Chassis fault maintenance method and device based on improved YOLOv5 algorithm, equipment and medium

Abstract

The invention provides a chassis fault maintenance method and device based on an improved YOLOv5 algorithm, equipment and a medium. The chassis fault maintenance method comprises the steps that sound signal features of a chassis system in the running process of a special vehicle are acquired; obtaining an improved YOLOv5 model obtained by replacing a preset network layer of the original YOLOv5 model, wherein the preset network layer comprises a down-sampling layer, an SPPF layer, a target detection layer and a multi-parallel network layer; inputting the sound signal features into an improved YOLOv5 model, and determining fault feature map data corresponding to a chassis system of the special vehicle based on an output result of the improved YOLOv5 model; and performing maintenance processing on the chassis system of the special vehicle based on the fault feature map data corresponding to the chassis system of the special vehicle. Therefore, the sound signal characteristics of the chassis system are collected in the running process of the special vehicle, fault recognition is carried out based on the improved YOLOv5 model, then fault maintenance is carried out, and the fault maintenance efficiency is effectively improved.

Description

Technical Field

[0001] Embodiments of the present disclosure relate to the technical field of fault repair, and in particular, to a chassis fault repair method, apparatus, device and medium based on an improved YOLOv5 algorithm. Background Art

[0002] Under the trend of coordinated development of science and technology and innovative technology, the design of special vehicles has gradually developed towards high speed, high safety and high load-bearing capacity, bringing convenience to people. At the same time, the probability of failure of special vehicle equipment has also increased accordingly. There are many types and numbers of internal components of special vehicles. During operation, all components need to cooperate to ensure the safety of special vehicle driving. Once a component in a special vehicle fails, it will form a certain safety hazard, affecting the normal driving of the special vehicle. In severe cases, it will cause major traffic safety accidents and endanger human life. Among the many components of special vehicles, the chassis system, as an important component, plays a load-bearing role. Its basic performance is directly related to the performance of special vehicle driving. Once a failure occurs, it will affect the stability, controllability and braking of the special vehicle.

[0003] In the related technology, when chassis engine inspection is carried out, it is mainly implemented based on manpower, with professionals manually investigating and troubleshooting faulty components and faulty locations, and then professionally repairing the faulty engine components.

[0004] However, when using existing methods to perform fault repair, the fault detection results are not accurate enough, resulting in low repair efficiency. Summary of the invention

[0005] The embodiments described herein provide a chassis fault repair method, apparatus, device and medium based on an improved YOLOv5 algorithm, which overcomes the above-mentioned problems.

[0006] In a first aspect, according to the content of the present disclosure, a chassis fault repair method based on an improved YOLOv5 algorithm is provided, comprising:

[0007] Acquiring the sound signal characteristics of the chassis system during the driving of the special vehicle, wherein the sound signal characteristics include: sound frequency characteristics, sound amplitude characteristics and sound time domain characteristics;

[0008] Obtain an improved YOLOv5 model obtained by replacing a preset network layer of the original YOLOv5 model, wherein the preset network layer includes: a downsampling layer, an SPPF layer, a target detection layer, and a multi-parallel network layer;

[0009] Inputting the sound signal feature into the improved YOLOv5 model, and determining the fault characteristic map data corresponding to the chassis system of the special vehicle based on the output result of the improved YOLOv5 model, wherein the fault characteristic map data includes: fault component, fault location and fault type;

[0010] Based on the fault characteristic diagram data corresponding to the chassis system of the special vehicle, the chassis system of the special vehicle is repaired.

[0011] In a second aspect, according to the content of the present disclosure, a chassis fault repair device based on an improved YOLOv5 algorithm is provided, comprising:

[0012] The first acquisition module is used to acquire the sound signal characteristics of the chassis system during the driving of the special vehicle, wherein the sound signal characteristics include: sound frequency characteristics, sound amplitude characteristics and sound time domain characteristics;

[0013] A second acquisition module is used to acquire an improved YOLOv5 model obtained by replacing a preset network layer of the original YOLOv5 model, wherein the preset network layer includes: a downsampling layer, an SPPF layer, a target detection layer, and a multi-parallel network layer;

[0014] A determination module, used for inputting the sound signal feature into the improved YOLOv5 model, and determining the fault characteristic map data corresponding to the chassis system of the special vehicle based on the output result of the improved YOLOv5 model, wherein the fault characteristic map data includes: fault component, fault location and fault type;

[0015] The maintenance module is used to perform maintenance processing on the chassis system of the special vehicle based on the fault characteristic diagram data corresponding to the chassis system of the special vehicle.

[0016] In a third aspect, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the steps of the chassis fault repair method based on the improved YOLOv5 algorithm in any of the above embodiments are implemented.

[0017] In a fourth aspect, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps of the chassis fault repair method based on the improved YOLOv5 algorithm in any of the above embodiments are implemented.

[0018] The chassis fault repair method based on the improved YOLOv5 algorithm provided in the embodiment of the present application obtains the sound signal characteristics of the chassis system during the driving of the special vehicle, and the sound signal characteristics include: sound frequency characteristics, sound amplitude characteristics and sound time domain characteristics; obtains the improved YOLOv5 model obtained by replacing the preset network layer of the original YOLOv5 model, and the preset network layer includes: downsampling layer, SPPF layer, target detection layer and multi-parallel network layer; inputs the sound signal characteristics into the improved YOLOv5 model, and determines the fault characteristic map data corresponding to the chassis system of the special vehicle based on the output result of the improved YOLOv5 model, and the fault characteristic map data includes: fault component, fault location and fault type; based on the fault characteristic map data corresponding to the chassis system of the special vehicle, the chassis system of the special vehicle is repaired. In this way, by collecting the sound signal characteristics of the chassis system during the driving of the special vehicle, and performing fault identification and then fault repair based on the improved YOLOv5 model, the fault repair efficiency is effectively improved.

[0019] The above description is only an overview of the technical solution of the embodiment of the present application. In order to more clearly understand the technical means of the embodiment of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the embodiment of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions of the embodiments of the present disclosure, the drawings of the embodiments will be briefly described below. It should be noted that the drawings described below only relate to some embodiments of the present disclosure, but are not intended to limit the present disclosure, wherein:

[0021] Figure 1 It is a flowchart of a chassis fault repair method based on an improved YOLOv5 algorithm provided by the present disclosure.

[0022] Figure 2A It is a schematic diagram of the structure of the first network layer in an improved YOLOv5 model provided by the present disclosure.

