Automobile steering wheel appearance defect detection device based on machine vision

Abstract

This invention belongs to the field of visual inspection technology, specifically a machine vision-based automotive steering wheel appearance defect detection device. It includes a steering wheel transport device and an inspection device, further comprising: a rolling loading mechanism configured to press against the surface of the steering wheel to be inspected with a preset constant pressure, and drive a ball to roll uniformly along the steering wheel surface to generate a continuous broadband mechanical excitation signal in the contact area between the ball and the steering wheel surface; broadband mechanical disturbance is applied by the rolling of the ball, and the dynamic physical response optical signal of the surface is continuously acquired by focusing on the recovery monitoring area after the ball leaves; this device can sensitively capture weak surface vibration and deformation responses caused by internal defects such as delamination and debonding, extending the detection capability from surface morphology to subcutaneous adhesive structures, and realizing automated online full inspection of hidden defects.

Description

Technical Field

[0001] This invention belongs to the field of visual inspection technology, specifically a machine vision-based device for detecting defects in the appearance of automotive steering wheels. Background Technology

[0002] As a core component directly controlled by the driver, the appearance quality and structural integrity of a car steering wheel directly affect the overall vehicle quality and driving safety. Currently, mid-to-high-end car steering wheels generally use leather material wrapped around an internal foam layer or frame. During the production process, due to factors such as fluctuations in adhesive processes and poor material matching, internal defects such as delamination, separation, and air bubbles can easily occur between the leather covering layer and the substrate. These defects are hidden beneath the leather surface and cannot be directly observed with the naked eye, but they gradually expand during use, eventually leading to bulging, cracking, or even peeling of the leather, posing a serious quality hazard.

[0003] For the detection of the aforementioned internal defects, existing methods mostly rely on traditional manual touch and tapping methods. This involves inspectors pressing the steering wheel surface with their fingers or tapping it with a small hammer, judging the presence of hollow areas or delamination based on feel or sound differences. This method depends on the inspector's subjective experience, making it difficult to quantify and standardize judgment criteria. Furthermore, the detection speed is slow and cannot meet the online full-inspection cycle requirements of automated production lines. Conventional 2D vision inspection methods use industrial cameras to image the steering wheel surface, identifying surface defects such as scratches, pinholes, and color differences through image analysis. However, its detection principle dictates that it can only capture surface texture and color information, and has almost no ability to detect internal defects that have not yet developed visible morphological changes such as surface wrinkles or bulges.

[0004] Therefore, the present invention provides a machine vision-based device for detecting defects in the appearance of automobile steering wheels. Summary of the Invention

[0005] In order to overcome the shortcomings of the prior art, at least one technical problem raised in the background art is solved.

[0006] The technical solution adopted by this invention to solve its technical problem is: the machine vision-based automotive steering wheel appearance defect detection device of this invention includes a steering wheel conveying device and a detection device, and further includes:

[0007] The rolling loading mechanism is configured to press against the surface of the steering wheel to be tested with a preset constant pressure and drive the ball to roll at a constant speed along the surface of the steering wheel to generate a continuous broadband mechanical excitation signal in the contact area between the ball and the surface of the steering wheel.

[0008] A vision acquisition module, whose imaging area covers at least a portion of the rolling path formed by the rolling loading mechanism on the steering wheel surface, is configured to continuously acquire the dynamic physical response optical signal of the steering wheel surface caused by the mechanical excitation signal within a preset recovery monitoring area after the ball leaves.

[0009] An image processing module is electrically connected to the vision acquisition module. The image processing module is configured to extract at least one local physical property anomaly feature caused by internal defects from the dynamic physical response optical signal, and to determine whether the steering wheel has internal defects based on the local physical property anomaly feature.

[0010] The beneficial effects of this invention are as follows:

[0011] 1. The machine vision-based automotive steering wheel appearance defect detection device of the present invention utilizes the rolling of a ball to apply a wide-band mechanical disturbance and focuses on the recovery monitoring area after the ball leaves to continuously collect the dynamic physical response optical signal of the surface. It can keenly capture the weak surface vibration and deformation response caused by defects such as internal delamination and delamination, and extend the detection capability from surface morphology to subcutaneous adhesive structure, realizing automated online full inspection of hidden defects.

[0012] 2. The machine vision-based automotive steering wheel appearance defect detection device of this invention utilizes an image processing module to perform time-frequency transformation on the temporal brightness change sequence of each pixel within the recovery monitoring area and extract vibration attenuation features, thereby deeply exploring the intrinsic dynamic physical properties of the material. The device not only analyzes the intrinsic response frequency shift caused by defects but also constructs an attenuation envelope through Hilbert transform, quantifying the dissipation time required for vibration energy to decay from its peak to a specified threshold. By comparing the measured frequency and dissipation time with a spatially resolved benchmark mapping, a two-dimensional physical layer diagnosis of the material is performed, significantly improving the certainty and anti-interference capability for identifying minute, latent defects.

[0013] 3. The machine vision-based automotive steering wheel appearance defect detection device of the present invention generates the final judgment result by fusing the confidence maps output by two channels. This overcomes the limitation of a single feature channel having a blind spot for specific defect types, enabling the device to capture various internal defects with different morphological appearances and physical causes in all directions, effectively balancing high detection sensitivity and low false alarm rate. Attached Figure Description

[0014] The invention will now be further described with reference to the accompanying drawings.

[0015] Figure 1 This is a perspective view of the present invention;

[0016] Figure 2 In this invention Figure 1Enlarged view of point A in the image;

[0017] Figure 3 This is a system flowchart of the present invention.

[0018] In the diagram: 1. Conveying equipment; 2. Testing equipment; 3. Positioning bracket; 4. Flexible connecting frame; 5. Sphere. Detailed Implementation

[0019] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0020] The machine vision-based automotive steering wheel appearance defect detection device described in this embodiment of the invention includes a steering wheel conveying device 1 and a detection device 2, and further includes:

[0021] The rolling loading mechanism is configured to press against the surface of the steering wheel to be tested with a preset constant pressure and drive the ball 5 to roll at a constant speed along the surface of the steering wheel so as to generate a continuous broadband mechanical excitation signal in the contact area between the ball 5 and the surface of the steering wheel.

