A wheel hub detection method and system

Abstract

The application provides a wheel hub detection method and system, which comprises the following steps: placing a wheel hub on a preset detection platform through a preset mechanical arm, and driving the wheel hub to rotate at a constant speed through the preset detection platform; during the rotation of the wheel hub, a to-be-detected area matched with the wheel hub is calibrated in real time, and the to-be-detected area is scanned in full through a preset laser ultrasonic detector; ultrasonic signals generated by the preset laser ultrasonic detector during the scanning are collected in real time, and a cross-sectional image corresponding to the wheel hub is simulated synchronously; the ultrasonic signals and the cross-sectional image are synchronously input into a preset CNN model, and whether the wheel hub has defects is judged in real time through the preset CNN model. The application can objectively and accurately judge whether the wheel hub has defects, and the detection efficiency is improved correspondingly.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of hub detection, and particularly relates to a hub detection method and system. BACKGROUND

[0002] With the progress of science and technology and the rapid development of productivity, automobiles have been popularized in people's daily life and have become one of the indispensable means of transportation for people's daily travel, greatly facilitating people's life.

[0003] Among them, the hub is one of the important parts of the existing automobile, which is used for installing tires and enabling the automobile to travel on the ground, so the performance of the hub can also correspondingly affect the performance of the automobile driving.

[0004] Further, the existing hub needs to be detected before leaving the factory to determine whether the produced hub has micro cracks and pores and other defects in real time. However, in the actual detection process, the existing technology needs the staff to hold the scanner to scan the hub point by point, which needs to consume a long detection time, and is prone to miss detection and misdiagnosis, which reduces the detection efficiency of the hub. SUMMARY

[0005] Therefore, the purpose of the present application is to provide a hub detection method and system to solve the problem that the existing technology needs the staff to hold the scanner to scan the hub point by point, which needs to consume a long time and is prone to miss detection and misdiagnosis.

[0006] The first aspect of the embodiment of the present application provides:

[0007] A hub detection method, wherein the method comprises:

[0008] placing the hub on a preset detection platform by a preset mechanical arm, and rotating the hub at a constant speed by the preset detection platform;

[0009] during the rotation of the hub, real-time calibrating a to-be-detected area adapted to the hub, and performing full-disk scanning on the to-be-detected area by a preset laser ultrasonic detector;

[0010] real-time collecting ultrasonic signals corresponding to the scanning process of the preset laser ultrasonic detector, and synchronously simulating a cross-sectional image corresponding to the hub;

[0011] synchronously inputting the ultrasonic signals and the cross-sectional image into a preset CNN model, and real-time determining whether the hub has defects by the preset CNN model.

[0012] The beneficial effects of the present application are: by placing the wheel hub on the corresponding detection platform, the detection process of the current wheel hub is immediately performed, based on this, in order to comprehensively detect whether the current wheel hub has defects, the first step is to calibrate the to-be-detected area prone to defects in the current wheel hub, based on this, full scanning is carried out, and the corresponding ultrasonic signal and cross-sectional image can be obtained during the scanning process, based on this, finally input into the inside of the preset CNN model, the current wheel hub can be objectively and accurately judged by the CNN model in real time whether it has defects, so as to save the process of manual point-by-point scanning, shorten the detection time, and improve the detection efficiency.

[0013] Further, the step of real-time calibrating the to-be-detected area matched with the wheel hub during the rotation of the wheel hub comprises:

[0014] real-time collection of continuous frame images corresponding to the wheel hub generated during the rotation of the wheel hub by a preset high-speed industrial camera, and real-time extraction of key structural features corresponding to the wheel hub in the continuous frame images by a preset edge detection algorithm;

[0015] real-time construction of a dynamic three-dimensional contour model corresponding to the wheel hub, and real-time extraction of a defect distribution heat map corresponding to the wheel hub in a historical detection database;

[0016] spatial matching of the dynamic three-dimensional contour model and the defect distribution heat map to real-time detect a high-defect area, and setting the high-defect area as the to-be-detected area.

[0017] Further, the step of spatial matching of the dynamic three-dimensional contour model and the defect distribution heat map to real-time detect a high-defect area comprises:

[0018] extracting a plurality of feature points from the dynamic three-dimensional contour model, and real-time marking a projection position corresponding to each feature point in the defect distribution heat map;

[0019] real-time calculation of a target heat value corresponding to the projection position of the feature point by a RANSAC algorithm, and real-time judgment of whether the target heat value is greater than a preset heat threshold;

[0020] If it is judged in real time that the target heat value is greater than the preset heat threshold, the area where the feature point corresponding to the current heat value exists is set as the high-defect area.

[0021] Further, the step of real-time collection of the ultrasonic signal corresponding to the wheel hub generated during the scanning process of the preset laser ultrasonic detector, and synchronous simulation of a cross-sectional image corresponding to the wheel hub comprises:

[0022] starting a dual-channel acquisition mode of the laser ultrasonic detector in real time, wherein a first channel is used to acquire an ultrasonic reflection signal, and a second channel is used to acquire an ultrasonic transmission signal;

[0023] constructing a corresponding polar coordinate system with a rotation center of the hub as an origin, and acquiring an initial image corresponding to the hub in real time;

[0024] dividing the initial image into a plurality of image grids according to the polar coordinate system, detecting signal amplitudes corresponding to the ultrasonic reflection signal and the ultrasonic transmission signal in real time, and filling corresponding pixel points in the plurality of image grids according to the signal amplitudes to generate the cross-sectional image.

