Product positioning and guiding method based on industrial vision

Abstract

The invention provides a product positioning and guiding method based on industrial vision, which comprises the following steps: selecting the type and damping coefficient of a shock absorber according to the vibration frequency characteristic of precision equipment in an industrial environment, and matching with the rigidity of a spring to form a shockproof structure meeting the shockproof requirement of the precision equipment; constructing an environment three-dimensional point cloud model containing the positions of the shock absorbers according to the extracted image feature information in combination with the layout of the shockproof device, and identifying and marking the shock absorbers in the environment three-dimensional point cloud model; based on the mark of the shock absorber in the environment three-dimensional point cloud model, acquiring equipment vibration acceleration data, and estimating the vibration response of the anti-vibration device to obtain corrected vibration displacement and speed; and according to the damping characteristic of the shock absorber and the corrected vibration displacement and speed, performing precise compensation on the original vibration signal, and filtering the compensated signal through a digital filter to obtain a stable vibration signal.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and in particular to an industrial visual product positioning and guiding method. Background Art

[0002] In complex industrial environments, visual positioning systems face severe challenges. In industrial operating environments, the continuous vibration and shock waves generated by large operating machines interfere with precision equipment, causing blurred and out-of-focus images, which seriously affect positioning accuracy. In such harsh environments, traditional visual positioning technology is difficult to meet the requirements of high precision and high reliability. However, industrial operations have an urgent need for precise positioning, especially in dangerous areas or remote control scenarios. At the same time, due to the complex and changeable industrial environment, on-site staff often find it difficult to respond to various emergencies independently. How to enable remote experts to provide real-time guidance to on-site personnel under limited communication conditions and provide accurate positioning and operation suggestions is also a thorny issue. This not only involves the optimization of communication technology, but also requires consideration of how to effectively transmit visual information in complex environments to ensure that remote experts can accurately understand the on-site situation and give appropriate guidance. Summary of the invention

[0003] The present invention provides an industrial visual product positioning and guiding method, which mainly includes:

[0004] According to the vibration frequency characteristics of precision equipment in industrial environments, select the shock absorber type and damping coefficient, and match the spring stiffness to form a shockproof structure that meets the shockproof requirements of precision equipment;

[0005] An industrial camera is placed according to the anti-vibration structure that meets the anti-vibration requirements of precision equipment. The image of the vibrating equipment on site is collected by the industrial camera. The image is pre-processed according to the characteristics of the anti-vibration material. The texture and shape features of the image are extracted through the convolutional neural network. At the same time, the network is fine-tuned by fine-tuning the large model to adapt to the characteristic changes of the vibrating equipment in the current industrial environment and extract the image feature information.

[0006] Based on the extracted image feature information and the layout of the anti-vibration device, a three-dimensional point cloud model of the environment including the location of the shock absorber is constructed, and the shock absorber in the three-dimensional point cloud model of the environment is identified and marked;

[0007] Based on the markers of the shock absorbers in the three-dimensional point cloud model of the environment, the vibration acceleration data of the equipment is obtained, the vibration response of the shockproof device is estimated, and the corrected vibration displacement and velocity are obtained;

[0008] According to the damping characteristics of the shock absorber, the corrected vibration displacement and velocity, the original vibration signal is accurately compensated, and the compensated signal is filtered through a digital filter to obtain a stable vibration signal;

[0009] Perform abnormal analysis on the stabilized vibration signal to identify abnormal vibration patterns, establish an equipment fault warning model based on the identification results, and transmit the warning information and on-site vibration data to the remote expert terminal in real time based on the vibration abnormality information in the fault warning model and the position of the shock absorber in the environmental 3D point cloud model;

[0010] Decompress and restore the received warning information and on-site vibration data, and remotely monitor and diagnose the equipment through the equipment's vibration status and fault warning information. If data transmission abnormalities or equipment failures occur, alarm prompts are given according to the exception handling mechanism, and the equipment is urgently shut down or maintained, thus realizing product positioning guidance with industrial vision.