[0022] Broadband mechanical excitation signals refer to mechanical vibration or stress wave signals that encompass a wide frequency range. These signals can effectively excite the response characteristics of materials at different frequencies, helping to reveal internal defects of different types and depths.

[0023] The conveying device 1 is equipped with a positioning bracket 3, and a swingable elastic connecting frame 4 is hinged to the positioning bracket 3. A ball 5 is rotatably mounted on the front end of the elastic connecting frame 4. The elastic connecting frame 4 is configured to apply a preset constant pressure to the surface of the steering wheel to be tested, so as to drive the ball 5 to always abut against the steering wheel surface with a constant elastic force and roll along its contour, thereby generating a continuous broadband mechanical excitation signal in the contact area between the ball 5 and the steering wheel surface. This excitation method can uniformly excite the dynamic response of the internal structure of the steering wheel.

[0024] The vision acquisition module, whose imaging area covers at least a portion of the rolling path formed by the rolling loading mechanism on the steering wheel surface, is configured to continuously acquire the dynamic physical response optical signal of the steering wheel surface caused by the mechanical excitation signal within a preset recovery monitoring area after the ball 5 leaves.

[0025] The vision acquisition module is used to capture optical information from the steering wheel surface. Its imaging area covers at least a portion of the rolling path formed by the rolling loading mechanism on the steering wheel surface. Within a preset recovery monitoring area after the ball 5 leaves, the module continuously acquires the dynamic physical response optical signals of the steering wheel surface caused by the aforementioned mechanical excitation signal. The imaging area of ​​the vision acquisition module covers at least a portion of the rolling path formed by the rolling loading mechanism on the steering wheel surface. This module can be a fixedly mounted industrial camera with a sufficiently large field of view to cover the entire scanning path of the rolling loading mechanism in a single operation.

[0026] Within a preset recovery monitoring zone after the sphere 5 leaves, the visual acquisition module continuously acquires the dynamic physical response optical signals of the steering wheel surface caused by the aforementioned mechanical excitation signal. High-speed image acquisition is immediately initiated and continues for a certain period. The duration of this recovery monitoring zone can be preset based on empirical values, for example, set to 0.1 to 1 second after the sphere 5 leaves. Continuous acquisition can be achieved by setting a high frame rate for the camera, for example, capturing images at a speed of hundreds of frames per second to capture minute dynamic changes on the surface. The optical signals can be minute deformations, vibrations, or changes in the intensity of reflected light on the steering wheel surface under excitation. For example, a high-resolution camera can capture minute blurring or displacement of the surface texture.

[0027] The recovery monitoring zone refers to the specific time period or area during which the surface of the sphere 5 recovers from its deformed state to its initial or stable state after rolling past a certain point. During this period, abnormal physical responses caused by internal defects will be more pronounced.

[0028] The image processing module is electrically connected to the vision acquisition module. The image processing module is configured to extract at least one local physical property abnormality feature caused by internal defects from the dynamic physical response optical signal, and to determine whether there are internal defects in the steering wheel based on the local physical property abnormality feature.

[0029] The image processing module is electrically connected to the vision acquisition module and is used to receive and analyze the aforementioned dynamic physical response optical signals. The core function of this module is to extract the abnormal features of local physical properties caused by internal defects from these optical signals.

[0030] Local physical property anomalies refer to physical characteristics of the steering wheel surface that differ from normal areas in specific regions, such as vibration frequency, damping characteristics, or local stiffness. These anomalies serve as the basis for determining the presence or absence of internal defects.

[0031] The image processing module is configured to extract at least one anomalous feature of local physical properties caused by internal defects from the aforementioned dynamic physical response optical signal. One extraction method is to analyze the brightness changes of specific regions in the image sequence. For example, after sphere 5 rolls past and leaves, the brightness changes in normal areas will quickly stabilize, while areas with internal defects may exhibit continuous brightness fluctuations or abnormal brightness decay patterns. Another extraction method is to analyze the displacement or deformation of pixels in the image sequence. For example, by calculating the minute displacement fields of each point on the surface using optical flow or digital image correlation (DIC) techniques, defective regions may exhibit abnormal displacement patterns or larger deformations. Texture changes in specific regions of the image sequence can also be analyzed; for example, defective regions may exhibit texture blurring or distortion under excitation.

[0032] Dynamic physical response optical signals refer to the visual signals generated by the deformation, vibration, or changes in optical properties of the steering wheel surface after being subjected to mechanical excitation. These signals can reflect the mechanical behavior of the internal structure of the steering wheel.

[0033] A controllable broadband mechanical excitation is applied to the surface of the steering wheel using a rolling loading mechanism. A vision acquisition module continuously captures dynamic physical response optical signals within the recovery monitoring area. An image processing module then extracts and analyzes local physical property anomalies, thereby enabling the detection of hidden defects inside the automotive steering wheel. This method effectively overcomes the subjectivity and inefficiency of traditional manual inspection, as well as the limitations of conventional 2D vision inspection in identifying internal defects. It significantly improves the accuracy, efficiency, and repeatability of the inspection, meeting the online full-inspection requirements of automated production lines and ensuring the overall quality of automotive steering wheels and driving safety.

[0034] The image processing module is configured to perform a differential operation on the current frame image acquired within the recovery monitoring area and the baseline image of the same position before the sphere 5 was rolled, to generate a differential image. When a local bright spot with a brightness value exceeding a preset threshold appears in the differential image, the area corresponding to the bright spot is determined to be a candidate defect area.

[0035] The image processing module performs differential operations, which highlight the differences between the images before and after rolling, amplify the changes in surface brightness caused by mechanical excitation, thereby enhancing the visibility of abnormal signals and effectively eliminating background interference.