[0025] Further, the step of filling corresponding pixel points in the plurality of image grids according to the signal amplitudes to generate the cross-sectional image comprises:

[0026] performing time domain slicing processing on the signal amplitudes to generate a plurality of slice signals;

[0027] detecting signal peak values corresponding to each of the slice signals in real time, and assigning an index value to each of the image grids;

[0028] generating a target pixel point according to the signal peak values, and filling the target pixel point into a target image grid according to a size of the index value.

[0029] Further, the step of inputting the ultrasonic signal and the cross-sectional image into a preset CNN model in real time and determining whether the hub has a defect through the preset CNN model comprises:

[0030] intercepting the ultrasonic signal into 1024 signal segments at intervals of 200 ns, extracting a peak value, a valley value and a number of zero-crossing points of each signal segment to form a signal feature matrix of 3x1024, and performing normalization processing on each element in the signal feature matrix;

[0031] scaling the cross-sectional image to 224x224 pixels, splitting it into an image feature matrix of 3x224x224 according to RGB three channels, wherein each channel pixel value is divided by 255 to complete normalization;

[0032] inputting the signal feature matrix and the image feature matrix into the preset CNN model to determine whether the hub has a defect in real time.

[0033] Further, the step of inputting the signal feature matrix and the image feature matrix into the preset CNN model to determine in real time whether the hub has defects comprises:

[0034] The signal feature matrix is input into a 1D convolution layer of the preset CNN model, 16 convolution kernels with a size of 3 are set in the layer, the step is set to 1, a ReLU activation function is used, and a signal feature map of 16*512 is output after maximum pooling; the image feature matrix is input into a 2D convolution layer of the preset CNN model, 32 convolution kernels with a size of 3*3 are set in the layer, the step is set to 1, a ReLU activation function is used, and an image feature map of 32*112*112 is output after maximum pooling;

[0035] The signal feature map and the image feature map are fused into a corresponding target feature vector, and are input into a full connection layer of the preset CNN model, and then are input into an output layer after a dropout layer, the output layer uses a sigmoid activation function, and outputs a probability value of 0-1, when the probability value is greater than or equal to 0.6, it is determined that the hub has defects, and when the probability value is less than 0.6, it is determined that the hub has no defects.

[0036] The second aspect of the embodiment of the application provides:

[0037] A hub detection system, wherein the system comprises:

[0038] A detection module is configured to place a hub on a preset detection platform by a preset mechanical arm, and drive the hub to rotate at a constant speed by the preset detection platform.

[0039] A calibration module is configured to calibrate a to-be-detected area adapted to the hub in real time during rotation of the hub, and perform full-plate scanning on the to-be-detected area by a preset laser ultrasonic detector.

[0040] A collection module is configured to collect ultrasonic signals generated by the preset laser ultrasonic detector in real time during scanning, and synchronously simulate a cross-sectional image corresponding to the hub.

[0041] A judgment module is configured to input the ultrasonic signals and the cross-sectional image into a preset CNN model synchronously, and determine in real time whether the hub has defects by the preset CNN model.

[0042] Further, the calibration module is specifically configured to:

[0043] Real-time collection of continuous frame images corresponding to the hub generated in the rotation process by a pre-set high-speed industrial camera, and real-time extraction of key structural features corresponding to the hub in the continuous frame images by a pre-set edge detection algorithm;

[0044] Real-time construction of a dynamic three-dimensional contour model corresponding to the hub, and real-time extraction of a defect distribution heat map corresponding to the hub in a historical detection database;

[0045] Spatial matching of the dynamic three-dimensional contour model and the defect distribution heat map to real-time detect a corresponding defect high-risk area, and corresponding setting of the defect high-risk area as the to-be-detected area.

[0046] Further, the calibration module is specifically used for:

[0047] Extraction of a plurality of feature points from the dynamic three-dimensional contour model, and real-time marking of a projection position corresponding to each feature point in the defect distribution heat map;

[0048] Real-time calculation of a target heat value corresponding to the feature point at the projection position by a RANSAC algorithm, and real-time judgment of whether the target heat value is greater than a pre-set heat threshold;

[0049] If it is real-time judged that the target heat value is greater than the pre-set heat threshold, the area where the feature point corresponding to the current heat value exists is set as the defect high-risk area.

[0050] Further, the collection module is specifically used for:

[0051] Real-time starting of a dual-channel collection mode of the laser ultrasonic detector, wherein a first channel is used to collect an ultrasonic reflection signal, and a second channel is used to collect an ultrasonic transmission signal;

[0052] Construction of a corresponding polar coordinate system with the center of rotation of the hub as the origin, and real-time collection of an initial image corresponding to the hub;

[0053] According to the polar coordinate system, the initial image is divided into a plurality of image grids, and the signal amplitudes corresponding to the ultrasonic reflection signal and the ultrasonic transmission signal are real-time detected, and according to the signal amplitudes, corresponding pixel points are filled in a plurality of image grids to correspondingly generate the cross-sectional image.

[0054] Further, the collection module is specifically used for:

[0055] Time domain slicing processing of the signal amplitudes to correspondingly generate a plurality of slice signals;

[0056] Real-time detection of each said slice signal respectively corresponding signal peak, and each said image grid is assigned to the corresponding index value;

[0057] According to the signal peak generates corresponding target pixel point, and according to the size of the index value fills the target pixel point to the inside of the corresponding target image grid.

