Traction motor base plate oil leakage fault identification method, device and equipment and medium

Abstract

The invention discloses an oil leakage fault identification method and device based on a traction motor bottom plate, equipment and a medium, and relates to the technical field of image detection. The target segmentation model obtained by training the combination of the backbone network and the compression excitation mechanism enhances the feature expression ability of the abnormal target area and improves the prediction precision. Determining the area of the oil leakage region based on the predicted segmentation result of the abnormal target region; and determining the oil leakage fault of the traction motor bottom plate according to the relationship between the oil leakage area and a preset value. And the oil leakage area is determined according to the finally obtained segmentation result, so that the accuracy of the oil leakage area is improved, missing detection and error detection are prevented, the recognition rate of oil leakage faults is improved, the reliability of fault detection is improved, and potential safety risks are reduced.

Description

Technical Field

[0001] The present application relates to the field of image detection technology, and in particular to a method, device, equipment and medium for identifying oil leakage faults in a traction motor base plate. Background Art

[0002] The traction motor is the core component of the EMU. If the oil leaks onto the bottom plate, it will not only affect the normal operation of the motor, but may also cause more serious safety hazards.

[0003] Conventional fault detection relies on manual review, with varying individual criteria and corresponding fault identification. This can also lead to missed or false positives, reducing fault detection reliability and increasing potential safety risks.

[0004] Therefore, how to improve the reliability of oil leakage fault detection is a technical problem that those skilled in the art urgently need to solve. Summary of the Invention

[0005] The purpose of this application is to provide a method, device, equipment and medium for identifying oil leakage faults in the traction motor base plate, so as to solve the problem that conventional fault detection uses manual inspection to cause missed reports or false reports, thereby reducing the reliability of fault detection.

[0006] To solve the above technical problems, the present application provides a method for identifying oil leakage faults in a traction motor base plate, comprising: Acquire an image of a traction motor base plate to be inspected; Call the target segmentation model trained by the backbone network and the compression incentive mechanism; Inputting the traction motor bottom plate image into the target segmentation model to predict the segmentation result of the abnormal target area; wherein the abnormal target area is at least the oil leakage area, the foreign matter area and the water flow area; Determine the area of the oil spill based on the segmentation results of the abnormal target area; The oil leakage fault of the traction motor base plate is determined according to the relationship between the area of the oil leakage region and a preset value.

[0007] On the one hand, the training process of the target segmentation model includes: Obtain the initial target segmentation model constructed by the backbone network and the compressed excitation mechanism, as well as the traction motor oil leakage dataset; Inputting the traction motor oil leakage dataset into the initial target segmentation model to predict and obtain an initial segmentation result; Determine the loss result by using the oil spill area corresponding to the initial segmentation result and the loss function; When the loss result is greater than or equal to the first threshold, fine-tuning the model parameters of the initial target segmentation model, and returning to the step of inputting the traction motor oil leakage dataset into the initial target segmentation model until the obtained loss result is less than the first threshold; When the loss result is less than a first threshold, the initial target segmentation model is determined to be the final target segmentation model.

[0008] On the other hand, the backbone network and the compression incentive mechanism are used to construct the initial target segmentation model, including: Based on the middle and last layers of the backbone network, feature extraction is performed on the traction motor baseplate image, and shallow and deep features are output accordingly. The deep features are input into the compression excitation mechanism to recalibrate the feature channels to obtain the feature map output by the encoder; Input the shallow features into the first convolutional layer of the decoder to obtain the first feature map; Upsampling the feature map output by the encoder to obtain an upsampled feature map; The first feature map and the upsampled feature map are concatenated, and the second convolutional layer is applied to refine the concatenated features to obtain the second feature. The size is restored by upsampling to obtain the initial segmentation result corresponding to the traction motor baseplate image.

[0009] On the other hand, the process of determining the loss function includes: Get multiple target loss functions; Determine the weight coefficient corresponding to the contribution of the balance loss term based on multiple objective loss functions; The final loss function is determined based on multiple target loss functions and their corresponding weight coefficients.

[0010] On the other hand, the process of determining the target loss function includes: Obtaining probability parameters of pixels in the traction motor bottom plate image predicted to be oil leakage areas; determining a first objective loss function based on the probability parameter, the focus parameter, and the number of pixels; Get the true labels of the pixels in the traction motor baseplate image; Determine the second objective loss function based on the probability parameter, the true label and the number of samples; Get the boundary area and number of boundary pixels in the true label; The third objective loss function is determined based on the probability parameter, the true label, the boundary area, and the number of boundary pixels.

[0011] On the other hand, the process of determining the traction motor oil leakage dataset includes: Obtain an initial dataset of traction motor bedplate oil leakage; Performing abnormal area positioning processing on the initial data set to obtain a positioning data set; intercepting the positioning data set to obtain a traction motor bottom plate area; The traction motor bottom plate area is locally enhanced to obtain a traction motor oil leakage dataset.