[0036] Differential analysis is an image processing technique that generates a new image by subtracting the brightness or grayscale values ​​of two images pixel by pixel. If the brightness values ​​of the two images are the same at a certain pixel location, the brightness value of the differential image at that location is zero (or close to zero). If the brightness values ​​are different, the differential image at that location will display a non-zero brightness value, the magnitude of which reflects the degree of difference. Image processing modules can use hardware acceleration units, such as Field-Programmable Gate Arrays (FPGAs) or Application-Specific Integrated Circuits (ASICs), or software algorithms, such as functions in image processing libraries (like OpenCV), to implement pixel-by-pixel differencing. For example, for grayscale images, the pixel values ​​of the differential image can be calculated as the absolute difference between the brightness values ​​of corresponding pixels in the current frame image and the baseline image. For color images, differential operations can be performed on the red, green, and blue (RGB) channels separately, or the color image can be converted to grayscale first before differential analysis.

[0037] In addition, the baseline image of the same position before the sphere 5 is rolled can be obtained by performing a complete scan of the steering wheel surface before the start of the inspection process and recording the image of each position in the unstressed state as the baseline; or by performing a pre-scan of the area to be rolled before each rolling operation to obtain the baseline image of that area.

[0038] Based on this, when a local bright spot with a brightness value exceeding a preset threshold appears in the differential image, the corresponding region of the bright spot is determined to be a candidate defect region. A local bright spot refers to a region in the differential image where the pixel brightness value is significantly higher than the brightness of its surrounding area or the overall background, and its brightness value exceeds a preset threshold. This threshold is usually determined based on experimental data or statistical analysis and is used to distinguish between normal fluctuations and abnormal responses caused by defects. The image processing module can employ various image segmentation or feature extraction algorithms to detect local bright spots. For example, the differential image can be binarized, marking pixels with brightness values ​​higher than the preset threshold as foreground (bright spots) and those lower as background. Subsequently, connected component analysis can be used to identify and locate these bright spot regions. The preset threshold can be set by statistical analysis of a large number of differential images of defect-free steering wheels, establishing an empirical value that can effectively distinguish between normal fluctuations and defect signals, or it can be adaptively adjusted using machine learning methods. The preset threshold can be a fixed value or an adaptive threshold that is dynamically adjusted based on the local characteristics of the image.

[0039] In addition, to improve robustness, morphological operations (such as opening and closing operations) can be combined to remove noise or fill small holes, ensuring that the detected bright spots have certain size and shape characteristics.

[0040] By performing a differential operation between the current frame image acquired within the recovery monitoring area and the baseline image of the same position before rolling of sphere 5, the inherent background noise such as texture and color unevenness on the steering wheel surface can be effectively eliminated, highlighting only the dynamic physical response of the surface caused by the mechanical excitation of sphere 5. Due to their abnormal mechanical properties, the surface deformation or vibration characteristics of internal defect areas will differ significantly from those of intact areas after being excited. These differences manifest as significant local changes in brightness values ​​in the differential image, forming local bright spots. By setting a preset threshold to detect these local bright spots with brightness values ​​exceeding the threshold, the precise identification and location of minute surface changes caused by internal defects can be achieved, avoiding misjudgments caused by background interference or subtle changes that are difficult to detect.

[0041] This method, based on differential imaging and thresholded bright spot detection, significantly improves the accuracy and reliability of detection, thereby enabling more effective extraction of local physical property anomalies caused by internal defects from dynamic physical response optical signals, and thus accurately determining whether the steering wheel has internal defects.

[0042] The local physical property anomaly is characterized by vibration spectrum anomalies; the image processing module reconstructs the temporal brightness change sequence of each pixel location within the monitoring area. Perform time-frequency transformation to obtain the time spectrum. The intrinsic response frequency of each pixel is determined by the following formula. :

[0043]

[0044] In the formula, For pixels At any moment ,frequency The time-frequency amplitude at that location This is the analysis time window after sphere 5 leaves the pixel region. The frequency sweep analysis range is preset based on the mechanical characteristics of the intact area of ​​the steering wheel.

[0045] For a typical steering wheel structure with a leather-wrapped foam layer. It can be set to 50Hz. It can be set to 2000Hz, which covers the typical frequency shift range caused by degumming / delamination defects.

[0046] Analysis time window This is a preset time period after sphere 5 leaves the pixel area, for example, It can be taken 0.05 seconds after the sphere leaves the pixel area. The main phase of material free vibration decay can be captured 0.3s after the sphere leaves the pixel area.

[0047] The time-frequency transformation can be performed using a short-time Fourier transform (STFT), with the Hanning window selected as the window function. The window length is determined based on the rolling speed of the sphere 5 and the frame rate of the visual acquisition module to achieve sufficient time resolution for transient vibration response.

[0048] Among these, abnormal vibration spectrum characteristics refer to the different frequency domain characteristics of the vibration response of the steering wheel surface under test after being subjected to mechanical excitation compared to the normal area. This anomaly typically manifests as an enhancement, weakening, shift, or the appearance of new frequency components in specific frequencies. For example, when defects such as delamination or peeling exist inside the steering wheel, the local stiffness of the defective area changes, causing a change in its natural vibration frequency after excitation, thus exhibiting significantly different characteristics in the vibration spectrum compared to the intact area. By analyzing these spectral differences, the presence of internal defects can be indirectly inferred.

[0049] Time-based brightness change sequence This refers to the sequence of brightness values ​​of a specific pixel on the steering wheel surface, continuously acquired by the vision acquisition module within the recovery monitoring area, over time. Because micro-vibrations in the normal direction of the surface cause periodic changes in the angle or path of local reflected light, therefore... The fluctuation frequency is consistent with the surface vibration frequency, and can be directly used for subsequent spectrum analysis.

