A disturbance-based interpretable waveform signal fault detection method

Abstract

The present invention relates to the field of computer vision and signal processing technology, and in particular to a method for interpretable waveform signal fault detection based on disturbance. The technical solution includes the following steps: converting the collected waveform signal data into image data, annotating the fault signals in the image, and constructing a training set, a validation set, and a test set; using the training set and the validation set to train the YOLOv8 target detection model so that it can identify the fault signals in the waveform signal image. The present invention effectively solves the problem of incorrect interpretation caused by the destruction of the continuity of the waveform signal in the traditional method through the innovative disturbance method of eliminating and splicing the middle area and truncating and compensating the two end areas, significantly improves the accuracy of interpretable analysis, and can accurately quantify the influence weight of different areas of the fault signal on the prediction of the YOLOv8 model, helping users understand the basis for model decision-making and improving the accuracy and reliability of fault detection.

Description

Technical Field

[0001] The present invention relates to the technical field of computer vision and signal processing, and in particular to a disturbance-based interpretable waveform signal fault detection method. Background Art

[0002] Currently, monitoring and analyzing relevant waveform signals, such as current, voltage, and electromagnetic waveforms, is crucial for timely detection of equipment failures and anomalies in fields such as industrial equipment and communications. The rapid development of deep learning has led to the widespread adoption of deep learning methods for fault warning and diagnosis. While these methods can achieve promising results, they also introduce the problem of deep learning being a black box. Perturbation-based interpretability methods can better explain the model's decision-making basis and behavior. Fault signals in the signal domain differ from general target detection tasks. Traditional perturbation-based interpretability methods can undermine the integrity and continuity of the fault signal, rendering the perturbed signal incomplete and discontinuous, leading to erroneous interpretations.

[0003] In summary, this application proposes a disturbance-based interpretable waveform signal fault detection method. Summary of the Invention

[0004] The purpose of the present invention is to propose a disturbance-based interpretable waveform signal fault detection method to address the problem in the background technology that the use of traditional disturbance-based interpretability methods will destroy the integrity and continuity of fault signals.

[0005] The technical solution of the present invention is a disturbance-based interpretable waveform signal fault detection method, comprising the following steps:

[0006] Convert the collected waveform signal data into image data, annotate the fault signals in the image, and construct training sets, validation sets, and test sets;

[0007] Use the training set and validation set to train the YOLOv8 target detection model to enable it to identify fault signals in waveform signal images;

[0008] Perform perturbation processing on the fault signal images in the test set;

[0009] The perturbed image is input into the trained YOLOv8 model for prediction, and the perturbation weights of the middle area and the two end areas are calculated respectively;

[0010] A visual heat map is generated based on the calculated weights and superimposed on the original image, with color depth indicating the influence weight of different regions on the model prediction.

[0011] Optionally, the fault signal images in the test set are disturbed, specifically including:

[0012] a. Eliminate splicing disturbances in the middle area: In the vertical center area of the fault signal detection frame, divide it according to the preset ratio and remove multiple heights of , and splice the removed areas to maintain the continuity of the waveform signal; is the height of the center area, and the ratio of the vertical range of the center area of the detection frame to the two end areas is 6:4;

[0013] b. Truncation compensation disturbance at both ends: The fault signal detection frame is gradually truncated and the truncation height is extended to the center boundary in the vertical direction of the two end areas. At the same time, the truncation points are connected by straight lines to compensate for the disturbance to ensure signal integrity. The above two disturbances are not performed simultaneously.

[0014] Optionally, the perturbation weights of the middle region and the two end regions are calculated separately, specifically including:

[0015] a. The formula for calculating the weight of the middle area is:

[0016]

[0017] in, is the raw confidence score, For the The confidence score after removing the splicing, For the The weight of the middle area corresponding to the secondary disturbance;

[0018] b. The formula for calculating the weight of the two end areas is:

[0019]

[0020] in, For the Confidence score after sub-truncation compensation.

[0021] Optionally, the specific implementation of removing the splicing disturbance in the middle area includes:

[0022] The height of the center area of the detection frame is:

[0023]

[0024] in, is the total vertical height of the fault signal detection frame, is the height of the center area, and the height of the culled block meets And move the range vertically Internal Execution perturbations, where is the number of splits to remove the block, is the height of the vertical sub-area to be removed each time, , is the original diagonal ordinate of the fault signal detection frame, In order to eliminate the vertical movement position of the block, it is limited to the central area; the above two disturbances are not performed at the same time.