[0012] On the other hand, the oil spill area is determined based on the segmentation results of the abnormal target area, including: Binarizing the regional image of the traction motor bottom plate image corresponding to the abnormal target area to obtain a first image; Set the pixels of the oil leak area in the first image to 1; The area of the oil spill is determined based on the pixel value of 1.

[0013] To solve the above technical problems, the present application further provides a device for identifying oil leakage faults in a traction motor base plate, comprising: An acquisition module, used for acquiring an image of a traction motor base plate to be inspected; The calling module is used to call the target segmentation model trained by the backbone network and the compression incentive mechanism; A prediction module is configured to input the traction motor bottom plate image into a target segmentation model to predict a segmentation result of an abnormal target area; wherein the abnormal target area is at least an oil leakage area, a foreign matter area, and a water flow area; A first determination module is used to determine the area of the oil leak region based on the segmentation result of the abnormal target region; The second determining module is used to determine the oil leakage fault of the traction motor base plate according to the relationship between the area of the oil leakage region and a preset value.

[0014] To solve the above technical problems, the present application further provides an electronic device, comprising: Memory for storing computer programs; The processor is configured to implement the steps of the method for identifying an oil leakage fault of a traction motor base plate when executing the computer program.

[0015] In order to solve the above technical problems, the present application also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the traction motor bottom plate oil leakage fault identification method as described above are implemented.

[0016] This application provides a method for identifying oil leakage faults in a traction motor baseplate. This method utilizes a target segmentation model trained using a backbone network and a compression excitation mechanism as a reference for predicting segmentation results. This unified reference avoids missed oil leakage faults due to different detection standards during manual fault detection. An image of the traction motor baseplate is input into the target segmentation model to predict segmentation results for abnormal target areas. These abnormal target areas include at least oil leakage areas, foreign matter areas, and water flow areas. This application uses the target segmentation model to predict segmentation results for abnormal target areas, improving prediction accuracy. The target segmentation model, trained using a backbone network and a compression excitation mechanism, enhances the feature representation of abnormal target areas and improves prediction accuracy. The area of the oil leakage area is determined based on the predicted segmentation results of the abnormal target area. The presence of an oil leakage fault in the traction motor baseplate is determined based on the relationship between the oil leakage area and a preset value. The oil leakage area is determined based on the final segmentation results, improving accuracy and preventing missed and false detections. This improves the recognition rate of oil leakage faults, enhances the reliability of fault detection, and reduces potential safety risks.

[0017] In addition, the present application also provides a traction motor bottom plate oil leakage fault identification device, electronic equipment and medium, which have the same beneficial effects as the above-mentioned traction motor bottom plate oil leakage fault identification method. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the embodiments of the present application, the following is a brief introduction to the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0019] Figure 1 A flowchart of a method for identifying oil leakage faults in a traction motor base plate provided in an embodiment of the present application; Figure 2 A model architecture diagram of an initial target segmentation model provided in an embodiment of the present application; Figure 3 A flowchart for determining a multi-objective segmentation weight coefficient provided in an embodiment of the present application; Figure 4 A fault diagram identified by a method for identifying a traction motor bottom plate oil leakage fault provided in an embodiment of the present application; Figure 5 A flowchart of another method for identifying oil leakage faults in a traction motor base plate provided in an embodiment of the present application; Figure 6A structural diagram of a traction motor bottom plate oil leakage fault identification device provided in an embodiment of the present application; Figure 7 A structural diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0020] The following will be combined with the accompanying drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0021] The core of this application is to provide a method, device, equipment and medium for identifying oil leakage faults in the traction motor base plate, so as to solve the problem of missed reports or false reports caused by manual inspection in conventional fault detection, thereby reducing the reliability of fault detection.

[0022] In order to enable those skilled in the art to better understand the present application, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0023] Conventional traction motor baseplate oil leakage fault detection cannot take into account both no missed faults and a small number of false alarms, resulting in low fault identification accuracy. The traction motor baseplate oil leakage fault identification method provided in this application can solve the above technical problems.

[0024] Figure 1 A flowchart of a method for identifying oil leakage faults in a traction motor base plate is provided in an embodiment of the present application. Figure 1 As shown, the method includes: S11: Acquire an image of the traction motor base plate to be inspected; S12: Call the target segmentation model trained by the backbone network and the compression incentive mechanism; S13: Inputting the traction motor bottom plate image into the target segmentation model to predict and obtain a segmentation result of an abnormal target area; wherein the abnormal target area is at least an oil leakage area, a foreign matter area, and a water flow area; S14: determining the area of the oil leak based on the segmentation result of the abnormal target area; S15: Determine the oil leakage fault of the traction motor base plate according to the relationship between the oil leakage area and the preset value.

[0025] Specifically, the traction motor floor image to be inspected in step S11 corresponds to the high-definition train image captured after the train passes through a high-definition industrial linear array camera, using trackside imaging equipment installed on both sides of the railway. Traction motor oil leaks on the floor are affected by factors such as train speed, weather, camera exposure, and the amount of oil leaked, resulting in variable shapes and areas, and thus significant variations in fault images. Therefore, during image data collection, every effort is made to capture images of various shapes to ensure data set diversity. The traction motor floor image of this embodiment corresponds to prior knowledge of hardware equipment, wheelbase information, and related locations. It can be obtained by initially extracting the traction motor floor area from the bottom camera image and then performing contrast enhancement processing to achieve accuracy and diversity during image acquisition, thereby facilitating accurate subsequent recognition.

