Precise part size online detection method and system based on machine vision

Abstract

The invention provides a precision part size online detection method and system based on machine vision, and relates to the technical field of machine vision and precision measurement. The method comprises the following steps: acquiring part weight data through a weighing sensor at a feeding end of a conveyor belt, and monitoring a vibration state in real time by using an acceleration sensor at a detection station; based on weight data, the system controls adjustable vibration reduction supports on the two sides of a conveying belt and grating encoders of detection stations to provide accurate relative displacement data, the minimum vibration moment is determined by combining vibration acceleration signals, an industrial camera is triggered at the optimal moment to collect images, coordinate correction is conducted on the images through the displacement data, and the image precision is improved. And distortion caused by vibration is eliminated. Based on the corrected target image, edge features are extracted, and the part size is calculated through calibration parameters. By implementing the method, accurate suppression of vibration is realized, and high-precision size detection is further realized.

Description

Technical Field

[0001] The present application relates to the field of machine vision and precision measurement technology, and in particular to a method and system for online detection of the dimensions of precision parts based on machine vision. Background Art

[0002] In today's industrial production, precision parts are increasingly used, and their dimensional accuracy is directly related to product quality and performance. Therefore, online dimensional detection methods for precision parts based on machine vision have emerged. Currently, the industry generally adopts an online inspection solution that uses a single industrial camera coupled with a preset template matching algorithm. This solution installs a high-speed industrial camera on the production line, captures images of precision parts, and then compares them with a standard template, enabling automated dimensional inspection of precision parts. Compared to traditional manual inspection, this method significantly improves inspection efficiency and reduces human error, meeting production needs to a certain extent.

[0003] However, because precision parts of varying weights generate varying vibration characteristics when moving on a conveyor belt, this vibration can cause irregular blurring in the captured image. In particular, when the system uses preset fixed exposure parameters, it cannot adaptively adjust to the vibration characteristics of different precision parts, resulting in unstable image quality and affecting the accuracy of subsequent dimensional measurements. Summary of the Invention

[0004] The present application provides a method and system for online detection of the dimensions of precision parts based on machine vision, which is used to solve the technical problems in the prior art that the traditional single industrial camera online detection solution is difficult to cope with the differences in vibration characteristics of precision parts of different weights, and that it is difficult to ensure image quality by using fixed exposure parameters.

[0005] In the first aspect, the present application provides an online detection method for the size of precision parts based on machine vision, which is applied to an online detection system. The method comprises: collecting weight data of the precision parts to be detected by a weighing sensor arranged at the feed end of a conveyor belt, and collecting vibration acceleration signals of the precision parts to be detected in real time by an acceleration sensor arranged at a detection station of the conveyor belt; according to the weight data, controlling adjustable vibration-damping brackets arranged on both sides of the conveyor belt, the adjustable vibration-damping brackets comprising a plurality of piezoelectric ceramic drive units, and forming a displacement in the opposite direction of the vibration acceleration signal by adjusting the expansion and contraction amount of each piezoelectric ceramic drive unit; A grating encoder installed at the inspection station acquires relative displacement data caused by vibration. The grating encoder is fixed to the conveyor belt base. The minimum vibration moment of the precision part to be inspected when passing through the inspection station is determined based on the vibration acceleration signal and the relative displacement data. At the minimum vibration moment, a high-speed industrial camera installed at the inspection station is triggered to perform image acquisition to obtain a part image. The relative displacement data is used to perform coordinate correction on the part image to eliminate image distortion caused by vibration and obtain a corrected target image. The edge features of the precision part are extracted based on the target image, and the size of the precision part is calculated according to preset calibration parameters.

[0006] By adopting the above technical solution, the weight and vibration information of the parts are first obtained in real time through weighing sensors and accelerometers, providing basic data for subsequent vibration suppression. Based on the weight data, the piezoelectric ceramic drive unit is controlled to produce a displacement opposite to the vibration, thereby achieving active vibration reduction. The introduction of the grating encoder provides accurate relative displacement data, which is combined with the vibration acceleration signal to determine the optimal imaging moment. By triggering image acquisition at the moment of minimum vibration and using the displacement data for coordinate correction, the impact of vibration on imaging quality is significantly reduced. This multi-sensor collaboration and closed-loop control method not only achieves precise vibration suppression, but also ensures the optimal timing of image acquisition, ultimately achieving high-precision dimensional detection.

[0007] In a second aspect, the present application provides an online detection system, which includes: one or more processors and a memory; the memory is coupled to the one or more processors, the memory is used to store computer program code, the computer program code includes computer instructions, and the one or more processors call the computer instructions to enable the online detection system to execute the method described in the first aspect and any possible implementation method of the first aspect.

[0008] In a third aspect, the present application provides a computer-readable storage medium comprising instructions, which, when executed on an online detection system, enable the online detection system to execute the method described in the first aspect and any possible implementation of the first aspect.

[0009] In a fourth aspect, the present application provides a computer program product, which, when executed on an online detection system, enables the online detection system to execute the method described in the first aspect and any possible implementation manner of the first aspect. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 This is a flow chart of a method for online detection of dimensions of precision parts based on machine vision in an embodiment of the present application; Figure 2 This is another flow chart of the method for online detection of dimensions of precision parts based on machine vision in an embodiment of the present application; Figure 3 It is a schematic diagram of the physical device structure of the online detection system in the embodiment of the present application. DETAILED DESCRIPTION

[0011] The terms used in the following examples of the present application are for the purpose of describing specific embodiments only and are not intended to limit the present application. As used in the specification and appended claims of the present application, the singular expressions "a," "an," "said," "above," "the," and "this" are intended to include plural expressions as well, unless the context clearly indicates otherwise. It should also be understood that the term "and / or" used in the present application refers to and encompasses any or all possible combinations of one or more of the listed items.

[0012] In the following, the terms "first" and "second" are used for descriptive purposes only and should not be understood to imply or suggest relative importance or implicitly indicate the number of the technical features indicated. Therefore, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of the embodiments of this application, unless otherwise specified, "plurality" means two or more.

[0013] For ease of understanding, the following describes the process of the method provided by this implementation. Figure 1 , which is a flow chart of the online detection method for the size of precision parts based on machine vision in an embodiment of the present application.

[0014] S101. The weight data of the precision parts to be inspected is collected via a weighing sensor installed at the feed end of the conveyor belt, and the vibration acceleration signal of the precision parts to be inspected is collected in real time via an accelerometer installed at the inspection station of the conveyor belt. The weighing sensor can be a high-precision strain gauge weighing sensor, whose operating principle is based on the resistance strain effect of metal strain gauges. When the precision parts are placed on the conveyor belt and pass through the feed end, the weight of the parts acts on the weighing sensor, causing the elastic body of the parts to deform. The strain gauge attached to the elastic body also deforms, resulting in a change in the resistance value of the strain gauge. By measuring the change in the resistance value of the strain gauge, the resistance change is converted into a voltage signal using a Wheatstone bridge circuit. The analog voltage signal is then amplified by an amplifier and converted into a digital signal through an analog-to-digital conversion circuit, and finally transmitted to the online detection system.