[0011] The technical solution provided by the embodiment of the present invention may have the following beneficial effects:

[0012] The present invention discloses a product positioning guidance method for industrial vision. The method selects appropriate shock absorbers and shockproof structures, and on the basis of collecting images of vibration equipment by industrial cameras, uses convolutional neural networks and large model fine-tuning to extract image features, and constructs a three-dimensional point cloud model of the environment including the shock absorber position. Based on the model, the present invention accurately compensates and filters the vibration signal to achieve stable vibration signal acquisition. By performing abnormal analysis on the signal, an equipment failure warning model is established, and the warning information and vibration data are transmitted to a remote expert terminal in real time, so as to achieve remote monitoring and diagnosis of the equipment. When an abnormality occurs, the present invention can promptly issue an alarm and perform emergency processing, effectively ensuring the safe operation of precision equipment and improving the stability and reliability of industrial production. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 This is a flow chart of an industrial vision product positioning guidance method of the present invention.

[0014] Figure 2 It is a schematic diagram of an industrial vision product positioning and guiding method of the present invention.

[0015] Figure 3 This is another schematic diagram of an industrial vision product positioning and guiding method of the present invention. DETAILED DESCRIPTION

[0016] In order 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 in conjunction with the drawings in the embodiments of this specification. Obviously, the described embodiments are only part of the embodiments of this specification, not all of the embodiments. Based on the embodiments in this specification, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of this specification.

[0017] like Figure 1 -3. In this embodiment, a product positioning guidance method of industrial vision may specifically include:

[0018] Step S101, according to the vibration frequency characteristics of the precision equipment in the industrial environment, the shock absorber type and the damping coefficient are selected, and the spring stiffness is matched to form a shockproof structure that meets the shockproof requirements of the precision equipment.

[0019] Receive a data acquisition instruction carrying a unique identification number of a precision device, the data acquisition instruction includes vibration frequency characteristic data, equipment weight data and installation space data, and records the temperature, humidity and corrosive gas of the industrial environment; calculate a main frequency range based on the vibration frequency characteristic data, and combine the main frequency range with the equipment weight data, installation space data and environmental characteristic parameters to select the type of shock absorber; for the selected shock absorber type, use a genetic algorithm to optimize and calculate the damping coefficient, and the optimization objective function of the genetic algorithm is set to minimize the system vibration transmissibility; obtain the damping coefficient optimized by the genetic algorithm, and combine the damping coefficient with the shock absorber type data to establish a finite element model of the shock absorption system; use a binary search method to optimize the spring stiffness, and the convergence condition of the binary search method is set to the difference between the system natural frequency and the equipment vibration frequency is less than a preset threshold; construct a seismic structure mathematical model from the shock absorber type, the damping coefficient and the optimal spring stiffness; implement time domain response analysis in the seismic structure mathematical model, and the time domain response analysis selects a random vibration excitation signal that meets the characteristics of the industrial environment to calculate the displacement response and acceleration response of the structure.

[0020] Specifically, the vibration frequency characteristic data, equipment weight data and installation space data of precision equipment are obtained, and the temperature, humidity and corrosive gas of the industrial environment are recorded at the same time. The main frequency range is calculated according to the vibration frequency characteristic data, and the appropriate shock absorber type is selected in combination with the equipment weight data, installation space data and environmental characteristic parameters. For the selected shock absorber type, the genetic algorithm is used to optimize the calculation of the damping coefficient, and the optimization objective function is set to minimize the system vibration transmissibility. The optimized damping coefficient value is combined with the shock absorber type data, and the finite element model of the shock absorption system is established through ANSYS software. According to the established finite element model of the shock absorption system, the modal analysis function is used to calculate the system natural frequency, and the system natural frequency is compared with the vibration frequency of the precision equipment. The dichotomy method is used to optimize the spring stiffness, and the convergence condition is set as the difference between the system natural frequency and the equipment vibration frequency is less than the preset threshold. The mathematical model of the seismic structure is constructed based on the data of the shock absorber type, damping coefficient and optimal spring stiffness. Time domain response analysis is implemented in MATLAB. Random vibration excitation signals that meet the characteristics of the industrial environment are selected to calculate the displacement response and acceleration response of the structure. The evaluation indicators include maximum displacement, maximum acceleration and root mean square acceleration. According to the results of the time domain response analysis, it is judged whether the shock absorption effect of the seismic structure on the precision equipment meets the preset requirements. If not, return to the shock absorber type selection step and re-optimize the parameters. If the requirements are met, the final shock absorber type, damping coefficient and spring stiffness values ​​are output. When obtaining precision equipment data, an acceleration sensor is used to measure the vibration frequency characteristics and record the frequency response in the range of 0-1000Hz. The equipment weighs 500kg and the installation space is 1m×1m×0.5m. The industrial environment temperature is 35℃, the humidity is 80%, and there is a trace amount of hydrogen sulfide gas. The main frequency range calculation result is 20-80Hz. According to these parameters, the spring damping shock absorber is selected as the most suitable type. When using the 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 is defined as minimizing the average transmissibility in the frequency range of 20-80 Hz. The optimized damping coefficient is 0.3. The finite element model of the shock absorption system is established in ANSYS, with a mesh unit size of 0.01m and a total number of about 10,000 units. The modal analysis calculates that the first three natural frequencies of the system are 15Hz, 45Hz and 85Hz respectively. The spring stiffness is optimized by the binary method, with an initial search range of 10^4-10^6N / m, and the convergence condition is set to the difference between the first-order natural frequency of the system and the main frequency of the equipment is less than 1Hz. After 10 iterations, the optimal spring stiffness is 3.5×10^5N / m. Time domain response analysis is performed in MATLAB, and Gaussian white noise is selected as the excitation signal, with a bandwidth of 0-200Hz and a power spectral density of 0.1g^2 / Hz. The structural response within 100s is 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. According to the preset shock absorption requirements of maximum displacement <1mm and maximum acceleration <1g, it is judged that the shockproof structure meets the requirements. The final output shock absorber type is a spring damping shock absorber with a damping coefficient of 0.3 and a spring stiffness of 3.5×10^5N / m.

