An industrial vision product positioning guidance method

Abstract

The application provides an industrial vision product positioning guide method, comprising the following steps: selecting a shock absorber type and a damping coefficient according to a precision equipment vibration frequency characteristic of an industrial environment, matching a spring stiffness, and forming an anti-vibration structure meeting anti-vibration requirements of the precision equipment; constructing an environment three-dimensional point cloud model containing a shock absorber position according to extracted image feature information and in combination with a layout of an anti-vibration device, identifying and marking the shock absorber in the environment three-dimensional point cloud model; obtaining equipment vibration acceleration data based on the marking of the shock absorber in the environment three-dimensional point cloud model, estimating a vibration response of the anti-vibration device, and obtaining a corrected vibration displacement and speed; accurately compensating an original vibration signal according to a damping characteristic of the shock absorber, the corrected vibration displacement and speed, filtering the compensated signal through a digital filter, and obtaining a stable vibration signal.

Description

Technical Field

[0001] This invention relates to the field of information technology, and in particular to a product positioning guidance method based on industrial vision. Background Technology

[0002] In complex industrial environments, visual positioning systems face severe challenges. The continuous vibrations and shock waves generated by large machines in industrial operations interfere with precision equipment, causing image blurring and defocusing, severely impacting positioning accuracy. In such harsh environments, traditional visual positioning technologies struggle to meet the requirements of high precision and reliability. However, industrial operations have an urgent need for accurate positioning, especially in hazardous areas or remote control scenarios. Furthermore, the complex and ever-changing industrial environment often makes it difficult for on-site personnel to independently handle various emergencies. How to enable real-time guidance from remote experts to on-site personnel, providing accurate positioning and operational advice under limited communication conditions, is also a challenging problem. This involves not only optimizing communication technologies but also considering how to effectively transmit visual information in complex environments to ensure that remote experts can accurately understand the situation and provide appropriate guidance. Summary of the Invention

[0003] This invention provides an industrial vision-based product positioning guidance method, mainly comprising:

[0004] Based on the vibration frequency characteristics of precision equipment in the industrial environment, select the type of shock absorber and damping coefficient, and match the spring stiffness to form a shockproof structure that meets the vibration protection requirements of precision equipment.

[0005] An industrial camera is placed in a vibration-proof structure that meets the vibration requirements of precision equipment. Images of the vibrating equipment in the field are captured by the industrial camera. The images are preprocessed according to the characteristics of the vibration-proof material. The texture and shape features of the images are extracted by a convolutional neural network. At the same time, the network is fine-tuned by a large model to adapt to the changes in the characteristics of the vibrating equipment in the current industrial environment and to extract image feature information.

[0006] Based on the extracted image feature information and the layout of the shock absorber, a three-dimensional point cloud model of the environment containing the location of the shock absorber is constructed, and the shock absorbers in the three-dimensional point cloud model of the environment are identified and marked.

[0007] Based on the markers of the shock absorbers in the 3D point cloud model of the environment, the vibration acceleration data of the equipment is obtained, the vibration response of the shock absorber is estimated, and the corrected vibration displacement and velocity are obtained.

[0008] Based on the damping characteristics of the shock absorber, the corrected vibration displacement and velocity, the original vibration signal is precisely compensated. The compensated signal is then filtered by a digital filter to obtain a stable vibration signal. Anomaly analysis is performed on the stable vibration signal to identify abnormal vibration modes. Based on the identification results, an equipment fault early warning model is established. Based on the vibration anomaly information in the fault early warning model, combined with the position of the shock absorber in the three-dimensional point cloud model of the environment, the early warning information and on-site vibration data are transmitted to a remote expert terminal in real time.

[0009] The received warning information and on-site vibration data are decompressed and restored. The equipment is remotely monitored and diagnosed through vibration status and fault warning information. If data transmission abnormalities or equipment failures occur, an alarm is triggered according to the abnormality handling mechanism, and the equipment is shut down or maintained, thus realizing product positioning guidance based on industrial vision.

[0010] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects:

[0011] This invention discloses an industrial vision-based product positioning guidance method. This method selects appropriate shock absorbers and vibration-damping structures, and based on images of vibrating equipment acquired by an industrial camera, utilizes convolutional neural networks and large-scale model fine-tuning to extract image features and construct a 3D point cloud model of the environment, including the location of the shock absorbers. Based on this model, this invention performs precise compensation and filtering of vibration signals to achieve stable vibration signal acquisition. By performing anomaly analysis on the signals, an equipment fault early warning model is established, and the early warning information and vibration data are transmitted in real time to a remote expert terminal, enabling remote monitoring and diagnosis of the equipment. When an anomaly occurs, this invention can promptly issue alarms and handle emergencies, effectively ensuring the safe operation of precision equipment and improving the stability and reliability of industrial production. Attached Figure Description

[0012] Figure 1 This is a flowchart of an industrial vision-based product positioning guidance method according to the present invention.

[0013] Figure 2 This is a schematic diagram of an industrial vision-based product positioning guidance method according to the present invention.

[0014] Figure 3 This is another schematic diagram of an industrial vision-based product positioning guidance method according to the present invention. Detailed Implementation

[0015] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this specification, and not all embodiments. Based on the embodiments in this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this specification.

[0016] like Figure 1-3 This embodiment of an industrial vision-based product positioning guidance method may specifically include:

[0017] Step S101: Based on the vibration frequency characteristics of precision equipment in the industrial environment, select the type of shock absorber and the damping coefficient, and match the spring stiffness to form a shock-resistant structure that meets the vibration protection requirements of precision equipment.

[0018] The system receives a data acquisition command carrying a unique identifier for precision equipment. This command includes vibration frequency characteristic data, equipment weight data, and installation space data, while simultaneously recording the temperature, humidity, and corrosive gases of the industrial environment. Based on the vibration frequency characteristic data, the system calculates the main frequency range. This main frequency range, combined with the equipment weight data, installation space data, and environmental characteristic parameters, is used to select the type of vibration damper. For the selected vibration damper type, a genetic algorithm is used to optimize the damping coefficient. The objective function of the genetic algorithm is set to minimize the system vibration transmissibility. The damping coefficient obtained from the genetic algorithm optimization is obtained. This damping coefficient, combined with the vibration damper type data, is used to establish a finite element model of the vibration damping system. A bisection method is used to optimize the spring stiffness. The convergence condition of the bisection method is set to the difference between the system's natural frequency and the equipment vibration frequency being less than a preset threshold. A mathematical model of the anti-vibration structure is constructed from the vibration damper type, damping coefficient, and optimal spring stiffness. Time-domain response analysis is performed in the mathematical model of the anti-vibration structure. The time-domain response analysis selects a random vibration excitation signal that conforms to the characteristics of the industrial environment to calculate the displacement and acceleration responses of the structure.

