient, and it only contains an operational amplifier, a few resistors and capacitors. Compared with other more complex experimental circuits, its structure is intuitive and easy to implement. The circuit is as follows Figure 4 The present invention not only performs diagnostic analysis on common single component faults in the circuit, but also further performs experimental simulation on multiple fault situations, which include the increase and decrease of component values ​​in the circuit.

[0070] The resistance tolerance in the circuit is set to 5%, the capacitance tolerance is set to 10%, the excitation source is a pulse signal with an amplitude of 5V, a pulse period of 1.96ms, and a pulse width of 10us. Figure 4 There are 10 fault types in total, see Table 1 for details;

[0071] Table 1 Fault types of Shallen-key bandpass filter circuit

[0072]

[0073] ↑: Component value exceeds nominal value

[0074] ↓: Component value is less than the nominal value

[0075] Pspice is used to perform Monte Carlo simulation on the analog circuit under test under the stimulus of the test stimulus signal. After data acquisition, since the input of the CNN part is a two-dimensional matrix, it is necessary to transform the processed fault data from one dimension to a two-dimensional matrix as input data, and the visual T-SNE diagram of the input data of the Shallen-key bandpass filter circuit is shown in Figure 2. Figure 5 ;

[0076] Figure 6 Yes Figure 5 A T-SNE diagram showing the visual classification effect of the output data after the input data is detected by the CNN-LSTM fault detection model of the present invention; Figure 5 and Figure 6 The horizontal and vertical coordinates represent the first main feature dimension and the second main feature dimension of the data point in the two-dimensional dimensionality reduction space, respectively. The main function is to intuitively display the distribution relationship of the sample points in the high-dimensional space. Figure 6 Each point in represents a sample, and colors distinguish samples of different categories.

[0077] Clustering of data points: Points of the same color tend to form tight clusters, indicating that T-SNE can effectively map high-dimensional features of similar samples to the same location. This shows that the features extracted by the model can distinguish different categories well.

[0078] Class separation: There is a clear gap between clusters of different colors, which means that samples of different categories are separated in the high-dimensional feature space. T-SNE intuitively expresses this separation in two-dimensional space through dimensionality reduction.

[0079] The visualization after classification shows the distribution of different fault types. The classification results show that different fault types are well separated in the feature space, most fault data points are distributed more concentratedly, and the distance between different fault categories is significantly increased, and the overlap phenomenon is greatly reduced.

[0080] In particular, in the classified t-SNE image, most fault types form independent cluster distributions, and each cluster corresponds to a specific fault type. This shows that the classification model can effectively distinguish different fault states, and even after the high-dimensional data is compressed into a two-dimensional space, each fault category still maintains good distinguishability.

[0081] Figure 7 This is a schematic diagram of the detection accuracy of each category of the Shallen-key bandpass filter circuit changing with the training rounds. Figure 7 Each curve in is the fluctuation curve of various fault categories changing with the training rounds. Figure 7 It clearly presents the changing trend of the fault diagnosis rate of different fault types as the number of training times increases. It is worth noting that for all faults, the fault diagnosis rate shows a significant upward trend, and eventually approaches or reaches an accuracy rate close to 100%. This result not only verifies the effectiveness and robustness of the CNN-LSTM fault detection model adopted by the present invention, but also further proves that by increasing the number of training times, the model can accurately identify and distinguish various fault modes. The realization of a high fault diagnosis rate is of great significance for improving system reliability, reducing downtime and maintenance costs caused by faults, and provides strong support for fast and accurate fault detection and diagnosis in practical applications.

[0082] Table 2 shows the single fault diagnosis accuracy of the Shallen-key bandpass filter circuit, as follows:

[0083] Table 2 Single fault diagnosis accuracy of Shallen-key bandpass filter circuit

[0084]

[0085] ↑: Component value exceeds nominal value

[0086] ↓: Component value is less than the nominal value

[0087] It can be concluded from Table 2 that the model designed by the present invention performs well in overall fault diagnosis, with an accuracy rate of over 95%, and achieves a high recognition level in the diagnosis of all fault types. This fully proves the reliability and practicality of the model in the identification of complex circuit states and multiple fault types.

[0088] Although the present invention is described herein with reference to specific embodiments, it should be understood that these embodiments are merely examples of the principles and applications of the present invention. It should therefore be understood that many modifications may be made to the exemplary embodiments and that other arrangements may be devised without departing from the spirit and scope of the present invention as defined by the appended claims. It should be understood that the various dependent claims and features described herein may be combined in a manner different from that described in the original claims. It should also be understood that features described in conjunction with individual embodiments may be used in other described embodiments.

Claims

1. The analog circuit fault detection method based on CNN-LSTM is characterized by: The steps include: S1. Data collection and preprocessing: The tolerance of each component of the analog circuit to be detected is set, and the voltage value of the circuit output node at each sampling time under the fault type corresponding to each sampling period is collected when the analog circuit to be detected is in operation state, and the voltage value of the circuit output node at each sampling time in the sampling period is used as the input data of the training sample, and the fault type corresponding to the sampling period is used as the label of the training sample; After linear interpolation of all training samples, the input data of each training sample is reshaped into a two-dimensional matrix; S2. Build a CNN-LSTM fault detection model and train it using training samples; S3. Use the trained CNN-LSTM fault detection model to perform fault detection on the analog circuit to be detected.

2. The analog circuit fault detection method based on CNN-LSTM according to claim 1, characterized in that: Each fault type is caused by one or more component failures in the analog circuit to be detected.

3. The analog circuit fault detection method based on CNN-LSTM according to claim 1, characterized in that: The CNN-LSTM fault detection model includes CNN, flattening layer, LSTM and SOFTMAX layers set in sequence; CNN, used to extract features from the input two-dimensional matrix to obtain a multi-dimensional feature map; A flattening layer, used to flatten the multi-dimensional feature map into a one-dimensional vector array; LSTM, used to extract time series correlation features from one-dimensional vector arrays to obtain time series feature vectors; The SOFTMAX layer is used to reduce the dimension of the time series feature vector output by LSTM.

4. The analog circuit fault detection method based on CNN-LSTM according to claim 3, characterized in that: The CNN includes a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer, a third convolutional layer and a third pooling layer which are arranged in sequence.

5. The analog circuit fault detection method based on CNN-LSTM according to claim 4, characterized in that: The dimension of the two-dimensional matrix is ​​28×28; The convolution kernel size of the first convolution layer is 3×3, the sampling window size of the first pooling layer is 2×2, the convolution kernel size of the second convolution layer is 3×3, the sampling window size of the second pooling layer is 2×2, the convolution kernel size of the third convolution layer is 2×2, and the sampling window size of the third pooling layer is 2×2.

6. The analog circuit fault detection method based on CNN-LSTM according to claim 1, characterized in that: During the training of the CNN-LSTM fault detection model, the back propagation algorithm is used to update the model parameters.

7. An analog circuit fault detection device based on CNN-LSTM, comprising a storage device, a processor, and a computer program stored in the storage device and executable on the processor, characterized in that: The processor executes a computer program to implement the analog circuit fault detection method based on CNN-LSTM as described in any one of claims 1 to 6.

8. A computer-readable storage device, characterized in that: The storage device stores a computer program, wherein the computer program, when executed, implements the analog circuit fault detection method based on CNN-LSTM as claimed in any one of claims 1 to 6.

9. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the analog circuit fault detection method based on CNN-LSTM as described in claims 1 to 6 is implemented.

Images