[0015] At the inspection station, an accelerometer can be securely mounted in advance at a specific location on the conveyor belt. It can accurately sense the vibration acceleration generated by parts moving on the conveyor belt. The sensor converts the acceleration signal into an electrical signal output. The analog-to-digital conversion (ADC) circuit then converts the analog signal into a digital signal for subsequent processing by the online inspection system's processor. To obtain more comprehensive vibration information, the accelerometer can be configured with multiple axes. For example, a common three-axis accelerometer can simultaneously measure the vibration acceleration of a part in three mutually perpendicular directions. The system collects this acceleration data in real time, stores it, and analyzes it at a certain frequency, so that subsequent processing can be performed based on the vibration situation.

[0016] S102: Based on the weight data, control adjustable vibration-damping brackets disposed on both sides of the conveyor belt, wherein the adjustable vibration-damping brackets include multiple groups of piezoelectric ceramic drive units, and generate a displacement in a direction opposite to the vibration acceleration signal by adjusting the expansion and contraction of each piezoelectric ceramic drive unit; The system compares the collected weight data with a pre-set weight-to-vibration damping parameter mapping table. This mapping table, derived from extensive experiments and data analysis, covers the optimal vibration damping parameters for precision parts in different weight ranges. For example, if a precision part is detected to weigh 50 grams, the mapping table indicates that at this weight, the piezoelectric ceramic drive unit must produce a displacement of a specific magnitude and direction to effectively suppress vibration. Based on this information, the system calculates the required expansion and contraction of each piezoelectric ceramic drive unit.

[0017] During the actual control process, the system applies a corresponding voltage signal to the piezoelectric ceramic drive unit. Piezoelectric ceramics have a unique piezoelectric effect. When voltage is applied, they produce expansion and contraction deformation related to the voltage's magnitude and polarity. Taking a certain type of piezoelectric ceramic drive unit as an example, when a voltage of a specific frequency and amplitude is input, it can accurately expand and contract to a specific length in a short period of time. Based on the calculation results, the system precisely controls the input voltage of each piezoelectric ceramic drive unit. If the vibration acceleration signal of the part indicates an upward vibration direction, the system controls the piezoelectric ceramic drive unit to produce a downward displacement, thereby generating a displacement in the opposite direction of the vibration acceleration signal, effectively offsetting the vibration of the part.

[0018] To ensure the stability and reliability of the vibration reduction effect, the system also monitors the vibration acceleration signals collected by the accelerometer in real time. If it finds that the vibration acceleration fails to achieve the expected vibration reduction effect, it will quickly adjust the control parameters of the piezoelectric ceramic drive unit. This closed-loop control mechanism can constantly adjust itself according to the actual vibration conditions. For example, at a certain moment, due to an unexpected situation on the production line, the vibration of the part suddenly intensifies. The accelerometer detects that the vibration acceleration exceeds the expected range. The system will immediately increase the expansion and contraction of the piezoelectric ceramic drive unit, increase the vibration reduction strength, and quickly restore the part vibration to an acceptable level.

[0019] S103, obtaining relative displacement data caused by vibration by using a grating encoder provided at a detection station, where the grating encoder is fixed on a conveyor belt base; Relative displacement data refers to the measurement of position changes caused by part vibration during the inspection process. At the inspection station, a grating encoder fixed to the conveyor base is responsible for acquiring this vibration-induced relative displacement data. The grating encoder primarily consists of a light source, a code disk, a photosensitive element, and a signal processing circuit. The code disk is engraved with a series of fine grating stripes, which act like precise scale markings, recording displacement changes.

[0020] When the conveyor belt is displaced by the vibration of the parts, the code disk moves synchronously. Light from a light source passes through the code disk's grating stripes and illuminates a photosensitive element. As the code disk rotates, the light signal received by the photosensitive element undergoes periodic changes. This change in light signal is converted into electrical pulses, and the number of pulses is directly related to the rotation angle or displacement of the code disk. By counting and processing these electrical pulses, the online detection system can accurately calculate the relative displacement of the conveyor belt during vibration. To improve measurement accuracy, the system uses a high-resolution grating encoder.

[0021] In some embodiments, during the online dimensional inspection of precision parts, to more effectively suppress part vibration and improve inspection accuracy, before acquiring relative displacement data caused by vibration using a grating encoder, the online inspection system performs a series of operations to actively control and monitor the part's vibration state. The system uses an audio exciter and an acoustic wave sensor located at the infeed end of a conveyor belt to acquire the natural frequency of the precision part being inspected. The audio exciter emits acoustic waves within a specific frequency range into the part, which propagate within the part and induce vibration. When the excitation frequency approaches the part's natural frequency, resonance occurs, significantly increasing the part's vibration amplitude. The acoustic wave sensor receives the reflected acoustic waves generated by the part's vibration and converts them into electrical signals. The system analyzes and processes these electrical signals, accurately identifying the part's natural frequency using algorithms such as fast Fourier transforms. Based on the acquired natural frequency, the system controls multiple sets of piezoelectric ceramic vibrators located within a set range of the conveyor belt to generate anti-phase vibration waves at a set frequency. Piezoelectric ceramic vibrators exhibit a piezoelectric effect, generating mechanical vibrations when an electrical signal is applied. The system generates an electrical signal at a frequency corresponding to the natural frequency and applies it to the piezoelectric ceramic resonator, causing it to generate vibration waves with the same frequency but opposite phase as the component's vibration. These anti-phase vibration waves intersect with the component's own vibration waves in space. Due to the principle of wave interference, they superimpose and form destructive interference. This interference effectively cancels out the component's vibration energy, thereby reducing its vibration amplitude.

[0022] To accurately monitor the part's motion in real time, the system utilizes three orthogonally arranged Hall effect sensor arrays located at the inspection station. Magnetic material can be pre-coated on the part surface. For non-magnetic parts, magnetic material can be applied to the inspection fixture or jig, allowing the Hall effect sensors to monitor the fixture's motion. As the part moves on the conveyor belt, the surrounding magnetic field changes. The Hall effect sensors sense these changes and convert them into electrical signals. Because the three Hall effect sensors are arranged orthogonally, they can detect magnetic field variations along the part's three perpendicular directions, thereby acquiring information about the part's three-dimensional motion, including translation and rotation. Finally, the system adjusts the operating parameters of the piezoelectric ceramic resonator based on the part's three-dimensional motion. If the Hall effect sensors detect a significant vibration amplitude in a particular direction or a change in the part's motion, the system adjusts the amplitude, frequency, and phase of the electrical signal applied to the piezoelectric ceramic resonator accordingly. For example, if the system detects increased vibration along the x-axis at a given moment, the system increases the vibration amplitude of the piezoelectric ceramic resonator in that direction, generating a stronger, anti-phase vibration wave to offset the vibration in that direction. Through this real-time monitoring and dynamic adjustment, the vibration of the parts is accurately offset, ensuring that the parts are in a relatively stable state during the detection process, providing good conditions for subsequent accurate acquisition of relative displacement data, image acquisition, and dimensional detection.