[0021] Step S102, placing an industrial camera according to the shockproof structure that meets the shockproof requirements of precision equipment, collecting images of the vibrating equipment on site through the industrial camera, preprocessing the images according to the characteristics of the shockproof material, extracting the texture and shape features of the images through a convolutional neural network, and fine-tuning the network through a large model fine-tuning method to adapt to the characteristic changes of the vibration equipment in the current industrial environment and extract image feature information.

[0022] The optimal shooting distance and angle of the industrial camera are calculated according to the frequency response characteristics of the shockproof structure, and the installation position of the industrial camera is adjusted to obtain a sequence of images of the vibration equipment. The image sequence is subjected to denoising by a median filtering algorithm to obtain a first image; the first image is subjected to contrast enhancement by an adaptive histogram equalization method to obtain a second image. A multi-layer convolutional neural network structure is constructed, and the initial weights are imported by using the ResNet50 pre-trained model; the convolutional neural network is fine-tuned using the second image to obtain a fine-tuned convolutional neural network. A multi-scale feature map is extracted from the fine-tuned convolutional neural network; the feature covariance matrix of the multi-scale feature map is calculated; the feature covariance matrix is ​​subjected to eigenvalue decomposition; the principal component whose contribution rate exceeds a preset threshold is selected to obtain a feature vector after dimensionality reduction. The feature vectors are grouped by a k-means clustering algorithm; if the result of the feature vector grouping meets a preset condition, the feature vector is determined to be a low-dimensional feature representation that characterizes the feature changes of the vibration equipment in the current industrial environment.

[0023] Specifically, the optimal shooting distance and angle of the industrial camera are calculated according to the frequency response characteristics of the shockproof structure, the camera installation position is adjusted, and the image sequence of the vibration equipment is collected by the industrial camera to obtain the image resolution, frame rate, exposure time, and record the type, thickness, and density of the shockproof material. The median filtering algorithm is used to reduce the noise of the collected image sequence, and the adaptive histogram equalization method is used to enhance the image contrast. According to the damping ratio of the shockproof material, sharpening filters of different strengths are selected to enhance the edge details. A multi-layer convolutional neural network structure is constructed, and the parameters of the convolution layer, pooling layer, and fully connected layer are set. The ResNet50 pre-trained model is used to import the initial weights. The network is fine-tuned using the vibration equipment image dataset in the industrial environment, and the last few layers of the network structure are adjusted according to the characteristics of the vibration equipment. Multi-scale feature maps are extracted from the fine-tuned convolutional neural network, the feature covariance matrix is ​​calculated, and the eigenvalue decomposition is performed. The principal components whose contribution rate exceeds the preset threshold are selected to obtain the feature vector after dimensionality reduction. The feature vectors are grouped by the k-means clustering algorithm to obtain a low-dimensional feature representation that characterizes the feature changes of the vibration equipment in the current industrial environment. The leave-one-out cross-validation method is used to evaluate the extracted feature vector's ability to characterize the changes in the characteristics of the vibration equipment under different working conditions, calculate the classification accuracy of the feature vector under different conditions, and verify the effectiveness of the feature extraction method. In an industrial environment, assuming that the main frequency response of the shockproof structure is 20-80Hz, based on this characteristic, it is calculated that the optimal shooting distance of the industrial camera is 1.5 meters, and the angle is perpendicular to the surface of the vibration equipment. Set the camera acquisition parameters, resolution = 1920×1080 pixels, frame rate = 60fps, exposure time