[0023] Figure 2B It is a structural diagram of a CA mechanism provided by the present disclosure.

[0024] Figure 2C It is a schematic diagram of the structure of the second network layer in an improved YOLOv5 model provided by the present invention.

[0025] Figure 2D It is a schematic diagram of the structure of the third network layer in an improved YOLOv5 model provided by the present invention.

[0026] Figure 2EIt is a schematic diagram of the structure of the fourth network layer in an improved YOLOv5 model provided by the present invention.

[0027] Figure 2F It is a structural schematic diagram of an improved YOLOv5 model provided by the present disclosure.

[0028] Figure 3 It is a structural schematic diagram of a chassis fault repair device based on an improved YOLOv5 algorithm provided by the present invention.

[0029] Figure 4 It is a structural schematic diagram of a computer device provided by the present disclosure.

[0030] It should be noted that the elements in the drawings are schematic and not drawn to scale. DETAILED DESCRIPTION

[0031] In order to make the purpose, technical solution and advantages of the embodiments of the present disclosure clearer, the technical solution of the embodiments of the present disclosure will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present disclosure, rather than all of the embodiments. Based on the described embodiments of the present disclosure, all other embodiments obtained by those skilled in the art without creative work also fall within the scope of protection of the present disclosure.

[0032] Unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by a person skilled in the art to which the subject matter of the present disclosure belongs. It will be further understood that terms such as those defined in commonly used dictionaries should be interpreted as having a meaning consistent with their meaning in the context of the specification and the relevant art, and will not be interpreted in an idealized or overly formal form unless otherwise explicitly defined herein. As used herein, a statement that two or more parts are "connected" or "coupled" together shall mean that the parts are joined together directly or through one or more intermediate components.

[0033] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase "embodiments" in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0034] The term "and / or" in this article is only a description of the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can mean: A exists, A and B exist at the same time, and B exists. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship. Terms such as "first" and "second" are only used to distinguish one component (or a part of a component) from another component (or another part of a component).

[0035] In the description of the present application, unless otherwise specified, "plurality" means more than two (including two), and similarly, "multiple groups" means more than two groups (including two).

[0036] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the accompanying drawings.

[0037] Figure 1 is a flow chart of a chassis fault repair method based on an improved YOLOv5 algorithm provided by an embodiment of the present disclosure, such as Figure 1 As shown, the specific process of the chassis fault repair method based on the improved YOLOv5 algorithm includes:

[0038] S110. Acquire sound signal characteristics of a chassis system during the driving of the special vehicle.

[0039] Among them, the acoustic signal technology collects and obtains the operating status parameters of each component of the special vehicle to identify whether there is any abnormality in the working status of the equipment. If the measured operating status changes dynamically, it means that the equipment has a certain degree of potential failure, which is convenient for detection and maintenance personnel to formulate solutions in time.

[0040] The sound signal features include: sound frequency features, sound amplitude features and sound time domain features. In the process of acquiring sound signal features, the sound signals (such as intermittent combination signals, body vibration signals and abnormal noise signals, etc.) emitted by the chassis of special vehicles during operation are collected in real time through sensors. The clarity and accuracy of the signals are improved through signal filtering, and then the collected sound signals are converted into analyzable parameters, namely sound frequency features, sound amplitude features and sound time domain features, through feature extraction technology. These extracted features can be used to construct a deep residual network to improve the accuracy of identifying fault sound signals.

[0041] It should be noted that before collecting chassis fault sound signals, all parameters that may affect the signals are adjusted to the optimal values ​​to minimize their impact on the signals. For example, the sound signal sampling frequency is set to be greater than the highest frequency in the signal spectrum to avoid distortion in signal collection. Based on the Shannon sampling principle, the fault sound signals of the chassis of the special vehicle to be repaired are collected in real time; the collected sound signals are subjected to noise reduction processing, including removing abnormal points of the sound signals, changing the average position of the sound signals, filtering out redundant sounds in the sound signals, and self-adjusting the output signals; the signal-to-noise ratio of the signal is improved through sound signal preprocessing.

[0042] On this basis, the characteristics of sound signals are identified and extracted from three analysis angles, namely amplitude domain, time domain and frequency domain. Based on the window function, the signal is divided into multiple different segments to obtain the signal power spectrum. According to the changing law of the power spectrum, the mutation signal is detected and the width of the time window and frequency window is adjusted to analyze the sound signal.

[0043] S120, obtaining an improved YOLOv5 model obtained by replacing a preset network layer of the original YOLOv5 model.

[0044] The preset network layers include: a downsampling layer, an SPPF layer, an object detection layer, and a multi-parallel network layer. It can be understood that the improved YOLOv5 model includes not only the network layer after replacing the preset network layer, but also other network layers in the original YOLOv5 model.

[0045] In some embodiments, obtaining an improved YOLOv5 model obtained by replacing a preset network layer of an original YOLOv5 model includes:

[0046] The first network layer is used to replace the downsampling layer of the original YOLOv5 model; the second network layer is used to replace the SPPF layer of the original YOLOv5 model; the third network layer is used to replace the target detection layer of the original YOLOv5 model; the fourth network layer is used to replace the multi-parallel network layer of the original YOLOv5 model to obtain an improved YOLOv5 model.

[0047] Among them, the first network layer enhances the fault features by introducing the attention mechanism to obtain multi-feature fault feature fusion data, the second network layer obtains multi-scale splicing data by multi-scale fusion and splicing of the fault feature fusion data, the third network layer adds a small-size detection layer to the target detection layer of the original YOLOv5 model to obtain fault feature detection data based on multi-scale splicing data, and the fourth network layer is used to perform multi-level parallel processing on the fault feature detection data to obtain fault feature parallel data.

[0048] In the original YOLOv5 network, the downsampling layer is a very important part. The downsampling layer in the YOLOv5 backbone network uses a convolution with a kernel size of 3×3 and a stride of 2. Each convolution operation in the downsampling layer will cause the size of the feature map to become smaller, thereby improving the training efficiency of the model. To this end, the receptive field of each pixel in the feature map can be expanded so that the model can better capture global information and improve detection accuracy. In order to enhance the invariance of features, a series of operations are performed on the image to make the model robust. However, the use of convolution for downsampling in YOLOv5 has the following disadvantages: the size of the feature map will become smaller during the convolution downsampling process, which will inevitably cause the loss of some important information; the resolution of the feature map will be reduced during the downsampling process, and the system's ability to detect targets will be reduced; the use of large convolution kernels for downsampling operations will increase the amount of calculation of the model and affect the model detection speed.