[0058] Further, the judging module is specifically used for:

[0059] The ultrasonic signal is intercepted as 1024 signal segments at 200ns intervals, and the peak value, valley value and zero-crossing point number of each signal segment are extracted to form a 3x1024 signal feature matrix, and each element in the signal feature matrix is normalized;

[0060] The cross-sectional image is scaled to 224x224 pixels, and is split into a 3x224x224 image feature matrix according to RGB three channels, wherein each channel pixel value is divided by 255 to complete normalization;

[0061] The signal feature matrix and the image feature matrix are input into the inside of the preset CNN model to judge whether the hub has defects in real time.

[0062] Further, the judging module is specifically used for:

[0063] The signal feature matrix is input into the 1D convolution layer of the preset CNN model, the layer is provided with 16 convolution kernels with a size of 3, the step is set to 1, the ReLU activation function is adopted, and a 16x512 signal feature map is output after maximum pooling; the image feature matrix is input into the 2D convolution layer of the preset CNN model, the layer is provided with 32 convolution kernels with a size of 3x3, the step is set to 1, the ReLU activation function is adopted, and a 32x112x112 image feature map is output after maximum pooling;

[0064] The signal feature map and the image feature map are fused into a corresponding target feature vector, and input into the full connection layer of the preset CNN model, and then input into the output layer after the dropout layer, the output layer adopts the sigmoid activation function, and outputs the probability value of 0-1, when the probability value is greater than or equal to 0.6, it is determined that the hub has defects, and when the probability value is less than 0.6, it is determined that there is no defect.

[0065] The third aspect of the embodiment of the application provides:

[0066] A computer comprises a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor implements the hub detection method as described above when executing the computer program.

[0067] The fourth aspect of the embodiments of the present application provides:

[0068] A readable storage medium, which stores a computer program, wherein the program is executed by a processor to implement the hub detection method as described above.

[0069] Additional aspects and advantages of the present application will be made apparent from the following description. BRIEF DESCRIPTION OF DRAWINGS

[0070] Figure 1 A flow chart of the hub detection method provided by the first embodiment of the present application;

[0071] Figure 2 A structural block diagram of the hub detection system provided by the third embodiment of the present application.

[0072] The following detailed description will further describe the present application with reference to the above-mentioned drawings. DETAILED DESCRIPTION

[0073] In order to facilitate the understanding of the present application, the present application will be described more fully below with reference to the accompanying drawings. The accompanying drawings show several embodiments of the present application. However, the present application can be realized in many different forms and is not limited to the embodiments described herein. On the contrary, these embodiments are provided so that the disclosure of the present application is more thorough and comprehensive.

[0074] It should be noted that when an element is referred to as being "fixedly attached" to another element, it can be directly on the other element or there can be an intervening element. When an element is referred to as being "connected" to another element, it can be directly connected to the other element or intervening elements can be present. As used herein, the terms "vertical", "horizontal", "left", "right", and the like are merely used for the purpose of illustration.

[0075] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used in the description of the application herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. As used herein, the term "and / or" includes any and all combinations of one or more of the associated listed items.

[0076] Reference will now be made to Figure 1The wheel hub detection method provided by the first embodiment of the application can objectively and accurately detect whether the wheel hub has defects, thereby eliminating the manual detection process, shortening the detection time, and improving the detection efficiency.

[0077] Specifically, the first embodiment provides:

[0078] A wheel hub detection method, specifically comprising the following steps:

[0079] Step S10, placing the wheel hub on a preset detection platform by a preset mechanical arm, and rotating the wheel hub at a constant speed by the preset detection platform;

[0080] It should be noted that the wheel hub to be detected is stably placed on the preset detection platform (such as a tool table with a rotating shaft) by the preset mechanical arm (with precise positioning function), and then the detection platform drives the wheel hub to rotate at a constant speed (such as 5r / min, ensuring uniform scanning data). The initial position of the wheel hub is ensured by the mechanical arm, and the constant rotation provides a stable dynamic detection scene for subsequent laser ultrasonic scanning, avoiding missing points or repetition caused by speed fluctuations.

[0081] Step S20, during the rotation of the wheel hub, real-time calibration of the detection area adapted to the wheel hub, and full scanning of the detection area by a preset laser ultrasonic detector;

[0082] It should be noted that during the rotation of the wheel hub, the algorithm is used to determine the detection area to be detected in real time (excluding the edge or mounting hole area without detection significance), and then the preset laser ultrasonic detector is started to perform full coverage and no-angle scanning on the area (laser excitation of ultrasonic waves, abnormal reflection / transmission of ultrasonic waves during propagation). It should be noted that "area calibration" reduces the invalid scanning range and improves the detection efficiency; "laser ultrasonic scanning" uses the high sensitivity of ultrasonic waves to internal defects to detect surface and internal defects of the wheel hub.