[0023] S104, determining the minimum vibration moment of the precision part to be inspected when it passes through the inspection station based on the vibration acceleration signal and the relative displacement data; The system pre-processes the vibration acceleration signal and relative displacement data. Since the actual collected data may contain noise and interference information, a filtering algorithm is used to reduce the noise, making the data smoother and more accurately reflecting the actual vibration state of the part.

[0024] The system will judge the vibration state by combining the change trend of vibration acceleration and relative displacement. Generally speaking, when the vibration acceleration amplitude gradually decreases and the relative displacement change tends to be flat, it indicates that the vibration of the part is weakening. For example, within a certain period of time, the vibration acceleration amplitude changes from 5m / s to 2 Gradually decrease to 1m / s 2 At the same time, the change in relative displacement between adjacent sampling moments decreases from 0.5mm to 0.1mm, indicating that the vibration is gradually weakening and may be approaching the minimum vibration moment. The system continuously monitors these data changes and preliminarily screens the possible minimum vibration moment range by setting reasonable thresholds.

[0025] In order to more accurately determine the moment of minimum vibration, the system uses a dynamic time warping (DTW) algorithm. This algorithm can find the best matching path in data from different time series, thereby comparing the similarity of vibration acceleration signals and relative displacement data. Taking the inspection of a certain precision part as an example, the system will perform DTW matching on the currently collected vibration acceleration and relative displacement data sequences with the pre-stored standard stable vibration data sequence. When the matching degree reaches the highest, the corresponding moment is considered to be the moment of minimum vibration. This is because the standard stable vibration data sequence represents an ideal low-vibration state, and a high degree of matching with it means that the current part vibration state is close to the ideal state, that is, it is at the moment of minimum vibration.

[0026] In practice, the system continuously optimizes its strategy for determining the moment of minimum vibration. By accumulating and analyzing extensive experimental data, it continuously adjusts the filtering algorithm's parameters, threshold settings, and DTW algorithm matching rules to accommodate the vibration characteristics of different types of precision parts. This allows the online inspection system to more accurately determine the moment of minimum vibration, providing the optimal timing for subsequent image acquisition, effectively reducing the impact of vibration on image quality, and laying a solid foundation for high-precision dimensional inspection.

[0027] S105, triggering a high-speed industrial camera located at the inspection station to perform image acquisition at the minimum vibration moment to obtain an image of the part; Once the online inspection system determines the moment of minimum vibration, it immediately triggers a high-speed industrial camera located at the inspection station to capture high-quality images of the part. The system sends a trigger signal to the camera, which is transmitted via a specialized communication interface to ensure stable and timely signal transmission. For example, a high-speed digital I / O interface enables accurate transmission of the trigger signal to the camera within microseconds. To ensure trigger accuracy, the system performs a series of preparatory steps before triggering, such as checking the camera's operating status and ensuring that parameters such as aperture, focal length, and exposure time are set to optimal values.

[0028] Upon receiving a trigger signal, a high-speed industrial camera rapidly initiates the image acquisition process. For example, a high-resolution CMOS industrial camera can rapidly complete image exposure, charge transfer, and digitization. During the exposure process, the camera controls the duration of light entering the lens according to a pre-set exposure time. For precision parts with minimal vibration, a shorter exposure time may be selected to minimize image blur caused by minute part movement.

[0029] In actual production, the system also considers production line continuity and inspection efficiency. To avoid production line stalls caused by camera image acquisition, multiple cameras are used in a collaborative approach. Multiple high-speed industrial cameras simultaneously capture parts from different angles, not only capturing multi-view images of the parts but also improving inspection efficiency. The system also synchronizes the images captured by multiple cameras to ensure temporal and spatial consistency between the images, providing comprehensive and accurate data support for subsequent image analysis and dimensional calculations.

[0030] In some embodiments, after acquiring the part image, the online detection system performs a series of operations to improve the image quality and detection accuracy. First, the system controls the annular light source array set at the detection station to emit light in sequence according to a preset timing. The annular light source array is set around the detection station and contains multiple LED light sources at different angles. Through precise timing control, each LED light source is lit in turn to emit light to the surface of the part. In this process, the part surface reflects light at different incident angles differently. The system uses high-precision light sensors to obtain multiple sets of reflected light intensity data at different incident angles. For example, from 0° to 360°, the reflected light intensity data is collected every certain angle (such as 10°). These data reflect the reflection characteristics of the part surface at different lighting angles. Based on these data, the system establishes a reflectivity distribution model for the part surface, which accurately characterizes the reflection characteristics of the part surface to light at different incident angles.

[0031] Next, based on the reflectivity distribution model, the system determines the optimal light source parameter combination. The system conducts an in-depth analysis of the model data, taking into account factors such as the part's shape, material, and inspection requirements. It determines light source parameters, such as intensity, color temperature, and exposure time, that highlight the part's surface features and minimize shadows and reflections. After determining the optimal parameters, the system controls each LED light source in the ring array to perform multiple exposures according to this combination, generating a multi-frame enhanced image sequence. During each exposure, the camera captures an image of the part according to the set parameters. The images contain varying levels of detail under different exposures, and the multiple frames complement each other, enhancing the overall image information. The system then performs registration and fusion on the multi-frame enhanced image sequence. During registration, the system uses feature point matching algorithms, such as the SIFT (Scale-Invariant Feature Transform) algorithm, to identify corresponding feature points in each frame. It then calculates the translation, rotation, and scaling relationships between the images to spatially align the frames. Fusion merges the registered images, combining the strengths of each frame to enhance image clarity and detail. After completing the registration and fusion, the system uses a photometric stereo vision algorithm to calculate the normal vector distribution of the part surface. This algorithm is based on the reflected light intensity data under different incident angles. Through complex mathematical operations, it determines the normal vector direction of each point on the part surface, and then reconstructs the three-dimensional contour data of the part surface based on the normal vector distribution. This data contains the height information and tilt angle information of the part surface.