[0019] Specifically, the process involves acquiring vibration frequency characteristic data, equipment weight data, and installation space data for the precision equipment, while simultaneously recording the temperature, humidity, and corrosive gases of the industrial environment. Based on the vibration frequency characteristic data, the main frequency range is calculated. Combining this with the equipment weight data, installation space data, and environmental characteristic parameters, a suitable vibration damper type is selected. For the selected vibration damper type, a genetic algorithm is used to optimize and calculate the damping coefficient, with the objective function set to minimize the system vibration transmissibility. The optimized damping coefficient value is combined with the vibration damper type data, and a finite element model of the vibration damping system is established using ANSYS software. Based on the established finite element model, modal analysis is used to calculate the system's natural frequency, which is then compared with the vibration frequency of the precision equipment. A bisection method is used to optimize the spring stiffness, with the convergence condition set as the difference between the system's natural frequency and the equipment's vibration frequency being less than a preset threshold. A mathematical model of the seismic-resistant structure is constructed based on data from the vibration damper type, damping coefficient, and optimal spring stiffness. This paper implements time-domain response analysis in MATLAB, selecting a random vibration excitation signal that conforms to the characteristics of an industrial environment. The displacement and acceleration responses of the structure are calculated, and evaluation indicators include maximum displacement, maximum acceleration, and root mean square acceleration. Based on the time-domain response analysis results, it is determined whether the vibration damping effect of the anti-vibration structure on the precision equipment meets the preset requirements. If not, the process returns to the damper type selection step and parameter optimization is performed again. If the requirements are met, the final damper type, damping coefficient, and spring stiffness values ​​are output. When acquiring data from the precision equipment, an accelerometer is used to measure the vibration frequency characteristics, recording the frequency response within the 0-1000Hz range. The equipment weighs 500kg, and the installation space is 1m×1m×0.5m. The industrial environment temperature is 35℃, humidity is 80%, and trace amounts of hydrogen sulfide gas are present. The calculated main frequency range is 20-80Hz. Based on these parameters, a spring-damped damper is selected as the most suitable type. When using a genetic algorithm to optimize the damping coefficient, the population size is set to 100, the number of iterations is 50, the crossover probability is 0.8, and the mutation probability is 0.1. The objective function was defined as minimizing the average transmissibility within the frequency range of 20-80Hz. The optimized damping coefficient was 0.3. A finite element model of the vibration reduction system was established in ANSYS, with a mesh size of 0.01m and a total of approximately 10,000 elements. Modal analysis yielded the first three natural frequencies of the system as 15Hz, 45Hz, and 85Hz. The spring stiffness was optimized using a bisection method, with an initial search range of 10^4-10^6 N / m. The convergence condition was set to the difference between the system's first natural frequency and the equipment's main frequency being less than 1Hz. After 10 iterations, the optimal spring stiffness was obtained as 3.5×10^5 N / m. Time-domain response analysis was performed in MATLAB, using Gaussian white noise as the excitation signal, with a bandwidth of 0-200Hz and a power spectral density of 0.1g^2 / Hz. The structural response over 100s was calculated, with a sampling frequency of 1000Hz.The evaluation index calculation results are: maximum displacement = 0.5mm, maximum acceleration = 0.8g, and root mean square acceleration = 0.3g. Based on the preset vibration reduction requirements of maximum displacement < 1mm and maximum acceleration < 1g, the vibration-damping structure is judged to meet the requirements. The final output vibration damper type is a spring-damped vibration damper with a damping coefficient of 0.3 and a spring stiffness of 3.5 × 10^5 N / m.

[0020] Step S102: Place an industrial camera according to the anti-vibration structure that meets the anti-vibration requirements of precision equipment, and collect images of the vibrating equipment in the field through the industrial camera. Preprocess the images according to the characteristics of the anti-vibration material, extract the texture and shape features of the images through a convolutional neural network, and fine-tune the network through a large model fine-tuning method to adapt it to the changes in the characteristics of the vibrating equipment in the current industrial environment, and extract image feature information.

[0021] The optimal shooting distance and angle of the industrial camera are calculated based on the frequency response characteristics of the vibration-damping structure. The installation position of the industrial camera is adjusted to acquire an image sequence of the vibrating equipment. The image sequence is denoised using a median filtering algorithm to obtain a first image. The first image is then contrast-enhanced using an adaptive histogram equalization method to obtain a second image. A multi-layer convolutional neural network structure is constructed, and initial weights are imported using a ResNet50 pre-trained model. The convolutional neural network is fine-tuned using the second image to obtain a fine-tuned convolutional neural network. Multi-scale feature maps are extracted from the fine-tuned convolutional neural network. The feature covariance matrix of the multi-scale feature maps is calculated. Eigenvalue decomposition is performed on the feature covariance matrix. Principal components with contribution rates exceeding a preset threshold are selected to obtain dimensionality-reduced feature vectors. The feature vectors are grouped using a k-means clustering algorithm. If the grouping results of the feature vectors meet preset conditions, the feature vectors are determined to be low-dimensional feature representations characterizing the feature changes of the vibrating equipment under the current industrial environment.

[0022] Specifically, the optimal shooting distance and angle of the industrial camera are calculated based on the frequency response characteristics of the vibration-damping structure. The camera installation position is adjusted, and image sequences of the vibrating equipment are acquired using the industrial camera. Image resolution, frame rate, and exposure time are obtained, and the type, thickness, and density of the vibration-damping material are recorded. Median filtering is used to reduce noise in the acquired image sequences, and adaptive histogram equalization is used to enhance image contrast. Sharpening filters of different intensities are selected based on the damping ratio of the vibration-damping material to improve edge details. A multi-layer convolutional neural network structure is constructed, with parameters set for convolutional, pooling, and fully connected layers. Initial weights are imported using a ResNet50 pre-trained model, and the network is fine-tuned using a dataset of vibration-damping equipment images from an industrial environment. The last few layers of the network structure are adjusted to suit the characteristics of the vibration-damping equipment. Multi-scale feature maps are extracted from the fine-tuned convolutional neural network, the feature covariance matrix is ​​calculated, eigenvalue decomposition is performed, and principal components with contribution rates exceeding a preset threshold are selected to obtain dimensionality-reduced feature vectors. The feature vectors are grouped using a k-means clustering algorithm to obtain a low-dimensional feature representation of the characteristic changes of the vibration-damping equipment under the current industrial environment. Leave-one-out cross-validation was used to evaluate the ability of the extracted feature vectors to represent the changes in the characteristics of vibrating equipment under different operating conditions. The classification accuracy of the feature vectors under different conditions was calculated to verify the effectiveness of the feature extraction method. In an industrial environment, assuming the main frequency response of the vibration-damping structure is 20-80Hz, the optimal shooting distance for the industrial camera was calculated to be 1.5 meters, with the angle perpendicular to the surface of the vibrating equipment. The camera acquisition parameters were set as follows: resolution = 1920×1080 pixels, frame rate = 60fps, and exposure time = 1 / 250 second. The vibration-damping material recorded was a rubber vibration damping pad with a thickness of 25mm and a density of 1.2g / cm³. 3 Median filtering with a 5×5 window size was applied to the acquired image sequences to eliminate salt-and-pepper noise. Adaptive histogram equalization used an 8×8 local region to improve image contrast. Based on the damping ratio of the rubber vibration damping pad (0.15), a Laplacian sharpening filter with a kernel size of 3×3 was selected to enhance edge details. The constructed convolutional neural network contained 5 convolutional layers, each followed by a max-pooling layer, and finally 3 fully connected layers. A ResNet50 model pre-trained on ImageNet was used as the initial weights. During fine-tuning, the first 40 layers were frozen, and the last 10 layers were retrained. The fine-tuning dataset contained 5000 images of vibration equipment under different operating conditions. Feature maps of the 3rd, 4th, and 5th convolutional layers were extracted from the fine-tuned network, resulting in a 2048-dimensional feature vector. The covariance matrix of the feature vectors was calculated, and eigenvalue decomposition was performed. The top 100 principal components with a cumulative contribution rate of 95% were selected. The k-means clustering algorithm was used, with k=5, to group the dimensionality-reduced features. The feature representation capability was evaluated using leave-one-out cross-validation, and the classification accuracy reached 92.5% under five different working conditions.

[0023] Step S103: Based on the extracted image feature information and the layout of the shock absorber, construct an environmental three-dimensional point cloud model containing the location of the shock absorber, and identify and mark the shock absorber in the environmental three-dimensional point cloud model.

[0024] A multi-frame environmental depth map is acquired using a structured light scanner. A dense point cloud data sequence is generated based on the multi-frame environmental depth map. This dense point cloud data sequence is obtained by mapping two-dimensional image pixels and their features to three-dimensional spatial coordinates using a back-projection algorithm. The dense point cloud data sequence is then registered to obtain a three-dimensional environmental point cloud model. The registration process employs an iterative nearest-point algorithm; iteration stops if the number of iterations reaches a preset maximum or the convergence error is less than a preset convergence threshold. Local geometric features and texture features are extracted from the three-dimensional environmental point cloud model. The local geometric features include… The texture features include normal vectors and curvature, and include spin maps and fast point feature histograms. A support vector machine (SVM) classifier is used to classify each point in the 3D point cloud model of the environment. The SVM classifier uses a radial basis function kernel and is trained on a labeled shock absorber point cloud dataset. Region growing is performed on the points classified as shock absorbers in the 3D point cloud model of the environment to obtain complete shock absorber entities. If the distance between adjacent points is less than a preset growth threshold, the adjacent point is added to the shock absorber entity. Spatial labeling is performed on the shock absorber entities using the minimum bounding box method.