[0083] Step S30, real-time acquisition of the ultrasonic signals generated by the preset laser ultrasonic detector during scanning, and synchronous simulation of the cross-sectional image corresponding to the wheel hub;

[0084] It should be noted that in actual application, two tasks are synchronously performed, one is to collect the ultrasonic signal (including the characteristic information of defect reflection / transmission) generated by the laser ultrasonic detector in the scanning process in real time; the other is to simulate the cross-sectional image of the hub based on the hub rotation track and the scanning data (to intuitively present the radial structure distribution of the hub). It should be noted that the ultrasonic signal can provide quantitative data of the defect (such as the signal amplitude related to the size and depth of the defect), and the cross-sectional image can provide spatial position information of the defect, and the two can realize multi-dimensional detection based on “data + image”.

[0085] In step S40, the ultrasonic signal and the cross-sectional image are synchronously input into a preset CNN model, and whether the hub has a defect is determined in real time by the preset CNN model.

[0086] It should be noted that the collected ultrasonic signal and the simulated cross-sectional image are synchronously input into a preset CNN model (convolutional neural network model, which has been trained by a large amount of defect / non-defect hub data), and the model extracts features and classifies to output the judgment result of whether the hub has a defect in real time. The present application uses the strong feature learning ability of the CNN model for images and time series signals to replace manual judgment, improves the accuracy and efficiency of defect recognition, and avoids human subjective errors. Through the above-mentioned manner, whether the hub has a defect can be objectively and accurately detected, and the detection efficiency is improved.

[0087] Second embodiment

[0088] Further, the step of calibrating the to-be-detected area adapted to the hub in real time during the rotation of the hub comprises:

[0089] The continuous frame images corresponding to the rotation of the hub are collected in real time by a preset high-speed industrial camera, and the key structural features corresponding to the hub in the continuous frame images are extracted in real time by a preset edge detection algorithm;

[0090] A dynamic three-dimensional contour model corresponding to the hub is constructed in real time, and a defect distribution heat map corresponding to the hub is extracted in real time from a historical detection database;

[0091] The dynamic three-dimensional contour model and the defect distribution heat map are spatially matched to detect the corresponding high-defect area in real time, and the high-defect area is set as the to-be-detected area.

[0092] It should be noted that the preset high-speed industrial camera (such as frame rate 100fps, which ensures the capture of dynamic details of the hub rotation) is started, and continuous frame images during hub rotation are collected; then through a preset edge detection algorithm (such as Canny algorithm), key structural features (such as rim edge, spoke contour, and hub inner / outer diameter boundary) of the hub are extracted from the image. Based on the extracted key structural features, a dynamic three-dimensional contour model of the hub is constructed in real time through a multi-view geometry algorithm (restoring the three-dimensional structure of the hub, including radial and axial dimensions); at the same time, the defect distribution heat map corresponding to the hub of the same type is extracted from the historical detection database (storing a large number of defect detection records of the same type of hub) (the higher the heat value, the higher the defect occurrence rate in this area). Based on this, the dynamic three-dimensional contour model and the defect distribution heat map are spatially aligned (to ensure one-to-one correspondence of their area positions), and the area with a higher heat value (defect-prone area) is found by matching, and this area is directly set as the area to be detected for subsequent processing.

[0093] Further, the step of spatially matching the dynamic three-dimensional contour model and the defect distribution heat map to detect the corresponding defect-prone area in real time comprises:

[0094] extracting a plurality of feature points from the dynamic three-dimensional contour model, and marking the projection positions corresponding to each of the feature points in the defect distribution heat map in real time;

[0095] calculating the target heat value generated by the feature points at the projection positions in real time through the RANSAC algorithm, and judging whether the target heat value is greater than a preset heat threshold in real time;

[0096] If it is judged in real time that the target heat value is greater than the preset heat threshold, the area where the feature point corresponding to the current heat value exists is set as the defect-prone area.

[0097] It should be noted that from the dynamic three-dimensional profile model, several representative feature points (such as the connection points of the rim and the spoke, the curvature change points of the hub surface, and the key structures of the hub are ensured to be covered) are selected; at the same time, in the defect distribution heat map, the projection position corresponding to each feature point is found (through space coordinate mapping, the feature points are ensured to be consistent with the position of the heat map). Based on this, the RANSAC algorithm (random sample consensus algorithm, which can effectively exclude abnormal value interference) is used to calculate the target heat value (i.e. the quantitative value of the defect occurrence probability of the position) corresponding to each feature point at its projection position. Then, the target heat value is compared with the preset heat threshold (such as 0.7, which is set according to the historical defect rate). It should be noted that the RANSAC algorithm ensures the accuracy of the heat value calculation, and the threshold judgment clearly defines the "high risk" standard, avoiding misjudgment of low defect rate areas as high incidence areas. Therefore, if the target heat value of a feature point is greater than the preset heat threshold, it indicates that the area where the feature point is located has a high defect occurrence rate, and the area (with the feature point as the center, and the preset range is expanded to 5mm) is set as a high defect area. Through the above method, the required detection area can be accurately circled, and the detection resources are focused on the high-risk parts, so as to facilitate subsequent processing.

[0098] Further, the step of real-time collecting the ultrasonic signals generated by the preset laser ultrasonic detector during scanning and synchronously simulating the cross-sectional image corresponding to the hub comprises:

[0099] Real-time starting the dual-channel acquisition mode of the laser ultrasonic detector, wherein the first channel is used to collect ultrasonic reflection signals, and the second channel is used to collect ultrasonic transmission signals;

[0100] Constructing a corresponding polar coordinate system with the center of rotation of the hub as the origin, and real-time collecting an initial image corresponding to the hub;

[0101] According to the polar coordinate system, the initial image is divided into a plurality of image grids, and the signal amplitudes contained in the ultrasonic reflection signals and the ultrasonic transmission signals are detected in real time, and according to the signal amplitudes, corresponding pixel points are filled in a plurality of image grids, to generate the cross-sectional image.