[0026] The preprocessing process of the traction motor base plate image in this embodiment is not limited and can be data enhancement corresponding to the data set and enhancement of the image itself, etc., to filter the noise data corresponding to the interference signal and achieve anti-interference at the source.

[0027] In step S12, the object segmentation model trained using the backbone network and the compressed excitation mechanism is called. The backbone network here is the core component responsible for extracting features from the input image, that is, extracting rich feature information. This includes not only low-level information such as the image's basic texture and color, but also high-level semantic information such as the object's shape and edges. The performance of the backbone network directly affects the accuracy and efficiency of the object detection task. In this embodiment, the backbone network structure can be a convolutional neural network (CNN) or a Transformer-based architecture. Both approaches are acceptable and are not limited here. The design can be based on actual conditions. It should be noted that if a CNN is used, a deep network can be constructed by stacking multiple convolutional and pooling layers, or a residual learning mechanism can be introduced to effectively alleviate the vanishing / exploding gradient problem in deep networks. A lightweight network is set up for mobile devices and edge computing scenarios to reduce model complexity and computational complexity. If a Transformer-based architecture is used, a self-attention mechanism is required to capture global information.

[0028] The compression excitation mechanism is a module used to enhance the performance of convolutional neural networks. By explicitly modeling the interdependencies between feature channels, it adaptively recalibrates the strength of channel feature responses, thereby enhancing the network's responsiveness to useful features. In this embodiment, the feature channels are recalibrated, and the importance of each feature channel is dynamically adjusted using global information in the channel dimension. This effectively enhances sensitivity to oil leak features, especially enabling adaptive feature enhancement for oil leak areas of different colors and shapes. In this embodiment, the encoder is constructed using a backbone network and a compression excitation mechanism to improve the recognition accuracy of the target segmentation model.

[0029] In step S13, the traction motor baseplate image is input into the target segmentation model to predict the segmentation results of abnormal target areas. Considering that the traction motor baseplate sometimes has traces of water flow and foreign matter, the abnormal target areas are at least divided into oil leakage areas, foreign matter areas, and water flow areas, so as to recognize images with various forms and improve the diversity and breadth of recognition.

[0030] Considering that this application only addresses oil leak fault identification, step S14 determines the area of the oil leak region from the segmentation results of the various abnormal target regions. The area calculation is based on the pixel values of the oil leak region after the oil leak region is identified in the image. This can be accomplished by pixel counting or complex algorithms (such as connected component labeling) to determine the actual area of each pixel.

[0031] In step S15, the oil leakage fault is determined based on the relationship between the area of the oil leakage region and the preset value. When the area of the oil leakage region is greater than the preset value, it is determined that the traction motor base plate is leaking oil.

[0032] An embodiment of the present application provides a method for identifying oil leakage faults in a traction motor baseplate. This method utilizes a target segmentation model trained using a backbone network and a compression excitation mechanism as a reference for predicting segmentation results. This unified reference avoids missed oil leakage faults due to different detection standards during manual fault detection. An image of the traction motor baseplate is input into the target segmentation model to predict segmentation results for abnormal target areas. These abnormal target areas include at least oil leakage areas, foreign matter areas, and water flow areas. This method improves prediction accuracy by predicting segmentation results for abnormal target areas using the target segmentation model. The target segmentation model, trained using a backbone network and a compression excitation mechanism, enhances the feature representation of abnormal target areas and improves prediction accuracy. The area of the oil leakage area is determined based on the predicted segmentation results for the abnormal target area. The presence of an oil leakage fault in the traction motor baseplate is determined based on the relationship between the oil leakage area and a preset value. The oil leakage area is determined based on the final segmentation results, improving accuracy and preventing missed and false detections. This improves the recognition rate of oil leakage faults, enhances the reliability of fault detection, and reduces potential safety risks.

[0033] In some embodiments, the object segmentation model training process includes: Obtain the initial target segmentation model constructed by the backbone network and the compressed excitation mechanism, as well as the traction motor oil leakage dataset; Input the traction motor oil leakage dataset into the initial target segmentation model to predict the initial segmentation results; The oil spill area corresponding to the initial segmentation result and the loss function are used to determine the loss result; When the loss result is greater than or equal to the first threshold, fine-tuning the model parameters of the initial target segmentation model, and returning to the step of inputting the traction motor oil leakage dataset into the initial target segmentation model until the loss result obtained is less than the first threshold; When the loss result is less than the first threshold, the initial target segmentation model is determined to be the final target segmentation model.

[0034] Specifically, an initial target segmentation model is constructed using a backbone network and a compression-excitation mechanism. A traction motor oil leakage dataset is obtained. This dataset can be obtained from a mature database or from field-collected images that have been preprocessed (e.g., denoising, contrast enhancement, color correction, etc.) to improve image recognizability. These two methods are not limited here and can be obtained based on actual conditions.