[0032] Finally, the system builds a posture compensation model based on the 3D contour data. This model describes the relationship between the part's posture changes and the 3D contour. The system synchronizes the posture compensation model with the vibration displacement data collected by the grating encoder and spatially registers it to generate comprehensive correction parameters. These parameters are used to compensate for the coupled effects of part posture changes and vibration displacement. Based on these comprehensive correction parameters, the system performs nonlinear correction on the part image. Using specific image transformation algorithms, such as spline-based transformations, the pixel positions and grayscale values of the image are adjusted to produce a high-precision corrected image.

[0033] In some embodiments, during the online inspection of precision part dimensions, to further improve image quality and effectively eliminate the effects of vibration on part images, the online inspection system performs the following sequence of operations after acquiring a part image: The system divides the acquired part image into multiple subregions to more closely analyze the image's local features. Each subregion undergoes multi-scale wavelet decomposition, which decomposes the image information into different frequency scales. High-frequency coefficients represent image edge and texture information, such as subtle surface lines and edge contours on the part; low-frequency coefficients represent the overall grayscale distribution of the image, reflecting the image's approximate brightness and basic shape. Based on previously acquired relative displacement data, the system establishes a part vibration trajectory model, which simulates the part's motion trajectory during image acquisition. Based on this model, the system calculates the instantaneous velocity and acceleration components of each subregion during image acquisition. These velocity and acceleration components reflect the subregion's motion state at different moments in time. Using these components, the system constructs a time-varying motion blur kernel function for each subregion. This function describes the blurring effect of vibration on the image of each subregion, providing a key basis for subsequent deblurring. Using the constructed motion blur kernel function, the system performs deconvolution operations on the high-frequency and low-frequency coefficients of each subregion. Deconvolution is a signal processing technique that can restore blurred signals to a certain extent. Deconvolution removes the blurring effect caused by vibration and restores edge and texture details in the high-frequency coefficients and overall grayscale information in the low-frequency coefficients.

[0034] The processed coefficients are then reconstructed into images for each subregion using an inverse wavelet transform. This is the inverse process of wavelet decomposition. The decomposed coefficients are reassembled into an image, and the reconstructed subregion images are stitched together to create a deblurred image. This image effectively eliminates the effects of vibration and more clearly presents the true shape of the part. To ensure optimal deblurring, the system uses the gradient magnitude and directional consistency of the deblurred image as clarity evaluation parameters. The gradient magnitude reflects the severity of grayscale changes in the image, while the directional consistency measures the consistency of edge directions. If the clarity evaluation parameter falls below a preset threshold, the image is still not clear enough and the deblurring effect needs improvement. In this case, the system uses the gradient descent method to optimize the parameters of the motion blur kernel function. Gradient descent is a commonly used optimization algorithm that iteratively adjusts parameters to reduce the objective function (here, the clarity evaluation parameter). After optimization, the system repeats the deconvolution and reconstruction steps to process the image again. This cycle continues until the clarity evaluation parameter is greater than or equal to the preset threshold. At this point, the deblurred image obtained meets the detection accuracy requirements and can provide high-quality data support for subsequent image-based part size detection, ensuring the accuracy and reliability of the detection results.

[0035] S106, using the relative displacement data to perform coordinate correction on the part image to eliminate image distortion caused by vibration, thereby obtaining a corrected target image; After acquiring a part image, the online inspection system uses the previously acquired relative displacement data to perform coordinate correction on the image to eliminate image distortion caused by vibration, resulting in a corrected target image. Specifically, the system uses the relative displacement data to determine the new position of each pixel in the image. As the part vibrates, its position in the image shifts, and the relative displacement data records this shift. For example, if the relative displacement data indicates that the part has shifted 5 pixels to the right horizontally and 3 pixels upward vertically, the system adjusts each pixel in the image by this offset. For example, a pixel in the upper left corner of the image originally represented the location of a feature on the part. However, due to vibration, this feature has shifted its position in the image. Based on the relative displacement data, the system shifts the pixel's coordinates 5 units to the right and 3 units upward, returning the pixel to its correct position when the part was not vibrating.

[0036] When performing coordinate correction, the system will use an efficient image transformation algorithm that can quickly and accurately transform the coordinates of a large number of pixels in the image. Taking the common bilinear interpolation algorithm as an example, it calculates the grayscale value of the transformed pixel by linearly interpolating the grayscale values of adjacent pixels, ensuring a smooth transition of the image during the transformation process and avoiding jagged edges or blurring. Suppose there is a pixel in the image that needs to be corrected, and the grayscale values of the four adjacent pixels around it are A, B, C, and D respectively. The bilinear interpolation algorithm will calculate the grayscale value of the pixel after correction based on the relative position relationship between the pixel and these four adjacent pixels, so that the corrected image can accurately reflect the true shape of the part while maintaining a good visual effect.

[0037] In actual application scenarios, due to the complex and diverse vibration conditions of different parts, which may involve nonlinear displacement changes, the system will combine multiple coordinate correction algorithms to address these situations. For parts with more severe vibration and complex displacement changes, the system may first use an affine transformation algorithm for preliminary correction, and then use a more complex spline interpolation-based algorithm to refine the image. In this way, the system can adapt to various complex vibration conditions, ensuring that the corrected target image accurately reflects the actual shape and position of the part, providing reliable image data support for subsequent precision dimensional inspection.

[0038] In some embodiments, after completing the part image correction to obtain the target image, the online inspection system utilizes a through-beam laser ranging sensor to further inspect the part to obtain more comprehensive part dimensional information. The through-beam laser ranging sensor, installed at the inspection station, emits a laser beam and measures the time it takes for the laser to reflect off the part surface before being received. Using the principle of the constancy of the speed of light, it calculates the distance between the sensor and the part surface. Due to the unevenness of the part surface, the reflected light returns at different locations with varying return times. This sensor acquires a series of distance data, which is integrated to form a height profile of the part surface, accurately depicting the surface's unevenness. For example, for a precision part with grooves and protrusions, a through-beam laser ranging sensor can clearly measure the groove depth and protrusion height, presenting them as data. The system then fuses this height profile data with the corrected target image. This fusion process, based on a specific algorithm, matches and integrates the height information in the height profile data with the two-dimensional dimensional information in the target image. At the image pixel level, height data is mapped to the corresponding position in the target image, so that each pixel not only contains the original 2D position information but also adds height information. This way, an image that originally only displayed the planar dimensions of a part is now combined with height data to form a comprehensive inspection result that includes both 2D dimensions and height information. For example, when inspecting a part with a complex curved surface, the fused result can simultaneously display the part's length, width, and the height of each surface point, providing the operator with more comprehensive information about the part's topography.

[0039] S107 , extracting edge features of the precision part based on the target image, and calculating the size of the precision part according to preset calibration parameters.