[0050] When the surface of a steering wheel experiences a dynamic physical response due to internal defects, this response causes minute deformations or vibrations, resulting in subtle variations in the brightness of that pixel at different times. Time-frequency transformation is a signal processing technique that can transform time-domain signals (such as...) Converted into a time spectrum that simultaneously contains time and frequency information. .

[0051] Time-frequency transformation (TFD) allows analysis of the frequency components of a signal across different time periods and their changes over time. Common TFD methods include Short-Time Fourier Transform (STFT) and Wavelet Transform. For example, STFT can be used to analyze the frequency components of a signal across different time periods and their changes over time. After the signal is windowed, a Fourier transform is performed, and the frequency distribution of the signal at different times is obtained as the time window slides.

[0052] Intrinsic response frequency This refers to the dominant frequency, or the frequency with the most significant or concentrated energy, in the vibration response of each pixel on the steering wheel surface after being subjected to mechanical excitation within a specific analysis time window. This frequency is usually closely related to the inherent vibration characteristics of the material and the stiffness of the local structure. It is calculated by measuring the time-frequency spectrum. Within the preset frequency sweep analysis range Within the analysis time window The frequency with the largest average amplitude on the pixel can quantify the intrinsic vibration characteristics of that pixel region. This calculation method can effectively extract the frequency components that contribute the most to the vibration response of the pixel within a specific time period, thus providing a key quantitative indicator for subsequent defect judgment.

[0053] This indicates that the image processing module is used for a specific pixel. After time-frequency transformation, at a specific moment and a specific frequency The vibration energy or intensity corresponding to the location. This value reflects the vibrational activity of that pixel at a specific time and frequency. For example, if a short-time Fourier transform is used, This can be represented as at time... Nearby, frequency The square of the Fourier transform amplitude at a given point is the power spectral density.

[0054] Frequency sweep analysis range This is a pre-determined frequency range based on experimental data or theoretical analysis of a large number of intact steering wheel areas. This range includes the main vibration frequencies that normal steering wheel materials may produce after being excited. By limiting the analysis range to this pre-determined range, noise frequencies or high-frequency harmonics unrelated to defects can be effectively eliminated, thereby improving the sensitivity and accuracy of defect detection. For example, by conducting experiments on multiple known intact steering wheels, measuring their vibration spectra under the same excitation, and statistically analyzing the distribution range of their main vibration frequencies, the frequency range can be determined. .

[0055] In this application, a rolling loading mechanism drives a ball 5 to roll uniformly along the steering wheel surface under a preset constant pressure. After the ball 5 leaves the surface, the visual acquisition module continuously acquires the dynamic physical response optical signals of the steering wheel surface caused by the mechanical excitation signal within a preset recovery monitoring area. This application defines local physical property anomalies as vibration spectrum anomalies, allowing detection to focus on unique vibration modes caused by internal defects. The image processing module performs time-frequency transformation on the temporal brightness change sequence of each pixel position within the recovery monitoring area, converting the dynamic optical signal from the time domain to the time-frequency domain. This enables simultaneous analysis of the frequency and time dimensions, which is crucial for capturing transient vibration responses. By obtaining the time-frequency spectrum... This laid the foundation for subsequent quantitative analysis. Furthermore, the intrinsic response frequency of each pixel was determined using a formula. That is, calculation within the preset frequency sweep analysis range Within the analysis time window The frequency with the largest average amplitude quantifies the dominant vibration characteristics. Internal defects typically cause local stiffness changes, leading to a shift in the intrinsic response frequency. By analyzing the recovery monitoring zone after sphere 5 leaves, interference during the loading process is eliminated, focusing on the material's free vibration response, thus more accurately capturing the frequency changes caused by defects. This detection method based on the abnormal characteristics of the vibration spectrum can fundamentally solve the technical problem of missing small, latent defects. Even before defects produce visible surface deformation, they can be reliably captured through the frequency shift signals they cause, significantly improving the accuracy and sensitivity of detecting internal defects in automotive steering wheels.

[0056] The image processing module further processes each pixel... A reference frequency mapping map with a pre-established, intact bonding region corresponding to a pixel location. A comparison is performed, and the pixel region is determined to have an internal defect if the following conditions are met:

[0057]

[0058] In the formula, This is a spatially resolved reference frequency map established through offline calibration or online self-learning. The preset frequency offset threshold has a value of [value]. 15% to 30%.

[0059] A pre-established reference frequency map of the intact bonding region corresponding to the pixel position. It is a two-dimensional data structure that stores the position of each pixel on the surface of the steering wheel. This map represents the intrinsic response frequency that the area should possess when it is in a well-bonded state. It reflects the inherent frequency differences in different areas of the steering wheel caused by factors such as material thickness, structural support, and curvature. This reference frequency map... The establishment of this can be achieved in several ways. For example, an offline calibration method can be used, in which a batch of known intact and defect-free steering wheels are selected from the production line as standard samples. These standard samples are then scanned using a rolling loading mechanism and images are acquired by a vision acquisition module. The intrinsic response frequency of each pixel position is then calculated using the method described above. .

[0060] Subsequently, statistical averaging or modeling is performed on the data from multiple standard samples to generate a reference frequency mapping map with spatial resolution. Alternatively, an online self-learning approach can be adopted. During the actual testing process, the system continuously collects a large amount of data from the steering wheel being tested. By performing cluster analysis, outlier detection, or machine learning algorithms on this data, it automatically identifies the frequency response patterns of most normal areas and dynamically updates or optimizes them. For example, methods such as moving average or Kalman filtering can be used to gradually adjust the reference frequency mapping based on recently detected normal sample data.

[0061] The comparison operation refers to the image processing module comparing each pixel on the steering wheel to be detected. Calculated intrinsic response frequencies With respect to a pre-established reference frequency at the corresponding location. Numerical comparisons are performed. The image processing module accesses data stored in memory or a database. Mapping graph, based on the coordinates of the current pixel The corresponding reference frequency value is retrieved, and then subtraction or comparison operations are performed to evaluate it. Compared to The deviation.