[0035] The traction motor oil leakage dataset is input into the initial target segmentation model to predict the initial segmentation result. The oil leakage area (prediction result) and the true label corresponding to the initial segmentation result are substituted into the loss function to calculate the loss result. The loss result here is the difference between the prediction result and the true value. By analyzing the size of the loss result and the first threshold, if the loss value is smaller, it means that the model's prediction result is closer to the true value and the model performance is better. In this embodiment, when the loss result is less than the first threshold, the initial target segmentation model is determined to be the final target segmentation model. When the loss result is greater than or equal to the first threshold, it is necessary to fine-tune the model parameters of the initial target segmentation model to obtain a new model for continued training until the loss result is less than the first threshold.

[0036] Additionally, you can continuously monitor the training loss and validation loss during training. The training loss should gradually decrease with the number of training rounds, while the validation loss is used to assess the model's performance on unseen data. If the training loss continues to decrease, but the validation loss begins to rise or stops decreasing, the model may be overfitting. If both the training loss and validation loss are high and decrease slowly, the model may be underfitting. When both the training loss and validation loss stabilize and no longer decrease significantly, the model can be considered converged. Use the above methods to determine the loss results and optimize and adjust the model.

[0037] In addition, based on the loss result, a limit on the number of iterations can be added. As long as the number of iterations reaches the preset number of iterations, the target initial model of the most recent iteration will be used as the final target segmentation model.

[0038] This embodiment provides a method for subsequent training by establishing an initial target segmentation model, using a loss function to evaluate key indicators of model performance, and monitoring and analyzing loss results to determine the quality of model training and ensure an improvement in the model's recognition rate.

[0039] In some embodiments, the backbone network and the compression excitation mechanism are used to construct an initial object segmentation model, including: Based on the middle and last layers of the backbone network, feature extraction is performed on the traction motor baseplate image, and shallow and deep features are output accordingly. The deep features are input into the compression excitation mechanism to recalibrate the feature channels to obtain the feature map output by the encoder; Input the shallow features into the first convolutional layer of the decoder to obtain the first feature map; Upsampling the feature map output by the encoder to obtain an upsampled feature map; The first feature map and the upsampled feature map are concatenated, and the second convolutional layer is applied to refine the concatenated features to obtain the second feature. The size is restored by upsampling to obtain the initial segmentation result corresponding to the traction motor baseplate image.

[0040] Figure 2 A model architecture diagram of an initial target segmentation model provided in an embodiment of the present application, such as Figure 2 As shown, the model mainly includes a network encoder and a decoder. The network encoder part uses Swin Transformer as the backbone network. Through the windowed self-attention mechanism and hierarchical structure, it effectively captures the long-distance dependencies and global context information in the traction motor base plate image. The middle layer of the backbone network extracts features from the image to obtain shallow features, and the last layer of the backbone network extracts features from the image to obtain deep features. This embodiment can capture long-range dependencies and multi-scale information in the traction motor base plate image through the windowed self-attention mechanism and hierarchical feature extraction, and has significant advantages in the recognition of complex structures such as oil leakage areas and base plate edges.

[0041] The deep features are fed into the module corresponding to the compression excitation mechanism, and the deep feature channels are adaptively weighted. This corresponds to recalibrating the feature channels, precisely enhancing the weights of feature channels related to oil leaks and suppressing the influence of irrelevant channels. This allows the model to focus more on the texture, color, and distribution characteristics of the oil leak area, effectively distinguishing between oil leaks and normal oil or water flow on the traction motor baseplate, and improving the robustness of oil leak detection in complex backgrounds and changing environments.

[0042] Specifically, in the compression excitation mechanism, each feature channel of the deep features is first compressed to extract the important information of the channel. This is mainly achieved by compressing the spatial dimension of the image into a scalar through global average pooling to retain the important information of the channel. Continuing through the excitation mechanism, the weight of each channel is calculated. This weight automatically changes according to the content of the input image, thereby enhancing the feature channels related to the oil leak and suppressing the channels unrelated to the oil leak. The nonlinear relationship between channels is mainly learned through two fully connected layers to generate weights. The generated weights are multiplied by the original feature map for recalibration, enhancing important features and suppressing unimportant features. The compression excitation mechanism dynamically adjusts the importance of each feature channel through the global information of the channel dimension, effectively enhancing the sensitivity to oil leak features, especially for oil leak areas of different colors and shapes, and can adaptively enhance features.

[0043] The encoder design can focus more on the texture, color and distribution characteristics of the oil leakage area, effectively distinguishing oil leakage from normal oil or water flow on the bottom plate surface, and improving recognition accuracy.

[0044] After the feature map output by the encoder is subjected to 1X1 convolution processing, the feature map is upsampled to restore the spatial resolution and obtain the upsampled features.

