Intermittent fault feature fast mining strategy for electronic circuit system

Abstract

The application discloses a strategy for extracting fault features of electronic circuit systems, named as SSEST strategy, which is used for perceiving global information and paying attention to notable local information, and mining important local information means realizing expression of intermittent fault features of electronic circuit systems, specifically, first, S transformation is performed on a circuit output time sequence signal to acquire time-frequency domain features, then a squeeze and excitation network attention module is used to distribute channel weights, subsequently, input into a Swin Transformer framework, and pay attention to local information related to faults from global signals, and deep mining is performed on fault features, and two electronic circuits are taken as experimental circuits, the proposed diagnostic strategy realizes rapid and high-precision diagnosis, and shows that the proposed multiple attention mechanism is efficient for feature mining of intermittent faults of electronic circuit systems.

Description

Technical Field

[0001] This invention relates to the field of intermittent faults in electronic circuit systems, specifically a rapid discovery strategy for intermittent fault characteristics in electronic circuit systems. Background Technology

[0002] Electronic circuits play an irreplaceable role in many fields, including aerospace, military, and medical. Ensuring the reliability of electronic circuits is essential for the safety of the systems they operate in. When intermittent faults occur in a system, they may trigger the alarm unit of the system or equipment. However, during maintenance operations, it is easy to find no faults. This problem is becoming increasingly prominent in the field of equipment testing and support. It is necessary to diagnose and eliminate the main cause of this phenomenon, namely intermittent faults, in order to avoid more serious damage to the system or equipment. Therefore, it is essential to study the diagnostic methods for intermittent faults in electronic circuit systems.

[0003] Due to factors such as the manufacturing process of electronic components, their failure and degradation characteristics, and external environmental stress, circuits are prone to intermittent faults. Intermittent faults are characterized by random occurrence, strong signal singularity, short duration, and ease of concealment, thus posing a potential hazard to electronic circuit systems.

[0004] Overall, machine learning or deep learning-based methods have seen significant application in fault diagnosis of circuits or other objects in recent years, achieving promising results. However, these methods essentially focus on building or improving models for feature extraction, or on constructing effective fault features, and are not suitable for the problem of intermittent, randomly occurring faults addressed in this paper. Summary of the Invention

[0005] The purpose of this invention is to provide a rapid detection strategy for intermittent fault characteristics in electronic circuit systems, thereby solving the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a rapid feature mining strategy for intermittent faults in electronic circuit systems, characterized by: performing an S-transform on the output signal of the electronic circuit system, analyzing the time-frequency characteristics of the input signal, then using the Squeeze and excitation network attention module for preliminary feature extraction and weighted attention analysis, and using the Swin Transformer framework for deep feature mining to achieve attention to fault-related local information from the global signal, thereby achieving fault determination;

[0007] The S-transform includes a Gaussian window function;

[0008] The Squeeze and excitation network attention module is a typical implementation method;

[0009] Swin Transformer is based on the Transformer architecture.

[0010] In a preferred embodiment of the present invention, a Gaussian window function is used, which eliminates the need for window function selection and improves the defect of fixed window width. The S-transform is used to process the data. The difference between the S-transform and the short-time Fourier transform is that the height and width of the Gaussian window change with the frequency, which overcomes the defect of fixed window height and width of the short-time Fourier transform. It has good time-frequency characteristics and is suitable for extracting some time-frequency and features of the signal using the S-transform.

[0011] In a preferred embodiment of the present invention, the implementation process of the attention module for extracting feature layer channel weights includes the following steps:

[0012] (1) Perform global average pooling on the input feature layer with size H×W×C, so that the data of the other two dimensions of the non-channel layer are averaged to obtain a 1×1×C output;

[0013] (2) Perform two full connections on the output of the first step. The first full connection selects fewer neurons to obtain an output of 1×1×C / r; the second full connection has the same number of neurons as the input feature layer to obtain an output of 1×1×C.