[0062] Judgment conditions This is the core criterion for determining the presence of internal defects in a pixel region. This condition indicates that when the actual measured intrinsic response frequency... Significantly higher than the intact reference frequency at this location And it exceeds a preset frequency offset threshold. Only when this condition is met is the region considered to have internal defects. This is based on the physical principle that internal defects (such as delamination or peeling) typically lead to a decrease in local structural stiffness, thereby increasing the vibration frequency of that region. The image processing module performs this inequality check on each pixel, first calculating... Then Compare this result with the calculated value. If the inequality holds, then mark the pixel as a candidate defect point.

[0063] Preset frequency offset threshold This is a tolerance value used to filter noise and normal material variations. It ensures that only frequencies that are abnormal to a certain extent are considered defects, thus improving detection robustness and reducing false alarms. Its value range is... 15% to 30%, which provides a reasonable dynamic threshold related to the reference frequency. It can be determined in advance through experiments or statistical analysis and stored as a [database name]. Maps with the same spatial resolution, or through a functional relationship (e.g.) ,in (A constant between 0.15 and 0.30) is dynamically calculated during runtime. During offline calibration or online self-learning, the frequency fluctuation range of intact samples can be analyzed to determine the value at each location. .

[0064] By introducing a spatially resolved reference frequency map The system can set personalized benchmarks based on the inherent physical differences in different areas of the steering wheel surface, effectively solving the problem of lacking reliable benchmarks and significantly improving the accuracy of defect detection. Simultaneously, by setting a preset frequency offset threshold... This effectively filters out slight frequency variations caused by environmental noise, minute material fluctuations, or measurement errors, avoiding false alarms and thus enhancing the robustness of the detection. This method, based on... The judgment criteria combine physical principles and statistical considerations, making the identification of internal defects more reliable. Furthermore, the reference frequency mapping diagram... The system can be established through offline calibration or online self-learning, ensuring that it can adapt to production variations of different batches and models of steering wheels, thus improving the system's versatility and adaptability.

[0065] The image processing module is also configured to extract vibration attenuation features, specifically: from the time spectrum. Extracted from Time-domain amplitude attenuation envelope at frequency And calculate using the following formula:

[0066]

[0067] In the formula, for The real-time signal after the frequency components have been bandpass filtered. This represents the Hilbert transform, which shifts the phase of each frequency component of the original real signal. This generates corresponding orthogonal imaginary part signals, which together with the original real signals constitute an analytic signal, facilitating the extraction of the instantaneous amplitude envelope;

[0068] A zero-phase Butterworth bandpass filter can be used, with its center frequency set to the intrinsic response frequency. Bandwidth set to 20% to 50%. The filter applies to the temporal brightness variation sequence at the pixel location. The output is Real time-domain signal after bandpass filtering of frequency components Based on this, the Hilbert transform is used. Construct an analytical signal, and then extract its time-domain amplitude attenuation envelope according to the formula. .

[0069] Among these, extracting vibration attenuation characteristics refers to assessing the internal structural integrity of a material by analyzing the decrease in vibration amplitude over time after being excited. Internal defects, such as delamination or separation, can alter the material's damping properties, leading to abnormal vibration energy dissipation.

[0070] From the time spectrum Extracted from Time-domain amplitude attenuation envelope at frequency The aim is to focus the analysis on the steering wheel material at specific intrinsic frequencies. The response at this frequency is typically closely related to the material's natural vibration modes. Through methods such as bandpass filtering, these frequencies can be separated from the complex time-spectrum. Frequency components allow for more accurate analysis of local vibration behavior related to defects.

[0071] The Hilbert transform can convert a real signal into its analytic form, thus facilitating the extraction of the instantaneous amplitude envelope. This envelope curve intuitively reflects the decay process of vibration energy over time. Besides the Hilbert transform, the amplitude decay envelope can also be approximated by rectifying the real signal and combining it with low-pass filtering, or by using a peak detection algorithm to track the continuous peaks of the vibration waveform to construct the envelope.

[0072] The image processing module further determines the attenuation envelope. From its peak amplitude decay to Required dissipation time , Let be the base of the natural logarithm, and determine that the pixel region has an internal defect when the following conditions are met:

[0073]

[0074] The dissipation time of the current pixel region. For offline establishment of a reference decay time mapping map of the intact bonding region corresponding to the pixel position, This is the preset dissipation anomaly amplification factor, with a value range of 1.5 to 3.0.

[0075] This judgment criterion compares the actual measured dissipation time with a preset benchmark value and introduces an amplification factor. This is used to set the tolerance range, thereby enabling sensitive detection of defects. Reference decay time mapping diagram. It can be obtained by statistical analysis and averaging of a large number of intact steering wheel samples, or by establishing it through finite element simulation, so as to reflect the inherent material and structural differences in different areas of the steering wheel.

[0076] The above technical solution introduces vibration attenuation characteristics as a basis for judging internal defects, effectively overcoming the limitations of relying solely on vibration spectrum characteristics, especially for defect types that primarily affect material damping characteristics rather than stiffness, such as micro-delamination or loose internal material structure. The image processing module focuses on... Time-domain amplitude attenuation envelope at frequency This method can accurately capture the energy dissipation behavior of materials at key response frequencies. The Hilbert transform is used to extract the decay envelope, ensuring an accurate description of the vibration amplitude decay process and avoiding noise interference. Furthermore, the decay envelope is quantized from the peak amplitude... decay to Required dissipation time This is directly related to the damping characteristics of the material. When defects such as delamination or debonding exist inside the steering wheel, the acoustic impedance matching between the leather and the substrate is disrupted, and vibration energy cannot be effectively conducted to the substrate and dissipated, resulting in a significant slowdown in vibration decay, i.e., the dissipation time. It will increase significantly. Ultimately, by measuring... Mapping of baseline decay time that takes into account spatial differences Linear comparison, combined with preset dissipation anomaly amplification factor It can adaptively determine defect areas, thereby improving the comprehensiveness, accuracy and reliability of defect detection, reducing the risk of misjudgment, and ensuring the quality of car steering wheels and driving safety.