[0024] =1 / 250 seconds. The recorded anti-vibration material is a rubber vibration isolation pad with a thickness of 25mm and a density of 1.2g / cm3. A median filter with a window size of 5×5 is applied to the collected image sequence to eliminate salt and pepper noise. Adaptive histogram equalization uses a local area of ​​8×8 to enhance image contrast. According to the damping ratio of the rubber vibration isolation pad of 0.15, a Laplacian sharpening filter with a kernel size of 3×3 is selected to enhance edge details. The constructed convolutional neural network contains 5 convolutional layers, each followed by a maximum pooling layer, and finally 3 fully connected layers. The ResNet50 model pre-trained on ImageNet is used as the initial weight. The first 40 layers are frozen during fine-tuning, and the last 10 layers are retrained. The fine-tuning dataset contains 5,000 images of vibrating equipment in different working conditions. The feature maps of the 3rd, 4th, and 5th convolutional layers are extracted from the fine-tuned network, and a 2048-dimensional feature vector is calculated. The covariance matrix of the eigenvectors was calculated, and the eigenvalue decomposition was performed to select the top 100 principal components with a cumulative contribution rate of 95%. The k-means clustering algorithm was used, and k=5 was set to group the features after dimensionality reduction. The leave-one-out cross-validation method was used to evaluate the feature characterization ability, and the classification accuracy under 5 different working conditions reached 92.5%.

[0025] Step S103, constructing an environmental three-dimensional point cloud model including the position of the shock absorber according to the extracted image feature information and in combination with the layout of the shockproof device, and identifying and marking the shock absorber in the environmental three-dimensional point cloud model.

[0026] Acquire a multi-frame environment depth map, which is collected by a structured light scanner; generate a dense point cloud data sequence based on the multi-frame environment depth map, and map the two-dimensional image pixels and their features to three-dimensional space coordinates through a back-projection algorithm; perform registration processing on the dense point cloud data sequence to obtain a three-dimensional point cloud model of the environment, and the registration processing adopts an iterative closest point algorithm, and if the number of iterations reaches a preset maximum number of iterations or the convergence error is less than a preset convergence threshold, the iteration is stopped; extract local geometric features and texture features from the three-dimensional point cloud model of the environment, and the local geometric features include The invention discloses a method for realizing a three-dimensional point cloud model of an environment, wherein the three-dimensional point cloud model of the environment is a three-dimensional point cloud model of an environmental image ...