[0040] After obtaining the corrected target image, the online inspection system uses advanced edge detection algorithms to extract the edge features of the part. The common Canny edge detection algorithm is a good choice, as it uses multiple steps to accurately identify and outline the edges of the part. First, the algorithm applies a Gaussian filter to the image to remove noise, resulting in a smoother image and preventing noise from interfering with edge detection. For example, images of some precision parts may contain tiny noise points due to uneven lighting or sensor noise. Without filtering, these noise points may be mistaken for part edges. After Gaussian filtering, the noise in the image is effectively suppressed, and the true features of the part are highlighted.

[0041] Next, the Canny algorithm calculates the gradient magnitude and direction for each pixel in the image. The gradient magnitude reflects the severity of the change in the pixel's grayscale value, while the gradient direction indicates the direction of the fastest grayscale change. In part images, pixels at the edge have larger grayscale changes, and their gradient magnitudes are correspondingly larger. By setting an appropriate threshold, the algorithm can identify pixels with a gradient magnitude greater than the threshold as possible edge points. To further improve the accuracy of edge detection, the Canny algorithm also performs non-maximum suppression, removing pixels that have large gradient magnitudes but are not true edges, thereby obtaining a more accurate edge outline.

[0042] After successfully extracting the part's edge features, the system calculates the part's dimensions based on preset calibration parameters. These calibration parameters are obtained by photographing and analyzing standard parts of known dimensions. They establish a correspondence between image pixels and actual physical dimensions. For example, during the calibration process, a square standard part with a side length of 10 mm is photographed, and the number of pixels corresponding to its side length measured in the image is 100. Therefore, the calibration parameter is determined to be 1 pixel representing an actual length of 0.1 mm. When calculating the dimensions of the actual part to be inspected, the system measures the pixel length of the part's edge in the image. For example, if an edge has an image length of 80 pixels, based on the calibration parameters, the actual length of that edge is 8 mm. In a real production environment, the system takes into account the batch inspection requirements of parts. To improve inspection efficiency, parallel computing technology is used to simultaneously perform edge feature extraction and dimension calculation on multiple part images. This not only meets the inspection efficiency requirements of the production line, but also ensures high accuracy in the dimensional inspection of each part.

[0043] In the embodiment of the present application, due to the use of a multi-sensor collaborative working method, the weighing sensor at the feed end of the conveyor belt obtains the part weight data, the acceleration sensor at the detection station monitors the vibration state in real time, and the grating encoder provides accurate relative displacement data, so it is possible to comprehensively and accurately obtain key information about the part during the detection process. Based on this data, the adjustable vibration reduction bracket is controlled to perform active vibration reduction, the minimum vibration moment is determined to trigger the industrial camera to capture the image, and the image is coordinate corrected and the size is calculated. This effectively solves the problem that the traditional single industrial camera online detection solution in the existing technology is difficult to cope with the differences in vibration characteristics of precision parts of different weights, and the use of fixed exposure parameters is difficult to ensure image quality. It then achieves precise vibration suppression and high-quality image acquisition, ensuring the high precision and stability of part size detection, and meeting the strict requirements of modern production lines for online detection of precision part dimensions.

[0044] In some embodiments, after completing precision part dimension calculations, the online inspection system performs a series of data-based analysis and prediction operations to further optimize the inspection process and improve subsequent part vibration control. The system builds a part density distribution model based on the acquired weight data and precision part dimensions. Part weight and dimensions are key factors in determining density distribution. By dividing the part into multiple small units, a specific algorithm is used to calculate the density of each unit, combining known weight data and corresponding unit dimensions. For example, for parts with regular shapes, a simple volume-to-weight ratio can be used to provide a preliminary estimate of the density of each unit. For parts with complex shapes, more complex methods such as finite element analysis may be required. This creates a model that accurately reflects the part's internal density distribution. Based on this density distribution model, the system can determine the part's center of gravity and moment of inertia. Determining the center of gravity is crucial for understanding the part's balance characteristics. By integrating the mass and position of each unit in the density distribution model, the coordinates of the part's center of gravity can be accurately calculated. The moment of inertia reflects the inertial characteristics of the part during rotation and is related to the part's mass distribution and the position of the rotation axis. Similarly, using the density distribution model and the integral formula, the moment of inertia of a part around a specific axis can be accurately calculated. These two parameters are important indicators for describing the mechanical properties of a part.

[0045] After obtaining the center of gravity position and moment of inertia data, the system inputs them into a pre-set vibration prediction model. This model, developed based on extensive experimental data and theoretical analysis, predicts the vibration characteristics of a part during subsequent inspection based on its mechanical parameters. The model considers various factors, including the part's material properties, its motion state, and the operating environment of the inspection equipment. Once the center of gravity position and moment of inertia data are input, the model uses a complex algorithm to simulate the part's motion on a conveyor belt, generating predicted vibration characteristics for the next inspection cycle, including key parameters such as acceleration and frequency. Finally, based on these predicted vibration characteristics, the system proactively adjusts the control parameters of the piezoelectric ceramic driver unit. The piezoelectric ceramic driver unit is the core component of the adjustable vibration damping bracket, and adjustment of its control parameters directly affects the vibration damping effect. If the predicted vibration acceleration of the part during the next inspection cycle is high, the system increases the input voltage amplitude of the piezoelectric ceramic driver unit, generating a larger reverse displacement and enhancing the vibration damping effect. If the predicted vibration frequency changes, the system adjusts the input voltage frequency to better match the piezoelectric ceramic driver unit's vibration frequency with the predicted part vibration frequency, achieving optimal vibration damping. In this way, the system realizes complete closed-loop control from part mechanical property analysis to vibration prediction and then to advance adjustment of vibration reduction parameters, effectively improving the ability to suppress part vibration and ensuring the accuracy and stability of subsequent precision part dimensional detection.

[0046] During the operation of the online detection system, in order to more comprehensively understand the status of the conveyor belt and further improve the control accuracy of the parts vibration, steps S201-S204 can be performed. In combination with the above content, the following is a more detailed description of the process of the method provided by this embodiment. Figure 2 , is another flow chart of the online detection method for precision parts size based on machine vision in an embodiment of the present application.

[0047] S201, collecting strain data in real time by disposing at least three rows of strain gauge arrays along the direction of movement of the conveyor belt, wherein each row of the strain gauge array includes a plurality of strain gauges evenly distributed along the width of the conveyor belt, and the strain data represents the deformation of a local area of the conveyor belt; During actual installation, technicians will place at least three rows of strain gauge arrays at equal intervals along the conveyor belt's direction of motion. For example, on a high-speed conveyor belt producing precision parts, a row of strain gauges is placed every 10 centimeters. Each row consists of 10 strain gauges evenly spaced across the width of the conveyor belt. These strain gauges utilize high-precision metal foil strain gauges, which work based on the metal's resistance strain effect. When a strain gauge is deformed by an external force, its resistance changes accordingly.