[0014] (3) Then, the Sigmoid operation is performed to fix the C values ​​between 0 and 1. Thus, the weights (between 0 and 1) of each channel of the input feature layer are obtained. The dimension is 1×1×C, which means that different weights are assigned to each input channel, and the important channels have higher weights.

[0015] (4) After obtaining this weight, multiply this weight by the original input feature layer to obtain the output H×W×C.

[0016] As a preferred embodiment of the present invention, the improved Swing Transformer structure can generate windows of different sizes through a hierarchical feature mapping strategy. Since the number of patches for different windows is consistent, the computational complexity is low. In contrast, the previous Transformer generated feature maps with a single resolution and had quadratic complexity.

[0017] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0018] Compared to the traditional Transformer structure, this invention is the first to introduce the Swin Transformer, which is adept at perceiving global information and can focus on important local information, to explore the representation of intermittent fault features. It uses attention mechanisms multiple times to filter and mine key information or important features layer by layer, avoiding the problem of difficulty in accurately extracting features by manually selecting features. It proposes an innovative fault diagnosis strategy, combining the advantages of multiple modules to achieve fast and high-precision diagnosis of circuits. It has successfully explored the effectiveness of using the Swin-Transformer as a means of mining fault features in electronic circuits at an early stage.

[0019] This invention utilizes a novel feature learning model, Swin Transformer, which fully leverages the model's advantage of being adept at global perception while still being able to notice local information. Experiments have shown that: (1) S-transformation of the timing data of the circuit can significantly improve the performance of image classification algorithms in diagnosing original signals that are timing signals; (2) The attention mechanism of the SENet module is an effective strategy for initially locking the features of two-dimensional data; (3) The Swin Transformer framework can focus on fault-related local information from the global signal, deeply mine fault features, and achieve the final diagnosis. Attached Figure Description

[0020] Other features, objects, and advantages of the present invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0021] Figure 1 It is the attention module of Squeeze-and-Excitation Networks;

[0022] Figure 2 This is a structural diagram of the proposed SSEST diagnostic strategy;

[0023] Figure 3 This is the schematic diagram of the experimental filter circuit. Detailed Implementation

[0024] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0025] Reference Figure 1The Squeeze and Excitation Network (SENet) attention module, proposed in 2017, is a typical implementation of a channel attention mechanism. This attention mechanism obtains the weights of each channel of the feature layer of the input object through a specific strategy. Therefore, subsequent processing of objects processed by the SENet attention module is more efficient. The implementation process of this attention module, which extracts the channel weights of the feature layer, includes the following steps:

[0026] 1. Perform global average pooling on the input feature layer with size H×W×C, so that the data in the other two dimensions of the non-channel layer are averaged to obtain a 1×1×C output;

[0027] 2. Perform two fully connected layers on the output of the first step. The first fully connected layer selects fewer neurons, resulting in an output of 1×1×C / r. The second fully connected layer has the same number of neurons as the input feature layer, resulting in an output of 1×1×C.

[0028] 3. Then, a Sigmoid operation is performed to fix the C values ​​between 0 and 1. At this point, the weights (between 0 and 1) of each channel of the input feature layer are obtained, with a dimension of 1×1×C. This means that different weights are assigned to each input channel, with more important channels having higher weights.

[0029] 4. After obtaining this weight, we multiply it by the original input feature layer to obtain the output H×W×C;

[0030] Reference Figure 2 A SSEST diagnostic strategy for dealing with intermittent faults in electronic circuit systems is proposed:

[0031] First, key characterization signals of the electronic circuit are acquired, and the time-series signal is transformed into a time-frequency diagram through S-transform processing. This step is used to extract time-frequency domain features from the time-series signal.

[0032] Subsequently, the SE attention module is used to explore the key channels of the preceding time-frequency plot, assigning larger weights to important channels to achieve the purpose of feature pre-extraction;

[0033] Finally, the adjusted time-frequency graph is input into the Swing Transformer framework. The difference between this framework and the Transformer is that it can generate windows of different sizes through a hierarchical feature mapping strategy. Since the number of patches in different windows is the same, the computational complexity is low. In contrast, the previous Transformer generates feature maps with a single resolution and has quadratic complexity.