[0045] The shallow features are fed into the encoder's first convolutional layer (1x1 convolution). After the upsampled features are concatenated with the shallow feature map, multi-scale information is fused (the deep features are fused with the shallow features) to generate the secondary features. The concatenated feature map undergoes a 3x3 convolution to extract richer features. Finally, the concatenated feature map is upsampled to restore the original size and output as the transformed data. This is the model's predicted segmentation result, also known as the initial segmentation result.

[0046] The target segmentation model establishment process provided in this embodiment, through the construction process of the backbone network and the compression excitation mechanism, effectively captures the long-distance dependencies in the image through the windowed self-attention mechanism, and accurately enhances the feature channel weights related to oil spills through the "compression-excitation" mechanism to improve recognition accuracy.

[0047] In some embodiments, the process of determining the loss function includes: Get multiple target loss functions; Determine the weight coefficient corresponding to the contribution of the balance loss term based on multiple objective loss functions; The final loss function is determined based on multiple target loss functions and their corresponding weight coefficients.

[0048] Specifically, in this embodiment, the setting of the loss function can be a single loss function or a combination of multiple loss functions. In this embodiment, multiple target loss functions are used to determine the weight coefficient corresponding to the contribution of the balanced loss term.

[0049] Treat weights as hyperparameters, determining them through manual adjustment or grid search. Alternatively, use adaptive weights to dynamically adjust weights to balance the contribution of loss terms. A sliding window strategy can be used to record the value of each loss term in the most recent iteration and adjust accordingly. Alternatively, use uncertainty metrics to balance loss terms. These are not specific here and can be set based on actual circumstances.

[0050] During adaptive weight adjustment, the value of each loss item in the most recent preset iterations is recorded during training, the average of each loss item is calculated, and the weight is dynamically adjusted based on the average loss item and the expected ratio. For example, if the average loss item is too high, its weight is reduced.

[0051] In the weight setting of this embodiment, Figure 3 A flow chart for determining a multi-objective segmentation weight coefficient is provided in an embodiment of the present application, such as Figure 3As shown, after inputting the image data, the target segmentation model is transformed based on the preset weight coefficient to obtain the prediction result, the prediction result is processed by the loss function to obtain the loss value, and the optimizer is optimized according to the actual label result and the loss value to obtain the updated weight coefficient.

[0052] The number of target loss functions is not specified here and can be set based on actual circumstances. The specific functions corresponding to the target loss functions can be set based on the image features of the traction motor floor image according to actual circumstances and are not limited here. The final loss function is determined based on multiple target loss functions and their corresponding weight coefficients, primarily by summing them.

[0053] The multiple objective loss functions and their corresponding weight coefficients provided in this embodiment determine the final loss function. By reasonably allocating weights and balancing the collinearity of different tasks or objectives, the overall performance, model robustness and training efficiency of the model are improved.

[0054] In some embodiments, the process of determining the target loss function includes: Obtaining probability parameters of pixels in the traction motor bottom plate image predicted to be oil leakage areas; determining a first objective loss function based on the probability parameter, the focus parameter, and the number of pixels; Get the true labels of the pixels in the traction motor baseplate image; Determine the second objective loss function based on the probability parameter, the true label and the number of samples; Get the boundary area and number of boundary pixels in the true label; The third objective loss function is determined based on the probability parameter, the true label, the boundary area, and the number of boundary pixels.

[0055] Specifically, this embodiment takes into account the problem of class imbalance and integrates loss functions to improve the accuracy of oil leak fault identification. The first objective loss function is based on the improvement of the Focal Loss function, the second objective loss function is based on the improvement of the Dice Loss function, and the third objective loss function is based on the improvement of the boundary-aware loss function. The specific loss functions are defined as follows: ; in, 、 and The weight coefficients used to balance the contribution of each loss item are 1.0, 1.0, and 0.5 respectively; is the first objective loss function, is the second objective loss function, is the third objective loss function.

[0056] The formula of the first objective loss function is as follows: ; in, is a pixel, is the number of pixels, Pixels The probability of being predicted as an oil spill area, is the focus parameter, set to 2.

[0057] This first objective loss function reduces the weight of easily identifiable examples and increases the weight of difficult examples, allowing the model to focus more on hard-to-identify oil leak areas. This is particularly important for detecting subtle, early-stage oil leaks or reflective oil leaks on the floor of a train.

[0058] The formula of the second objective loss function is as follows: ; in, Pixels The true label.

[0059] This second objective loss function focuses on the overlap of segmented regions and is unaffected by class imbalance, making it particularly suitable for improving segmentation accuracy in small oil spills. It can also effectively improve detection sensitivity when the oil spill area is extremely small or discretely distributed.

[0060] The formula of the third objective loss function is as follows: ; in, represents the boundary region in the true label, is the number of boundary pixels.

[0061] The third objective loss function places special emphasis on the boundary pixels of the oil leak area, improving the accuracy of edge segmentation. In the detection of oil leaks in the traction motor baseplate, accurate segmentation of the oil leak boundary is crucial for determining the extent of the leak and making subsequent repair decisions.