[0025] Specifically, based on the extracted image feature information, a structured light scanner is used to acquire multiple frames of environmental depth maps. A back-projection algorithm maps the pixels and features of the two-dimensional images to three-dimensional spatial coordinates, generating a dense point cloud data sequence containing shock absorber location information. The iterative nearest-point algorithm is used to register the multiple frames of point cloud data, setting the maximum number of iterations to 50 and the convergence threshold to 0.001 meters to eliminate positional deviations during the scanning process. This results in a complete 3D environmental point cloud model while preserving the color and depth information of each point. Local geometric and texture features are extracted from the 3D environmental point cloud model, and the normal vector, curvature, spin map, and fast point feature histogram for each point are calculated to construct a high-dimensional feature vector database. A support vector machine (SVM) classifier is used to classify each point in the point cloud, trained using a labeled shock absorber point cloud dataset. A radial basis function kernel function is selected, and the penalty parameter C is set to 10 to identify the shock absorber region. A region growing algorithm is used to segment the complete shock absorber entity, setting the growth threshold to 0.05 meters. Finally, minimum bounding boxes are used to spatially label the identified shock absorbers. By combining the pre-acquired layout map of the shock absorber devices, the identification results are verified and corrected to improve the accuracy of the shock absorber markings. In an industrial environment, a structured light scanner with a resolution of 1920×1080 pixels is used to acquire a sequence of environmental depth maps at a rate of 30 frames per second. Each frame of the depth map is converted into a point cloud using a back-projection algorithm, while the previously extracted SIFT feature points are mapped to their corresponding 3D coordinates. The iterative nearest-point algorithm is used to register consecutive frame point clouds, with a maximum number of iterations of 50, a convergence threshold of 0.001 meters, and an average registration time of 0.5 seconds per frame. The fused complete point cloud model contains approximately 5 million points, each storing XYZ coordinates, RGB color values, and depth information. Local features are calculated for each point in the point cloud model, including normal vectors, curvature, spin maps, and fast point feature histograms. The total size of the constructed feature vector database is approximately 2GB. A support vector machine (SVM) classifier was trained using the libsvm library. The training set contained 10,000 labeled point cloud samples of shock absorbers and 40,000 non-shock absorber samples. A radial basis function (RBF) kernel was selected, and the penalty parameter C was set to 10. The cross-validation accuracy reached 95%. Region growing was performed on points classified as shock absorbers, with a growth threshold of 0.05 meters, resulting in an average of 2,000 points per shock absorber. Principal component analysis (PCA) was used to calculate the minimum bounding box, accurately locating the spatial position and orientation of the shock absorbers. Finally, the recognition results were compared with a pre-drawn CAD layout map with an accuracy of 0.01 meters. Markers with errors exceeding 0.1 meters were corrected, achieving a final shock absorber positioning accuracy of ±0.05 meters.

[0026] By stitching together depth image features from different perspectives, a complete 3D point cloud model of the environment surrounding the shock absorber is obtained. Based on a pre-established 3D CAD model library of shock absorbers, regions similar to the CAD models are searched in the environmental point cloud to preliminarily determine the coarse location of the shock absorber. A subset of the point cloud representing the coarse location of the shock absorber is extracted, and the precise boundary and pose information of the shock absorber are further refined using a point cloud segmentation algorithm, marking the shock absorber as identified.

[0027] Depth images of the environment surrounding the shock absorber are acquired using a multi-view depth camera. Bilateral filtering is applied to these depth images to remove noise, resulting in filtered depth images. Based on these filtered depth images, the transformation matrix between adjacent viewpoints is calculated using SIFT feature point matching and the RANSAC algorithm. An iterative nearest-point algorithm is then used to register and fuse multiple depth images, yielding a 3D point cloud model of the environment. Key geometric features, including surface normal vector distribution, curvature distribution, and local shape descriptors, are extracted from a pre-established 3D CAD model library for the shock absorber, constructing feature description vectors. Regions similar to these feature description vectors are searched in the 3D point cloud model using a sliding window. If high-similarity regions are identified, cluster analysis is performed on these regions. For the point cloud subsets obtained from the cluster analysis, a region growing algorithm is applied for fine segmentation. The precise boundary of the shock absorber is obtained through iterative expansion. Principal component analysis is used to calculate the principal axis direction and center position of the shock absorber, determining its precise pose.

[0028] Specifically, multi-view depth cameras are used to acquire depth images of the environment surrounding the shock absorber. Bilateral filtering is applied to each depth image to remove noise, and morphological operations are used to fill small occluded areas. SIFT feature point matching and the RANSAC algorithm are used to calculate the transformation matrix between adjacent views. An iterative nearest-point algorithm is used to register and fuse multiple depth images to obtain a complete 3D point cloud model of the environment. Key geometric features, including surface normal vector distribution, curvature distribution, and local shape descriptors, are extracted from a pre-established 3D CAD model library of the shock absorber to construct feature description vectors. In the environmental point cloud model, a sliding window is used to search for regions similar to the features of the CAD model. The window size is set to 1.5 times the shock absorber size, and the step size is 1 / 4 of the window size. Cosine similarity is used to calculate the similarity between feature vectors. Cluster analysis is performed on the identified high-similarity regions. The density clustering algorithm DBSCAN is used to determine the potential location of the shock absorber, with a neighborhood radius of 0.1 meters and a minimum number of points of 50. A subset of the point cloud within a fixed radius around each cluster center is extracted as a coarse location of the shock absorber. For the extracted point cloud subset, a region growing algorithm was applied for fine segmentation, with a normal vector angle threshold of 10 degrees, a curvature change threshold of 0.05, and a neighborhood search radius of 0.02 meters. The precise boundary of the shock absorber was obtained through iterative expansion. Principal component analysis was used to calculate the principal axis direction and center position of the shock absorber to determine its precise pose. The segmentation results were precisely matched with the CAD model using the iterative closestpoint algorithm, and a geometric consistency score was calculated. If the score was higher than a preset threshold, the shock absorber was marked as identified. In an industrial environment, three 1280x720 resolution depth cameras were used to acquire depth images of the environment surrounding the shock absorber at a rate of 15 frames per second. Each depth image was first subjected to bilateral filtering with a 5x5 window size, a spatial standard deviation of 10, and a range standard deviation of 0.1 to remove noise, and then a 3x3 structuring element closing operation was applied to fill small areas of occlusion. Approximately 1000 feature points were extracted from each image using the SIFT algorithm. Feature matching was performed using FLANN (Fast Nearest Neighbor) search, and outlier matching points were removed using the RANSAC algorithm. The transformation matrix between adjacent viewpoints was calculated. Point cloud registration was performed using an iterative nearest-neighbor algorithm, with a maximum iteration count of 50 and a convergence threshold of 0.001 meters, resulting in a complete 3D environmental point cloud model containing approximately 5 million points. Geometric features of 10 common shock absorbers were extracted from a CAD model library, and a 128-dimensional FPFH feature descriptor was calculated for each model. In the environmental point cloud, a 0.3m x 0.3m x 0.3m sliding window with a step size of 0.075 meters was used to search for similar regions. A cosine similarity threshold of 0.85 was set, resulting in approximately 100 candidate regions. The DBSCAN algorithm was applied to these regions with a parameter ε of 0.1 meters and MinPts of 50, yielding 10-20 clusters.A subset of point clouds within a 0.2-meter radius is extracted from each cluster center. A region growing algorithm is used for fine segmentation, with a normal vector angle threshold of 10 degrees, a curvature change threshold of 0.05, and a neighborhood radius of 0.02 meters. The principal axis direction and center position are calculated using PCA to obtain the 6-DOF pose of the shock absorber. Finally, the ICP algorithm is used to match the segmentation results with the CAD model, setting a maximum of 100 iterations and a convergence threshold of 0.0001 meters. If the root mean square error after matching is less than 0.005 meters, the shock absorber is marked as identified, achieving sub-millimeter accuracy.

