A method and system for rapid detection of wire harness wire connection integrity in an automobile

Abstract

This invention discloses a rapid detection method and system for the integrity of automotive wiring harness connections, relating to the field of automotive electronic testing technology. The method includes: processing a standardized terminal image input to a depth edge detection network to extract wire and pixel features, and generating a wire connection region mask based on pixel connectivity; extracting pixel grayscale values ​​from the wire connection region mask, statistically analyzing the grayscale distribution to form a histogram, and normalizing the histogram to generate a terminal connection feature vector; inputting the terminal connection feature vector into a pre-trained convolutional neural network for feature classification to obtain a connection state prediction result, matching it with a connection level standard to generate an initial wire connection integrity label; fusing the functional response feature set with the initial wire connection integrity label to construct a comprehensive wire connection integrity feature vector, performing analysis and calculation, and generating a wire connection integrity detection report. This invention improves the accuracy and efficiency of wire connection state determination and detection.

Description

Technical Field

[0001] This invention relates to the field of automotive electronics testing technology, and in particular to a rapid testing method and system for the integrity of automotive wiring harness connections. Background Technology

[0002] In modern automotive manufacturing, wiring harnesses, as a key component of vehicle electronic and electrical systems, undertake the dual functions of energy transmission and signal transmission. With the increasing electrification and intelligence of automobiles, the integrity of wiring harness connections has become a crucial factor in ensuring vehicle safety, reliability, and functionality. Currently, the detection of wiring harness connection integrity mainly relies on manual visual inspection or functional testing based on traditional electrical parameters. For example, tools such as multimeters and oscilloscopes are used to measure the resistance, voltage, or current response of the wires to determine whether the connection between the wires and terminals is good.

[0003] Furthermore, existing vision- or electrical inspection-based technologies typically lack the ability to comprehensively evaluate the functionality of wire connections. They rely solely on appearance or a single electrical parameter for judgment, failing to integrate image features and functional response information simultaneously, thus limiting the reliability and accuracy of the detection results. This is particularly prominent in modern intelligent vehicles, where complex electronics place high demands on the integrity of wiring harness connections, and functional abnormalities can affect vehicle safety or the normal operation of critical electronic equipment. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a rapid detection method for the integrity of automotive wiring harness wire connections, solving the problem of rapid and accurate detection of wire connections.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides a rapid detection method for the integrity of automotive wiring harness wire connections, comprising, Acquire terminal connection images and perform preprocessing to generate standardized terminal images; The standardized terminal image input depth edge detection network is processed to extract wire and pixel features, and a wire connection region mask is generated according to pixel connectivity. Pixel grayscale values ​​are extracted from the mask of the wire connection area, the grayscale distribution is statistically analyzed to form a histogram, and the histogram is normalized to generate the terminal connection feature vector. The terminal connection feature vector is input into a pre-trained convolutional neural network for feature classification to obtain the connection status prediction result, and then matched with the connection level standard to generate an initial wire connection integrity label. The initial wire connection integrity label is matched and compared with the functional test signal. The resistance, voltage and current response values ​​in the functional test signal are extracted, and normalized and time-synchronized processing and integration are performed to generate a functional response feature set. The functional response feature set is fused with the initial conductor connection integrity label to construct a comprehensive feature vector of conductor connection integrity. The vector is then analyzed and calculated to generate a conductor connection integrity detection report.

[0007] As a preferred embodiment of the rapid detection method for the integrity of automotive wiring harness connections according to the present invention, the specific steps for generating standardized terminal images are as follows: Distortion correction and geometric correction are performed on the terminal connection image to generate a geometrically corrected terminal image, and multi-scale illumination equalization processing is performed to generate an illumination equalized terminal image. The terminal images with uniform illumination are subjected to noise reduction filtering and edge enhancement processing to generate terminal images with clear edges. The size is then normalized and the coordinates are standardized to generate standardized terminal images.

[0008] As a preferred embodiment of the rapid detection method for the integrity of automotive wiring harness connections according to the present invention, the specific steps for generating the wire connection area mask are as follows: The standardized terminal image is input into the depth edge detection network to extract multi-scale wire contour features and calculate edge intensity, generating a preliminary edge probability map. Non-maximum suppression and binarization are applied to the preliminary edge probability map to generate a binary edge mask; Pixel connectivity analysis and morphological closure calculation are performed on the binary edge mask to generate a wire connection region mask.

[0009] As a preferred embodiment of the rapid detection method for the integrity of automotive wiring harness connections according to the present invention, the specific steps for generating the terminal connection feature vector are as follows: The mask of the wire connection area is mapped to the standardized terminal image at the pixel level, grayscale information is extracted and organized to generate a grayscale set of wire connection pixels. The grayscale distribution of the pixel grayscale set of the wire connection is statistically analyzed, a grayscale histogram is constructed, and numerical normalization is performed to generate the terminal connection feature vector.

[0010] As a preferred embodiment of the rapid detection method for the integrity of automotive wiring harness wire connections according to the present invention, the pre-trained convolutional neural network is obtained by supervised learning training on terminal connection feature vectors and feature classification optimization based on the integrity status of wire connections.

[0011] As a preferred embodiment of the rapid detection method for the integrity of automotive wiring harness connections according to the present invention, the specific steps for generating initial wiring connection integrity tags are as follows: The terminal connection feature vector is input into a pre-trained convolutional neural network, and the terminal connection feature vector is subjected to layer-by-layer convolution and non-linear activation to generate an intermediate feature representation map. The intermediate feature representation map is mapped and classified using a fully connected layer to obtain the predicted probability distribution of the connection state. The predicted probability distribution of connection status is compared and matched with the preset connection level standard to generate an initial conductor connection integrity label.