[0077] The local physical property anomaly is a local equivalent stiffness feature; the image processing module calculates the maximum downward displacement of each point on the steering wheel surface during the rolling process of the sphere 5. Residual displacement after release And calculate the local stiffness exponent using the following formula. :

[0078]

[0079] This calculation process directly quantifies the stiffness characteristics of the material during the elastic deformation stage, whereby... This represents the elastic deformation of the material. In this way, the physical and mechanical properties are transformed into quantifiable image processing results, providing an objective basis for subsequent defect determination.

[0080] Specifically, a high-resolution vision acquisition module continuously captures images of the steering wheel surface before, during, and after the sphere 5 is rolled. Digital image correlation (DIC) technology is used, which compares the grayscale features of sub-regions in the images before and after deformation to track the displacement of minute surface textures or preset marker points. For leather steering wheels with less surface texture, a random speckle pattern can be applied to the steering wheel surface before inspection. Alternatively, the natural texture features of the leather surface itself can be used as speckle. In this case, the vision acquisition module needs to be equipped with an industrial camera with a resolution of at least 5 megapixels to ensure that each analysis sub-region contains a sufficient number of texture feature points for matching calculations, thereby accurately measuring the downward displacement of each point on the surface when the force is maximum. And the residual displacement when the surface of sphere 5 returns to a stable state after it leaves. .

[0081] In the formula, A preset constant pressure is applied to sphere 5;

[0082] When the measuring point If the stiffness is lower than a preset ratio of the stiffness benchmark value obtained from the statistics of the surrounding normal area, it is determined that there is an internal defect at that point.

[0083] The image processing module processes images within a certain neighborhood of the current measured point. The values ​​are statistically analyzed, such as calculating the mean or median, and used as the stiffness benchmark for that region. If the measured point... If the value is below a preset percentage of the benchmark, the point is determined to have an internal defect. As an alternative implementation, a spatially resolved stiffness benchmark mapping can be established by scanning multiple known intact steering wheels before testing for comparison. During actual testing, the image processing module will... The value is compared to the reference value at the corresponding location in the mapping diagram. If it is lower than the reference value by a preset percentage, it is considered a defect. This method can accommodate normal stiffness differences that may exist in different areas of the steering wheel.

[0084] Local equivalent stiffness refers to a material's ability to resist deformation in a localized area. In the context of automotive steering wheel inspection, it quantifies the elasticity or rigidity exhibited by the internal structure (such as the adhesion between the leather layer and the foam layer or skeleton) of the steering wheel surface under external pressure. Internal defects, such as delamination, separation, or bubbles, can significantly reduce the stiffness of the local material, making that area more susceptible to deformation under stress. By capturing this stiffness change, structural defects hidden beneath the surface can be effectively identified.

[0085] This application directly quantifies the mechanical properties of materials by introducing local equivalent stiffness characteristics. The image processing module accurately calculates the maximum downward displacement of each point on the steering wheel surface during the rolling process of the sphere 5. Residual displacement after release These displacements reflect the deformation and recovery behavior of the material under stress, and are key data for assessing local stiffness. Based on this, a constant pressure is preset... With elastic deformation The ratio of these values ​​is used to objectively calculate the local stiffness index. .

[0086] When the measuring point When the stress level is significantly lower than the statistical baseline value of the surrounding normal area, an internal defect can be accurately identified. This method can effectively identify localized material softening caused by internal defects such as delamination, delamination, or bubbles, overcoming the shortcomings of traditional visual inspection and vibration analysis in detecting stiffness-related defects. Furthermore, this method determines defects by comparing the relative stiffness difference between the tested area and its surrounding normal area, decoupling defect identification from absolute pressure values. Even if slight fluctuations occur in the actual output of constant pressure due to the production line environment, anomalies can be robustly identified, thereby reducing reliance on precision force control hardware and maintenance costs, and improving the stability and reliability of the inspection.

[0087] The vision acquisition module uses a CMOS image sensor with a rolling shutter mode; when the rolling loading mechanism is scanning, the image processing module is configured to demodulate the surface vibration frequency by analyzing the oscillation period of bright and dark stripes along the rolling direction in a single frame image.

[0088] Its characteristic is that the sensor exposes and reads out row by row or column by column, rather than exposing the entire image simultaneously. This working mode allows different parts of the image to correspond to different exposure time points when capturing moving objects, thus recording the spatial-temporal information of motion in a single frame.

[0089] When the sphere 5 rolls at a constant speed on the steering wheel surface and a continuous mechanical excitation signal is applied, the steering wheel surface will produce minute vibrations. Because the vision acquisition module uses a rolling shutter mode and the scanning direction has a specific relationship with the rolling direction of the sphere 5, these minute vibrations will appear as periodic bright and dark stripes along the rolling direction in a single frame image. The image processing module can use frequency domain analysis methods based on Fourier transform (FFT) or wavelet transform to analyze the pixel brightness profile along the rolling direction in a single frame image, identify the main periodic brightness changes, and thus determine the stripe period.

[0090] The vibration frequency is calculated using the following formula. :

[0091]

[0092] This formula represents the most fundamental physical relationship between vibration frequency and period, and is used to calculate the period time obtained through image analysis. Converted to actual vibration frequency The image processing module obtains... Then, simply perform the reciprocal operation.

[0093] It should be noted that when using the aforementioned rolling shutter mode to measure vibration frequency, the line readout rate of the image sensor... The Nyquist sampling theorem must be satisfied, i.e. ,in Let be the vibration frequency of the surface to be measured. For the highest predicted frequency... The row read rate should be selected. At least greater than An image sensor is used to ensure that vibration signals are fully sampled.

[0094] In the formula, The measured fringe period time is calculated as follows: [Detection continuous...] Total number of pixel rows per brightness period Divide by the line readout rate of the image sensor With the number of cycles The product of, i.e. .