[0049] The SPPF layer of YOLOv5 uses multiple 5×5 Maxpooling layers in series, and before performing the CBS (convolution) operation, it performs a splicing operation on the output of the three Maxpooling (pooling) layers to fuse local features and global features. Maximum pooling can avoid distortion of some images and reduce some redundant information, but it will also lose some detail information. This embodiment performs multi-scale fusion by changing the number of channels and the number of pooling layers, which can effectively extract target features in the operation of the chassis system and improve the model detection capability.

[0050] The original YOLOv5 model has only three detection layers, corresponding to three sets of initial anchor box values. The input image size defaults to 640×640. The network outputs 80×80, 40×40, and 20×20 feature maps after three downsamplings, which can detect small targets of size 8×8, medium targets of size 16×16, and large targets of size 32×32, respectively. In actual application scenarios, most of the detections are small or extremely small targets, and the detection targets are widely distributed, the targets are dense, and the scenes are complex. This embodiment increases the depth of the network by adding a small-sized target detection layer, which facilitates the extraction of important features deep in the network.

[0051] The original YOLOv5 C3 structure divides the original input into two branches, reduces the number of channels by half through convolution, connects them through splicing, and finally outputs a new feature map, where the number of input and output channels is the same in the bottleneck network. In order to allow the model to learn more features, this embodiment draws on the idea of ​​dense networks, improves the C3 module before the detection head, connects all layers, and each layer can receive the output of all previous layers, allowing the network to flow freely and enhancing the reuse of features.

[0052] In some embodiments, the first network layer includes: a convolution module, a downsampling module, an attention module and a fusion module; the convolution module generates initial feature map data by performing a convolution operation on the sound signal features, the downsampling module obtains downsampled feature map data by downsampling the initial feature map data, the attention module obtains feature enhanced map data by performing feature enhancement on the downsampled feature map data, and the fusion module is used to perform convolution fusion processing on the feature enhanced map data to obtain fault feature fusion data.

[0053] This embodiment uses the CPC module to construct the first network layer. The structure of the first network layer is as follows: Figure 2A As shown in the figure, the first layer is the convolution module, i.e., the input layer. The input information is the preprocessed special vehicle chassis fault sound signal image data (i.e., sound signal features). These data contain different types of chassis fault features, such as noise and jitter. The output of this layer is the preliminary feature map data (i.e., initial feature map data) generated by the convolution operation. The size of the feature map will be adjusted according to the size and step size of the convolution kernel, providing basic feature information.

[0054] The second layer is a downsampling module, which uses convolution CBS or pooling Maxpooling operations with a convolution kernel size of 3×3 and a stride of 2 to further reduce the size of the feature map and expand the receptive field. The output of this layer is the data of the generated downsampled feature map (i.e., downsampled feature map data), which is smaller in size but contains more complex features, improving the model's ability to capture chassis fault features.

[0055] The third layer is the attention module, which introduces the CA (coordinate attention) mechanism. In this layer, the CA mechanism is used to further enhance the expressiveness of features, and the number of channels of the output feature map is reduced to 1 / r (r is 32). The attention weights obtained by global average pooling of this feature map can effectively enhance the model's attention to important features and improve the robustness and accuracy of the model.

[0056] The last layer is the fusion module, which outputs fault feature fusion data after operations such as shared convolution, normalization and activation function. It effectively integrates the characteristic information of chassis faults and can accurately reflect the status of special vehicle chassis. It can be used for subsequent fault detection and dynamic maintenance decisions to ensure the safety and reliability of special vehicles.

[0057] Through the input and output information of each layer in the first network layer, it can be seen that each layer is constantly extracting and strengthening the characteristics of special vehicle chassis failures. The layered structural design enables the YOLOv5 network to extract features at multiple scales, and ultimately achieve efficient and accurate fault detection; multiple layers are connected in a cascade manner, and the output of each layer provides more in-depth feature information for the next layer, thereby enhancing the learning and reasoning capabilities of the model.

[0058] Combining two commonly used downsampling methods, the convolution with a convolution kernel size of 3×3 and a stride of 2 and the maximum pooling with a pooling kernel size of 2×2 and a stride of 2 are used. The downsampling operation is performed in two branches, and the coordinate attention mechanism is added after the two branches are fused.

[0059] When the downsampling method uses a convolutional layer or a maximum pooling layer, the calculation formulas are shown in formula (1) and formula (2).

[0060]

[0061] In formula (1) and formula (2), (H', W') is the size of the output feature map, (H, W) is the size of the input feature map, p is the padding, k is the convolution kernel or pooling kernel size, and s is the step size.

[0062] After the two branches are fused, the CA mechanism is added. The attention mechanism is a cognitive mechanism used to better process information. It can better concentrate attention, process information, remember and learn.

[0063] The CA mechanism can enhance the expressiveness of mobile network learning features. It can convert any intermediate tensor Input = [x 1 ,x 2 ,...,x c ]∈R H×W×C After transformation, the output is the same tensor Ouput = [y 1 ,y 2 ,...,y c ]∈R H×W×C .

[0064] The CA mechanism structure diagram is as follows Figure 2B The CA mechanism performs global average pooling in the horizontal and vertical directions respectively to obtain two feature maps, and then fuses the feature maps in the two directions, and obtains the feature map f after shared convolution, normalization and activation function operations, as shown in formula (3).

[0065] f=δ(F 1 ([z h ,z w ])) (3)

[0066] In formula (3), δ is the activation function, F 1 is a shared convolution, z h ,z w They are the feature maps in the vertical and horizontal directions respectively.

[0067] The purpose of using shared convolution on the aggregated feature map is to reduce the number of channels to the original 1 / r, where r is 32. The feature map f is then split for convolution to reduce the number of channels to C. A sigmoid operation is then performed to obtain the attention weights in two directions. Finally, the attention weights in the two directions are expanded to obtain the output result.

[0068] The CPC module can well improve the generalization ability and robustness of the model, reduce the loss of important features, process information more effectively, focus on important features, and improve the efficiency and accuracy of the system.

[0069] In some embodiments, the second network layer includes: a compression module, a pooling module and a splicing module, the pooling module is formed by connecting multiple pooling units in series; the compression module compresses the fault feature fusion data to obtain fault feature compressed data, the pooling module performs multi-scale pooling on the fault feature compressed data to obtain fault feature pooled data, and the splicing module obtains multi-scale splicing data by splicing the fault feature compressed data and the fault feature pooling data.