[0029] Step S104: Based on the markers of the shock absorbers in the three-dimensional point cloud model of the environment, obtain the vibration acceleration data of the equipment, estimate the vibration response of the shock absorber, and obtain the corrected vibration displacement and velocity.

[0030] The positions of the shock absorbers are determined based on the 3D point cloud model of the environment. After obtaining the positions, a triaxial accelerometer is installed on each shock absorber. A fast Fourier transform algorithm is used to perform spectral analysis on the acceleration signals collected by the triaxial accelerometers. From the spectral analysis results, the main vibration frequencies and amplitudes of each shock absorber in the x, y, and z directions are extracted to construct vibration feature vectors. Based on a pre-established mass-spring-damping dynamic model of the shock absorber, the vibration feature vectors are used as input, and the vibration response of the shock absorber is estimated in real time using a Kalman filter algorithm to obtain the instantaneous displacement and velocity states of each shock absorber. A Bayesian fusion algorithm is used to integrate the instantaneous displacement and velocity states with the shock absorber position information in the 3D point cloud model of the environment, and the instantaneous displacement and velocity states are corrected using a weighted least squares method. A high-speed camera is used to collect the motion trajectory of the marked points on the surface of the shock absorbers, and the positions of the marked points are extracted from the motion trajectory. The positions of the marked points are compared with the corrected instantaneous displacement and velocity states to determine the accuracy of the corrected instantaneous displacement and velocity states.

[0031] Specifically, based on the marked positions of the shock absorbers in the 3D point cloud model of the environment, a triaxial accelerometer is installed on each shock absorber. A high-precision data acquisition card is used to synchronously acquire the vibration acceleration data of each shock absorber at a sampling frequency of 10kHz. The raw signal is preprocessed using a bandpass filter of 10Hz-1kHz to filter out high-frequency noise and low-frequency drift, and the processed data is stored as a time series matrix. The preprocessed acceleration signal is then subjected to spectral analysis using a Fast Fourier Transform algorithm to extract the main vibration frequencies and amplitudes of each shock absorber in the x, y, and z directions, constructing vibration feature vectors. Principal component analysis is used to reduce the feature dimensionality, calculating eigenvalues ​​and eigenvectors, and selecting the top principal components with a cumulative contribution rate of 95% as the dimensionality-reduced features. Based on a pre-established mass-spring-damping dynamic model of the shock absorber, the dimensionality-reduced vibration feature vectors are used as input, and the vibration response of the shock absorber is estimated in real time using a Kalman filter algorithm to obtain the instantaneous displacement and velocity state of each shock absorber. A Bayesian fusion algorithm was employed to integrate the estimated displacement and velocity with the shock absorber position information from the 3D point cloud model of the environment. Weighted least squares was used to correct the vibration displacement and velocity, with weights determined based on the measurement error covariance matrix. A high-speed camera was used to capture the motion trajectories of marked points on the shock absorber surface at a rate of 1000 frames per second. Image processing algorithms were used to extract the positions of these marked points, which were then compared with the corrected vibration displacement and velocity to verify the accuracy of the results. In an industrial environment, a PCB356A15 triaxial accelerometer with a sensitivity of 100mV / g was installed on each of the four shock absorbers located according to the point cloud model. A NIPXIe-4499 data acquisition card was used, with a sampling rate of 10kHz and a acquisition duration of 60 seconds. The acquired raw signals were processed using a Butterworth bandpass filter with a passband of 10Hz-1kHz and an order of 4. The filtered data was organized into a 12-column time series matrix representing the four shock absorbers and three directions. An 8192-point FFT analysis was performed on each data column to obtain the spectrum in the range of 0-5kHz. The top 5 frequencies with the largest amplitudes in each direction and their amplitudes were extracted to form a 4×3×5=60-dimensional feature vector. Dimensionality reduction was performed using PCA, and the top 10 principal components with a cumulative contribution rate of 95% were selected. Based on a single-degree-of-freedom vibration model with a mass of 50kg, a spring stiffness of 5×10^5N / m, and a damping coefficient of 1000N·s / m, a state-space equation was constructed. A Kalman filter was used, with the process noise covariance Q set as a diagonal matrix [1e-6, 1e-4] and the measurement noise covariance R as 1e-4, updating the state estimate at a frequency of 100Hz. A Bayesian fusion algorithm was used to fuse the displacement and velocity estimated by the Kalman filter with the position information in the point cloud model, with the fusion weights determined according to their respective uncertainties. Weighted least squares was used for final correction, and the weight matrix was determined by the inverse covariance matrix of the measurement error.Meanwhile, a Phantom V2512 high-speed camera, with a resolution of 1280×800 pixels and a frame rate of 1000fps, was used to capture images of 4mm diameter reflective markers on the surface of the shock absorber. The trajectory of the markers was extracted through threshold segmentation and centroid calculation, and compared with the corrected displacement; the average error was less than 0.1mm.

[0032] Based on different types and locations of vibration dampers, a vibration transmission model is established. By combining the stiffness and damping parameters of the vibration dampers, the propagation and attenuation characteristics of vibration signals in the equipment structure are simulated to obtain comprehensive equipment vibration response evaluation results.

[0033] A geometric model of the equipment structure is constructed using the finite element method, simplifying the shock absorber into a spring-damping element. The stiffness and damping parameters of the shock absorber are measured using a dynamic stiffness tester to obtain its physical characteristics. Based on these physical characteristics, corresponding physical parameters are assigned to each element in the geometric model. Modal analysis is used to calculate the natural frequencies and mode shapes of the geometric model, obtaining the dynamic characteristics of the equipment structure. A vibration transfer function is constructed, converting the input vibration signal into responses at key nodes to obtain the vibration propagation characteristics of the equipment structure. The vibration propagation characteristics are fitted using the response surface methodology to obtain the vibration response of the equipment structure under different operating conditions. A central composite experimental design method is used to determine the sampling points for the response surface methodology, obtaining the mapping relationship between the vibration response and the shock absorber parameters and vibration input. Monte Carlo simulation samples are generated using Latin hypercube sampling, and statistical analysis is performed on the mapping relationship to obtain a comprehensive evaluation result of the equipment structure's vibration response.

[0034] Specifically, based on the type and installation location of the vibration damper, a geometric model of the equipment structure is constructed using the finite element method. The vibration damper is simplified into a spring-damped element. Key structural nodes are determined by calculating the modal participation factor, and mass elements are placed at these nodes, forming a complete vibration transmission network through node connections. The stiffness and damping parameters of the vibration damper are measured using a dynamic stiffness tester. For rubber vibration dampers, nonlinear parameters are introduced to characterize their properties. Combined with structural material parameters from a material property database, corresponding physical properties are assigned to each element in the finite element model. Modal analysis is used to calculate the natural frequencies and mode shapes of the structure. Based on structural dynamics theory, a vibration transfer function is constructed. The vibration signal at the input end is converted into the response of each key node through the transfer function. Numerical integration is performed using the fourth-order Runge-Kutta method, with the time step set to 1 / 20 of the reciprocal of the highest frequency of interest. The time-domain response is solved to obtain the propagation and attenuation characteristics of vibration in the structure. The vibration response under different operating conditions was fitted using the response surface methodology. A central composite experimental design method was employed to determine sampling points. A second-order polynomial was used to fit the mapping relationship between the vibration response and the damper parameters and vibration input. 10,000 Monte Carlo simulation samples were generated through Latin hypercube sampling. Statistical analysis yielded a comprehensive evaluation of the equipment's vibration response, including the maximum response value, root mean square value, and frequency response function of each key node. In the vibration analysis of a certain industrial equipment, a finite element model of the equipment was constructed using ANSYS Workbench, containing 23,456 elements and 34,567 nodes. Four dampers were installed on the equipment: two metal spring dampers and two rubber dampers. The parameters of the shock absorbers were measured using an MTS831.50 dynamic stiffness tester. The stiffness of the metal spring shock absorber was 5 × 10⁵ N / m, and the damping coefficient was 1000 N·s / m. For the rubber shock absorber, a nonlinear parameter was introduced, with the stiffness expressed as k = k₀ + k₁x + k₂x², where k₀ = 3 × 10⁵ N / m, k₁ = 1 × 10⁶ N / m², and k₂ = 5 × 10⁷ N / m³. By calculating the modal participation factor, 15 key nodes with a contribution rate exceeding 5% were selected. Modal analysis was performed, yielding the first three natural frequencies of 15 Hz, 28 Hz, and 43 Hz. A vibration transfer function was constructed, with input random vibrations from 0 to 200 Hz and a PSD of 0.1 g² / Hz. The fourth-order Runge-Kutta method was used to solve the time-domain response, with a time step of 0.25 ms. A central composite test design was employed with 23 operating points, and a second-order polynomial was used to fit the response surface. 10,000 Monte Carlo simulation samples were generated using Latin hypercube sampling, and statistical analysis was performed to obtain the maximum response value, root mean square value, and frequency response function of each key node. The final evaluation results show that, within the 95% confidence interval, the maximum vibration displacement of the equipment does not exceed 0.8 mm, and the maximum acceleration does not exceed 5 g.