[0012] As a preferred embodiment of the rapid detection method for the integrity of automotive wiring harness connections described in this invention, the functional test signal is obtained by applying standard voltage and current conditions to the automotive wiring harness under test at the testing station to obtain voltage excitation signal and current excitation signal, and by collecting the resistance, voltage and current responses of each terminal of the wiring harness under different load conditions.

[0013] As a preferred embodiment of the rapid detection method for the integrity of automotive wiring harness connections according to the present invention, the specific steps for generating the functional response feature set are as follows: Based on the initial wire connection integrity label, the corresponding detection channel in the functional test signal is matched, and the original resistance, voltage and current signals of each channel are extracted to generate the original functional response dataset. The original functional response dataset is subjected to noise filtering and baseline correction, and time alignment and resampling are performed according to the sampling time of each channel to generate a time synchronization signal set; Multidimensional feature extraction is performed on the time synchronization signal set, and the time-domain and frequency-domain statistical parameters of each channel are calculated to generate a preliminary functional response feature matrix. The preliminary functional response feature matrix is ​​standardized and integrated, and then arranged according to the order of the conductor channels to generate a functional response feature set.

[0014] As a preferred embodiment of the rapid detection method for the integrity of automotive wiring harness connections according to the present invention, the specific steps for generating a wiring connection integrity detection report are as follows: The functional response feature set and the initial conductor connection integrity label are associated and integrated according to the conductor channel order to generate a channel-level fusion feature matrix; The channel-level fusion feature matrix is ​​numerically mapped and weighted to form a comprehensive feature vector of conductor connection integrity. Using anomaly detection methods, the comprehensive feature vector of conductor connection integrity is analyzed and calculated to generate conductor connection integrity status judgment results, which are then compiled and summarized to generate a conductor connection integrity detection report.

[0015] Secondly, the present invention provides a rapid detection system for the integrity of automotive wiring harness connections, comprising, The image acquisition module is used to acquire terminal connection images, perform preprocessing, and generate standardized terminal images; The edge extraction module is used to process the input depth edge detection network of the standardized terminal image, extract the features of the wires and pixels, and generate a mask of the wire connection area according to the pixel connectivity. The feature generation module is used to extract pixel grayscale values ​​from the mask of the wire connection area, statistically analyze the grayscale distribution to form a histogram, and normalize the histogram to generate the terminal connection feature vector. The status determination module is used to input the terminal connection feature vector into the pre-trained convolutional neural network for feature classification, obtain the connection status prediction result, match the connection level standard, and generate the initial wire connection integrity label. The signal processing module is used to match and compare the initial wire connection integrity label with the functional test signal, extract the resistance, voltage and current response values ​​in the functional test signal, and perform normalization and time synchronization processing and integration to generate a functional response feature set. The comprehensive analysis module is used to fuse the functional response feature set with the initial conductor connection integrity label to construct a comprehensive feature vector of conductor connection integrity, and perform analysis and calculation to generate a conductor connection integrity detection report.

[0016] The beneficial effects of this invention are as follows: By inputting standardized terminal images into a depth edge detection network, multi-scale conductor contour features are extracted and edge strength is calculated. Non-maximum suppression, binarization, pixel connectivity analysis, and morphological closure calculation are performed, achieving precise localization of conductor connection areas in terminal images. This effectively distinguishes conductors from the background and suppresses noise interference, ensuring the integrity and continuity of conductor connectivity features, achieving high precision and robustness. It provides reliable spatial information and grayscale data for the generation of terminal connection feature vectors, improving the accuracy and detection efficiency of conductor connection status determination. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 A flowchart of a rapid detection method for the integrity of automotive wiring harness wire connections.

[0019] Figure 2This is a schematic diagram of a rapid testing system for the integrity of automotive wiring harness connections.

[0020] Figure 3 A flowchart for generating standardized terminal images.

[0021] Figure 4 A flowchart for generating a mask for the wire connection area. Detailed Implementation

[0022] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0023] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0024] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0025] Reference Figures 1-4 As one embodiment of the present invention, this embodiment provides a rapid detection method for the integrity of automotive wiring harness connections, comprising the following steps: S1. Acquire terminal connection images and perform preprocessing to generate standardized terminal images.

[0026] S1.1 Perform distortion correction and geometric correction on the terminal connection image to generate a geometrically corrected terminal image, and perform multi-scale illumination equalization processing to generate an illumination equalized terminal image.

[0027] Specifically, a standard light source and an industrial camera are set up at the inspection station, and the lighting angle and brightness are adjusted to ensure uniform lighting in the terminal area; the automotive wiring harness to be tested is fixed on a standard fixture, so that the terminal position and the camera optical axis maintain a fixed angle; high-resolution images of the terminal connection area are acquired using the camera, for example, with a resolution of 1920×1080 pixels, and the image frame with the highest clarity is selected; the acquired terminal connection images are labeled and stored according to the inspection number and channel position to generate a terminal connection image dataset; Based on the imaging parameters recorded in the terminal connection image, the camera intrinsic parameter matrix and distortion coefficients are extracted, and a back projection algorithm is used to perform remapping calculations on the coordinates of each pixel to compensate for radial and tangential distortion, generating a geometrically corrected terminal image. According to the boundary coordinates of the terminal area, perspective transformation is used to adjust the proportion and correct the angle of the spatial geometry of the geometrically corrected terminal image, so that the structural shape of the terminal area is consistent with the actual physical structure. Multi-scale illumination equalization processing is performed on the geometrically corrected terminal image to generate an illumination-equalized terminal image.