[0095] The image processing module first performs one-dimensional signal processing on the brightness profile along the scrolling direction in a single frame image, for example, by finding local maxima and minima to determine the boundaries of the brightness period. Then, it statistically analyzes the continuous... Number of pixel rows spanned per cycle Image sensor line readout rate This is a fixed hardware parameter, usually found in the sensor's datasheet or queried and configured through the camera's SDK (Software Development Kit). The image processing module pre-loads or acquires this parameter in real-time during calculations. This method incorporates the spatial information of the image (number of pixels and rows). ) and the sensor's time characteristics (line readout rate) By combining these methods, the spatial fringe period can be precisely converted into a time period. This allows for the accurate calculation of vibration frequency.

[0096] Through the above technical solution, the vision acquisition module adopts a CMOS image sensor with a rolling shutter mode, which can directly capture the bright and dark stripes formed by the dynamic physical response optical signal generated by mechanical excitation on the surface of the steering wheel in a single frame image by utilizing its line-by-line exposure characteristics when the sphere 5 is scanning.

[0097] This approach avoids the complexity of traditional methods that require acquiring multiple frames for time-series analysis, significantly simplifying the data acquisition and processing flow. The image processing module analyzes the oscillation period of bright and dark stripes along the scrolling direction in a single frame image, combined with the line readout rate of the image sensor. It can efficiently and accurately demodulate the surface vibration frequency. This provides a basis for subsequent extraction of vibration spectrum anomalies (such as intrinsic response frequencies). This provides real-time, accurate input, greatly improving detection efficiency and real-time performance. The solution effectively solves the problem of efficiently capturing surface vibration frequencies during rolling scanning, enabling vibration spectrum analysis-based internal defect detection to better adapt to the online inspection needs of automated production lines, thereby improving the accuracy and efficiency of internal defect detection in automotive steering wheels.

[0098] The image processing module contains parallel morphological feature processing channels and vibration feature processing channels;

[0099] Parallel operation means that the two processing channels can execute their respective feature extraction and analysis tasks simultaneously and independently without waiting for the other channel to complete. This parallel processing can be achieved in various ways. For example, different computing tasks can be allocated on a multi-core processor or a graphics processing unit (GPU) so that the calculation of morphological and vibrational features can be performed synchronously; or, dedicated hardware accelerators, such as field-programmable gate arrays (FPGAs) or application-specific integrated circuits (ASICs), can be used to provide independent computing resources for each channel, thereby achieving efficient parallel data stream processing.

[0100] The morphological feature processing channel is configured to extract morphological afterimages and / or texture flow features from the dynamic physical response optical signal and output a first defect confidence map.

[0101] Morphological ghosting refers to the optical traces left after the sphere 5 leaves the steering wheel surface due to minute surface deformations caused by internal defects or hysteresis in deformation recovery. Morphological ghosting features can be extracted by comparing image differences before and after sphere 5 rolling, or by analyzing the optical attenuation process of surface deformation over a period of time after rolling. For example, inter-frame differencing can be performed on continuously acquired image sequences, or digital image correlation (DIC) technology can be used to track micro-displacements on the surface. Texture flow features refer to the changes in local texture direction, density, or arrangement caused by internal structural inhomogeneities or defects when the steering wheel surface texture is subjected to mechanical excitation. Texture flow features can be extracted by applying optical flow algorithms to analyze the motion vectors of surface textures, or by using texture analysis algorithms such as Gabor filters and Local Binary Patterns (LBP) to quantify texture changes. The first defect confidence map is a two-dimensional image generated based on these morphological features, using a preset threshold judgment or machine learning model. The value of each pixel represents the probability or confidence of the presence of a morphological defect in that area.

[0102] The vibration feature processing channel is configured to extract vibration spectrum features and / or vibration attenuation features from the dynamic physical response optical signal and output a second defect confidence map.

[0103] Vibration spectrum characteristics refer to the specific frequency components and their intensity contained in the optical signal of the dynamic physical response of the steering wheel surface after being subjected to mechanical excitation. Internal defects (such as delamination and bubbles) can alter the local stiffness and damping characteristics of the material, thereby affecting its intrinsic vibration frequencies and resonance behavior. Vibration spectrum characteristics can be extracted by performing Fourier transform or wavelet transform on the pixel brightness time series acquired by the vision acquisition module to analyze its frequency distribution and energy spectrum. Vibration decay characteristics refer to the rate at which the vibration amplitude of the steering wheel surface decays over time after being subjected to mechanical excitation. Internal defects often lead to abnormal vibration energy dissipation, manifested as prolonged or shortened vibration decay time. Vibration decay characteristics can be extracted by analyzing the envelope of the vibration signal and calculating the time constant or damping ratio required for it to decay from its peak to a specific proportion. The second defect confidence map is a two-dimensional image generated based on these vibration characteristics, using similar threshold judgments or machine learning models. The value of each pixel represents the probability or confidence level of the presence of a defect with abnormal vibration characteristics in that area.

[0104] The image processing module further fuses the first and second defect confidence maps to generate the final defect determination result. The fusion process aims to comprehensively utilize the advantages of two different types of features to overcome the limitations of a single feature. Fusion methods may include, but are not limited to: weighted summation, where different weights are assigned to the two confidence maps based on experience or training data, and then they are superimposed; logical judgment, for example, determining a defect when either confidence map reaches a preset threshold, or requiring both to reach a certain threshold; or employing more complex machine learning algorithms, such as support vector machines (SVM), decision trees, or neural networks, using the two confidence maps as input to train the model and output the final defect determination result. The final defect determination result can be binary (defective / no defect), or it can include information such as the type, location, and severity of the defect.