[0070] Among them, this embodiment uses the S-SPPF module to construct the second network layer, and its structure is as follows Figure 2C As shown. Use a 1×1 convolution to compress the number of channels to obtain the feature f 1 The original SPPF module uses three 5×5 Maxpooling series, which easily ignores some targets and affects the detection accuracy. This embodiment adjusts the number of Maxpooling to add a new scale of features, which is integrated with the previous features of different scales and global features, effectively solving the problem of relatively small detection targets and improving target detection accuracy. 1 As input, the feature f is obtained by using SPPF with 4 Maxpooling sequences. 2 , f 2 With f 1 Perform concatenation operations to enhance the ability of feature aggregation and optimize information so that the number of channels in the output feature map is the same as the number of channels in the original input feature map. By reducing and increasing the number of channels, the number of gradients is increased to achieve high cardinality and sparse connections. Finally, integrate through 1×1 convolution to output the feature fcbs(f 1 ②f 2 ), ② represents the concatenation operation, and fcbs represents the convolution operation.

[0071] In some embodiments, the third network layer includes: four detection layers, the four detection layers include: three target detection layers in the original YOLOv5 model and a small-size detection layer; the detection size corresponding to the small-size detection layer is smaller than any detection size corresponding to the target detection layer in the original YOLOv5 model.

[0072] Among them, the structure in the third network layer is as follows Figure 2D As shown. In this embodiment, a small target detection head is added on the basis of the three detection heads of the original YOLOv5 model, the output feature map size is 160×160, and the initial anchor box value size is set to [5,6,8,14,15,11]. In the original YOL-Ov5 network, the neck network is upsampled twice and then downsampled to output three detection layers. In this embodiment, the neck needs to be upsampled three times, that is, upsampling continues after the seventeenth layer, the twentieth layer is concat-operated with the features of the second layer, and scale fusion is performed to splice the bottom feature map with the high-level feature map, so that the high-level has stronger location information and semantic information at the same time. When the feature map reaches 160×160, the downsampling operation is performed again until a detection feature map of 20×20 scale appears, so that there are four target detection layers, which increases the depth of the network and extracts important features deep in the network. As a result, the network can become sensitive to the features of small targets and extremely small targets. The small target detection layer contains rich texture and contour information, and can learn multi-scale feature information in complex scenes, improve the accuracy of small target detection, and reduce missed detections and false detections.

[0073] In some embodiments, the fourth network layer includes: a C3 structure module, a CrossConv module, a residual network module and an output module; the C3 structure module obtains fault feature extraction data by performing data extraction processing on the fault feature detection data, the CrossConv module performs correlation analysis on the fault feature extraction data by constructing a similarity matrix to obtain fault feature related data, the residual network module obtains fault feature enhanced data by splicing the fault feature extraction data and the fault feature related data, and the output module obtains fault feature parallel data by splicing the fault feature extraction data, the fault feature related data and the fault feature enhanced data.

[0074] Among them, this embodiment uses a multi-parallel network CC3 to construct the fourth network layer, and its structure is as follows Figure 2EAs shown in the figure. The main function of the input layer is to prepare the feature map in the previous layer for subsequent processing; the second layer is the C3 structure module, and the input feature map data is processed by convolution, usually using a convolution kernel of 3 and a step size of 1. The main function of this convolution operation is to extract features and halve the number of channels, making the flow of information in the network more efficient. The output feature map of this layer retains important chassis fault feature information. The number of channels is halved, but its spatial features are still rich.

[0075] The third layer is the CrossConv module, which captures the correlation between features by performing dot product calculations on the feature vectors at each position of the two feature maps. The function of this layer is to optimize the matching relationship between feature maps by constructing a similarity matrix and using the Sinkhorn algorithm. The output feature vector has an updated weight, which makes the similarity between feature maps higher. This process helps to reduce false detection and missed detection and improve the accuracy of feature extraction.

[0076] The fourth layer is the residual network module, which concatenates the output of the C3 structure and the output of CrossConv through the concatenation operation, which can maintain the original information while enhancing the feature expression ability. By combining the features of multiple branches, the network can obtain richer feature information in different feature maps, thereby improving the accuracy and performance of the entire network.

[0077] The last layer is the output module, which not only contains information that has undergone multiple feature extractions and similarity enhancements, but also includes rich features from each branch, forming a comprehensive feature representation. Such an output feature map can effectively support subsequent fault detection and classification tasks, ensuring the effectiveness and reliability of the model in practical applications.

[0078] In this embodiment, a parallel form of cross convolution (CrossConv) with a convolution kernel of 3 and a step size of 1 is added to the CC3 module. CrossConv performs dot product on the feature vectors at each position of the two feature maps, calculates their similarity to construct a matrix, captures the correlation between the features, obtains the matching relationship between the two similarity matrices through the Sinkhorn algorithm, and predicts the similarity span between the two feature maps. By updating the weights, the similarity features between the two feature maps are more similar, which is conducive to feature extraction and can effectively solve the problems of false detection and missed detection. See formula (4) and formula (5) for details.

[0079]

[0080] In formula (4) and formula (5), i∈v 1 ,j∈v 2 is the output k-layer feature, is the adjacent rank matrix, is the conversion of the original information, v 1 is from or Select from the k-1 layers, v 2 is the feature vector obtained by CrossConv To update the weights.

[0081] Therefore, by adding a residual network to the CC3 module and using Concat to form a multi-parallel network, the network effect will be more accurate than the original one, while avoiding gradient disappearance and improving feature expression capabilities.

[0082] S130, inputting the sound signal features into the improved YOLOv5 model, and determining the fault characteristic diagram data corresponding to the chassis system of the special vehicle based on the output result of the improved YOLOv5 model.

[0083] Among them, the fault characteristic map data includes: faulty components, fault location and fault type. The fault type records different types of faults, such as intermittent connection, body shaking and abnormal noise, etc. This information helps to quickly identify possible problems with the vehicle chassis. The fault position can indicate the specific component where the fault occurred, such as the suspension system, steering system or other key components, which is convenient for targeted inspection.

[0084] Inputting the sound signal features into the improved YOLOv5 model, and determining the fault feature map data corresponding to the chassis system of the special vehicle based on the output result of the improved YOLOv5 model, may include: inputting the sound signal features into the first network layer in the improved YOLOv5 model to obtain corresponding fault feature fusion data; inputting the corresponding fault feature fusion data into the second network layer to obtain corresponding multi-scale splicing data; inputting the corresponding multi-scale splicing data into the third network layer to obtain corresponding fault feature detection data; inputting the corresponding fault feature detection data into the fourth network layer to obtain corresponding fault feature parallel data; and then processing the fault feature parallel data in combination with the output layer and other network layers inherited from the original YOLOv5 model in the improved YOLOv5 model to obtain the fault feature map data corresponding to the chassis system of the special vehicle.