[0035] Step S105: Based on the damping characteristics of the shock absorber, the corrected vibration displacement and velocity, the original vibration signal is precisely compensated, and the compensated signal is filtered by a digital filter to obtain a stable vibration signal.

[0036] Obtain the damping characteristic curve of the shock absorber, and establish a nonlinear mapping relationship between damping force and vibration displacement and velocity based on the damping characteristic curve; process the nonlinear characteristics using a piecewise linearization method, dividing the damping characteristic curve into multiple linear intervals; fit the linear intervals using the least squares method to obtain a local linear damping characteristic function; substitute the corrected vibration displacement and velocity into the local linear damping characteristic function to calculate the actual damping force; convert the original vibration signal to the frequency domain using a fast Fourier transform, and construct a frequency domain compensation function based on the actual damping force; correct the amplitude of the original signal, where the amplitude compensation coefficient is the ideal damping force. The ratio of the damping force to the actual damping force is used; the compensated frequency domain signal is converted back to the time domain using inverse fast Fourier transform to obtain the preliminarily compensated time domain vibration signal; the preliminarily compensated time domain vibration signal is analyzed by wavelet decomposition at multiple scales to identify and remove high-frequency noise and unstable components introduced by the compensation; a Butterworth low-pass filter is designed to filter the denoised compensated signal to eliminate residual high-frequency interference; the variance ratio and signal-to-noise ratio improvement of the signal before and after filtering are calculated to evaluate the stability of the compensated and filtered signals; if the variance ratio decreases by more than a preset threshold and the signal-to-noise ratio improvement exceeds a preset threshold, a stable vibration signal is determined to be obtained.

[0037] Specifically, based on the damping characteristic curve of the shock absorber, a nonlinear mapping relationship between damping force and vibration displacement and velocity is established. For rubber shock absorbers, a piecewise linearization method is used to handle the nonlinear characteristics, dividing the characteristic curve into multiple linear intervals. Within each interval, a local linear damping characteristic function is obtained by fitting using the least squares method. The corrected vibration displacement and velocity are substituted into the characteristic function of the corresponding interval to calculate the actual damping force. The original vibration signal is converted to the frequency domain using a fast Fourier transform. Based on the calculated actual damping force, a frequency domain compensation function is constructed to correct the amplitude of the original signal. The amplitude compensation coefficient is the ratio of the ideal damping force to the actual damping force. Phase compensation is achieved by calculating the phase difference between the ideal damping force and the actual damping force. To avoid introducing discontinuities through compensation, a smoothing window function is used to smooth the compensation function. The compensated frequency domain signal is converted back to the time domain using an inverse fast Fourier transform to obtain a preliminarily compensated time domain vibration signal. Multi-scale analysis of the signal is performed using 3-level db4 wavelet decomposition to identify and remove high-frequency noise and unstable components introduced by compensation. An 8th-order Butterworth low-pass filter was designed, with the cutoff frequency set to 1.5 times the main frequency component of the signal. This filter was used to filter the denoised compensated signal, eliminating residual high-frequency interference. The stability of the compensated and filtered signals was evaluated by calculating the variance ratio and signal-to-noise ratio (SNR) improvement before and after filtering. A stable vibration signal was considered obtained if the variance ratio decreased by more than 50% and the SNR improved by more than 6 dB. In the vibration analysis of industrial equipment, the nonlinear damping characteristic curve of the shock absorber was measured using an MTS831.50 dynamic stiffness tester. The curve was divided into five linear intervals, and a local linear function was obtained within each interval using the least squares method. The corrected peak vibration displacement was 0.5 mm, and the peak velocity was 0.05 m / s. Substituting these values ​​into the corresponding interval function, the actual damping force was calculated to be 850 N. The original 10kHz sampling rate vibration signal was converted to the frequency domain using an 8192-point FFT, and a frequency domain compensation function was constructed. The amplitude compensation coefficient decreased from 1.2 to 1.0 within the 0-100Hz range, and the phase compensation decreased from 15° to 0° within the 0-100Hz range. A Hanning window function was used to smooth the compensation function, with a window length of 5% of the compensation function length. The compensated frequency domain signal was converted back to the time domain using an IFFT to obtain the preliminarily compensated vibration signal. A 3-level db4 wavelet decomposition was performed using the Matlab wavelet toolbox to remove coefficients in the highest frequency band whose absolute values ​​were less than 1.5 times the mean. An 8th-order Butterworth low-pass filter with a cutoff frequency of 150Hz was designed, and zero-phase filtering was performed using the `filtfilt` function. The variance ratio of the signal before and after filtering was calculated to be 0.45, and the signal-to-noise ratio improvement was 8.3dB, indicating a stable vibration signal. The final stable vibration signal had a peak-to-peak value of 1.2mm and an amplitude of 0.4mm at a dominant frequency of 100Hz.

[0038] Step S106: Perform anomaly analysis on the stabilized vibration signal to identify abnormal vibration modes, and establish an equipment fault early warning model based on the identification results. Based on the vibration anomaly information in the fault early warning model, and combined with the location of the shock absorbers in the environmental 3D point cloud model, transmit the early warning information and on-site vibration data to a remote expert terminal in real time.

[0039] After obtaining a stable vibration signal, time-frequency analysis is performed on the signal. A three-level wavelet packet transform is conducted using db4 wavelets to obtain the time-frequency characteristics of the vibration signal. Based on these characteristics, the energy percentage and entropy of eight sub-bands are calculated to construct a 16-dimensional feature vector. Principal component analysis is used to reduce the dimensionality of the feature vector, resulting in low-dimensional features representing the vibration mode. Anomaly detection is performed on these low-dimensional features using the isolated forest algorithm to identify abnormal vibration modes. These abnormal vibration modes are matched against a pre-established fault type library to determine potential fault types and severity. Based on the abnormal vibration modes and fault types, a multi-class fault warning model based on an RBF kernel support vector machine is constructed. The precise position coordinates of the shock absorber are extracted from the 3D point cloud model of the environment. The position coordinates and the output of the fault warning model are fused using a Kalman filter to obtain fused warning information. OPC is then used to perform the final warning. The UA protocol transmits the fused early warning information and real-time vibration data to a remote expert terminal via an industrial Ethernet network. If the remote expert terminal receives the early warning information and real-time vibration data, it uses the LZ4 algorithm to compress the early warning information and real-time vibration data in real time. The compressed data is then encrypted using the AES-256 encryption algorithm, and the encrypted data is transmitted via the UDP protocol.