[0025] Optionally, the specific implementation of the truncation compensation disturbance at both end regions includes:

[0026] The initial cut-off height is After each disturbance, the cutoff height is expanded to , , until it reaches the boundary of the central area and satisfies ,in, is the initial split times of the truncation block, is the height of the initial truncation block, For the The height of the truncated block after the perturbation.

[0027] Optionally, the generation of the visualized heat map includes:

[0028] Weight normalization: The weights of different regions of each fault signal are independently normalized to map them to the interval 0 and 1, where the maximum weight value corresponds to 1 and the minimum weight value corresponds to 0;

[0029] Color mapping rules: The normalized weight values are converted to RGB color values through linear interpolation. A weight value of 1 corresponds to red (RGB: 255, 0, 0), and a weight value of 0 corresponds to blue (RGB: 0, 0, 255). Intermediate weight values are gradually transitioned proportionally.

[0030] Transparency adjustment: adjust the transparency of the heat map and the original image to 50%, and then overlay the two;

[0031] Dynamic range adaptation: For different fault signals, their normalized maximum and minimum weight values are calculated independently, ensuring that the thermal color mapping of each fault signal is based only on its own weight distribution and is not interfered with by other signals.

[0032] Optionally, when the waveform signal is converted into image data, wavelet transform or Fourier transform is used to perform noise reduction processing.

[0033] Optionally, the diagonal coordinates of the fault signal detection box output by the YOLOv8 model are and , whose confidence score is .

[0034] Compared with the prior art, this application has at least one of the following beneficial technical effects:

[0035] The problem of perturbation-based interpretability methods providing incorrect explanations in target detection models for waveform signal image detection is solved.

[0036] This interpretability method can be used to calculate the influence weights of different regions in the fault signal on the target detection model.

[0037] This invention effectively solves the problem of incorrect interpretation caused by the destruction of waveform signal continuity in traditional methods through the innovative disturbance method of removing and splicing the middle area and truncating and compensating the two end areas. It significantly improves the accuracy of interpretability analysis, can accurately quantify the influence weight of different areas of the fault signal on the YOLOv8 model prediction, and intuitively display it through a visual heat map to help users understand the basis of model decision-making. In addition, based on the results of weight analysis, the model training strategy can be optimized in a targeted manner to improve the accuracy and reliability of fault detection. This method is suitable for signal detection in various industrial and communication fields such as current, voltage, and electromagnetic waveforms, and has broad practicality and promotion value. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 The flowchart of a disturbance-based interpretable waveform signal fault detection method. DETAILED DESCRIPTION

[0039] The technical solution of the present invention is further described below with reference to the accompanying drawings and specific embodiments.

[0040] Example

[0041] See also Figure 1 ,This method consists of three parts: the preparation of waveform signal image ,dataset and YOLOv8 model training, the perturbation of the test set using ,the proposed perturbation method, and the fault signal weight ,visualization.

[0042] The first part involves annotating waveform signal images and training the YOLOv8 model. After collecting waveform signal data from the sensor, the data is processed using wavelet transform and Fourier transform to reduce noise. The processed signal data is then converted into an image, and the fault signals or specific types of signals are annotated to create a data set. The training and validation sets are used for YOLOv8 model training, while the test set is used for perturbation and prediction.

[0043] The second part is to use the proposed perturbation method to perturb the test set, which is also the core part of the entire patent. For the fault signal, the perturbation is divided into two parts. One is the perturbation of the central area of the fault signal, that is, removal and splicing, and the other is the perturbation of the areas at both ends of the fault signal, that is, truncation compensation. The two perturbations are performed separately and not at the same time. The two parts are described in detail below.

[0044] First, the model is used to predict the image to obtain the position information and confidence score of the detection frame of the fault signal. The diagonal coordinates of the detection frame are set to and , and its corresponding confidence score is P. The ratio of the vertical range of the central area of the detection frame to the two end areas is 6:4.