[0048] The online detection system collects data on the resistance changes of strain gauges in real time at an extremely high frequency. Since the conveyor belt is subject to various forces during operation, such as the weight of the components, its own tension, and friction with the transmission mechanism, these forces can cause deformation in localized areas of the conveyor belt. When a certain area of the conveyor belt experiences tension or compression, the strain gauge attached to that area also deforms, causing a change in resistance. The system measures the change in strain gauge resistance and converts the resistance change into a voltage signal using a Wheatstone bridge circuit. This signal is then amplified by an amplifier and converted to a digital signal using an analog-to-digital conversion circuit for storage and analysis.

[0049] S202, calculating a strain difference based on the strain data collected by each two adjacent rows of strain gauge arrays, and determining a bending curvature and a torsion angle of the conveyor belt based on the strain difference, wherein the bending curvature represents the degree of bending of the conveyor belt in the vertical direction, and the torsion angle represents the rotation angle of the conveyor belt around the longitudinal axis; After acquiring the strain data of the conveyor belt, the system will compare and analyze the data collected by each adjacent row of strain gauge arrays one by one. Taking two adjacent rows of strain gauge arrays as an example, assuming that the strain data collected by the first row of strain gauge arrays are a1, a2...an, and the strain data collected by the second row of strain gauge arrays are a1 ′ 、a2 ′ ……an ′, the system will calculate the strain difference of the strain gauge at the corresponding position, that is, Δa1=a1-a1 ′ , Δa2=a2-a2 ′ ...Δan=an-an ′ When calculating curvature, the system applies specialized algorithms based on the principles of material mechanics and geometric relationships. Because vertical bending of the conveyor belt causes differences in strain at different locations, the system can calculate the degree of vertical bending by analyzing the distribution of these strain differences. For example, at a specific moment, the system analyzed the collected strain difference data and found that the strain difference between two adjacent rows of strain gauges in a certain section of the conveyor belt exhibited a linear trend. Based on beam bending theory, the system used this data to calculate the curvature k of the conveyor belt in that area. The value of this curvature k directly reflects the degree of belt bending in that area; larger k values indicate more severe belt bending. If the curvature exceeds a certain allowable range, it may affect the smooth transportation of parts on the conveyor belt, thereby affecting detection accuracy. The system also uses strain difference analysis to determine the torsion angle. When the conveyor belt twists, the strain at different locations along the width will vary, and this difference will also be reflected in the strain data collected by the two adjacent rows of strain gauge arrays. By analyzing the characteristics of these differences and applying mathematical methods such as tensor analysis, the system converts the strain differences into parameters that describe torsion, thereby calculating the rotation angle of the conveyor belt around the longitudinal axis.

[0050] S203. Inputting the bending curvature and torsion angle into a preset conveyor belt dynamic deformation prediction model to obtain a vibration state prediction result, wherein the vibration state prediction result includes the expected vibration parameters when the part arrives at the inspection station. The conveyor belt dynamic deformation prediction model is pre-trained using machine learning based on the strain data of the conveyor belt under different working conditions and the vibration parameters at corresponding times; During the training phase of the conveyor belt dynamic deformation prediction model, technicians collected a large and diverse dataset, covering strain data for conveyor belts under different operating conditions, such as varying loads, operating speeds, ambient temperatures, and belt materials. At the same time, the vibration parameters of the parts at the corresponding moments were recorded, including vibration acceleration, vibration frequency, and vibration displacement. This data was used as a training set and fed into a machine learning model for repeated training. For example, using a neural network algorithm from deep learning, the model learned the complex nonlinear relationship between strain data and vibration parameters by continuously adjusting the network's weights and thresholds. After training and optimizing on this massive amount of data, the model gradually acquired the ability to accurately predict part vibration parameters based on the input bending curvature and torsion angle.

[0051] When the online inspection system inputs the real-time acquired bending curvature and torsion angle into the trained model, the model rapidly performs calculations and analysis. Based on the previously learned relationship between strain and vibration, combined with the currently input bending curvature and torsion angle data, the model predicts the expected vibration parameters when the part arrives at the inspection station. For example, the model might predict that the vibration acceleration of a part upon arrival at the inspection station will be a, the vibration frequency will be f, and the vibration displacement will be d. These predicted parameters are crucial for the online inspection system to prepare for vibration reduction in advance. If a high vibration acceleration is predicted, it means that the part may experience significant vibration during the inspection, affecting the accuracy of image acquisition and dimensional inspection. Based on these predicted results, the system can adjust relevant equipment parameters in advance, such as increasing the vibration reduction strength of the piezoelectric ceramic drive unit, to ensure that the part's vibration during inspection is within an acceptable range, thereby ensuring inspection accuracy.

[0052] In the actual production process, since the working conditions may change, the system will regularly update and optimize the model. By continuously collecting new strain data and vibration parameters, the model will be retrained to enable it to better adapt to new working conditions.

[0053] S204 , performing data fusion processing on the vibration state prediction result and the actual vibration acceleration signal collected by the acceleration sensor, and adaptively adjusting the control parameters of the piezoelectric ceramic drive unit according to the fused vibration information to compensate for the vibration caused by the conveyor belt deformation in advance.

[0054] After obtaining the vibration state prediction results, the system first performs data fusion between the predicted vibration state results and the actual vibration acceleration signals collected by the accelerometer. Since the prediction results are estimated values based on the conveyor belt deformation, while the actual vibration acceleration signals reflect the actual vibration conditions of the current part, combining the two provides a more comprehensive and accurate understanding of the part's vibration state. For example, the system uses a weighted averaging method for data fusion, assigning different weights based on the reliability of the predicted results and the actual measured values. If the prediction model has been trained with a large amount of data and performs stably under the current operating conditions, the prediction result may be given a higher weight; conversely, if the accelerometer data is more reliable, its weight may be appropriately increased. Through this fusion method, the system obtains more accurate vibration information, providing a more reliable basis for subsequent control decisions.

[0055] Based on the fused vibration information, the system automatically adjusts the control parameters of the piezoelectric ceramic drive unit. The piezoelectric ceramic drive unit is a key component of the adjustable vibration damping bracket. By adjusting its control parameters, such as the amplitude, frequency, and phase of the input voltage, the expansion and contraction and vibration characteristics of the piezoelectric ceramic can be changed, thereby achieving precise control of the vibration of the part. For example, if the fused vibration information shows that the vibration acceleration of the part is too large, the system will increase the input voltage amplitude of the piezoelectric ceramic drive unit to produce a larger reverse displacement and enhance the vibration reduction effect. At the same time, if the vibration frequency is predicted to change, the system will adjust the frequency of the input voltage accordingly to better match the vibration frequency of the piezoelectric ceramic drive unit with the vibration frequency of the part to achieve the best vibration reduction effect.