[0040] Specifically, time-frequency analysis is performed on the stabilized vibration signal. A three-level wavelet packet transform is conducted using the db4 wavelet to extract the signal's time-frequency features. The energy percentage and entropy of eight sub-bands are calculated to construct a 16-dimensional feature vector. Principal component analysis is used to reduce the feature dimensionality, retaining principal components with a cumulative contribution rate of 95%, resulting in low-dimensional features representing the vibration modes. Anomaly detection is performed on the dimensionality-reduced features using the isolated forest algorithm. With 100 trees, a subsampling size of 256, and an anomaly threshold of -0.5, abnormal vibration modes are identified. These abnormal modes are matched with a pre-established fault type library to determine potential fault types and severity, generating labeled data for training the fault warning model. Based on the identified abnormal vibration modes and fault types, a multi-class fault warning model based on the RBF kernel is constructed. The kernel parameter γ and penalty coefficient C are optimized through grid search and 5-fold cross-validation to improve the model's generalization ability and prediction accuracy. The precise location coordinates of the shock absorbers are extracted from the 3D point cloud model of the environment. The location information and the output of the fault early warning model are fused using a Kalman filter. The fused early warning information and real-time vibration data are transmitted to a remote expert terminal via industrial Ethernet using the OPCUA protocol. The LZ4 algorithm is used for real-time data compression, and AES-256 encryption is used to ensure data transmission security. Low-latency real-time data transmission is achieved via the UDP protocol. In a vibration monitoring system for industrial equipment, a stable vibration signal with a sampling rate of 10kHz is analyzed. The PyWavelets library is used to implement a 3-level wavelet packet transform of db4 wavelets, resulting in eight frequency bands with frequency ranges of 0-625Hz, 625-1250Hz, 1250-1875Hz, 1875-2500Hz, 2500-3125Hz, 3125-3750Hz, 3750-4375Hz, and 4375-5000Hz. The energy percentage and Shannon entropy of each frequency band are calculated, and a 16-dimensional feature vector is constructed. Dimensionality reduction was performed using the PCA algorithm from the scikit-learn library, retaining four principal components with a cumulative contribution rate of 96.8%. Anomaly detection was performed using an isolated forest model with 100 trees and a subsampling size of 256, with an anomaly threshold set to -0.5, identifying three abnormal vibration patterns. These abnormal patterns were compared with a library containing 10 fault types, confirming a bearing inner race fault with moderate severity. An RBF kernel SVM multi-classification model was implemented using scikit-learn's SVC library, with parameter optimization via GridSearchCV. The optimal parameters were C=10 and γ=0.01, achieving a 92% accuracy with 5-fold cross-validation. Coordinates of four shock absorbers were extracted from the point cloud model with sub-millimeter accuracy. A Kalman filter was implemented using the FilterPy library to fuse location information and fault warning results. An OPCUA server was implemented using the python-opcua library, with a data update frequency of 100Hz.The lz4 library is used to compress the transmitted data, achieving a compression ratio of 1:5. The AES-256 algorithm from the pycryptodome library is used to encrypt the data. Asynchronous transmission based on UDP is implemented using the asyncio library, with an average latency of less than 10ms. The remote expert terminal receives 100kb of compressed and encrypted data per second, containing complete vibration waveforms and fault warning information.

[0041] Step S107: Decompress and restore the received warning information and on-site vibration data, and remotely monitor and diagnose the equipment based on its vibration status and fault warning information. If data transmission abnormalities or equipment malfunctions occur, an alarm will be triggered according to the abnormality handling mechanism, and the equipment will be shut down or maintained to achieve product positioning guidance using industrial vision.

[0042] The system receives compressed and encrypted warning information and vibration data, decrypts the warning information and vibration data using the AES-256 decryption algorithm, and recovers the original data using the LZ4 decompression algorithm. The decompressed original data is stored in a Redis high-performance cache database. The latest equipment vibration status and fault warning information are extracted from the Redis cache database, and historical data is stored using the InfluxDB time-series database. The root mean square value and peak factor of the vibration signal are calculated using a 60-second sliding time window, and an equipment health index is constructed based on the warning information. The equipment operating status is determined based on the equipment health index; if data transmission anomalies or equipment faults are detected, an anomaly handling mechanism is triggered. A processing strategy is automatically selected based on a preset decision tree algorithm, where the decision tree includes nodes for data integrity, vibration amplitude, and frequency characteristics. The system communicates with the field PLC via the OPCUA protocol to execute corresponding processing operations. Based on real-time images acquired by an industrial camera, the YOLOv5 target detection algorithm is used to identify the product location. Combined with the equipment's 3D model and vibration data, the precise coordinates of the product relative to the equipment are calculated. Product positioning information and equipment status indicators are overlaid and displayed on the remote monitoring interface using WebGL technology.

[0043] Specifically, the system receives compressed and encrypted early warning information and on-site vibration data. It decrypts the data using the AES-256 decryption algorithm, recovers the original data using the LZ4 decompression algorithm, verifies data integrity using the CRC32 checksum algorithm, and stores the decompressed data in a Redis high-performance cache database. The latest equipment vibration status and fault early warning information are extracted from the Redis cache, and historical data is stored in the InfluxDB time-series database. The root mean square value and peak factor of the vibration signal are calculated using a 60-second sliding time window and a 50% overlap rate. Combined with the early warning information, an equipment health index is constructed, and the equipment operating status is determined by threshold judgment. If data transmission anomalies or equipment faults are detected, an anomaly handling mechanism is triggered. A pre-set decision tree algorithm automatically selects a processing strategy. The decision tree includes nodes for data integrity, vibration amplitude, and frequency characteristics, with branch conditions covering three levels: normal, warning, and severe. The system executes actions such as sending alarm prompts, generating maintenance suggestions, or issuing emergency shutdown commands. It communicates with the on-site PLC via the OPCUA protocol to execute a shutdown operation and simultaneously records the abnormal event in the MongoDB log database. Based on real-time images captured by industrial cameras, the YOLOv5 target detection algorithm is used to identify product positions. Combined with the equipment's 3D model and vibration data, the precise coordinates of the product relative to the equipment are calculated. WebGL technology is used to overlay and display product positioning information and equipment status indicators on a remote monitoring interface, achieving industrial vision-based product positioning guidance and equipment status visualization. In a remote monitoring system for industrial equipment, 100KB of compressed and encrypted data is received per second, decrypted using the AES-256-CBC mode of the OpenSSL library, with a key length of 256 bits. The LZ4 algorithm decompresses data in 4KB blocks, achieving an average decompression speed of 5GB / s. CRC32 checksums use the IEEE 802.3 standard polynomial 0x04C11DB7, verifying 32-bit data blocks. After decompression, the data is stored in a Redis 6.0 cluster using 3 master nodes, 3 slave nodes, and a total memory of 24GB. InfluxDB 2.0 stores 30 days of historical data using the TSM engine, achieving a compression ratio of 10:1. Vibration signals were collected at a 10kHz sampling rate, and RMS value, peak factor, skewness, and kurtosis were calculated in a 60-second window. The equipment health index ranged from 0 to 100, with thresholds set to Warning = 70 and Severe = 40. The decision tree had a depth of 4, containing 7 leaf nodes, and was trained using the CART algorithm, achieving an accuracy of 95%. Communication with the Siemens S7-1500 PLC was achieved via the OPCUA protocol, with a response time of [missing information].

[0044] <100ms. The YOLOv5 model runs on an NVIDIA T4 GPU at 30fps, with mAP@0.5 reaching 0.89. WebGL uses the Three.js library to render 3D scenes, displaying product locations and device status smoothly at 60fps. The entire remote monitoring system has an average latency of <500ms, a reliability of 99.99%, and supports simultaneous online monitoring of 100 devices.

[0045] The above are only some preferred embodiments of the present invention, but the present invention is not limited thereto, and many improvements and modifications can be made. Any improvements and modifications made based on the basic principles of the present invention should be considered to fall within the protection scope of the present invention.