[0027] Specifically, according to the extracted image feature information, a structured light scanner is used to obtain a multi-frame environmental depth map. The two-dimensional image pixels and their features are mapped to three-dimensional space coordinates through the back-projection algorithm to generate a dense point cloud data sequence containing the shock absorber position information. The iterative closest point algorithm is used to align the multi-frame point cloud data. The maximum number of iterations is set to 50 and the convergence threshold is set to 0.001 meters. The position deviation during the scanning process is eliminated and the complete three-dimensional point cloud model of the environment is obtained by fusion, while the color and depth information of each point is retained. Local geometric features and texture features are extracted from the three-dimensional point cloud model of the environment, and the normal vector, curvature, spin map and fast point feature histogram of each point are calculated to construct a high-dimensional feature vector database. The support vector machine classifier is used to classify each point in the point cloud. The labeled shock absorber point cloud dataset is used for training. The radial basis kernel function is selected, and the penalty parameter C is set to 10 to identify the shock absorber area. The complete shock absorber entity is segmented by the region growing algorithm, and the growth threshold is set to 0.05 meters. Finally, the minimum bounding box is used to spatially mark the identified shock absorber. Combined with the pre-acquired layout of the anti-vibration device, the recognition results are verified and corrected to improve the accuracy of the shock absorber marking. In the industrial environment, a structured light scanner with a resolution of 1920×1080 pixels is used to obtain a sequence of environmental depth maps at a speed of 30 frames per second. Each frame of the depth map is converted into a point cloud by a back-projection algorithm, and the previously extracted SIFT feature points are mapped to the corresponding 3D coordinates. The iterative closest point algorithm is used to align the point clouds of consecutive frames, setting the maximum number of iterations to 50 and the convergence threshold to 0.001 meters. The average alignment takes 0.5 seconds. The fused complete point cloud model contains about 5 million points, each of which stores 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 about 2GB. The support vector machine classifier was trained using the libsvm library. The training set contained 10,000 labeled shock absorber point cloud samples and 40,000 non-shock absorber samples. The radial basis kernel function was selected, the penalty parameter C was set to 10, and the cross-validation accuracy reached 95%. Regional growth was performed on the points classified as shock absorbers, and the growth threshold was set to 0.05 meters. On average, each shock absorber contained 2,000 points. The principal component analysis method was used to calculate the minimum bounding box to accurately locate the spatial position and orientation of the shock absorber. Finally, the recognition results were compared with the pre-drawn CAD layout with an accuracy of 0.01 meters, and the marks with errors exceeding 0.1 meters were corrected. The final shock absorber positioning accuracy reached

[0028] ±0.05 m.

[0029] The depth image features from different perspectives are stitched together to obtain a complete 3D point cloud model of the shock absorber's surroundings. Based on the pre-established 3D CAD model library of the shock absorber, the area similar to the CAD model is searched in the environmental point cloud to preliminarily determine the rough position of the shock absorber, extract the point cloud subset of the rough position of the shock absorber, and further refine the precise boundary and posture information of the shock absorber through the point cloud segmentation algorithm, and mark the shock absorber as recognized.

[0030] A depth image of the environment around the shock absorber acquired by a multi-view depth camera is obtained, and a bilateral filter is applied to the depth image to remove noise to obtain a filtered depth image. According to the filtered depth image, the transformation matrix between adjacent viewpoints is calculated using SIFT feature point matching and RANSAC algorithm, and multiple depth images are registered and fused using iterative closest point algorithm to obtain a three-dimensional point cloud model of the environment. Key geometric features are extracted from a pre-established three-dimensional CAD model library of the shock absorber, and the key geometric features include surface normal vector distribution, curvature distribution and local shape descriptor to construct a feature description vector. In the three-dimensional point cloud model of the environment, a region similar to the feature description vector is searched through a sliding window. If a high-similarity region is identified, a cluster analysis is performed on the high-similarity region. For the point cloud subset obtained by the cluster analysis, a regional growing algorithm is applied for fine segmentation, and the precise boundary of the shock absorber is obtained by iterative expansion. The principal axis direction and center position of the shock absorber are calculated using the principal component analysis method to determine the precise position and posture of the shock absorber.