[0085] This embodiment uses the improved YOLOv5 algorithm to enhance the performance of target detection, extracts and processes the special vehicle chassis fault feature map, performs feature fusion processing, ensures that effective fault information is not lost, and enhances the expression ability of the chassis fault feature map at multiple scales. The structure of the improved YOLOv5 model is as follows: Figure 2F shown.

[0086] The improved YOLOv5 algorithm is used to perform multi-scale fusion prediction in detection to extract feature maps of chassis fault features of different sizes. The network and convolutional layer of the improved YOLOv5 algorithm residual unit are added, and the pooling layer is discarded to reduce the negative impact of pooling in fault feature fusion, so as to achieve the purpose of transforming single label into multi-label classification. Based on the multi-scale fusion function of the improved YOLOv5 algorithm, the number of chassis fault detection layers is increased from 1 to 3, and feature maps with sizes of 13×13, 26×26, and 52×52 are generated respectively.

[0087] The number of chassis fault detection layers is increased from 1 to 3, and multiple feature maps of different sizes are generated, mainly to enhance the ability of feature extraction. The primary function of this process is to achieve multi-scale feature extraction. By setting feature maps of different sizes (such as 13×13, 26×26, and 52×52), the model can obtain chassis fault features at different perceptions, which means that the features of small and large faults can be captured at the same time, improving the ability to identify various faults.

[0088] Increasing the number of layers and the size of the feature map helps improve the detection accuracy of the model. Smaller feature maps are more suitable for capturing subtle fault features, while larger feature maps can identify larger structural changes. This design ensures that the model can effectively cope with various complex situations when dealing with chassis faults, avoid information loss, and thus improve the overall detection performance.

[0089] By enhancing the robustness of the model, this structural design enables the model to better adapt to changes under different working conditions. The increased number of layers and multi-scale feature maps not only improve the overall performance of the model, but also ensure that it can better cope with the challenges of chassis fault detection in practical applications and ensure the safe and reliable operation of special vehicles.

[0090] When identifying approach faults, other image features can also be fused, such as the shape, edge, and texture information of chassis components. This diverse feature fusion process enables the final feature map to fully reflect the operating status and potential faults of the chassis. The feature map obtained after fusion is a multi-dimensional feature representation that can more accurately capture the fault characteristics of the chassis, thereby improving the efficiency and accuracy of fault detection.

[0091] When performing feature fusion, a fault feature fusion prediction frame is constructed. The prediction frame is based on the improved YOLOv5 network and calculates the special vehicle chassis fault feature fusion confidence, as shown in formula (6).

[0092] S=P(C i |O a )P(O a )I(6)

[0093] In formula (6), P(Ci |O a ) is the confidence score of the grid prediction i-type chassis fault feature fusion, which represents the detection credibility of the chassis fault feature in each feature map. By calculating the confidence score for each feature map, the fusion effect of features of different sizes can be evaluated, thereby judging the overall performance of the fault feature. This association makes multi-scale feature fusion more effective in fault detection, ensuring the capture and accuracy of important information. I is the intersection-over-union ratio of fault feature fusion. P(O a )=1, it means that the special vehicle chassis fault feature is integrated into the prediction frame; P(O a )=0, it means that the special vehicle chassis fault feature fusion is not in the prediction box.

[0094] Each feature map (13×13, 26×26, 52×52) corresponds to a different scale, allowing YOLOv5 to perform object detection at different levels. Specifically, each feature map is responsible for detecting fault features of different sizes: smaller feature maps (such as 13×13) are more suitable for capturing small-sized fault features, while larger feature maps (such as 52×52) can detect larger structural faults.

[0095] Confidence scores of chassis fault features in each feature map are calculated. These scores reflect the credibility of the fault features in a specific feature map, where P(C i |O a ) is 1, indicating that the fault feature is valid in the prediction frame, P(C i |O a ) is 0, which means it is invalid. By calculating the overlapping area between the predicted box and the true box, the fusion effect of the features can be determined. If the IoU exceeds the set threshold, it indicates that the feature fusion effect is good, otherwise there may be problems. By combining the confidence score and IoU results, a comprehensive evaluation is formed. Based on these data, the effectiveness and accuracy of feature fusion can be judged to ensure the necessary adjustment and optimization of feature extraction and fusion strategies in subsequent processing.

[0096] There are many ways to adjust the running state of the YOLOv5 backbone network according to the fusion situation. For example, if the confidence score is found to be consistently low and the IoU is not up to standard in a specific detection task, you can choose to reduce the learning rate to help the model better adapt to feature fusion, prevent overfitting, and ensure that the model learns more effective features.

[0097] In addition, if the detection effect of a certain feature map is not ideal, for example, if the confidence score of this layer is lower than expected when processing small target faults, you can consider increasing the number of residual units of this layer or adjusting the size of the convolution kernel to enhance its feature extraction ability. This optimization can make the model more effective when dealing with faults of a specific scale. If it is found that the feature fusion effect of small targets is not good during detection, the number of layers of the feature map can be appropriately increased, or a new attention mechanism can be introduced to improve the expressiveness of the features and enhance the detection effect. Through these adjustments, the backbone network of YOLOv5 can more flexibly adapt to the detection needs of different scenarios and features, and improve the overall performance.

[0098] S140: Based on the fault characteristic diagram data corresponding to the chassis system of the special vehicle, repair the chassis system of the special vehicle.

[0099] In some embodiments, the chassis system of the special vehicle is repaired based on the fault characteristic diagram data corresponding to the chassis system of the special vehicle, including:

[0100] Based on the fault location and fault type of the fault component in the fault characteristic diagram data corresponding to the chassis system of the special vehicle, the fault repair level of the fault component is determined; based on the fault repair level of the fault component, the fault component is repaired.

[0101] Among them, the fault severity data is measured by the fault location and fault type of the fault component to provide an assessment of the impact of the fault, and the fault is classified into light, medium and severe levels to obtain the fault repair level, which helps to determine the priority and urgency of the repair. In addition, real-time monitoring data can be used to obtain vehicle operating status information such as vibration, temperature and oil pressure through sensors. The comprehensive analysis of these data can fully reflect the health status of the special vehicle chassis and provide a basis for subsequent fault diagnosis and repair.