[0062] This embodiment provides three objective loss functions: Focal Loss, which adjusts the weighting of easy and difficult samples to enhance learning of subtle oil leak areas; Dice Loss, which improves the segmentation accuracy of small oil leaks; and Boundary Perception Loss, which enhances the precise localization of leak edges. This multi-objective optimization loss function design leverages the global modeling advantages of SwinTransformer and the feature enhancement capabilities of the SE module. While addressing the imbalance between oil leak and background categories in EMU floor images, it also effectively improves the model's ability to identify oil leaks of varying forms and severity.

[0063] In some embodiments, the process of determining the traction motor oil leakage dataset includes: Obtain an initial dataset of traction motor bedplate oil leakage; The initial data set is processed to locate the abnormal area to obtain the positioning data set; The positioning data set is intercepted and processed to obtain the traction motor base plate area; The traction motor baseplate area is locally enhanced to obtain the traction motor oil leakage dataset.

[0064] Specifically, during the dataset creation process, trackside imaging equipment was installed on both sides of the railway. High-definition industrial linear array cameras were used to capture high-definition train images after the trains passed through them. Traction motor oil leaks on the floorboards can vary in shape and area due to factors such as train speed, weather, camera exposure, and the amount of oil leaked. This results in significant variation in the fault images. Therefore, during the image data collection process, efforts were made to include images of various shapes to ensure dataset diversity.

[0065] The initial dataset is processed to locate abnormal areas, generating localization data. This localization can be performed using a localization model or other methods, which are not limited here. The initial dataset used for localization consists of a grayscale image dataset and a corresponding labeled image dataset. Because the traction motor baseplate sometimes exhibits water flow and foreign objects, the images are labeled into three categories: oil leakage, water flow, and foreign objects. A marking tool is used to mark the areas of oil leakage, foreign objects, and water flow on the traction motor baseplate. Polygons are drawn to cover the component areas. After labeling, a corresponding binary image is generated for each image. Pixel values for oil leakage areas are 1, pixels for foreign objects are 2, pixels for water flow are 3, and all other areas are 0. The localized traction motor baseplate dataset is used to detect traction motor baseplate oil leakage faults. The sample dataset needs to cover images with various forms of baseplate oil leakage. To improve the generalization of the algorithm, data augmentation is still required. Augmentation methods include image rotation, translation, scaling, and mirroring. Each augmentation is performed under random conditions to maximize sample diversity and applicability.

[0066] The positioning dataset is intercepted to preliminarily extract the traction motor floor area. Based on prior knowledge of the hardware, wheelbase information, and related locations, the traction motor floor area can be preliminarily intercepted from the bottom camera image. The traction motor floor area is locally enhanced to obtain this dataset. Due to differences in camera angles and distances at different sites, the brightness levels of the acquired images vary. Some images are too dark to clearly observe the oil leaking area on the traction motor floor. Therefore, before inputting the image into the target segmentation model, it is necessary to perform local adaptive contrast enhancement on it. The enhancement process is not limited here and can use histogram equalization, linear transformation, nonlinear transformation, etc., which can be set according to actual conditions.

[0067] The process of determining the traction motor oil leakage dataset provided in this embodiment ensures that the collected dataset covers images of various shapes to the greatest extent possible, thereby improving diversity and applicability.

[0068] In some embodiments, determining the area of the oil leak region based on the segmentation result of the abnormal target region includes: Binarizing the regional image of the traction motor bottom plate image corresponding to the abnormal target area to obtain a first image; Set the pixels of the oil leak area in the first image to 1; The area of the oil spill is determined based on the pixel value of 1.

[0069] Specifically, after predicting and determining the segmentation results of the abnormal target area, the regional image is binarized to obtain a first image, where pixels corresponding to the oil leak area are set to 1, and pixels corresponding to the abnormal target area in the non-oil leak area are set to 0. The area of the region with a pixel value of 1 is calculated as the area of the oil leak area. A preset value is also set. If the area of the oil leak area is greater than the preset value, a fault alarm is issued for this part of the traction motor baseplate area. If it is not greater than the preset value, the next traction motor baseplate image is processed.

[0070] Figure 4 The present application provides a fault diagram identified by a method for identifying a traction motor bottom plate oil leakage fault, such as Figure 4 As shown, the box corresponds to the oil leakage area.

[0071] The oil leak area determination process provided in this embodiment uses binarization processing to determine the area corresponding to the pixel value, simplifying the pixel value into two states, reducing the complexity of the image data, speeding up the subsequent oil leak area determination process, and highlighting the key features of the image, facilitating subsequent image analysis and processing, thereby improving the efficiency of oil leak fault identification.

[0072] Figure 5 A flowchart of another method for identifying oil leakage faults in a traction motor base plate provided in an embodiment of the present application is shown in FIG. Figure 5 Shown, including: S21: collect images; S22: Create a traction motor baseplate oil leakage sample dataset; S23: Data enhancement; S24: Calculate the weight of the sample data set; S25: Obtaining a segmentation result of the traction motor base plate image after model recognition and prediction; S26: Obtain the oil leak location and perform binarization processing; S27: Calculate the area of oil spill; S28: Determine whether the oil leakage area is greater than a preset value. If yes, proceed to step S29; if not, proceed to step S30; S29: Fault alarm; S30: Acquire the next traction motor base plate image.