[0031] Specifically, a multi-view depth camera is used to collect depth images of the environment around the shock absorber. Bilateral filtering is applied to each depth image to remove noise, and small occluded areas are filled by morphological operations. SIFT feature point matching and RANSAC algorithm are used to calculate the transformation matrix between adjacent viewpoints. The iterative closest point algorithm is used to align 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 the pre-established 3D CAD model library of the shock absorber to construct a feature description vector. In the environmental point cloud model, a sliding window is used to search for areas with similar features to the CAD model. The window size is set to 1.5 times the size of the shock absorber, and the step size is 1 / 4 of the window size. The cosine similarity is used to calculate the similarity between feature vectors. Cluster analysis is performed on the identified high-similarity areas, and the density clustering algorithm DBSCAN is used to determine the potential shock absorber location. The neighborhood radius is set to 0.1 meters and the minimum number of points is set to 50. A subset of the point cloud within a fixed radius around each cluster center is extracted as the rough location of the shock absorber. For the extracted point cloud subset, the region growing algorithm is applied for fine segmentation, and the normal vector angle threshold is set to 10 degrees, the curvature change threshold is set to 0.05, and the neighborhood search radius is set to 0.02 meters. The precise boundary of the shock absorber is obtained by iterative expansion, and the principal axis direction and center position of the shock absorber are calculated by principal component analysis to determine its precise posture. The segmentation result is accurately matched with the CAD model by the iterative closest point algorithm, and the geometric consistency score is calculated. If the score is higher than the preset threshold, the shock absorber is marked as recognized. In an industrial environment, three depth cameras with a resolution of 1280x720 are used to collect depth images of the environment around the shock absorber at a rate of 15 frames / second. Each depth image is first subjected to bilateral filtering with a window size of 5x5, a spatial standard deviation of 10, and a range standard deviation of 0.1 to remove noise, and then a closing operation of a 3x3 structural element is applied to fill small area occlusions. The SIFT algorithm was used to extract about 1000 feature points in each image. The feature matching was performed through the FLANN fast nearest neighbor search. The RANSAC algorithm was then used to remove abnormal matching points and calculate the transformation matrix between adjacent viewpoints. The iterative nearest point algorithm was used for point cloud registration. The maximum number of iterations was set to 50 and the convergence threshold was set to 0.001 meters. A complete three-dimensional point cloud model of the environment containing about 5 million points was obtained. The geometric features of 10 common shock absorbers were extracted from the CAD model library, and the 128-dimensional FPFH feature descriptor was calculated for each model. In the environmental point cloud, a sliding window of 0.3 meters x 0.3 meters x 0.3 meters was used with a step size of 0.075 meters to search for similar areas. The cosine similarity threshold was set to 0.85, and about 100 candidate areas were screened out. The DBSCAN algorithm was applied to these areas with the parameter ε set to 0.1 meters and MinPts set to 50, and 10-20 clusters were obtained.A point cloud subset within a radius of 0.2 meters is extracted for 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 main axis direction and center position are calculated by PCA to obtain the 6-DOF pose of the shock absorber. Finally, the ICP algorithm is used to match the segmentation result with the CAD model, with a maximum number of iterations of 100 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, with sub-millimeter accuracy.

[0032] Step S104, based on the mark of the shock absorber in the three-dimensional point cloud model of the environment, the equipment vibration acceleration data is obtained, the vibration response of the shockproof device is estimated, and the corrected vibration displacement and velocity are obtained.

[0033] The shock absorber mark position is determined according to the three-dimensional point cloud model of the environment, and a three-axis acceleration sensor is installed on each shock absorber after obtaining the shock absorber mark position; the acceleration signal collected by the three-axis acceleration sensor is subjected to spectrum analysis by using a fast Fourier transform algorithm, and the main vibration frequency and amplitude of each shock absorber in the three directions of x, y and z are extracted from the spectrum analysis results to construct a vibration characteristic vector; according to the pre-established mass-spring-damping dynamic model of the shockproof device, the vibration characteristic vector is used as input, and the vibration response of the shockproof device is estimated in real time by using a Kalman filter algorithm to obtain the instantaneous displacement and velocity state of each shock absorber; the instantaneous displacement and velocity state are integrated with the shock absorber position information in the three-dimensional point cloud model of the environment by using a Bayesian fusion algorithm, and the instantaneous displacement and velocity state are corrected by using a weighted least squares method; the motion trajectory of the mark point on the surface of the shock absorber is collected by using a high-speed camera, the mark point position is extracted from the motion trajectory, the mark point position is compared with the corrected instantaneous displacement and velocity state, and the accuracy of the corrected instantaneous displacement and velocity state is judged.