[0045] That is, the high of the fault signal is , width is The height of the central area is

[0046]

[0047] in, is the height of the central area, The total vertical height of the fault signal detection frame. For the disturbance mode of elimination and splicing, the width of the elimination block is set to be the same as the detection frame. The height of the fault signal center area is an integer multiple of the elimination block. Since an excessively high elimination block will cause a fault in the splicing, the signals at both ends of the splicing will not be connected. Therefore, the height of the elimination block needs to be limited. The basic information of the elimination block is as follows:

[0048]

[0049]

[0050] in, is the number of splits to remove the block, is the height of the vertical sub-area to be removed each time, and its moving range is: , let the diagonal coordinates of the fault signal detection frame after removing the splicing disturbance each time be ( , ),( , ), then:

[0051]

[0052]

[0053] in, , is the original diagonal coordinate of the fault signal detection frame, To remove the vertical movement of the block, limit it to the central area. and are the new diagonal coordinates of the detection box after perturbation, and the horizontal coordinates remain unchanged.

[0054] Because the center area of the original detection frame is as high as the removal block times, , so the disturbance method of removing splicing will be executed n times. Let the confidence score of the model prediction after removing splicing disturbance each time be , , then the weight of the area after each culling is:

[0055]

[0056] in, is the raw confidence score, For the The confidence score after removing the splicing, For the The weight of the middle area corresponding to the perturbation. For the perturbation method of truncation compensation, the truncation block moves in the vertical direction, and the height of the truncation block doubles after each perturbation until it reaches the boundary of the central area. The compensation function is to keep the truncated signal continuous. Specifically, pairs of truncation points are found at the truncation point and then connected with a straight line. Assume that the initial information of the truncation block is as follows:

[0057]

[0058]

[0059] After each truncation compensation disturbance, the height of the truncation block changes as follows: is the number of perturbations:

[0060]

[0061] in, To cut off the initial height, is the initial split times of the truncation block, is the height of the initial truncation block, For the The height of the truncation block after the disturbance. The diagonal coordinate position of the fault signal after truncation compensation is, Indicates passing Secondary disturbance:

[0062]

[0063] Assume that the confidence score of the model prediction output after each truncation compensation disturbance is , is the total number of disturbances at both ends of the fault signal, and the corresponding weight after each disturbance is:

[0064]

[0065] in, For the The confidence score after truncation compensation is given by and The resulting disturbance weights are the weights of how each region of the fault signal in the image affects the model. Even if the image contains multiple fault signals, this method can still be used to calculate the weights of how the weights of all the fault signal regions affect the model.

[0066] The third part uses a visualization method based on the weights obtained in the second part. Specifically, the weight of each area of the fault signal can be superimposed on the original image in the form of a heat map. The larger the weight, the darker and redder the area, and vice versa. This can intuitively reflect the impact of different areas of the fault signal on the model. The generation of a visual heat map specifically includes:

[0067] Weight normalization: The weights of different regions of each fault signal are independently normalized to map them to the interval 0 and 1, where the maximum weight value corresponds to 1 and the minimum weight value corresponds to 0;

[0068] Color mapping rules: The normalized weight values are converted to RGB color values through linear interpolation. A weight value of 1 corresponds to red (RGB: 255, 0, 0), and a weight value of 0 corresponds to blue (RGB: 0, 0, 255). Intermediate weight values are gradually transitioned proportionally.

[0069] Transparency adjustment: adjust the transparency of the heat map and the original image to 50%, and then overlay the two;

[0070] Dynamic range adaptation: For different fault signals, their normalized maximum and minimum weight values are calculated independently, ensuring that the thermal color mapping of each fault signal is based only on its own weight distribution and is not interfered with by other signals.

[0071] This method uses a perturbation method that eliminates and splices the middle region and compensates by truncating the ends. This avoids the disruption of waveform signal continuity caused by traditional occlusion methods, ensuring that the perturbed signal retains its original characteristics and reducing misinterpretations. Based on the difference in confidence scores before and after the perturbation, the influence of different regions of the fault signal on the model prediction is accurately calculated, clarifying the role of key areas. A heat map is superimposed on the original image, visually displaying the importance of different regions using color depth (red to blue), helping users quickly understand the model's decision logic.

[0072] It is worth noting that, based on the results of weight analysis, model training strategies or signal processing methods can be adjusted to improve fault detection accuracy and model reliability. This approach is applicable to a variety of signal detection scenarios in industrial equipment and communications, including current, voltage, and electromagnetic waveforms, addressing the need for fault diagnosis for diverse waveform signals. The elimination and splicing method does not introduce new features such as occlusions, reducing the possibility of model misjudgment due to perturbations and enhancing the objectivity of interpretable analysis.