[0028] It should also be noted that multi-scale illumination equalization processing specifically involves: performing layered filtering, contrast compensation, and brightness normalization on the brightness channels at multiple scales; balancing brightness differences through local dynamic range compression and global brightness compensation, for example, controlling local brightness variations within ±20% of the original brightness; local dynamic range compression refers to compressing overly bright or dark local pixels within a local area of ​​the image based on the average brightness and brightness fluctuation of that area, keeping brightness variations within a controllable range, thereby suppressing abrupt changes in local highlights or shadows; global brightness compensation refers to performing a uniform boost or debuff operation on the brightness channels based on the overall brightness distribution of the entire image after local compression, making the brightness between different areas more consistent, thereby restoring overall visual balance.

[0029] S1.2. Perform noise reduction filtering and edge enhancement processing on the illumination equalization terminal image to generate a terminal image with clear edges, and perform size normalization and coordinate standardization to generate a standardized terminal image.

[0030] Specifically, a bilateral filtering method is used to smooth the illumination-equalized terminal image, removing random noise while preserving the edge structure. In the example, the filtering radius is set to 5 pixels, and the ratio of color variance to spatial variance is set to 1.2. A high-pass filtering method is used to extract the high-frequency components in the illumination-equalized terminal image, and the high-frequency components are superimposed on the original illumination-equalized terminal image with an edge enhancement weight coefficient of, for example, 0.3, to enhance the contrast of the wire edge contour and generate a terminal image with clear edges. The terminal image with clear edges is then normalized by scaling the pixel matrix to the standard terminal size (e.g., 256 pixels wide and 256 pixels high) to maintain a consistent aspect ratio. Based on the pixel coordinate boundary range of the terminal target area in the illumination-equalized terminal image, the coordinate center point is calculated and the coordinates are standardized to align the spatial coordinates of the terminal area with the detection reference coordinates, generating a standardized terminal image.

[0031] It should also be noted that the pixel coordinate boundary range refers to the rectangular area defined by the minimum and maximum horizontal and vertical coordinate values ​​of the outermost pixel in the image coordinate system.

[0032] S2. Process the standardized terminal image input depth edge detection network to extract wire and pixel features, and generate wire connection area masks according to pixel connectivity.

[0033] S2.1 Input the standardized terminal image into the depth edge detection network, extract multi-scale conductor contour features and calculate edge intensity to generate a preliminary edge probability map.

[0034] Specifically, the standardized terminal images are pixel-normalized to a fixed size to ensure that the pixel values ​​of the input images are distributed within the range of 0 to 1. The normalized terminal images are then sequentially input into the convolutional layers of a deep edge detection network to extract multi-scale conductor contour features. Shallow convolutional layers are used to extract detailed edge features, while deep convolutional layers are used to extract global contour information. The feature maps output from each convolutional layer are upsampled and concatenated pixel-by-pixel to form a multi-scale conductor contour feature set. The edge response intensity of each pixel is calculated using a feature fusion convolutional kernel, expressed as: ; in, Indicates the pixel position of the standardized terminal image. Edge response strength, The pixel x-coordinate of the standardized terminal image. Represents the pixel ordinate of the standardized terminal image. Indicates the number of convolutional layers. Indicates the first Each convolutional layer at the normalized terminal image at pixel location Feature map pixel values, The range of values ​​for the index of the convolutional layer is: arrive , Indicates feature fusion convolution kernel, This represents the convolution operation. This represents a normalization function that maps edge response strengths to a range of 0 to 1. The edge response intensity is normalized, and a preliminary edge probability map is generated based on the normalized edge response intensity values.

[0035] It should also be noted that the feature fusion convolution kernel refers to the set of convolution weights used to perform weighted convolution operations on multi-scale feature maps at corresponding pixel positions. By weighting and summing the feature values ​​from different convolutional layers at the same pixel position, it is possible to fuse detailed edge features and global contour features simultaneously in a single convolution operation, thereby forming a unified edge response intensity. The specific steps for pre-training a deep edge detection network are as follows: Standardized terminal images are paired with corresponding terminal edge annotations to construct a training sample set; the training sample set is divided into batches, for example, each batch contains 32 standardized terminal images and their corresponding annotations; each batch of standardized terminal images is input into the input layer of the deep edge detection network, sequentially passing through convolutional layers, activation layers, and pooling layers to extract multi-layer edge feature representations; a loss function, such as binary cross-entropy loss, is calculated based on the network output and the corresponding annotation; the backpropagation algorithm and optimizer, such as stochastic gradient descent, are used to iteratively update the convolutional kernel weights; batch training is repeated until the loss function converges, forming the deep edge detection network.

[0036] S2.2 Perform non-maximum suppression and binarization on the preliminary edge probability map to generate a binary edge mask.