[0034] Specifically, according to the shock absorber mark position in the three-dimensional point cloud model of the environment, a three-axis acceleration sensor is installed on each shock absorber, and a high-precision data acquisition card is used to synchronously collect the vibration acceleration data of each shock absorber at a sampling frequency of 10kHz. The original signal is preprocessed through a 10Hz-1kHz bandpass filter 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 spectrally analyzed using the fast Fourier transform algorithm to extract the main vibration frequency and amplitude of each shock absorber in the three directions of x, y, and z, and construct the vibration feature vector. The principal component analysis method is used to reduce the feature dimension, calculate the eigenvalue and eigenvector, and select the first few principal components with a cumulative contribution rate of 95% as the features after dimensionality reduction. Based on the pre-established mass-spring-damper dynamic model of the shockproof device, the vibration feature vector after dimensionality reduction is used as input, and the vibration response of the shockproof device is estimated in real time through the Kalman filter algorithm to obtain the instantaneous displacement and velocity state of each shock absorber. The Bayesian fusion algorithm is used to integrate the estimated displacement and velocity with the shock absorber position information in the three-dimensional point cloud model of the environment. The vibration displacement and velocity are corrected by the weighted least squares method, and the weight is determined according to the measurement error covariance matrix. The motion trajectory of the marker points on the surface of the shock absorber is collected by a high-speed camera at a rate of 1000 frames / second. The position of the marker points is extracted by the image processing algorithm and compared with the corrected vibration displacement and velocity to verify the accuracy of the results. In the industrial environment, a PCB356A15 triaxial accelerometer with a sensitivity of 100mV / g is installed on each of the four shock absorbers located according to the point cloud model. The NIPXIe-4499 data acquisition card is used, the sampling rate is set to 10kHz, and the acquisition duration is 60 seconds. The collected raw signal is processed by a Butterworth bandpass filter with a passband of 10Hz-1kHz and an order of 4. The filtered data is organized into a time series matrix of 4 shock absorbers×3 directions with a total of 12 columns. An 8192-point FFT analysis was performed on each column of data to obtain a spectrum within the range of 0-5kHz. The top five frequencies with the largest amplitude in each direction and their amplitudes were extracted to form a 4×3×5=60-dimensional feature vector. The top ten principal components with a cumulative contribution rate of 95% were selected through PCA dimensionality reduction. The state space equation was constructed 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 Kalman filter was used, the process noise covariance Q was set to a diagonal matrix [1e-6, 1e-4], the measurement noise covariance R was 1e-4, and the state estimate was updated at a frequency of 100Hz. The 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, and the fusion weight was determined according to their respective uncertainties. The weighted least squares method was used for final correction, and the weight matrix was determined by the inverse covariance matrix of the measurement error.At the same time, a Phantom v2512 high-speed camera was used with a resolution of 1280 × 800 pixels and a frame rate of 1000 fps to capture the 4 mm diameter reflective markers on the surface of the shock absorber. The marker trajectory was extracted through threshold segmentation and centroid calculation and compared with the corrected displacement, with an average error of less than 0.1 mm.

[0035] According to the different types and positions of shock absorbers, a vibration transfer model is established. Combined with the stiffness and damping parameters of the shock absorber, the propagation and attenuation characteristics of the vibration signal in the equipment structure are simulated to obtain a comprehensive equipment vibration response evaluation result.

[0036] The finite element method is used to construct a geometric model of the equipment structure, and the shock absorber is simplified into a spring-damper unit; the stiffness and damping parameters of the shock absorber are measured using a dynamic stiffness tester to obtain the physical characteristics of the shock absorber; according to the physical characteristics, corresponding physical parameters are assigned to each unit in the geometric model; the modal analysis method is used to calculate the natural frequency and vibration mode of the geometric model to obtain the dynamic characteristics of the equipment structure; a vibration transfer function is constructed, and the vibration signal at the input end is converted into the response of each key node through the transfer function to obtain the vibration propagation characteristics of the equipment structure; the response surface method is used to fit the vibration propagation characteristics to obtain the vibration response of the equipment structure under different working conditions; the sampling points of the response surface method are determined using the central composite test design method to obtain the mapping relationship between the vibration response and the shock absorber parameters and the vibration input; Monte Carlo simulation samples are generated by Latin hypercube sampling, and the mapping relationship is statistically analyzed to obtain a comprehensive evaluation result of the vibration response of the equipment structure.