[0034] Furthermore, by moving window partitions between consecutive self-attention layers, the shifted windows bridge the windows of the previous layer, providing connections between them. By introducing local window interactions, the Swin Transformer possesses stronger global awareness capabilities, enabling it to extract fault-related local features from global signals, thus achieving fault diagnosis.

[0035] Reference Figure 3 Two electronic circuits, an S-filter circuit and a B-filter circuit, were selected, and different fault types were set for the two circuits to verify the proposed fault diagnosis strategy.

[0036] Based on the sensitivity analysis of the S bandpass circuit, in addition to setting the fault-free state NF, fault types such as intermittent fault of R1, intermittent fault of R2, permanent fault of R2, intermittent fault of U1, and intermittent fault of C2 are also set, for a total of 6 circuit states. For the B high-pass circuit, in addition to the NF state, fault types such as intermittent fault of R1, intermittent fault of C1, intermittent fault of U1, intermittent fault of U2, intermittent fault of U3, intermittent fault of U4, and permanent fault of U4 are set, for a total of 8 circuit states.

[0037] The entire experimental procedure is as follows: First, power is supplied to the circuit and an excitation is input to keep all relays in the normally closed state. The output waveform under normal conditions can be seen through the oscilloscope. Then, pulse signals are applied to the relays at each fault setting point. This experiment mainly focuses on analyzing intermittent fault data with a duty cycle of 99.4%. At this time, the circuit output can be seen to produce intermittent faults through the oscilloscope. Then, the output signal of the circuit is acquired through the acquisition card, and the acquired data is saved to disk.

[0038] The diagnostic results and analysis first present the results after performing S-transform on the timing signals of the circuit, and then elaborate on the diagnostic results of the proposed SSEST strategy on the two experimental electronic circuits.

[0039] It should be noted that multiple comparative experiments were also conducted to explore the superiority of the proposed diagnostic strategy.

[0040] In summary, compared to the traditional Transformer structure, this invention is the first to introduce the Swin Transformer, which excels at perceiving global information and can focus on important local information, to explore the representation of intermittent fault features. It utilizes attention mechanisms multiple times to filter and mine key information or important features layer by layer, avoiding the problem of difficulty in accurately extracting features by manually selecting features. It proposes an innovative fault diagnosis strategy, combining the advantages of multiple modules to achieve rapid and high-precision diagnosis of circuits. It has successfully explored the effectiveness of using the Swin-Transformer as a means of mining fault features in electronic circuits at an early stage.

[0041] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered illustrative and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the scope of the invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

[0042] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A method for rapidly identifying intermittent fault characteristics in electronic circuit systems, comprising the following steps: (1) Signal time-frequency analysis: Perform S-transform on the output signal of the electronic circuit system and analyze the time-frequency characteristics of the input signal. The S-transform is a Gaussian window function. (2) Primary channel attention feature extraction: The Squeeze and excitationnetwork attention module is used to perform preliminary feature extraction on the time-frequency analysis results of the signal; (3) Deep global context feature mining: The output of the channel attention-weighted feature layer is used for deep feature mining through the SwinTransformer framework; (4) Fault determination: Based on the deep features output by the Swing Transformer framework, the intermittent faults of the electronic circuit system are identified and classified by a classifier; The Squeeze and excitation network attention module is a typical implementation, and the SwinTransformer is based on the Transformer structure. In step (2), the time-frequency information is subjected to global average pooling, two full connections and Sigmoid operation in sequence to obtain channel weights and multiply them with the original feature layer to achieve channel attention recalibration; In step (3), the recalibrated features are input into the Swin Transformer framework. By calculating self-attention in a local window and combining it with a shift window operation, cross-window information interaction and deep feature mining are achieved.

Images