[0037] Specifically, the Sobel operator is used to calculate the gradient components in the horizontal and vertical directions of the preliminary edge probability map, thereby obtaining the gradient magnitude and gradient direction of each pixel. Based on the gradient direction of each pixel, two adjacent pixels are selected along the gradient direction and their gradient magnitudes are compared. If the gradient magnitude of the center pixel is less than that of any adjacent pixel, the center pixel is suppressed to zero, thus completing non-maximum suppression and outputting the suppressed edge intensity map. The suppressed edge intensity map is binarized using a double-threshold hysteresis connection method. Specifically, a high threshold and a low threshold are set (e.g., a high threshold of 0.20 and a low threshold of 0.08). Pixels with strong responses greater than the high threshold are marked as strong edges, pixels with responses less than the low threshold are set as background, and pixels between the high and low thresholds are first marked as weak edges and retained only if they are connected to strong edge pixels; otherwise, they are set as background. Small connected component removal (e.g., removing connected regions with fewer than 3 pixels) is performed on the binarization result, and a binary edge mask is output.

[0038] It should also be noted that the steps for setting the high and low thresholds are as follows: based on the gradient magnitude distribution of the preliminary edge probability map, select the upper percentile (e.g., the 90th percentile) as the high threshold candidate and the lower percentile (e.g., the 40th percentile) as the low threshold candidate, and normalize them to the range of 0 to 1. The high threshold is set to 0.15~0.30 and the low threshold is set to 0.05~0.10. The high threshold is about 2 to 3 times the low threshold and is used for binarization processing.

[0039] S2.3 Perform pixel connectivity analysis and morphological closure calculation on the binary edge mask to generate a wire connection region mask.

[0040] Specifically, using the 8-neighborhood connectivity algorithm, each pixel in the binary edge mask is scanned, and pixels that are connected and have a pixel value of 1 are marked as the same connected region. At the same time, the number of pixels in each connected region is recorded. Regions with a pixel count of less than 20 pixels are removed to avoid isolated noise interference. A circular structuring element with a radius of 3 pixels is selected, and morphological closure calculation is performed on the remaining connected regions. That is, first dilation and then erosion operations are performed to fill the gaps between wires and smooth the edge contours to ensure the continuity of the region and generate a wire connection region mask, in which the part with a pixel value of 1 represents the wire connection region.

[0041] It should be noted that by performing deep edge segmentation on standardized terminal images, combined with multi-scale wire contour feature extraction, edge strength calculation, non-maximum suppression, binarization, pixel connectivity analysis, and morphological closing operations, accurate localization of the wire connection region is achieved. Even in complex environments, it can still provide highly reliable wire connection region features, including the spatial coordinates of the wire end, the contour shape of the wire-terminal contact boundary, the edge connectivity structure corresponding to the wire insertion force direction, the gap width range between the wire and the terminal, and the compactness and continuity indicators of the connection region pixels, laying a solid data foundation for the entire rapid detection method.

[0042] S3. Extract pixel grayscale values ​​from the mask of the wire connection area, statistically analyze the grayscale distribution to form a histogram, and normalize the histogram to generate the terminal connection feature vector.

[0043] S3.1. Pixel-level correspondence is established between the mask of the wire connection area and the standardized terminal image, grayscale information is extracted and organized to generate a grayscale set of wire connection pixels.

[0044] Specifically, the wire connection area mask and the standardized terminal image are processed to correspond at the pixel level. That is, the gray values ​​of the positions where the pixel value of the wire connection area mask is 1 are compared with the gray values ​​of the standardized terminal image pixel by pixel. All corresponding gray values ​​are arranged according to the spatial order of the wire connection area to generate a gray set of wire connection pixels, where each gray value corresponds to a pixel in the wire connection area.

[0045] S3.2 Perform gray-level distribution statistics on the gray-level set of the wire connection pixels, construct a gray-level histogram, and perform numerical normalization processing to generate terminal connection feature vectors.

[0046] Specifically, the gray values ​​of the pixels in the wire connection grayscale set are statistically counted and divided into several grayscale intervals according to the grayscale level. For example, the grayscale value of 0 to 255 is divided into 256 intervals, and the number of pixels in each interval is calculated to construct a grayscale histogram. Each value in the grayscale histogram is normalized, for example, by dividing the number of pixels in each interval by the total number of pixels, so that the histogram value is in the range of 0 to 1, and a terminal connection feature vector is generated, where each feature value corresponds to the normalized pixel ratio of a grayscale interval in the grayscale histogram.

[0047] S4. Input the terminal connection feature vector into the pre-trained convolutional neural network for feature classification, obtain the connection status prediction result, match the connection level standard, and generate the initial wire connection integrity label.

[0048] S4.1 The pre-trained convolutional neural network is obtained by supervised learning training on the terminal connection feature vector and feature classification optimization based on the integrity status of the wire connection.

[0049] Specifically, the terminal connection feature vectors are paired with the corresponding wire connection integrity status labels to construct a training sample set. The training sample set is then divided into batches, for example, each batch contains 32 terminal connection feature vectors and labels. Each batch of terminal connection feature vectors is input into the input layer of a convolutional neural network, passing through convolutional layers, activation layers, pooling layers, and fully connected layers in sequence to extract multi-layer feature representations. A loss function, such as cross-entropy loss, is calculated based on the output of the fully connected layer and the corresponding wire connection integrity status labels. Backpropagation algorithm and optimizer, such as stochastic gradient descent, are used to iteratively update the convolutional kernel weights and fully connected layer weights of the convolutional neural network. Batch training is repeated until the loss function converges, forming a pre-trained convolutional neural network.