[0102] It should be noted that the application starts the dual-channel acquisition mode of the laser ultrasonic detector, specifically, the first channel is used to acquire ultrasonic reflection signals (reflecting the reflection of defects in the hub, the larger the defect, the stronger the reflection signal); the second channel is used to acquire ultrasonic transmission signals (reflecting the energy change of ultrasonic waves after penetrating the hub, defects will cause the transmission signal to weaken). Based on this, the center of the hub is taken as the origin to establish a polar coordinate system (the polar radius corresponds to the radial distance of the hub, and the polar angle corresponds to the rotation angle); at the same time, the initial image of the hub (such as the front image of the hub when it is static, containing initial structure information) is collected. Finally, according to the polar coordinate system, the initial image is divided into several uniform image grids (such as 0.1mm×0.1mm grid, the smaller the grid, the higher the image accuracy); then the signal amplitude (amplitude size is related to defect characteristics) in the ultrasonic reflection signal and the transmission signal is extracted, and the pixel points are filled in the corresponding image grid according to the amplitude size (such as large amplitude corresponding to dark pixels, representing possible defects), and finally the cross-sectional image is generated. Through the above-mentioned mode, the abstract ultrasonic signal can be converted into an intuitive image, so as to facilitate subsequent processing.

[0103] Further, the step of filling corresponding pixel points in a plurality of image grids according to the signal amplitude to correspondingly generate the cross-sectional image comprises:

[0104] Performing time domain slicing processing on the signal amplitude to correspondingly generate a plurality of slice signals;

[0105] Real-time detection of the signal peak value corresponding to each slice signal, and allocation of the corresponding index value to each image grid;

[0106] Generating a corresponding target pixel point according to the signal peak value, and filling the target pixel point into the inside of the corresponding target image grid according to the size of the index value.

[0107] It should be noted that the amplitude of the collected ultrasonic signal is subjected to time domain slicing processing (such as slicing at 100 ns intervals), and the continuous time domain signal is divided into a plurality of discrete slice signals (each slice corresponds to the signal amplitude at a time point, and is associated with the information of a certain depth of the hub). By time domain slicing, the one-dimensional continuous signal is converted into a multi-dimensional discrete signal, which is convenient for corresponding to different depth grids of the cross-sectional image. Based on this, for each slice signal, the signal peak value (the peak value is the most significant signal feature of the defect, representing the maximum amplitude at the time point) is detected in real time; at the same time, a unique index value (such as the “polar radius-polar angle” combined index in the polar coordinate system, which ensures the uniqueness of the grid position) is assigned to each image grid. Finally, according to the size of the signal peak value, the corresponding target pixel point (such as the larger the peak value, the lower the pixel gray value, and the darker the color) is generated; and according to the index value of the image grid, the target pixel point is accurately filled into the corresponding target image grid, and finally a complete cross-sectional image is formed. Through the above method, the defect information in the ultrasonic signal can be accurately reflected, so as to facilitate subsequent processing.

[0108] Further, the step of inputting the ultrasonic signal and the cross-sectional image into a preset CNN model in real time to determine whether the hub has defects comprises:

[0109] The ultrasonic signal is intercepted at 200 ns intervals to form 1024 signal segments, and the peak value, valley value and zero-crossing point number of each signal segment are extracted to form a 3x1024 signal feature matrix, and each element in the signal feature matrix is subjected to normalization processing;

[0110] The cross-sectional image is scaled to 224x224 pixels, and is split into a 3x224x224 image feature matrix according to the RGB three channels, wherein each channel pixel value is divided by 255 to complete normalization;

[0111] The signal feature matrix and the image feature matrix are input into the inside of the preset CNN model to determine whether the hub has defects in real time.

[0112] It should be noted that the ultrasonic signals are intercepted at intervals of 200 ns to obtain 1024 signal segments (covering the complete ultrasonic propagation period); the peak value, the valley value, and the number of zero-crossing points (three key features that comprehensively reflect signal fluctuations) are extracted for each segment to form a signal feature matrix of 3x1024 (3 rows correspond to the peak value / valley value / zero-crossing point, and 1024 columns correspond to the 1024 segments); and finally, each element in the matrix is normalized (e.g., mapped to the interval [0, 1] to eliminate signal amplitude level differences). The cross-sectional image is scaled to 224x224 pixels (to adapt to the conventional input size of the CNN model); and is split into an image feature matrix of 3x224x224 according to the RGB three channels (3 channels correspond to RGB, and 224x224 corresponds to the image pixel size); and the pixel value of each channel is divided by 255 (to map the pixel value from [0, 255] to [0, 1]) to complete the normalization. Finally, the processed signal feature matrix and image feature matrix are input into the preset CNN model, and the model outputs the judgment result of whether the hub has defects in real time through internal convolution, pooling, full connection, and other operations. The above method can form a standardized feature matrix for subsequent processing.