[0102] In this embodiment, the sound signal characteristics of the chassis system during the driving of the special vehicle are obtained, and the sound signal characteristics include: sound frequency characteristics, sound amplitude characteristics and sound time domain characteristics; the improved YOLOv5 model obtained by replacing the preset network layer of the original YOLOv5 model is obtained, and the preset network layer includes: downsampling layer, SPPF layer, target detection layer and multi-parallel network layer; the sound signal characteristics are input into the improved YOLOv5 model, and the fault characteristic map data corresponding to the chassis system of the special vehicle is determined based on the output result of the improved YOLOv5 model, and the fault characteristic map data includes: fault component, fault location and fault type; based on the fault characteristic map data corresponding to the chassis system of the special vehicle, the chassis system of the special vehicle is repaired. In this way, by collecting the sound signal characteristics of the chassis system during the driving of the special vehicle, and performing fault identification and then fault repair based on the improved YOLOv5 model, the fault repair efficiency is effectively improved.

[0103] In addition, based on the fusion results of chassis fault characteristics, it is also possible to dynamically detect whether there are intermittent chassis engagement, body shaking and abnormal noise problems after the special vehicle is started.

[0104] For example, by real-time monitoring of vehicle status and using sensors to collect dynamic data of the chassis after startup, such as vibration, sound, and pressure, the sensors can accurately capture the state changes of the vehicle when it is started, providing basic data for subsequent analysis; through data analysis, technicians can determine whether there are intermittent coupling, body shaking, or abnormal noise problems. For example, by analyzing the data of the vibration sensor, if the detected vibration amplitude and frequency exceed the normal range, it may indicate that the chassis has intermittent coupling problems. At the same time, abnormal noise captured by the acoustic sensor can also indicate potential faults; based on the analysis results, the technicians will compare the fault characteristics with the known database to confirm whether the chassis has the above problems. If an abnormality is detected, the corresponding components should be inspected and repaired to ensure the safety and reliability of special vehicles.

[0105] If the above problems exist, the components in the chassis should be tightened and aging components should be replaced in time. Check whether the chassis oil pipe joints are not tightly sealed, tighten the bolts according to the specified torque, and check the replacement oil pump pressure value to avoid the risk of oil pump leakage. Use professional tools to measure the tire pressure of special vehicle wheels, observe whether the steering wheel can work normally, whether the steering amplitude is consistent, and control the direction of the special vehicle chassis. Use a dial indicator to measure the bending deformation of the chassis shaft, and replace it immediately if it exceeds the standard. Through dynamic maintenance, reasonable detection and maintenance of special vehicle chassis component failures can be carried out to comprehensively improve the operating performance of the special vehicle chassis.

[0106] When extracting fault features, influenced by the deep residual learning network, it is necessary to establish a deep residual network so that the engine fault features can be extracted. Since the deep residual network is composed of smaller residual error blocks, each small residual error block contains two branches. Between two adjacent residual error blocks, it is necessary to stack them in order, so that a plus sign will be generated between the two residual blocks in the network structure, which represents a skip connection. In general, in the network structure composed of residual blocks, in addition to the different numbers and scales of convolution kernels, each small residual block also has similarities to a certain extent. Therefore, random sampling is performed in the convolution layer on each convolution kernel number, with a step size of two pixels. Since the convolution layer of the network structure adds a normalization layer, the resolution selected using the YOLOv5 algorithm should be within the range of 64×64-2×2. The feature map within this range can be fused with the feature map of the next layer, and then the fault feature extraction can be completed according to the basic feature structure.

[0107] In summary, this embodiment realizes real-time fault detection of special vehicle engines and chassis systems by improving the YOLOv5 algorithm. Traditional detection methods perform poorly in complex scenarios, especially for the detection of small targets, which often leads to problems such as missed detection and slow detection speed. The improved YOLOv5 algorithm enhances the feature extraction capability by introducing a new module to replace the residual block structure. Combined with the full convolutional network structure and the logistic evaluation method, this improvement significantly improves the speed, accuracy and efficiency of fault detection of special vehicles in dynamic operating environments, especially in small target fault detection.

[0108] By introducing intelligent dynamic detection technology, a new method for special vehicle chassis fault repair is proposed. Traditional repair methods have the problems of low repair efficiency and limited fault range identification, and it is difficult to identify potential faults of special vehicle chassis components in real time and accurately. The detection system based on the improved YOLOv5 algorithm can dynamically monitor various components of the chassis during the driving of special vehicles, detect faults in time, and provide accurate repair instructions. The combination of intelligent detection and maintenance makes troubleshooting more efficient, greatly reduces the need for manual intervention, and enhances the automation level of the system.

[0109] The engine and chassis systems of special vehicles have complex structures, and their fault characteristics are changeable and difficult to capture. This embodiment can more comprehensively capture the potential fault characteristics in the engine and chassis systems. This feature extraction method has stronger adaptability than the traditional single classification algorithm and can meet the fault identification needs of special vehicles in different usage scenarios. By introducing a three-scale detection mechanism, fault characteristics at different scales can be accurately captured, significantly improving the detection capability of the system.

[0110] In the design and maintenance of special vehicles, the chassis system is a key load-bearing and safety component, and the engine is the core driving force. Once these systems have problems, serious safety hazards may occur. This embodiment can monitor the operating status of the engine and chassis in real time and warn of potential faults. Compared with traditional fault detection systems, it not only improves the accuracy and efficiency of detection, but also can continuously optimize the detection model through deep learning technology to ensure the stability and reliability of the system in long-term use.

[0111] Improved YOLOv5 algorithm is used for fault detection and integrated with maintenance system to realize integrated fault diagnosis and maintenance. Traditional maintenance process usually relies on manual experience and post-test, which has the limitations of low efficiency and slow response. The new maintenance process combines real-time online detection, automatic diagnosis and preventive maintenance, so that special vehicles can quickly identify faults during driving and provide corresponding repair solutions, which greatly shortens maintenance time and improves safety.

[0112] Figure 3 A schematic diagram of the structure of a chassis fault repair device based on an improved YOLOv5 algorithm provided in this embodiment. The chassis fault repair device based on the improved YOLOv5 algorithm may include:

[0113] The first acquisition module 310 is used to acquire the sound signal characteristics of the chassis system of the special vehicle during driving, and the sound signal characteristics include: sound frequency characteristics, sound amplitude characteristics and sound time domain characteristics.