[0050] It should also be noted that the steps until the loss function converges are as follows: After each batch training round, the cross-entropy loss method is used to calculate the loss value of the convolutional neural network output and the integrity state label of the wire connection. The gradient is calculated through backpropagation and the weights are updated using the optimizer. After all batch training is completed, the total loss value is calculated and compared with the previous total loss value. When the total loss value no longer changes significantly, the loss function is determined to have converged. This process continues until the convergence condition is met and pre-training is completed. Specifically, when the change in the total loss value between two adjacent training rounds is extremely small and maintains a stable trend, the total loss value is determined to no longer change significantly. The wire connection integrity status label is the result of manual annotation of whether the wire connection in the standardized terminal image is complete. It comes from the manual judgment process of real terminal connection field images. The label content is usually "complete", "abnormal", "loose" and "detached" as the label content, which is used to train the convolutional neural network to identify the supervision signal of wire connection integrity.

[0051] S4.2 Input the terminal connection feature vector into the pre-trained convolutional neural network, perform layer-by-layer convolution and non-linear activation processing on the terminal connection feature vector, and generate an intermediate feature representation map.

[0052] Specifically, the terminal connection feature vector is input into the pre-trained convolutional neural network. Convolution operations and non-linear activation processing are performed on the terminal connection feature vector in sequence. This includes using convolution kernels to perform local weighted summation on the terminal connection feature vector to extract local features, and performing non-linear mapping through the ReLU activation function. Then, sliding convolution calculation is performed on the one-dimensional feature index of the terminal connection feature vector with a set stride, for example, the stride is set to 1 or 2 feature positions. Multiple layers of convolution and activation operations are repeated to gradually extract multi-scale features and generate an intermediate feature representation map. The output of each layer is used as the input of the next layer's convolution operation until the final intermediate feature representation map is obtained.

[0053] It should also be noted that the specific steps for setting the stride are to select the horizontal and vertical movement spacing based on the size of the terminal connection feature vector and the size of the convolution kernel, for example, to set the stride to 1 or 2 pixels, in order to determine the number of pixels the convolution kernel moves on the terminal connection feature vector in each convolution operation. Local features refer to the fine-grained connectivity information extracted by the convolution kernel within the local region of the terminal connectivity feature vector, such as the intensity variation pattern near the wire contact point, the gradient structure of the terminal edge, or local texture changes. Multi-scale features refer to the feature information extracted layer by layer in multi-layer convolution by using different receptive field sizes, from fine to coarse. For example, shallow convolutions yield detailed texture features and edge details, while deep convolutions yield overall connectivity features and global distribution features.

[0054] S4.3 Perform fully connected layer mapping and classification calculation on the intermediate feature representation map to obtain the predicted probability distribution of the connection state.

[0055] Specifically, the intermediate feature representation map is flattened into a one-dimensional vector, and a linear mapping is performed according to the number of neurons in the fully connected layer and the weight matrix. The Softmax activation function is then applied sequentially to the mapping results, mapping the output to the predicted probability values ​​of each wire connection state, generating a connection state prediction probability distribution, expressed as: ; in, This represents the probability distribution of connection state predictions. This represents the weight matrix of the fully connected layer. This represents the vector obtained by flattening the intermediate feature representation graph. This represents the bias vector of the fully connected layer.

[0056] S4.4 Compare and match the predicted probability distribution of connection status with the preset connection level standard to generate an initial conductor connection integrity label.

[0057] Specifically, the probability values ​​of each category in the predicted probability distribution of connection status are compared one by one with the corresponding category judgment criteria in the preset connection level standard; the category with the highest probability value and meeting the level standard is selected as the matching result of the terminal connection status based on the comparison results; the matching result is mapped to the corresponding wire connection integrity level code, for example, "good" is 1, "loose" is 2 and "disconnected" is 3; the matching results of all terminal connection statuses are sorted in sequence to generate an initial wire connection integrity label sequence, where each label corresponds to the connection integrity status of a wire.

[0058] It should also be noted that the specific steps for setting the connection level standard are as follows: Based on the automotive wiring harness industry testing specifications or historical testing data, determine the classification standards for different wire connection states. For example, define terminals with tight connections and good contact as the "good" level, terminals with slight looseness as the "loose" level, and terminals with unconnected or broken wires as the "disconnect" level. Assign a unique numerical code to each level, for example, "good" is 1, "loose" is 2, and "disconnect" is 3. Record the terminal connection feature range corresponding to each level, including terminal pixel grayscale features, edge features, and functional test response features, forming a standardized data table as a reference for subsequent connection state prediction probability distribution comparison and matching.

[0059] S5. Match and compare the initial wire connection integrity label with the functional test signal, extract the resistance, voltage and current response values ​​in the functional test signal, and perform normalization and time synchronization processing and integration to generate a functional response feature set.

[0060] S5.1 The functional test signal is obtained by applying standard voltage and current conditions to the automotive wiring harness under test at the testing station to obtain voltage excitation signal and current excitation signal, and collecting the resistance, voltage and current response of each terminal of the wiring harness under different load conditions.

[0061] Specifically, at the testing station, each terminal of the automotive wiring harness under test is connected sequentially to a standard voltage source and a standard current source. Voltage excitation signals of different amplitudes and current excitation signals of fixed amplitudes are applied according to a preset test sequence, such as a voltage range of 0~12 volts and a current range of 0~2 amperes. Under each load condition, the voltage response, current response, and resistance value of each terminal under the corresponding load are recorded. The collected resistance, voltage, and current data are numbered and organized according to the terminal sequence and load conditions to form functional test signals.