Claims

1. A product positioning guidance method based on industrial vision, characterized in that, The method includes: selecting the type and damping coefficient of shock absorbers based on the vibration frequency characteristics of precision equipment in an industrial environment, and matching the spring stiffness to form a shock-resistant structure that meets the vibration requirements of precision equipment; placing an industrial camera on the shock-resistant structure to capture images of the vibrating equipment in vibration; preprocessing the images according to the characteristics of the shock-resistant material; extracting texture and shape features of the images using a convolutional neural network; fine-tuning the network using a large model to adapt to the changing characteristics of the vibrating equipment in the current industrial environment; extracting image feature information; constructing a 3D point cloud model of the environment including the locations of shock absorbers based on the extracted image feature information and the layout of the shock-resistant device; identifying and marking the shock absorbers in the 3D point cloud model; and obtaining equipment vibration acceleration data based on the markings of the shock absorbers in the 3D point cloud model to analyze the vibration response of the shock-resistant device. The system estimates and corrects the vibration displacement and velocity. Based on the damping characteristics of the shock absorber and the corrected vibration displacement and velocity, it performs precise compensation on the original vibration signal. The compensated signal is then filtered using a digital filter to obtain a stable vibration signal. Anomaly analysis is performed on the stable vibration signal to identify abnormal vibration modes. Based on the identification results, a fault early warning model is established. Based on the vibration anomaly information in the fault early warning model and the position of the shock absorber in the 3D point cloud model of the environment, the early warning information and on-site vibration data are transmitted to a remote expert terminal in real time. The received early warning information and on-site vibration data are decompressed and restored. Remote monitoring and diagnosis of the equipment are performed based on the vibration status and fault early warning information. If data transmission anomalies or equipment malfunctions occur, an alarm is triggered according to the anomaly handling mechanism, and the equipment is shut down or maintenance is performed to achieve product positioning guidance using industrial vision.

2. The method according to claim 1, wherein, The process of selecting shock absorber type and damping coefficient based on the vibration frequency characteristics of precision equipment in an industrial environment, and matching spring stiffness to form a shock-resistant structure that meets the vibration protection requirements of precision equipment, includes: receiving a data acquisition command carrying a unique identifier for the precision equipment, the data acquisition command containing vibration frequency characteristic data, equipment weight data, and installation space data, while simultaneously recording the temperature, humidity, and corrosive gases of the industrial environment; calculating the main frequency range based on the vibration frequency characteristic data, the main frequency range being combined with the equipment weight data, installation space data, and environmental characteristic parameters to select the shock absorber type; and using a genetic algorithm to optimize the calculation of the damping coefficient for the selected shock absorber type. The genetic algorithm's objective function is set to minimize the system's vibration transmissibility. The damping coefficient obtained from the genetic algorithm is acquired, and this coefficient, combined with damper type data, is used to establish a finite element model of the vibration damping system. A bisection method is used to optimize the spring stiffness, with the convergence condition set to the difference between the system's natural frequency and the equipment's vibration frequency being less than a preset threshold. A mathematical model of the earthquake-resistant structure is constructed from the damper type, damping coefficient, and optimal spring stiffness. Time-domain response analysis is implemented in the earthquake-resistant structure mathematical model, using random vibration excitation signals that conform to industrial environmental characteristics to calculate the structure's displacement and acceleration responses.

3. The method according to claim 1, wherein, The process involves placing an industrial camera within a vibration-resistant structure that meets the anti-vibration requirements of precision equipment. The camera captures images of the vibrating equipment in action. Based on the characteristics of the vibration-resistant material, the images are preprocessed. A convolutional neural network is used to extract texture and shape features from the images. Simultaneously, the network is fine-tuned using a large-model fine-tuning method to adapt to the changing characteristics of the vibrating equipment in the current industrial environment. Image feature information is extracted, including: calculating the optimal shooting distance and angle of the industrial camera based on the frequency response characteristics of the vibration-resistant structure; adjusting the installation position of the industrial camera; acquiring a sequence of images of the vibrating equipment; using a median filtering algorithm to denoise the image sequence to obtain a first image; and using an adaptive histogram equalization method to further refine the first image. The contrast ratio is enhanced to obtain a second image; a multi-layer convolutional neural network structure is constructed, and the initial weights are imported using a ResNet50 pre-trained model; the convolutional neural network is fine-tuned using the second image to obtain a fine-tuned convolutional neural network; multi-scale feature maps are extracted from the fine-tuned convolutional neural network; the feature covariance matrix of the multi-scale feature maps is calculated; eigenvalue decomposition is performed on the feature covariance matrix; principal components with contribution rates exceeding a preset threshold are selected to obtain dimensionality-reduced feature vectors; the feature vectors are grouped using the k-means clustering algorithm; if the feature vector grouping results meet preset conditions, the feature vectors are determined to be low-dimensional feature representations characterizing the feature changes of the vibration equipment under the current industrial environment.

4. The method according to claim 1, wherein, The process involves constructing a 3D point cloud model of the environment, including the locations of the shock absorbers, based on the extracted image feature information and the layout of the shock absorbers. Identification and labeling of the shock absorbers in the 3D point cloud model includes: acquiring multiple frames of environmental depth maps using a structured light scanner; generating a dense point cloud data sequence based on the multiple frames of environmental depth maps, where the dense point cloud data sequence is obtained by mapping two-dimensional image pixels and their features to three-dimensional spatial coordinates using a back-projection algorithm; performing registration processing on the dense point cloud data sequence to obtain the 3D point cloud model of the environment, where the registration processing employs an iterative nearest-neighbor algorithm, stopping iteration if the number of iterations reaches a preset maximum number of iterations or the convergence error is less than a preset convergence threshold; extracting local geometric features and texture features from the 3D point cloud model of the environment, where the local geometric features include normal vectors and curvature, and the texture features include spin maps and fast point feature histograms; and using support vector machines for classification. The system classifies each point in the 3D point cloud model of the environment. The support vector machine classifier uses a radial basis function kernel function and is trained using a labeled shock absorber point cloud dataset. It performs region growing on the points classified as shock absorbers in the 3D point cloud model of the environment to obtain complete shock absorber entities. If the distance between adjacent points is less than a preset growing threshold, the adjacent point is added to the shock absorber entity. The system also performs spatial labeling on the shock absorber entities, which is achieved using the minimum bounding box method. Furthermore, it stitches together depth image features from different perspectives to obtain a complete 3D point cloud model of the environment surrounding the shock absorber. Based on a pre-established 3D CAD model library of shock absorbers, it searches for regions similar to the CAD models in the environmental point cloud to initially determine the coarse location of the shock absorber. It extracts a subset of the point cloud representing the coarse location of the shock absorber and further refines the precise boundary and pose information of the shock absorber using a point cloud segmentation algorithm, marking the shock absorber as identified.

5. The method according to claim 4, wherein, The process involves stitching together depth image features from different perspectives to obtain a complete 3D point cloud model of the environment surrounding the shock absorber. Based on a pre-established 3D CAD model library for the shock absorber, regions similar to the CAD model are searched in the environmental point cloud to initially determine the coarse location of the shock absorber. A subset of the point cloud representing the coarse location of the shock absorber is extracted, and a point cloud segmentation algorithm is used to further refine the precise boundary and pose information of the shock absorber. The shock absorber is then marked as identified. This process includes: acquiring depth images of the environment surrounding the shock absorber from multi-view depth cameras; applying bilateral filtering to the depth images to remove noise, obtaining a filtered depth image; and calculating the transformation matrix between adjacent perspectives using SIFT feature point matching and the RANSAC algorithm based on the filtered depth image. Multiple depth images are registered and fused using an iterative nearest-point algorithm to obtain a 3D point cloud model of the environment. Key geometric features, including surface normal vector distribution, curvature distribution, and local shape descriptors, are extracted from a pre-established 3D CAD model library of shock absorbers to construct feature description vectors. Regions similar to the feature description vectors are searched in the 3D point cloud model of the environment using a sliding window. If high-similarity regions are identified, cluster analysis is performed on these regions. For the point cloud subsets obtained from the cluster analysis, a region growing algorithm is applied for fine segmentation. The precise boundary of the shock absorber is obtained through iterative expansion. Principal component analysis is used to calculate the principal axis direction and center position of the shock absorber to determine its precise pose.