[0114] The second acquisition module 320 is used to obtain an improved YOLOv5 model obtained by replacing the preset network layers of the original YOLOv5 model, where the preset network layers include: a downsampling layer, an SPPF layer, a target detection layer, and a multi-parallel network layer.

[0115] The determination module 330 is used to input the sound signal features into the improved YOLOv5 model, and determine the fault characteristic map data corresponding to the chassis system of the special vehicle based on the output result of the improved YOLOv5 model, wherein the fault characteristic map data includes: fault component, fault location and fault type.

[0116] The maintenance module 340 is used to perform maintenance processing on the chassis system of the special vehicle based on the fault characteristic diagram data corresponding to the chassis system of the special vehicle.

[0117] In this embodiment, optionally, the second acquisition module 320 is specifically configured to:

[0118] The first network layer is used to replace the downsampling layer of the original YOLOv5 model; the second network layer is used to replace the SPPF layer of the original YOLOv5 model; the third network layer is used to replace the target detection layer of the original YOLOv5 model; the fourth network layer is used to replace the multi-parallel network layer of the original YOLOv5 model to obtain an improved YOLOv5 model; wherein the first network layer enhances the fault features by introducing an attention mechanism to obtain multi-feature fault feature fusion data, the second network layer performs multi-scale fusion and splicing on the fault feature fusion data to obtain multi-scale splicing data, the third network layer adds a small-size detection layer to the target detection layer of the original YOLOv5 model to obtain fault feature detection data based on multi-scale splicing data, and the fourth network layer is used to perform multi-stage parallel processing on the fault feature detection data to obtain fault feature parallel data.

[0119] In this embodiment, optionally, the first network layer includes: a convolution module, a downsampling module, an attention module and a fusion module; the convolution module generates initial feature map data by performing a convolution operation on the sound signal features, the downsampling module obtains downsampled feature map data by downsampling the initial feature map data, the attention module obtains feature enhanced map data by performing feature enhancement on the downsampled feature map data, and the fusion module is used to perform convolution fusion processing on the feature enhanced map data to obtain fault feature fusion data.

[0120] In this embodiment, optionally, the second network layer includes: a compression module, a pooling module and a splicing module, the pooling module is formed by connecting multiple pooling units in series; the compression module compresses the fault feature fusion data to obtain fault feature compressed data, the pooling module performs multi-scale pooling on the fault feature compressed data to obtain fault feature pooled data, and the splicing module obtains multi-scale splicing data by splicing the fault feature compressed data and the fault feature pooled data.

[0121] In this embodiment, optionally, the third network layer includes: four detection layers, the four detection layers include: three target detection layers in the original YOLOv5 model and a small-size detection layer; the detection size corresponding to the small-size detection layer is smaller than any detection size corresponding to the target detection layer in the original YOLOv5 model.

[0122] In this embodiment, optionally, the fourth network layer includes: a C3 structure module, a CrossConv module, a residual network module and an output module; the C3 structure module obtains fault feature extraction data by performing data extraction processing on the fault feature detection data, the CrossConv module performs correlation analysis on the fault feature extraction data by constructing a similarity matrix to obtain fault feature related data, the residual network module obtains fault feature enhanced data by splicing the fault feature extraction data and the fault feature related data, and the output module obtains fault feature parallel data by splicing the fault feature extraction data, the fault feature related data and the fault feature enhanced data.

[0123] In this embodiment, optionally, the maintenance module 340 is specifically used to:

[0124] Based on the fault location and fault type of the fault component in the fault characteristic diagram data corresponding to the chassis system of the special vehicle, the fault repair level of the fault component is determined; based on the fault repair level of the fault component, the fault component is repaired.

[0125] The chassis fault repair device based on the improved YOLOv5 algorithm provided in the present disclosure can execute the above method embodiments. Its specific implementation principles and technical effects can be found in the above method embodiments, and the present disclosure will not repeat them here.

[0126] The present application also provides a computer device. Figure 4 , Figure 4 This is a basic structural block diagram of the computer device in this embodiment.

[0127] The computer device includes a memory 410 and a processor 420 that are connected to each other through a system bus. It should be noted that the figure only shows a computer device with a memory 410 and a processor 420, but it should be understood that it is not required to implement all the components shown, and more or fewer components can be implemented instead. Among them, those skilled in the art can understand that the computer device here is a device that can automatically perform numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes but is not limited to microprocessors, application specific integrated circuits (Application Specific Integrated Circuit, ASIC), programmable gate arrays (Field-Programmable Gate Array, FPGA), digital processors (Digital Signal Processor, DSP), embedded devices, etc.

[0128] Computer devices can be computing devices such as desktop computers, notebooks, PDAs, and cloud servers. Computer devices can interact with users through keyboards, mice, remote controls, touch pads, or voice control devices.

[0129] The memory 410 includes at least one type of readable storage medium, and the readable storage medium includes a non-volatile memory or a volatile memory, such as a flash memory, a hard disk, a multimedia card, a card-type memory (e.g., SD or DX memory, etc.), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), a programmable read-only memory (PROM), a magnetic memory, a magnetic disk, an optical disk, etc., and the RAM may include a static RAM or a dynamic RAM. In some embodiments, the memory 410 may be an internal storage unit of a computer device, such as a hard disk or a memory of the computer device. In other embodiments, the memory 410 may also be an external storage device of a computer device, such as a plug-in hard disk, a smart memory card (SMC), a secure digital (SD) card, or a flash card (Flash Card), etc., equipped on the computer device. Of course, the memory 410 may also include both an internal storage unit of the computer device and an external storage device thereof. In this embodiment, the memory 410 is generally used to store an operating system and various application software installed on the computer device, such as the program code of the above method, etc. In addition, the memory 410 may also be used to temporarily store various data that have been output or are to be output.

[0130] The processor 420 is generally used to perform the overall operation of the computer device. In this embodiment, the memory 410 is used to store program codes or instructions, the program code includes computer operation instructions, and the processor 420 is used to execute the program codes or instructions stored in the memory 410 or process data, such as running the program code of the above method.

[0131] In this article, the bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. The bus system can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, only one thick line is used in the figure, but it does not mean that there is only one bus or one type of bus.

[0132] Another embodiment of the present application also provides a computer-readable medium, which may be a computer-readable signal medium or a computer-readable medium. A processor in a computer reads a computer-readable program code stored in the computer-readable medium, so that the processor can execute the functional actions specified in each step or a combination of steps in the above method; and generate a device for implementing the functional actions specified in each block or a combination of blocks in the block diagram.