[0113] Further, the step of inputting the signal feature matrix and the image feature matrix into the preset CNN model to judge in real time whether the hub has defects includes:

[0114] The signal feature matrix is input into a 1D convolution layer of the preset CNN model, which is provided with 16 convolution kernels with a size of 3, a step length of 1, and a ReLU activation function, and outputs a signal feature map of 16x512 after maximum pooling; and the image feature matrix is input into a 2D convolution layer of the preset CNN model, which is provided with 32 convolution kernels with a size of 3x3, a step length of 1, and a ReLU activation function, and outputs an image feature map of 32x112x112 after maximum pooling.

[0115] The signal feature map and the image feature map are fused into a corresponding target feature vector, which is input into a full connection layer of the preset CNN model, and then input into an output layer after a dropout layer, and the output layer adopts a sigmoid activation function to output a probability value of 0-1, and when the probability value is greater than or equal to 0.6, it is determined that the hub has defects, and when the probability value is less than 0.6, it is determined that there are no defects.

[0116] It should be noted that the signal feature matrix is input into a 1D convolution layer (adapted to one-dimensional time sequence signal): 16 convolution kernels with a size of 3 are set (local signal features are extracted), the step length is 1 (the features are not missed by point-by-point sliding), and a ReLU activation function is used (nonlinearity is introduced to improve the expression ability of the model); after maximum pooling (2*2 pooling kernel, key features are retained and dimensionality is reduced), a signal feature map with a size of 16*512 is output (16 is the number of channels, and 512 is the length of the features after dimensionality reduction).

[0117] The image feature matrix is input into a 2D convolution layer (adapted to a two-dimensional image): 32 convolution kernels with a size of 3*3 are set (local spatial features of the image are extracted), the step length is 1, and a ReLU activation function is used; after maximum pooling (2*2 pooling kernel), an image feature map with a size of 32*112*112 is output (32 is the number of channels, and 112*112 is the image size after dimensionality reduction). Based on this, the signal feature map (16*512) and the image feature map (32*112*112) are spliced into a unified target feature vector after a flatten operation (the two-dimensional image feature map is converted into a one-dimensional vector); the target feature vector is input into a fully connected layer (to realize global association of the features), and then input into an output layer after a dropout layer (some neurons are randomly discarded to prevent model overfitting); the output layer uses a sigmoid activation function (outputs a probability value of [0, 1] corresponding to the probability of the defect existing). Finally, in the actual application process, the probability threshold is set to 0.6, and when the probability value of the output layer is greater than or equal to 0.6, it is determined that the hub has a defect, and correspondingly, when the probability value of the output layer is less than 0.6, it is determined that the hub does not have a defect, so that whether each hub has a defect can be objectively and accurately detected, and the detection efficiency is improved.

[0118] Please refer to Figure 2 The third embodiment of the present application provides:

[0119] A hub detection system, wherein the system comprises:

[0120] A detection module for placing a hub on a predetermined detection platform by a predetermined mechanical arm, and rotating the hub at a constant speed by the predetermined detection platform;

[0121] A calibration module for calibrating a to-be-detected area adapted to the hub in real time during rotation of the hub, and performing full-scan on the to-be-detected area by a predetermined laser ultrasonic detector;

[0122] The acquisition module is configured to acquire ultrasonic signals generated by the preset laser ultrasonic detector in real time during scanning, and to simulate a cross-sectional image corresponding to the wheel hub in real time.

[0123] The judgment module is configured to input the ultrasonic signals and the cross-sectional image into a preset CNN model in real time, and to determine whether the wheel hub has defects by using the preset CNN model.

[0124] Further, the calibration module is specifically configured to:

[0125] The acquisition module is configured to acquire continuous frame images generated by the wheel hub in real time during rotation by using a preset high-speed industrial camera, and to extract key structural features corresponding to the wheel hub in the continuous frame images in real time by using a preset edge detection algorithm.

[0126] The judgment module is configured to input the ultrasonic signals and the cross-sectional image into a preset CNN model in real time, and to determine whether the wheel hub has defects by using the preset CNN model.

[0127] The judgment module is configured to input the ultrasonic signals and the cross-sectional image into a preset CNN model in real time, and to determine whether the wheel hub has defects by using the preset CNN model.

[0128] Further, the calibration module is specifically configured to:

[0129] The judgment module is configured to input the ultrasonic signals and the cross-sectional image into a preset CNN model in real time, and to determine whether the wheel hub has defects by using the preset CNN model.

[0130] The judgment module is configured to input the ultrasonic signals and the cross-sectional image into a preset CNN model in real time, and to determine whether the wheel hub has defects by using the preset CNN model.

[0131] If it is determined in real time that the target thermal value is greater than the preset thermal threshold, a region in which the feature point corresponding to the current thermal value exists is set as the defect high-occurrence region.

[0132] Further, the acquisition module is specifically configured to:

[0133] The judgment module is configured to input the ultrasonic signals and the cross-sectional image into a preset CNN model in real time, and to determine whether the wheel hub has defects by using the preset CNN model.

[0134] The judgment module is configured to input the ultrasonic signals and the cross-sectional image into a preset CNN model in real time, and to determine whether the wheel hub has defects by using the preset CNN model.

[0135] According to the polar coordinate system, the initial image corresponding is divided into a plurality of image grids, and the signal amplitudes corresponding contained in the ultrasonic reflection signal and the ultrasonic projection signal are detected in real time, and the corresponding pixel points are filled in the plurality of image grids according to the signal amplitudes, so as to correspondingly generate the cross-sectional image.