[0105] Through the above technical solution, the image processing module can process morphological and vibrational features in parallel, making full use of the multi-dimensional information contained in the dynamic physical response optical signals. The morphological feature processing channel can effectively identify surface micro-deformation, afterimages, or texture changes caused by internal defects, and has high sensitivity to defects that exhibit obvious deformation characteristics after mechanical excitation. Meanwhile, the vibrational feature processing channel focuses on analyzing the dynamic mechanical response of materials, such as intrinsic frequency shifts or abnormal vibration attenuation, and has unique detection capabilities for defects that mainly affect the internal stiffness or damping characteristics of materials (such as hollowing and delamination). By fusing these two complementary features through a confidence map, various internal defects hidden beneath the corrugated surface can be identified more reliably.

[0106] The terms "front," "back," "left," "right," "top," and "bottom" all refer to the figures in the accompanying drawings. Figure 1 Based on the perspective of the observer, the side of the device facing the observer is defined as the front, the left side of the observer is defined as the left, and so on.

[0107] In the description of this invention, it should be understood that the terms "center", "longitudinal", "lateral", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limiting the scope of protection of this invention.

[0108] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A machine vision-based device for detecting appearance defects on a steering wheel of a vehicle, comprising a conveying device (1) for transporting the steering wheel and a detection device (2), characterized in that, Also includes: The rolling loading mechanism is configured to press against the surface of the steering wheel to be tested with a preset constant pressure and drive the ball (5) to roll at a constant speed along the surface of the steering wheel so as to generate a continuous broadband mechanical excitation signal in the contact area between the ball (5) and the surface of the steering wheel. The visual acquisition module, whose imaging area covers at least a portion of the rolling path formed by the rolling loading mechanism on the steering wheel surface, is configured to continuously acquire the dynamic physical response optical signal of the steering wheel surface caused by the mechanical excitation signal within a preset recovery monitoring area after the ball (5) leaves. An image processing module is electrically connected to the vision acquisition module. The image processing module is configured to extract at least one local physical property anomaly feature caused by internal defects from the dynamic physical response optical signal, and to determine whether the steering wheel has internal defects based on the local physical property anomaly feature.

2. The machine vision based automobile steering wheel appearance defect detection device according to claim 1, characterized in that: The image processing module is configured to perform a differential operation on the current frame image collected in the recovery monitoring area and the baseline image of the same position before the sphere (5) is rolled, to generate a differential image; when a local bright spot with a brightness value exceeding a preset threshold appears in the differential image, the area corresponding to the bright spot is determined to be a candidate defect area.

3. The machine vision-based automotive steering wheel appearance defect detection device according to claim 1, characterized in that: The local physical property anomaly feature is a vibration spectrum anomaly feature; the image processing module obtains the time brightness change sequence of each pixel position in the recovery monitoring area Perform time-frequency transformation to obtain a time-frequency spectrum And the intrinsic response frequency of each pixel is determined according to the following formula : ； wherein is the pixel At the time , the frequency of the time-frequency spectrum amplitude, is the analysis time window after the sphere (5) leaves the pixel area, is the preset frequency sweep analysis range according to the mechanical properties of the steering wheel intact area.

4. The machine vision-based automotive steering wheel appearance defect detection device according to claim 3, characterized in that: The image processing module further processes each pixel... A reference frequency mapping map with a pre-established, intact bonding region corresponding to a pixel location. A comparison is performed, and the pixel region is determined to have an internal defect if the following conditions are met: ； In the formula, This is a spatially resolved reference frequency map established through offline calibration or online self-learning. The preset frequency offset threshold has a value of [value]. 15% to 30%.

5. The machine vision-based automotive steering wheel appearance defect detection device according to claim 4, characterized in that: The image processing module is also configured to extract vibration attenuation features, specifically: from the time spectrum. Extracted from Time-domain amplitude attenuation envelope at frequency And calculate using the following formula: ； In the formula, for The real-time signal after the frequency components have been bandpass filtered. This represents the Hilbert transform, which shifts the phase of each frequency component of the original real signal. To generate the corresponding orthogonal imaginary part signal; The image processing module further determines the attenuation envelope. From its peak amplitude decay to Required dissipation time , Let be the base of the natural logarithm, and determine that the pixel region has an internal defect when the following conditions are met: ； The dissipation time of the current pixel region. For offline establishment of a reference decay time mapping map of the intact bonding region corresponding to the pixel position, This is the preset dissipation anomaly amplification factor, with a value range of 1.5 to 3.

0.

6. The machine vision-based automotive steering wheel appearance defect detection device according to claim 1, characterized in that: The local physical property anomaly is a local equivalent stiffness feature; the image processing module calculates the maximum downward displacement of each point on the steering wheel surface during the rolling process of the sphere (5). Residual displacement after release And calculate the local stiffness exponent using the following formula. : ； In the formula, A preset constant pressure is applied to the sphere (5); When the measuring point If the stiffness is lower than a preset ratio of the stiffness benchmark value obtained from the statistics of the surrounding normal area, it is determined that there is an internal defect at that point.

7. The machine vision-based automotive steering wheel appearance defect detection device according to claim 3, characterized in that: The visual acquisition module employs a CMOS image sensor with a rolling shutter mode; during scanning by the rolling loading mechanism, the image processing module is configured to demodulate the surface vibration frequency by analyzing the oscillation period of bright and dark stripes along the rolling direction in a single frame image, specifically calculating the vibration frequency using the following formula. : ； In the formula, The measured fringe period time is calculated as follows: [Detection continuous...] Total number of pixel rows per brightness period Divide by the line readout rate of the image sensor With the number of cycles The product of, i.e. .

8. The machine vision-based automotive steering wheel appearance defect detection device according to claim 1, characterized in that: The image processing module contains parallel morphological feature processing channels and vibration feature processing channels. The morphological feature processing channel is configured to extract morphological afterimages and / or texture flow features from the dynamic physical response optical signal and output a first defect confidence map. The vibration feature processing channel is configured to extract vibration spectrum features and / or vibration attenuation features from the dynamic physical response optical signal and output a second defect confidence map. The image processing module further fuses the first defect confidence map and the second defect confidence map to generate the final defect determination result.

Images