[0062] It should also be noted that the specific steps of the preset test sequence are as follows: According to the terminal arrangement and electrical connection relationship of the wiring harness of the vehicle under test, arrange each terminal sequentially starting from 1; apply standard voltage and standard current conditions to each terminal sequentially to ensure that each terminal has undergone full-range voltage and current testing; form an orderly test list according to the terminal number sequence and load conditions, and record the corresponding terminal number, applied voltage amplitude and current amplitude as the basis for the sequence of functional test signal acquisition.

[0063] S5.2. Based on the initial wire connection integrity label, match the corresponding detection channel in the functional test signal, extract the original resistance, voltage and current signals of each channel, and generate the original functional response dataset.

[0064] Specifically, based on the initial wire connection integrity label, the corresponding detection channel is found in the functional test signal according to the terminal number and connection status recorded in the label; the resistance value, voltage value and current value recorded in the functional test signal are extracted channel by channel, and the original resistance, voltage and current signals corresponding to each terminal are sorted in the order of terminal number to generate the original functional response dataset containing all terminal channels. For example, the resistance, voltage and current signals of terminals 1 to N are summarized in sequence to form a matrix structure.

[0065] S5.3. Perform noise filtering and baseline correction on the original functional response dataset, and perform time alignment and resampling according to the sampling time of each channel to generate a time synchronization signal set.

[0066] Specifically, each channel signal in the original functional response dataset is first low-pass filtered to remove high-frequency noise, and baseline correction is performed on each channel signal by subtracting the average value or moving average baseline value of the initial sampling points. All channel signals are time-aligned according to the sampling time of each channel, and resampled using a linear interpolation method to make each channel signal correspond on a unified time axis, generating a time synchronization signal set containing the resistance, voltage and current of each terminal. For example, each channel signal from terminal 1 to terminal N is resampled and aligned to the time sequence at 0.1 seconds.

[0067] S5.4 Perform multi-dimensional feature extraction on the time synchronization signal set, calculate the time-domain and frequency-domain statistical parameters of each channel, and generate a preliminary functional response feature matrix.

[0068] Specifically, for each channel signal, the arithmetic mean method is used to calculate the average value according to the sampling time series, the variance and standard deviation are calculated using the analysis of variance method, the maximum and minimum values ​​are obtained using the extreme value search method, the peak-to-peak value is calculated using the peak-to-peak value analysis method, and the root mean square value is calculated using the root mean square analysis method to obtain the time-domain statistical parameters. For each channel signal, the local maximum and minimum values ​​and their durations in the time series are extracted using the local extremum detection method. A fast Fourier transform is performed on each channel signal to calculate the amplitude of the frequency components and identify the dominant frequency component. At the same time, the frequency center is obtained using the frequency weighting method, and the spectral energy is obtained through spectral energy analysis to form the frequency-domain statistical parameters. The power spectral density is calculated based on the spectral amplitude. The time-domain statistical parameters and frequency-domain statistical parameters of each channel are arranged into row vectors in a fixed order. The resistance signal, voltage signal, and current signal are processed separately. The row vectors of all channels are concatenated column by column to generate a preliminary functional response feature matrix. For example, the resistance, voltage, and current signals of terminals 1 to N are extracted and combined sequentially.

[0069] S5.5 Standardize and integrate the preliminary functional response feature matrix, and organize it according to the order of the conductor channels to generate a functional response feature set.

[0070] Specifically, the preliminary functional response feature matrix is ​​standardized, including linearly normalizing the minimum and maximum values ​​of each column so that the values ​​of each column fall within the range of 0 to 1, and centering the mean of each feature to obtain a standardized feature column with a mean of zero. The standardized feature columns are then arranged according to the conductor channel order, and the standardized feature columns corresponding to the resistance signal, voltage signal, and current signal channels are sequentially concatenated to form the row vector of each channel. The row vectors of all channels are then combined according to the column order to complete feature integration and generate a functional response feature set arranged according to the conductor channel order.

[0071] It should also be noted that the wire channel sequence refers to the sequential numbering and arrangement of the acquisition channels for resistance signals, voltage signals, and current signals according to the order in which the wires in the wiring harness under test are arranged at the terminals. This is used to ensure that the functional response feature set matches the actual wire layout.

[0072] S6. Perform feature fusion between the functional response feature set and the initial conductor connection integrity label to construct a comprehensive feature vector of conductor connection integrity, and perform analysis and calculation to generate a conductor connection integrity detection report.

[0073] S6.1. Associate and integrate the functional response feature set with the initial conductor connection integrity label according to the conductor channel order to generate a channel-level fusion feature matrix.

[0074] Specifically, for each conductor channel, the corresponding time-domain and frequency-domain feature vectors are extracted from the functional response feature set according to the conductor channel order, and matched sequentially with the connection status labels of the corresponding channels in the initial conductor connection integrity label. The functional response feature vectors and connection status labels are merged sequentially to form row vectors. All conductor channels are processed repeatedly, and the row vectors are concatenated column by column according to the conductor channel order to generate a channel-level fusion feature matrix, where each column corresponds to the functional response feature and connection integrity label of a channel.

[0075] S6.2. Perform numerical range mapping and weighted combination on the channel-level fusion feature matrix to form a comprehensive feature vector of conductor connection integrity.