[0136] Further, the acquisition module is specifically used for:

[0137] The signal amplitudes are subjected to time domain slicing processing, so as to correspondingly generate a plurality of slice signals;

[0138] The signal peak values corresponding to each slice signal are detected in real time, and the corresponding index values are allocated to each image grid;

[0139] According to the signal peak values, the corresponding target pixel points are generated, and the target pixel points are filled into the inside of the corresponding target image grid according to the size of the index values.

[0140] Further, the judgment module is specifically used for:

[0141] The ultrasonic signal is intercepted at intervals of 200 ns to form 1024 signal segments, the peak value, the valley value and the number of zero-crossing points of each signal segment are extracted, a signal feature matrix of 3*1024 is formed, and each element in the signal feature matrix is subjected to normalization processing;

[0142] The cross-sectional image is scaled to 224*224 pixels, and is split into an image feature matrix of 3*224*224 according to RGB three channels, wherein each channel pixel value is divided by 255 to complete normalization;

[0143] The signal feature matrix and the image feature matrix are input into the inside of the preset CNN model, so as to judge in real time whether the hub has defects.

[0144] Further, the judgment module is specifically used for:

[0145] The signal feature matrix is input into a 1D convolution layer of the preset CNN model, the layer is provided with 16 convolution kernels with a size of 3, the step is set to 1, adopts a ReLU activation function, and outputs a signal feature map of 16*512 after maximum pooling; the image feature matrix is input into a 2D convolution layer of the preset CNN model, the layer is provided with 32 convolution kernels with a size of 3*3, the step is set to 1, adopts a ReLU activation function, and outputs an image feature map of 32*112*112 after maximum pooling;

[0146] The signal feature map and the image feature map are fused into a corresponding target feature vector, and are input to a full connection layer of the preset CNN model, and are input to an output layer after a dropout layer, the output layer adopts a sigmoid activation function, and outputs a probability value of 0-1, when the probability value is greater than or equal to 0.6, it is determined that the hub has a defect, and when the probability value is less than 0.6, it is determined that there is no defect.

[0147] The fourth embodiment of the present application provides a computer comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the hub detection method as described above when executing the computer program.

[0148] The fifth embodiment of the present application provides a readable storage medium having a computer program stored thereon, wherein the program implements the hub detection method as described above when executed by a processor.

[0149] In summary, the hub detection method and system provided by the above embodiments of the present application can objectively and accurately detect whether the hub has a defect, so as to save the manual detection process, shorten the detection time and improve the detection efficiency.

[0150] It should be noted that the above-mentioned modules can be functional modules or program modules, which can be implemented by software or hardware. For the modules implemented by hardware, the above-mentioned modules can be located in the same processor, or the above-mentioned modules can be located in different processors in any combination.

[0151] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered a list of executable instructions for implementing logic functions, and can be specifically embodied in any computer-readable medium for use by an instruction execution system, apparatus or device, such as a computer-based system, a system including a processor or other system that can fetch instructions from the instruction execution system, apparatus or device and execute the instructions, or in conjunction with these instruction execution systems, apparatus or devices. For this specification, the "computer-readable medium" can be any device that can contain, store, communicate, propagate or transport programs for use by an instruction execution system, apparatus or device, or in conjunction with these instruction execution systems, apparatus or devices.

[0152] More specific examples (a non-exhaustive list) of the computer-readable medium include the following: an electrical connection (electronic) having one or more wires, a portable computer diskette (magnetic), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, and a portable compact disc read-only memory (CDROM). Additionally, the computer-readable medium can also be paper or another suitable medium upon which the program is printed, as the program can be electronically captured, for example, via an optical scanner, then compiled, interpreted, or otherwise processed, and stored in a computer memory in a form that is then employable by a computer. Examples of computer-readable media that are further within the spirit of the present application are a computer program product, a computer readable storage medium, and a computer.

[0153] It should be understood that aspects of the application can be implemented in hardware, software, firmware, or combinations thereof. In the embodiments described above, various steps or methods can be implemented, for example, by software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, and in another embodiment, any of the following techniques can be used to implement the hardware: discrete logic circuits having logic gates for implementing logic functions upon data signals, application specific integrated circuits having logic gates for implementing logic functions upon data signals, programmable gate arrays (PGA), field programmable gate arrays (FPGA), and the like.

[0154] In the description of the specification, the description of the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" and the like means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. In the specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0155] The above-described embodiments only express several implementation manners of the present application, which are described in a more specific and detailed manner, but cannot be understood as a limitation on the scope of the present application. It should be noted that, for those skilled in the art, several modifications and improvements can be made without departing from the concept of the present application, which are all within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the appended claims.