[0073] The above describes in detail various embodiments corresponding to the method for identifying oil leakage faults in the traction motor base plate. On this basis, the present application also discloses a device for identifying oil leakage faults in the traction motor base plate corresponding to the above method. Figure 6 This is a structural diagram of a traction motor bottom plate oil leakage fault identification device provided in an embodiment of the present application. Figure 6 As shown, the traction motor bottom plate oil leakage fault identification device includes: An acquisition module 11 is used to acquire an image of a traction motor base plate to be detected; A calling module 12 is used to call the target segmentation model trained by the backbone network and the compression excitation mechanism; The prediction module 13 is configured to input the traction motor bottom plate image into the target segmentation model to predict the segmentation result of the abnormal target area; wherein the abnormal target area is at least the oil leakage area, the foreign matter area and the water flow area; A first determining module 14 is configured to determine the area of the oil leak based on the segmentation result of the abnormal target area; The second determining module 15 is configured to determine the oil leakage fault of the traction motor base plate according to the relationship between the oil leakage area and a preset value.

[0074] Since the embodiments of the device part correspond to the above embodiments, the embodiments of the device part please refer to the description of the embodiments of the method part, and will not be repeated here.

[0075] For an introduction to a traction motor bottom plate oil leakage fault identification device provided in this application, please refer to the above method embodiment. This application will not go into details here. It has the same beneficial effects as the above-mentioned traction motor bottom plate oil leakage fault identification method.

[0076] Figure 7 A structural diagram of an electronic device provided in an embodiment of the present application, such as Figure 7 As shown, the electronic device includes: Memory 21, for storing computer programs; The processor 22 is configured to implement the steps of the traction motor bottom plate oil leakage fault identification method when executing the computer program.

[0077] The electronic device provided in this embodiment may include but is not limited to a tablet computer, a laptop computer, or a desktop computer.

[0078] The processor 22 may include one or more processing cores, such as a quad-core processor or an octa-core processor. The processor 22 may be implemented in at least one of the following hardware forms: a digital signal processor (DSP), a field-programmable gate array (FPGA), or a programmable logic array (PLA). The processor 22 may also include a main processor and a coprocessor. The main processor is a processor for processing data in the awake state, also known as a central processing unit (CPU); the coprocessor is a low-power processor for processing data in the standby state. In some embodiments, the processor 22 may be integrated with a graphics processing unit (GPU), which is responsible for rendering and drawing the content required to be displayed on the display screen. In some embodiments, the processor 22 may also include an artificial intelligence (AI) processor, which is responsible for processing computing operations related to machine learning.

[0079] The memory 21 may include one or more computer-readable storage media, which may be non-transitory. The memory 21 may also include high-speed random access memory, and non-volatile memory, such as one or more disk storage devices, flash memory storage devices. In this embodiment, the memory 21 is at least used to store the following computer program 211, wherein, after the computer program is loaded and executed by the processor 22, it can implement the relevant steps of the method for identifying the oil leakage fault of the traction motor base plate disclosed in any of the aforementioned embodiments. In addition, the resources stored in the memory 21 may also include an operating system 212 and data 213, etc., and the storage method may be temporary storage or permanent storage. Among them, the operating system 212 may include Windows, Unix, Linux, etc. The data 213 may include but is not limited to data involved in the method for identifying the oil leakage fault of the traction motor base plate, etc.

[0080] In some embodiments, the electronic device may further include a display screen 23 , an input / output interface 24 , a communication interface 25 , a power supply 26 , and a communication bus 27 .

[0081] Those skilled in the art will understand that Figure 7 The structure shown in the figure does not constitute a limitation of the electronic device, and may include more or fewer components than shown in the figure.

[0082] The processor 22 implements the traction motor bottom plate oil leakage fault identification method provided by any of the above embodiments by calling the instructions stored in the memory 21.

[0083] For an introduction to an electronic device provided by this application, please refer to the above method embodiment, which will not be repeated here. It has the same beneficial effects as the above method for identifying oil leakage faults based on the traction motor bottom plate.

[0084] Furthermore, the present application also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by the processor 22, the steps of the above-mentioned method for identifying oil leakage fault based on the traction motor base plate are implemented.

[0085] It is understandable that if the method in the above embodiment is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium and executes all or part of the steps of the methods of each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, and other media that can store program code.

[0086] For an introduction to a computer-readable storage medium provided in this application, please refer to the above method embodiment, which will not be repeated here in this application. It has the same beneficial effects as the above method for identifying oil leakage faults based on the traction motor base plate.

[0087] The above is a detailed introduction to the method, device, equipment and medium for identifying oil leakage faults based on the traction motor bottom plate provided by the present application. The various embodiments in the specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same and similar parts between the various embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part description. It should be pointed out that for ordinary technicians in this technical field, without departing from the principles of this application, several improvements and modifications can be made to the present application, and these improvements and modifications also fall within the scope of protection of this application.