[0056] In actual production, this adaptive adjustment mechanism responds in real time to changes in conveyor belt deformation and part vibration. For example, if a sudden change in the production line's operating speed causes increased conveyor belt deformation and part vibration, the system quickly integrates the predicted results with the actual vibration data and promptly adjusts the control parameters of the piezoelectric ceramic drive unit. At a certain moment, the system predicts that the vibration acceleration will increase significantly when the part arrives at the inspection station. Simultaneously, the accelerometer detects that the current vibration acceleration is also increasing. The system immediately increases the input voltage amplitude of the piezoelectric ceramic drive unit and adjusts its frequency to match the predicted vibration frequency. This operation preemptively compensates for vibration caused by conveyor belt deformation, ensuring that part vibration remains within an acceptable range while at the inspection station. Through this closed-loop control process, the online inspection system achieves preemptive compensation and precise control of vibration caused by conveyor belt deformation, effectively improving inspection accuracy and stability, ensuring the accuracy of online dimensional inspection of precision parts, and meeting the stringent high-precision inspection requirements of modern production lines.

[0057] In the embodiment of the present application, due to the use of a strain gauge array to monitor the deformation of the conveyor belt in real time, determine the bending curvature and torsion angle of the conveyor belt based on the strain difference, use a conveyor belt dynamic deformation prediction model trained by machine learning, and integrate the prediction results with the actual vibration data to adjust the vibration reduction parameters, it is possible to accurately obtain the deformation information of the conveyor belt, predict the vibration state of the parts when they arrive at the inspection station in advance, and timely and adaptively adjust the control parameters of the piezoelectric ceramic drive unit, which effectively solves the problems in the prior art that the deformation state of the conveyor belt cannot be predicted and it is difficult to compensate for the vibration caused by the deformation of the conveyor belt in advance, thereby achieving accurate prediction of the deformation of the conveyor belt and early suppression of the vibration, greatly improving the system's control ability over the vibration caused by the deformation of the conveyor belt, and ensuring the stable transportation and high-precision detection of parts during the online detection of the size of precision parts.

[0058] The following describes the online detection system in the embodiment of the present invention from the perspective of hardware processing. Figure 3, is a schematic diagram of the physical device structure of the online detection system in an embodiment of the present application.

[0059] It should be noted that Figure 3 The structure of the online detection system shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present invention.

[0060] like Figure 3 As shown, the online detection system includes a central processing unit (CPU) 301, which can perform various appropriate actions and processes according to the program stored in the read-only memory (ROM) 302 or the program loaded from the storage part 308 to the random access memory (RAM) 303, such as executing the method described in the above embodiment. Various programs and data required for system operation are also stored in the RAM 303. The CPU 301, ROM 302 and RAM 303 are connected to each other via a bus 304. An input / output (I / O) interface 305 is also connected to the bus 304.

[0061] The following components are connected to the I / O interface 305: an input section 306 including an audio input device, a push button switch, and the like; an output section 307 including a liquid crystal display (LCD), an audio output device, an indicator light, and the like; a storage section 308 including a hard disk and the like; and a communication section 309 including a network interface card such as a LAN (Local Area Network) card or a modem. The communication section 309 performs communication processing via a network such as the Internet. A drive 310 is also connected to the I / O interface 305 as needed. Removable media 311, such as a magnetic disk, an optical disk, a magneto-optical disk, or a semiconductor memory, is installed in the drive 310 as needed, so that computer programs read therefrom can be installed into the storage section 308 as needed.

[0062] In particular, according to an embodiment of the present invention, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present invention includes a computer program product that includes a computer program carried on a computer-readable medium, the computer program including a computer program for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network via the communication section 309 and / or installed from a removable medium 311. When the computer program is executed by the central processing unit (CPU) 301, the various functions defined in the present invention are performed.

[0063] It should be noted that specific examples of computer-readable storage media may include, but are not limited to, an electrical connection having one or more conductors, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM), a flash memory, an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present invention, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0064] The flowcharts and block diagrams in the accompanying drawings illustrate the possible architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present invention. Each box in the flowchart or block diagram can represent a module, program segment, or part of the code, and the above-mentioned module, program segment, or part of the code contains one or more executable instructions for implementing the specified logical functions. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in an order different from that marked in the accompanying drawings.

[0065] Specifically, the online detection system of this embodiment includes a processor and a memory. The memory stores a computer program. When the computer program is executed by the processor, the online detection method for the size of precision parts based on machine vision provided by the above embodiment is implemented.

[0066] As another aspect, the present invention further provides a computer-readable storage medium, which may be included in the online inspection system described in the above embodiments, or may exist independently and not be incorporated into the online inspection system. The storage medium carries one or more computer programs, which, when executed by a processor of the online inspection system, enable the online inspection system to implement the machine vision-based online inspection method for precision part dimensions provided in the above embodiments.

[0067] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present application.

[0068] As used in the above embodiments, the term “when…” may be interpreted to mean “if…” or “after…” or “in response to determining…” or “in response to detecting…”, depending on the context. Similarly, the phrases “upon determining…” or “if (stated condition or event) is detected” may be interpreted to mean “if determining…” or “in response to determining…” or “upon detecting (stated condition or event)” or “in response to detecting (stated condition or event)”, depending on the context.

[0069] Those skilled in the art will appreciate that all or part of the process steps in the above-described method embodiments can be implemented by a computer program instructing the relevant hardware. The program can be stored in a computer-readable storage medium, and when executed, the program can include the process steps in the above-described method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as ROM or random access memory (RAM), magnetic disks, or optical disks.

Claims

1. A method for online detection of precision parts dimensions based on machine vision, applied to an online detection system, characterized in that: The method comprises: The weight data of the precision parts to be tested is collected by a weighing sensor arranged at the feeding end of the conveyor belt, and the vibration acceleration signal of the precision parts to be tested is collected in real time by an acceleration sensor arranged at the detection station of the conveyor belt; Based on the weight data, the adjustable vibration-damping brackets provided on both sides of the conveyor belt are controlled. The adjustable vibration-damping brackets include multiple groups of piezoelectric ceramic drive units. By adjusting the expansion and contraction of each piezoelectric ceramic drive unit, a displacement in the opposite direction of the vibration acceleration signal is generated. The relative displacement data caused by vibration is obtained by using a grating encoder provided at the detection station, wherein the grating encoder is fixed on the conveyor belt base; Determining the minimum vibration moment when the precision part to be inspected passes through the inspection station according to the vibration acceleration signal and the relative displacement data; At the minimum vibration moment, a high-speed industrial camera disposed at the inspection station is triggered to perform image acquisition to obtain an image of the part; Using the relative displacement data to perform coordinate correction on the part image to eliminate image distortion caused by vibration, thereby obtaining a corrected target image; The edge features of the precision part are extracted based on the target image, and the size of the precision part is calculated according to preset calibration parameters.