[0133] Computer-readable media include but are not limited to electronic, magnetic, optical, electromagnetic, infrared memory or semiconductor systems, devices or apparatuses, or any appropriate combination of the foregoing, the memory is used to store program codes or instructions, the program codes include computer operating instructions, and the processor is used to execute the program codes or instructions of the above methods stored in the memory.

[0134] For the definitions of memory and processor, please refer to the description of the aforementioned computer device embodiment and will not be repeated here.

[0135] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of modules or units is only a logical function division, and there may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0136] Each functional unit or module in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0137] If the integrated unit 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 is essentially 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, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) or a processor (processor) to perform all or part of the steps of each embodiment method of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (Read-Only Memory, ROM), random access memory (Random Access Memory, RAM), disk or optical disk and other media that can store program code.

[0138] In the claims, any reference symbols placed between brackets shall not be construed as limiting the claims. The word "comprising" described in the present application does not exclude the presence of elements or steps not listed in the claims. The word "a" or "an" preceding an element does not exclude the presence of a plurality of such elements. The present application can be implemented with the aid of hardware comprising several different elements and with the aid of a suitably programmed computer. In a unit claim that lists a number of devices, several units of these devices may be embodied by the same hardware item. The use of first, second, and third, etc. does not indicate any order, and these words may be interpreted as names. The steps in the above embodiments, unless otherwise specified, should not be understood as limitations on the order of execution.

[0139] The above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A chassis fault repair method based on an improved YOLOv5 algorithm, characterized in that: include: Acquiring the sound signal characteristics of the chassis system during the driving of the special vehicle, wherein the sound signal characteristics include: sound frequency characteristics, sound amplitude characteristics and sound time domain characteristics; Obtain an improved YOLOv5 model obtained by replacing a preset network layer of the original YOLOv5 model, wherein the preset network layer includes: a downsampling layer, an SPPF layer, a target detection layer, and a multi-parallel network layer; Inputting the sound signal feature into the improved YOLOv5 model, and determining the fault characteristic map data corresponding to the chassis system of the special vehicle based on the output result of the improved YOLOv5 model, wherein the fault characteristic map data includes: fault component, fault location and fault type; Based on the fault characteristic diagram data corresponding to the chassis system of the special vehicle, the chassis system of the special vehicle is repaired.

2. The method according to claim 1, characterized in that The step of obtaining an improved YOLOv5 model obtained by replacing a preset network layer of the original YOLOv5 model includes: The downsampling layer of the original YOLOv5 model is replaced by the first network layer; The SPPF layer of the original YOLOv5 model is replaced by the second network layer; The target detection layer of the original YOLOv5 model is replaced by a third network layer; The multi-parallel network layers of the original YOLOv5 model are replaced by the fourth network layer to obtain the improved YOLOv5 model; Among them, the first network layer enhances the fault features by introducing the attention mechanism to obtain multi-feature fault feature fusion data, the second network layer obtains multi-scale splicing data by performing multi-scale fusion splicing on the fault feature fusion data, the third network layer adds a small-size detection layer to the target detection layer of the original YOLOv5 model to obtain fault feature detection data based on the multi-scale splicing data, and the fourth network layer is used to perform multi-stage parallel processing on the fault feature detection data to obtain fault feature parallel data.

3. The method according to claim 2, characterized in that The first network layer includes: a convolution module, a downsampling module, an attention module and a fusion module; The convolution module generates initial feature map data by performing a convolution operation on the sound signal features, the downsampling module obtains downsampled feature map data by downsampling the initial feature map data, the attention module obtains feature enhanced map data by performing feature enhancement on the downsampled feature map data, and the fusion module is used to perform convolution fusion processing on the feature enhanced map data to obtain the fault feature fusion data.

4. The method according to claim 3, characterized in that The second network layer includes: a compression module, a pooling module and a splicing module, wherein the pooling module is formed by connecting a plurality of pooling units in series; The compression module compresses the fault feature fusion data to obtain fault feature compression data, the pooling module performs multi-scale pooling on the fault feature compression data to obtain fault feature pooling data, and the splicing module splices the fault feature compression data and the fault feature pooling data to obtain the multi-scale splicing data.

5. The method according to claim 4, characterized in that The third network layer includes: four detection layers, the four detection layers include: three target detection layers in the original YOLOv5 model and one small-size detection layer; The detection size corresponding to the small-size detection layer is smaller than any detection size corresponding to the target detection layer in the original YOLOv5 model.

6. The method according to claim 5, characterized in that The fourth network layer includes: a C3 structure module, a CrossConv module, a residual network module and an output module; The C3 structure module obtains fault feature extraction data by performing data extraction processing on the fault feature detection data, the CrossConv module obtains fault feature related data by performing correlation analysis on the fault feature extraction data by constructing a similarity matrix, the residual network module obtains fault feature enhanced data by splicing the fault feature extraction data and the fault feature related data, and the output module obtains the fault feature parallel data by splicing the fault feature extraction data, the fault feature related data and the fault feature enhanced data.

7. The method according to claim 1, characterized in that The repairing process of the chassis system of the special vehicle based on the fault characteristic diagram data corresponding to the chassis system of the special vehicle comprises: Determining a fault repair level of the fault component based on the fault location and fault type of the fault component in the fault characteristic diagram data corresponding to the chassis system of the special vehicle; Based on the fault repair level of the fault component, the fault component is repaired.

8. A chassis fault repair device based on an improved YOLOv5 algorithm, characterized in that: include: The first acquisition module is used to acquire the sound signal characteristics of the chassis system during the driving of the special vehicle, wherein the sound signal characteristics include: sound frequency characteristics, sound amplitude characteristics and sound time domain characteristics; A second acquisition module is used to acquire an improved YOLOv5 model obtained by replacing a preset network layer of the original YOLOv5 model, wherein the preset network layer includes: a downsampling layer, an SPPF layer, a target detection layer, and a multi-parallel network layer; A determination module, used for inputting the sound signal feature into the improved YOLOv5 model, and determining the fault characteristic map data corresponding to the chassis system of the special vehicle based on the output result of the improved YOLOv5 model, wherein the fault characteristic map data includes: fault component, fault location and fault type; The maintenance module is used to perform maintenance processing on the chassis system of the special vehicle based on the fault characteristic diagram data corresponding to the chassis system of the special vehicle.

9. A computer device, characterized in that: The invention comprises a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, a chassis fault repair method based on the improved YOLOv5 algorithm as claimed in any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the chassis fault repair method based on the improved YOLOv5 algorithm as described in any one of claims 1 to 7 is implemented.

Images