6. The method according to claim 1, wherein, The process of acquiring equipment vibration acceleration data based on the markers of the shock absorbers in the environmental 3D point cloud model, estimating the vibration response of the shock-absorbing device, and obtaining corrected vibration displacement and velocity includes: determining the position of the shock absorber markers according to the environmental 3D point cloud model; installing a triaxial accelerometer on each shock absorber after acquiring the position of the shock absorber markers; performing spectral analysis on the acceleration signals collected by the triaxial accelerometers using a fast Fourier transform algorithm; extracting the main vibration frequencies and amplitudes of each shock absorber in the x, y, and z directions from the spectral analysis results to construct a vibration feature vector; and using the vibration feature vector as input based on a pre-established mass-spring-damping dynamic model of the shock-absorbing device, performing real-time analysis of the vibration response of the shock-absorbing device using a Kalman filter algorithm. The process involves estimating the instantaneous displacement and velocity states of each shock absorber; integrating these states with the shock absorber position information in the 3D point cloud model of the environment using a Bayesian fusion algorithm; correcting the instantaneous displacement and velocity states using a weighted least squares method; capturing the motion trajectories of marker points on the shock absorber surface using a high-speed camera; extracting the marker point positions from the motion trajectories; comparing the marker point positions with the corrected instantaneous displacement and velocity states to determine their accuracy; and further, establishing vibration transmission models based on different types and locations of shock absorbers, and simulating the propagation and attenuation characteristics of vibration signals in the equipment structure by combining the stiffness and damping parameters of the shock absorbers to obtain comprehensive equipment vibration response evaluation results.

7. The method according to claim 6, wherein, The method involves establishing a vibration transmission model based on different types and locations of vibration dampers. Combining the stiffness and damping parameters of the dampers, it simulates the propagation and attenuation characteristics of vibration signals within the equipment structure to obtain comprehensive equipment vibration response evaluation results. This includes: constructing a geometric model of the equipment structure using the finite element method, simplifying the vibration dampers to spring-damped elements; measuring the stiffness and damping parameters of the vibration dampers using a dynamic stiffness tester to obtain their physical characteristics; assigning corresponding physical parameters to each element in the geometric model based on these physical characteristics; and calculating the natural frequencies and mode shapes of the geometric model using modal analysis to obtain the... The dynamic characteristics of the equipment structure are described; a vibration transfer function is constructed to convert the vibration signal at the input end into the response of each key node, thereby obtaining the vibration propagation characteristics of the equipment structure; the vibration propagation characteristics are fitted using the response surface methodology to obtain the vibration response of the equipment structure under different operating conditions; the sampling points of the response surface methodology are determined using the central composite experimental design method to obtain the mapping relationship between the vibration response and the damper parameters and vibration input; Monte Carlo simulation samples are generated through Latin hypercube sampling, and the mapping relationship is statistically analyzed to obtain a comprehensive evaluation result of the vibration response of the equipment structure.

8. The method according to claim 1, wherein, The process involves precisely compensating the original vibration signal based on the damping characteristics of the shock absorber, the corrected vibration displacement, and velocity, and then filtering the compensated signal using a digital filter to obtain a stable vibration signal. This includes: acquiring the damping characteristic curve of the shock absorber; establishing a nonlinear mapping relationship between damping force and vibration displacement and velocity based on the damping characteristic curve; processing the nonlinear characteristics using a piecewise linearization method, dividing the damping characteristic curve into multiple linear intervals; fitting a local linear damping characteristic function using the least squares method based on the linear intervals; substituting the corrected vibration displacement and velocity into the local linear damping characteristic function to calculate the actual damping force; and converting the original vibration signal to the frequency domain using a fast Fourier transform, and then calculating the actual damping force based on the corrected vibration displacement and velocity. A frequency domain compensation function is constructed using damping force; the amplitude of the original signal is corrected, with the amplitude compensation coefficient being the ratio of the ideal damping force to the actual damping force; the compensated frequency domain signal is converted back to the time domain using inverse fast Fourier transform to obtain a preliminarily compensated time domain vibration signal; multi-scale analysis is performed on the preliminarily compensated time domain vibration signal using wavelet decomposition to identify and remove high-frequency noise and unstable components introduced by the compensation; a Butterworth low-pass filter is designed to filter the denoised compensated signal to eliminate residual high-frequency interference; the variance ratio and signal-to-noise ratio improvement of the signal before and after filtering are calculated to evaluate the stability of the compensated and filtered signals; if the variance ratio decreases by more than a preset threshold and the signal-to-noise ratio improvement exceeds a preset threshold, a stable vibration signal is determined to be obtained.

9. The method according to claim 1, wherein, The process involves anomaly analysis of the stabilized vibration signal to identify abnormal vibration modes. Based on the identification results, a fault early warning model is established. Using the vibration anomaly information in the fault early warning model, combined with the location of the shock absorbers in the environmental 3D point cloud model, the early warning information and on-site vibration data are transmitted in real time to a remote expert terminal. This includes: acquiring the stabilized vibration signal; performing time-frequency analysis on the vibration signal; using db4 wavelet to perform a 3-level wavelet packet transform to obtain the time-frequency characteristics of the vibration signal; calculating the energy percentage and entropy value of 8 sub-bands based on the time-frequency characteristics to construct a 16-dimensional feature vector; reducing the dimensionality of the feature vector through principal component analysis to obtain low-dimensional features representing the vibration modes; using the isolated forest algorithm to perform anomaly detection on the low-dimensional features to identify abnormal vibration modes; and comparing the abnormal vibration modes with a pre-established... The system matches the fault type library to determine potential fault types and severity. Based on the abnormal vibration modes and fault types, a multi-class fault early warning model based on RBF kernel support vector machine is constructed. The precise position coordinates of the shock absorber are extracted from the 3D point cloud model of the environment. The position coordinates and the output of the fault early warning model are fused using a Kalman filter to obtain fused early warning information. The fused early warning information and real-time vibration data are transmitted to a remote expert terminal via industrial Ethernet using the OPCUA protocol. If the remote expert terminal receives the early warning information and real-time vibration data, it uses the LZ4 algorithm to compress the early warning information and real-time vibration data in real time. The compressed data is encrypted using the AES-256 encryption algorithm. The encrypted data is transmitted via UDP protocol.

10. The method according to claim 1, wherein, The process involves decompressing and restoring received early warning information and on-site vibration data. Remote monitoring and diagnosis of the equipment are achieved through its vibration status and fault warning information. If data transmission anomalies or equipment malfunctions occur, an alarm is triggered according to the anomaly handling mechanism, prompting emergency shutdown or maintenance operations. This enables product positioning guidance based on industrial vision. The process includes: receiving compressed and encrypted early warning information and vibration data; decrypting the early warning information and vibration data using the AES-256 decryption algorithm; restoring the original data using the LZ4 decompression algorithm; storing the decompressed original data in a Redis high-performance cache database; and retrieving the latest equipment vibration status and fault warning information from the Redis cache database and storing it using an InfluxDB time-series database. Historical data; the root mean square value and peak factor of the vibration signal are calculated using a 60-second sliding time window, and an equipment health index is constructed based on the early warning information; the equipment operating status is judged based on the equipment health index, and an abnormality handling mechanism is triggered if abnormal data transmission or equipment failure is detected; a processing strategy is automatically selected according to a preset decision tree algorithm, the decision tree containing data integrity, vibration amplitude, and frequency characteristic nodes; communication with the field PLC via the OPCUA protocol is used to execute corresponding processing operations; based on real-time images acquired by industrial cameras, the YOLOv5 target detection algorithm is used to identify the product position, and the precise coordinates of the product relative to the equipment are calculated by combining the equipment 3D model and vibration data; the product positioning information and equipment status indication are overlaid and displayed on the remote monitoring interface using WebGL technology.

Images