[0088] It should also be noted that, in this specification, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus comprising the element.

Claims

1. A method for identifying oil leakage faults in a traction motor base plate, characterized in that: include: Acquire an image of a traction motor base plate to be inspected; Call the target segmentation model trained by the backbone network and the compression incentive mechanism; Inputting the traction motor bottom plate image into the target segmentation model to predict the segmentation result of the abnormal target area; wherein the abnormal target area is at least the oil leakage area, the foreign matter area and the water flow area; Determine the area of the oil spill based on the segmentation results of the abnormal target area; The oil leakage fault of the traction motor base plate is determined according to the relationship between the area of the oil leakage region and a preset value.

2. The method for identifying oil leakage fault based on the traction motor base plate according to claim 1 is characterized in that: The training process of the target segmentation model includes: Obtain the initial target segmentation model constructed by the backbone network and the compressed excitation mechanism, as well as the traction motor oil leakage dataset; Inputting the traction motor oil leakage dataset into the initial target segmentation model to predict and obtain an initial segmentation result; Determine the loss result by using the oil spill area corresponding to the initial segmentation result and the loss function; When the loss result is greater than or equal to the first threshold, fine-tuning the model parameters of the initial target segmentation model, and returning to the step of inputting the traction motor oil leakage dataset into the initial target segmentation model until the obtained loss result is less than the first threshold; When the loss result is less than a first threshold, the initial target segmentation model is determined to be the final target segmentation model.

3. The method for identifying oil leakage fault based on the traction motor base plate according to claim 2 is characterized in that: The initial target segmentation model is constructed by the backbone network and the compression incentive mechanism, including: Based on the middle and last layers of the backbone network, feature extraction is performed on the traction motor baseplate image, and shallow and deep features are output accordingly. The deep features are input into the compression excitation mechanism to recalibrate the feature channels to obtain the feature map output by the encoder; Input the shallow features into the first convolutional layer of the decoder to obtain the first feature map; Upsampling the feature map output by the encoder to obtain an upsampled feature map; The first feature map and the upsampled feature map are concatenated, and the second convolutional layer is applied to refine the concatenated features to obtain the second feature. The size is restored by upsampling to obtain the initial segmentation result corresponding to the traction motor baseplate image.

4. The method for identifying oil leakage fault based on the traction motor base plate according to claim 2 is characterized in that: The process of determining the loss function includes: Get multiple target loss functions; Determine the weight coefficient corresponding to the contribution of the balance loss term based on multiple objective loss functions; The final loss function is determined based on multiple target loss functions and their corresponding weight coefficients.

5. The method for identifying oil leakage fault based on the traction motor base plate according to claim 4 is characterized in that: The process of determining the target loss function includes: Obtaining probability parameters of pixels in the traction motor bottom plate image predicted to be oil leakage areas; determining a first objective loss function based on the probability parameter, the focus parameter, and the number of pixels; Get the true labels of the pixels in the traction motor baseplate image; Determine the second objective loss function based on the probability parameter, the true label and the number of samples; Get the boundary area and number of boundary pixels in the true label; The third objective loss function is determined based on the probability parameter, the true label, the boundary area, and the number of boundary pixels.

6. The method for identifying oil leakage fault based on the traction motor base plate according to claim 2 is characterized in that: The process of determining the traction motor oil leakage dataset includes: Obtain an initial dataset of traction motor bedplate oil leakage; Performing abnormal area positioning processing on the initial data set to obtain a positioning data set; intercepting the positioning data set to obtain a traction motor bottom plate area; The traction motor bottom plate area is locally enhanced to obtain a traction motor oil leakage dataset.

7. The method for identifying oil leakage fault based on the traction motor base plate according to claim 1 is characterized in that: Determine the oil spill area based on the segmentation results of the abnormal target area, including: Binarizing the regional image of the traction motor bottom plate image corresponding to the abnormal target area to obtain a first image; Set the pixels of the oil leak area in the first image to 1; The area of the oil spill is determined based on the pixel value of 1.

8. A device for identifying oil leakage faults based on the traction motor base plate, characterized in that: include: An acquisition module, used for acquiring an image of a traction motor base plate to be inspected; The calling module is used to call the target segmentation model trained by the backbone network and the compression incentive mechanism; A prediction module is configured to input the traction motor bottom plate image into a target segmentation model to predict a segmentation result of an abnormal target area; wherein the abnormal target area is at least an oil leakage area, a foreign matter area, and a water flow area; A first determination module is used to determine the area of the oil leak region based on the segmentation result of the abnormal target region; The second determining module is used to determine the oil leakage fault of the traction motor base plate according to the relationship between the area of the oil leakage region and a preset value.

9. An electronic device, characterized in that: include: Memory for storing computer programs; A processor is configured to implement the steps of the method for identifying oil leakage fault based on the traction motor base plate as described in any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and when the computer program is executed by the processor, the steps of the method for identifying oil leakage fault based on the traction motor bottom plate according to any one of claims 1 to 7 are implemented.

Images