[0073] The above specific embodiments are merely several optional embodiments of the present invention. Based on the technical solutions of the present invention and the relevant inspirations of the above embodiments, those skilled in the art may make various alternative improvements and combinations to the above specific embodiments.

Claims

1. A disturbance-based interpretable waveform signal fault detection method, characterized in that: The following steps are involved: Convert the collected waveform signal data into image data, annotate the fault signals in the image, and construct training sets, validation sets, and test sets; Use the training set and validation set to train the YOLOv8 target detection model to enable it to identify fault signals in waveform signal images; Perform perturbation processing on the fault signal images in the test set; The perturbed image is input into the trained YOLOv8 model for prediction, and the perturbation weights of the middle area and the two end areas are calculated respectively; Generate a visual heat map based on the calculated weights and overlay it on the original image, using color depth to indicate the influence of different regions on the model prediction. Perform disturbance processing on the fault signal images in the test set, specifically including: Eliminate splicing disturbances in the middle area: In the vertical center area of the fault signal detection frame, divide it according to the preset ratio and remove multiple heights of , and splice the removed areas to maintain the continuity of the waveform signal, where is the height of the center area, and the ratio of the vertical range of the center area of the detection frame to the two end areas is 6:4; Truncation compensation disturbance at both ends: The two end areas of the fault signal detection frame in the vertical direction are gradually truncated and the truncation height is extended to the center boundary. At the same time, the truncation points are connected by straight lines to compensate for the disturbance to ensure signal integrity. The above two disturbances are not performed at the same time.

2. The method for detecting faults in a waveform signal based on disturbance interpretability according to claim 1, wherein: Calculating the disturbance weights of the middle area and the two end areas separately includes: The formula for calculating the weight of the middle area is: in, is the raw confidence score, For the The confidence score after removing the splicing, For the The weight of the middle area corresponding to the secondary disturbance; The formula for calculating the weight of the two end areas is: ,in, For the Confidence score after sub-truncation compensation.

3. The method for detecting faults in a disturbance-based interpretable waveform signal according to claim 1, wherein: The specific implementation of removing the splicing disturbance in the middle area includes: The height of the center area of the detection frame is: ,in, is the total vertical height of the fault signal detection frame, is the height of the center area, and the height of the culled block meets , , and move the range vertically Internal Execution perturbations, where is the number of splits to remove the block, is the height of the vertical sub-area to be removed each time, , is the original diagonal ordinate of the fault signal detection frame, The vertical movement position of the culling block is limited to the central area.

4. The method for detecting faults in a disturbance-based interpretable waveform signal according to claim 1, wherein: The specific implementation of the truncation compensation disturbance at both end regions includes: The initial cut-off height is After each disturbance, the cutoff height is expanded to , , until it reaches the boundary of the central area and satisfies ,in, is the initial split times of the truncation block, is the height of the initial truncation block, For the The height of the truncated block after the perturbation.

5. The method for detecting faults in a waveform signal based on disturbance interpretability according to claim 1, wherein: The generation of the visual heat map includes: Weight normalization: The weights of different regions of each fault signal are independently normalized to map them to the interval 0 and 1, where the maximum weight value corresponds to 1 and the minimum weight value corresponds to 0; Color mapping rules: The normalized weight values are converted to RGB color values through linear interpolation. A weight value of 1 corresponds to red (RGB: 255, 0, 0), and a weight value of 0 corresponds to blue (RGB: 0, 0, 255). Intermediate weight values are gradually transitioned proportionally. Transparency adjustment: adjust the transparency of the heat map and the original image to 50%, and then overlay the two; Dynamic range adaptation: For different fault signals, their normalized maximum and minimum weight values are calculated independently, ensuring that the thermal color mapping of each fault signal is based only on its own weight distribution and is not interfered with by other signals.

6. The method for detecting faults in a waveform signal based on disturbance interpretability according to claim 1, wherein: When the waveform signal is converted into image data, wavelet transform or Fourier transform is used for noise reduction.

7. The method for detecting faults in a waveform signal based on disturbance interpretability according to claim 1, wherein: The diagonal coordinates of the fault signal detection box output by the YOLOv8 model are and , whose confidence score is .

Images