2. The method according to claim 1, characterized in that Before the step of acquiring relative displacement data caused by vibration by using a grating encoder provided at the detection station, the method further includes: The natural frequency of the precision part to be inspected is obtained by an audio exciter and an acoustic wave sensor provided at the feeding end of the conveyor belt; According to the natural frequency, a plurality of piezoelectric ceramic vibrators arranged within a set range of the conveyor belt are controlled to generate anti-phase vibration waves of a set frequency, thereby forming destructive interference with the vibration of the parts; The distribution of magnetic material pre-coated on the surface of the part is detected by three groups of orthogonally arranged Hall sensor arrays set at the detection station to obtain the three-dimensional motion state of the part; The working parameters of the piezoelectric ceramic vibrator are adjusted according to the three-dimensional motion state of the part to achieve precise vibration cancellation.

3. The method according to claim 1, characterized in that After extracting edge features of the precision part based on the target image and calculating the size of the precision part according to preset calibration parameters, the method further includes: Establishing a part density distribution model based on the weight data and the size of the precision parts; Determining the center of gravity position and moment of inertia of the part based on the density distribution model; Inputting the center of gravity position and the moment of inertia data into a preset vibration prediction model to generate a vibration characteristic prediction value for the next detection cycle; According to the vibration characteristic prediction value, the control parameters of the piezoelectric ceramic drive unit are adjusted in advance.

4. The method according to claim 1, wherein After the step of triggering a high-speed industrial camera disposed at the inspection station to perform image acquisition at the minimum vibration moment to obtain a part image, the method further includes: Controlling the annular light source array set at the inspection station to emit light in sequence according to a preset time sequence, obtaining multiple sets of reflected light intensity data at different incident angles, and establishing a reflectivity distribution model of the part surface. The reflectivity distribution model characterizes the reflectivity characteristics of the part surface for light at different incident angles; Determining an optimal light source parameter combination according to the reflectivity distribution model, and controlling each LED light source in the annular light source array to perform multiple exposure acquisitions according to the optimal light source parameter combination to obtain a multi-frame enhanced image sequence; Performing registration and fusion processing on the multi-frame enhanced image sequence, calculating the normal vector distribution of the part surface using a photometric stereo vision algorithm, and reconstructing three-dimensional contour data of the part surface based on the normal vector distribution, wherein the three-dimensional contour data includes height information and tilt angle information of the part surface; Establishing a posture compensation model based on the three-dimensional contour data, performing time synchronization and spatial registration on the posture compensation model and the vibration displacement data collected by the grating encoder to generate comprehensive correction parameters, wherein the comprehensive correction parameters are used to compensate for the coupling effect of the part posture change and the vibration displacement; The part image is subjected to nonlinear correction based on the comprehensive correction parameters to obtain a high-precision corrected image.

5. The method according to claim 1, wherein After the step of collecting the vibration acceleration signal of the precision part to be inspected in real time by an acceleration sensor provided at the inspection station of the conveyor belt, the method further includes: Strain data is collected in real time by means of at least three rows of strain gauge arrays arranged along the direction of movement of the conveyor belt, wherein each row of the strain gauge array includes a plurality of strain gauges evenly distributed along the width of the conveyor belt, and the strain data represents the deformation of a local area of the conveyor belt; Calculating a strain difference based on the strain data collected by each two adjacent rows of the strain gauge arrays, and determining a bending curvature and a torsion angle of the conveyor belt according to the strain difference, wherein the bending curvature represents a degree of bending of the conveyor belt in a vertical direction, and the torsion angle represents a rotation angle of the conveyor belt around a longitudinal axis; Inputting the bending curvature and the torsion angle into a preset conveyor belt dynamic deformation prediction model to obtain a vibration state prediction result, wherein the vibration state prediction result includes the expected vibration parameters when the part arrives at the inspection station. The conveyor belt dynamic deformation prediction model is previously obtained through machine learning training based on the strain data of the conveyor belt under different working conditions and the vibration parameters at corresponding times; The vibration state prediction result is fused with the actual vibration acceleration signal collected by the acceleration sensor, and the control parameters of the piezoelectric ceramic drive unit are adaptively adjusted according to the fused vibration information to compensate for the vibration caused by the deformation of the conveyor belt in advance.

6. The method according to claim 1, characterized in that After the step of using the relative displacement data to perform coordinate correction on the part image to eliminate image distortion caused by vibration and obtain a corrected target image, the method further includes: The height profile data of the part surface is obtained by a through-beam laser ranging sensor set at the detection station. The height profile data is used to characterize the concave-convex changes of the part surface; The height profile data is fused with the corrected target image to obtain a comprehensive detection result including two-dimensional size and height information.

7. The method according to claim 1, characterized in that After the step of triggering a high-speed industrial camera disposed at the inspection station to perform image acquisition at the minimum vibration moment to obtain a part image, the method further includes: Divide the part image into multiple sub-regions, perform multi-scale wavelet decomposition on each sub-region, and extract high-frequency coefficients and low-frequency coefficients at each scale, wherein the high-frequency coefficients represent the edge and texture information of the image, and the low-frequency coefficients represent the overall grayscale distribution of the image; Establishing a part vibration trajectory model based on the relative displacement data, calculating the instantaneous velocity and acceleration components of each sub-region during the image acquisition process, and constructing a time-varying motion blur kernel function for each sub-region based on the velocity and acceleration components; Performing deconvolution operations on the high-frequency coefficients and the low-frequency coefficients of each sub-region using the motion blur kernel function, reconstructing each sub-region image through inverse wavelet transform, and splicing the reconstructed sub-region images to obtain a deblurred image, wherein the deblurred image represents the image of the part after eliminating the influence of vibration; The gradient amplitude and directional consistency index of the deblurred image are used as clarity evaluation parameters. When the clarity evaluation parameter is less than a preset threshold, the parameters of the motion blur kernel function are optimized using the gradient descent method, and the deconvolution and reconstruction steps are repeated until the clarity evaluation parameter is greater than or equal to the preset threshold.

8. An online detection system, characterized in that: The online detection system includes: one or more processors and a memory; the memory is coupled to the one or more processors, the memory is used to store computer program code, the computer program code includes computer instructions, and the one or more processors call the computer instructions to enable the online detection system to execute the method described in any one of claims 1 to 7.

9. A computer-readable storage medium comprising instructions, characterized in that: When the instruction is executed on an online detection system, the online detection system is caused to execute the method according to any one of claims 1 to 7.

10. A computer program product, characterized in that When the computer program product is run on an online detection system, the online detection system is enabled to perform the method according to any one of claims 1 to 7.

Images