[0076] Specifically, the values ​​in each column of the channel-level fusion feature matrix are normalized to map the values ​​to a fixed range, such as the interval between 0 and 1. Weight coefficients are assigned to each column according to the order of the conductor channels, such as the example range of 0.2 to 0.5. After multiplying each column value with the corresponding weight coefficient, all columns are superimposed in the order of the conductor channels to form a single row vector. The single row vector is used as a whole to represent the weighted feature information of each conductor channel. The numerical integration is completed through sequential arrangement and weighted superposition to generate a comprehensive feature vector of conductor connection integrity, which contains a comprehensive representation of the functional response features of all conductor channels and the initial conductor connection integrity information.

[0077] S6.3 Using anomaly detection methods, analyze and calculate the comprehensive feature vector of conductor connection integrity, generate conductor connection integrity status judgment results, and compile and summarize them to generate a conductor connection integrity detection report.

[0078] Specifically, the Mahalanobis distance or local density method is used to detect anomalies in the comprehensive feature vector of conductor connection integrity, and an anomaly score is obtained for each feature vector. The anomaly scores are then judged according to preset classification rules, such as setting multiple levels of anomaly intervals and dividing the scores into normal, slightly abnormal, and severely abnormal levels. The judgment results of each conductor channel are summarized to form the conductor connection integrity status judgment result. The judgment results of each channel are organized according to the order of conductor channels, and the status of each channel is summarized by row or column to form a complete record, generating a conductor connection integrity detection report. Each record includes the conductor channel identifier and the corresponding connection integrity status, such as "complete" or "abnormal".

[0079] It should also be noted that the preset classification rule refers to the method of mapping the abnormal score corresponding to the comprehensive feature vector of the conductor connection integrity to different abnormal levels according to the abnormal score division standard. For example, the normal, slight abnormal and severe abnormal levels are determined by multi-level abnormal intervals. The local density method characterizes the degree of anomaly by calculating the neighborhood density of the comprehensive feature vector of conductor connectivity integrity in the feature space. Specifically, in the feature space, several nearest historical normal feature vectors are selected with the comprehensive feature vector of conductor connectivity integrity as the center, and the average neighborhood density between the comprehensive feature vector of conductor connectivity integrity and the historical normal feature vectors is calculated. When the average neighborhood density is relatively low, it is mapped to a higher anomaly value, and when the average neighborhood density is relatively high, it is mapped to a lower anomaly value. A local density anomaly score is generated through the inverse correspondence between density and anomaly value, and together with the statistical distance score obtained by Mahalanobis distance, an anomaly score is formed. Historical normal feature distribution refers to the statistical characteristics and numerical distribution of the comprehensive feature vector of conductor connection integrity collected in the order of conductor channels under normal conductor connection conditions, and is used as a reference benchmark for anomaly judgment.

[0080] It should be noted that by fusing the functional response feature set with the initial wire connection integrity label, a comprehensive feature vector of wire connection integrity is constructed. This vector is then analyzed and calculated using anomaly detection methods, enabling a multi-dimensional comprehensive evaluation of the wiring harness connection status. This achieves high-precision, multi-dimensional, and rapid wire connection integrity detection, significantly outperforming traditional methods that rely on only a single technical approach.

[0081] This embodiment also provides a rapid detection system for the integrity of automotive wiring harness wire connections, including: an image acquisition module for acquiring terminal connection images and preprocessing them to generate standardized terminal images; an edge extraction module for processing the standardized terminal images using a depth edge detection network, extracting wire and pixel features, and generating a wire connection region mask based on pixel connectivity; a feature generation module for extracting pixel grayscale values ​​from the wire connection region mask, statistically analyzing the grayscale distribution to form a histogram, and normalizing the histogram to generate a terminal connection feature vector; and a state determination module for inputting the terminal connection feature vector into... After pre-training, the convolutional neural network performs feature classification to obtain connection state prediction results, matches them with connection level standards, and generates initial conductor connection integrity labels. The signal processing module is used to match and compare the initial conductor connection integrity labels with functional test signals, extract the resistance, voltage, and current response values ​​from the functional test signals, and perform normalization, time synchronization processing, and integration to generate a functional response feature set. The comprehensive analysis module is used to fuse the functional response feature set with the initial conductor connection integrity labels to construct a comprehensive feature vector of conductor connection integrity, and perform analysis and calculation to generate a conductor connection integrity detection report.

[0082] In summary, this invention achieves precise localization of wire connection regions in terminal images by inputting standardized terminal images into a depth edge detection network, extracting multi-scale wire contour features and calculating edge strength, performing non-maximum suppression, binarization, pixel connectivity analysis, and morphological closure calculation. This effectively distinguishes wires from the background and suppresses noise interference, ensuring the integrity and continuity of wire connectivity features, achieving high precision and robustness. It also provides reliable spatial information and grayscale data for generating terminal connection feature vectors, improving the accuracy and detection efficiency of wire connection status determination.