Claims

1. A wheel hub inspection method, characterized in that, The method includes: The wheel hub is placed on a preset detection platform by a preset robotic arm, and the wheel hub is rotated at a constant speed by the preset detection platform. During the rotation of the wheel hub, the detection area that is compatible with the wheel hub is calibrated in real time, and the detection area is fully scanned by a preset laser ultrasonic detector. The ultrasonic signals generated by the preset laser ultrasonic detector during the scanning process are collected in real time, and a cross-sectional image corresponding to the wheel hub is simulated simultaneously. The ultrasonic signal and the cross-sectional image are synchronously input into a preset CNN model, and the preset CNN model is used to determine in real time whether the wheel hub has defects. The step of calibrating the detection area that matches the wheel hub in real time during the wheel hub's rotation includes: The system uses a preset high-speed industrial camera to capture continuous frame images of the wheel hub during its rotation in real time, and uses a preset edge detection algorithm to extract key structural features corresponding to the wheel hub in the continuous frame images in real time. A dynamic three-dimensional contour model corresponding to the wheel hub is constructed in real time, and a defect distribution heat map corresponding to the wheel hub is extracted in real time from the historical inspection database. The dynamic three-dimensional contour model is spatially matched with the defect distribution heat map to detect the corresponding high-incidence areas of defects in real time, and the high-incidence areas of defects are set as the areas to be detected. The step of acquiring the ultrasonic signals generated by the preset laser ultrasonic detector during the scanning process in real time and simultaneously simulating a cross-sectional image corresponding to the wheel hub includes: The dual-channel acquisition mode of the laser ultrasonic detector is activated in real time, wherein the first channel is used to acquire ultrasonic reflection signals and the second channel is used to acquire ultrasonic transmission signals. A polar coordinate system is constructed with the rotation center of the wheel hub as the origin, and an initial image corresponding to the wheel hub is acquired in real time. The initial image is divided into several image grids according to the polar coordinate system, and the signal amplitudes contained in the ultrasonic reflected signal and the ultrasonic transmitted signal are detected in real time. The corresponding pixels are filled in the several image grids according to the signal amplitudes to generate the cross-sectional image.

2. The wheel hub inspection method according to claim 1, characterized in that: The step of spatially matching the dynamic three-dimensional contour model with the defect distribution heatmap to detect the corresponding high-incidence areas of defects in real time includes: Several feature points are extracted from the dynamic three-dimensional contour model, and the projection positions corresponding to each feature point are marked in real time in the defect distribution heat map. The RANSAC algorithm is used to calculate the target thermal value generated by the feature point at the projection position in real time, and to determine in real time whether the target thermal value is greater than the preset thermal threshold. If it is determined in real time that the target thermal value is greater than the preset thermal threshold, then the area where the feature point corresponding to the current thermal value exists is set as the high-incidence area of ​​the defect.

3. The wheel hub inspection method according to claim 1, characterized in that: The step of filling corresponding pixels in a plurality of image grids according to the signal amplitude to generate the cross-sectional image includes: The signal amplitude is subjected to time-domain slicing to generate several slice signals. The signal peak value corresponding to each slice signal is detected in real time, and a corresponding index value is assigned to each image grid. The corresponding target pixel is generated based on the signal peak value, and the target pixel is filled into the corresponding target image grid according to the index value.

4. The wheel hub inspection method according to claim 1, characterized in that: The step of synchronously inputting the ultrasonic signal and the cross-sectional image into a preset CNN model, and using the preset CNN model to determine in real time whether the wheel hub has defects, includes: The ultrasonic signal is segmented into 1024 signal segments at 200ns intervals, and the peak value, valley value and zero crossing point of each signal segment are extracted to form a 3×1024 signal feature matrix. Each element in the signal feature matrix is ​​normalized. The cross-sectional image is scaled to 224×224 pixels and split into a 3×224×224 image feature matrix according to the RGB three channels. The pixel value of each channel is normalized by dividing by 255. The signal feature matrix and the image feature matrix are input into the preset CNN model to determine in real time whether the wheel hub has defects.

5. The wheel hub inspection method according to claim 4, characterized in that: The step of inputting the signal feature matrix and the image feature matrix into the preset CNN model to determine in real time whether the wheel hub has defects includes: The signal feature matrix is ​​input into the 1D convolutional layer of the preset CNN model. This layer has 16 convolutional kernels of size 3, a stride of 1, and uses the ReLU activation function. After max pooling, it outputs a 16×512 signal feature map. The image feature matrix is ​​input into the 2D convolutional layer of the preset CNN model. This layer has 32 convolutional kernels of size 3×3, a stride of 1, and uses the ReLU activation function. After max pooling, it outputs a 32×112×112 image feature map. The signal feature map and the image feature map are fused into a corresponding target feature vector, which is then input into the fully connected layer of the preset CNN model, and then into the output layer after the dropout layer. The output layer uses the sigmoid activation function to output a probability value of 0-1. When the probability value is greater than or equal to 0.6, the wheel hub is determined to have a defect; when the probability value is less than 0.6, it is determined to have no defect.

6. A wheel hub inspection system, characterized in that, For implementing the wheel hub detection method as described in any one of claims 1 to 5, the system comprises: The detection module is used to place the wheel hub on a preset detection platform through a preset robotic arm, and drive the wheel hub to rotate at a constant speed through the preset detection platform; The calibration module is used to calibrate the detection area that is compatible with the wheel hub in real time during the rotation of the wheel hub, and to perform a full scan of the detection area by a preset laser ultrasonic detector. The acquisition module is used to acquire the ultrasonic signals generated by the preset laser ultrasonic detector during the scanning process in real time, and simultaneously simulate the cross-sectional image corresponding to the wheel hub. The judgment module is used to synchronously input the ultrasonic signal and the cross-sectional image into a preset CNN model, and use the preset CNN model to judge in real time whether the wheel hub has defects.

7. A computer comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the wheel hub detection method as described in any one of claims 1 to 5.

8. A readable storage medium having a computer program stored thereon, characterized in that, When executed by the processor, the program implements the wheel hub detection method as described in any one of claims 1 to 5.

Images