[0083] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for rapid detection of wire harness wire connection integrity in an automobile, characterized by: include, Acquire terminal connection images and perform preprocessing to generate standardized terminal images; The standardized terminal image input depth edge detection network is processed to extract wire and pixel features, and a wire connection region mask is generated according to pixel connectivity. Pixel grayscale values ​​are extracted from the mask of the wire connection area, the grayscale distribution is statistically analyzed to form a histogram, and the histogram is normalized to generate the terminal connection feature vector. The terminal connection feature vector is input into a pre-trained convolutional neural network for feature classification to obtain the connection status prediction result, and then matched with the connection level standard to generate an initial wire connection integrity label. The initial wire connection integrity label is matched and compared with the functional test signal. The resistance, voltage, and current response values ​​in the functional test signal are extracted, normalized, and time-synchronized and integrated to generate a functional response feature set. The specific steps are as follows. Based on the initial wire connection integrity label, the corresponding detection channel in the functional test signal is matched, and the original resistance, voltage and current signals of each channel are extracted to generate the original functional response dataset. The original functional response dataset is subjected to noise filtering and baseline correction, and time alignment and resampling are performed according to the sampling time of each channel to generate a time synchronization signal set; Multidimensional feature extraction is performed on the time synchronization signal set, and the time-domain and frequency-domain statistical parameters of each channel are calculated to generate a preliminary functional response feature matrix. The preliminary functional response feature matrix is ​​standardized and integrated, and then arranged according to the conductor channel order to generate a functional response feature set. The functional response feature set and the initial conductor connection integrity label are fused to construct a comprehensive feature vector for conductor connection integrity. This vector is then analyzed and calculated to generate a conductor connection integrity detection report. The specific steps are as follows: The functional response feature set and the initial conductor connection integrity label are associated and integrated according to the conductor channel order to generate a channel-level fusion feature matrix; The channel-level fusion feature matrix is ​​numerically mapped and weighted to form a comprehensive feature vector of conductor connection integrity. Using anomaly detection methods, the comprehensive feature vector of conductor connection integrity is analyzed and calculated to generate conductor connection integrity status judgment results, which are then compiled and summarized to generate a conductor connection integrity detection report.

2. The method of claim 1, wherein: The specific steps for generating the standardized terminal image are as follows: Distortion correction and geometric correction are performed on the terminal connection image to generate a geometrically corrected terminal image, and multi-scale illumination equalization processing is performed to generate an illumination equalized terminal image. The terminal images with uniform illumination are subjected to noise reduction filtering and edge enhancement processing to generate terminal images with clear edges. The size is then normalized and the coordinates are standardized to generate standardized terminal images.

3. The rapid detection method for the integrity of automotive wiring harness connections as described in claim 1, characterized in that: The specific steps for generating the conductor connection area mask are as follows: The standardized terminal image is input into the depth edge detection network to extract multi-scale wire contour features and calculate edge intensity, generating a preliminary edge probability map. Non-maximum suppression and binarization are applied to the preliminary edge probability map to generate a binary edge mask; Pixel connectivity analysis and morphological closure calculation are performed on the binary edge mask to generate a wire connection region mask.

4. The rapid detection method for the integrity of automotive wiring harness connections as described in claim 1, characterized in that: The specific steps for generating the terminal connection feature vector are as follows: The mask of the wire connection area is mapped to the standardized terminal image at the pixel level, grayscale information is extracted and organized to generate a grayscale set of wire connection pixels. The grayscale distribution of the pixel grayscale set of the wire connection is statistically analyzed, a grayscale histogram is constructed, and numerical normalization is performed to generate the terminal connection feature vector.

5. The rapid detection method for the integrity of automotive wiring harness connections as described in claim 1, characterized in that: The pre-trained convolutional neural network is obtained by supervised learning training on the terminal connection feature vector and feature classification optimization based on the integrity status of the wire connection.

6. The rapid detection method for the integrity of automotive wiring harness connections as described in claim 1, characterized in that: The specific steps for generating the initial wire connection integrity label are as follows. The terminal connection feature vector is input into a pre-trained convolutional neural network, and the terminal connection feature vector is subjected to layer-by-layer convolution and non-linear activation to generate an intermediate feature representation map. The intermediate feature representation map is mapped and classified using a fully connected layer to obtain the predicted probability distribution of the connection state. The predicted probability distribution of connection status is compared and matched with the preset connection level standard to generate an initial conductor connection integrity label.

7. The rapid detection method for the integrity of automotive wiring harness connections as described in claim 1, characterized in that: The functional test signals are obtained by applying standard voltage and current conditions to the automotive wiring harness under test at the testing station to obtain voltage excitation signals and current excitation signals, and by collecting the resistance, voltage and current responses of each terminal of the wiring harness under different load conditions.

8. A rapid detection system for the integrity of automotive wiring harness connections, based on the rapid detection method for the integrity of automotive wiring harness connections according to any one of claims 1 to 7, characterized in that: include, The image acquisition module is used to acquire terminal connection images, perform preprocessing, and generate standardized terminal images; The edge extraction module is used to process the input depth edge detection network of the standardized terminal image, extract the features of the wires and pixels, and generate a mask of the wire connection area according to the pixel connectivity. The feature generation module is used to extract pixel grayscale values ​​from the mask of the wire connection area, statistically analyze the grayscale distribution to form a histogram, and normalize the histogram to generate the terminal connection feature vector. The status determination module is used to input the terminal connection feature vector into the pre-trained convolutional neural network for feature classification, obtain the connection status prediction result, match the connection level standard, and generate the initial wire connection integrity label. The signal processing module is used to match and compare the initial wire connection integrity label with the functional test signal, extract the resistance, voltage and current response values ​​in the functional test signal, and perform normalization and time synchronization processing and integration to generate a functional response feature set. The comprehensive analysis module is used to fuse the functional response feature set with the initial conductor connection integrity label to construct a comprehensive feature vector of conductor connection integrity, and perform analysis and calculation to generate a conductor connection integrity detection report.

Images