[0037] Specifically, according to the type and installation position of the shock absorber, the finite element method is used to construct the geometric model of the equipment structure, and the shock absorber is simplified into a spring-damper unit. The key nodes of the structure are determined by calculating the modal participation factor, and mass units are set at these nodes. A complete vibration transmission network is formed through node connection. The stiffness and damping parameters of the shock absorber are measured using a dynamic stiffness tester. For the rubber shock absorber, nonlinear parameters are introduced to characterize its characteristics. Combined with the structural material parameters in the material property database, the corresponding physical properties are assigned to each unit in the finite element model, and the modal analysis method is used to calculate the natural frequency and vibration mode of the structure. Based on the theory of structural dynamics, a vibration transfer function is constructed, and the vibration signal at the input end is converted into the response of each key node through the transfer function. The fourth-order Runge-Kutta method is used for numerical integration, and the time step is set to 1 / 20 of the inverse of the highest frequency of interest. The time domain response is solved to obtain the propagation and attenuation characteristics of the vibration in the structure. The response surface method is used to fit the vibration response under different working conditions. The sampling points are determined by the central composite test design method. The second-order polynomial is used to fit the mapping relationship between the vibration response and the shock absorber parameters and vibration input. 10,000 sets of Monte Carlo simulation samples are generated by Latin hypercube sampling. The comprehensive evaluation results of the equipment vibration response are obtained by statistical analysis, including the maximum response value, root mean square value and frequency response function of each key node. In the vibration analysis of an industrial equipment, ANSYS Workbench is used to build a finite element model of the equipment, which contains 23,456 units and 34,567 nodes. Four shock absorbers are installed on the equipment, of which two are metal spring shock absorbers and two are rubber shock absorbers. The shock absorber parameters are measured using the MTS831.50 dynamic stiffness tester. The stiffness of the metal spring shock absorber is 5×10^5N / m, and the damping coefficient is 1000N·s / m; the rubber shock absorber introduces nonlinear parameters, and the stiffness is expressed as

[0038] k=k0+k1x+k2x^2, where k0=3×10^5N / m, k1=1×10^6N / m^2, k2=5×10^7N / m^3. By calculating the modal participation factor, 15 key nodes with a contribution rate of more than 5% were selected. Modal analysis was performed, and the first three natural frequencies were obtained to be 15Hz, 28Hz, and 43Hz respectively. The vibration transfer function was constructed, and the input was a random vibration of 0-200Hz, with a PSD of 0.1g^2 / Hz. The fourth-order Runge-Kutta method was used to solve the time domain response, and the time step was set to 0.25ms. A central composite test design of 23 operating points was adopted, and a second-order polynomial was used to fit the response surface. 10,000 sets of Monte Carlo simulation samples were generated by Latin hypercube sampling, and the maximum response value, root mean square value and frequency response function of each key node were obtained by statistical analysis. The final evaluation results showed that within the 95% confidence interval, the maximum vibration displacement of the equipment did not exceed 0.8mm and the maximum acceleration did not exceed 5g.

[0039] Step S105, accurately compensating the original vibration signal according to the damping characteristics of the shock absorber, the corrected vibration displacement and velocity, filtering the compensated signal through a digital filter to obtain a stable vibration signal.

[0040] The damping characteristic curve of the shock absorber is obtained, and a nonlinear mapping relationship between the damping force and the vibration displacement and velocity is established according to the damping characteristic curve; the nonlinear characteristics are processed by a piecewise linearization method, and the damping characteristic curve is divided into a plurality of linear intervals; a local linear damping characteristic function is obtained by fitting the linear interval using the least square method; the corrected vibration displacement and velocity are substituted into the local linear damping characteristic function to calculate the actual damping force; the original vibration signal is converted into the frequency domain by using the fast Fourier transform, and a frequency domain compensation function is constructed according to the actual damping force; the amplitude of the original signal is corrected, and the amplitude compensation coefficient is an ideal The ratio of the damping force to the actual damping force; the compensated frequency domain signal is converted back to the time domain by inverse fast Fourier transform to obtain a preliminary compensated time domain vibration signal; the preliminary compensated time domain vibration signal is subjected to multi-scale analysis by wavelet decomposition to identify and remove the 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 signals before and after filtering are calculated to evaluate the stability of the compensated and filtered signals; if the variance is reduced by more than a preset threshold and the signal-to-noise ratio is improved by more than a preset threshold, it is determined that a stable vibration signal is obtained.

[0041] Specifically, according to the damping characteristic curve of the shock absorber, the nonlinear mapping relationship between the damping force and the vibration displacement and velocity is established. For the rubber shock absorber, the piecewise linearization method is used to process the nonlinear characteristics, and the characteristic curve is divided into multiple linear intervals. The local linear damping characteristic function is fitted in each interval by the least square 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 by fast Fourier transform. According to the calculated actual damping force, the 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. The phase compensation is achieved by calculating the phase difference between the ideal damping force and the actual damping force. In order to avoid discontinuity introduced by compensation, the smoothing window function is used to smooth the compensation function. The compensated frequency domain signal is converted back to the time domain by inverse fast Fourier transform to obtain the initial compensated time domain vibration signal. The signal is multi-scale analyzed by 3-layer db4 wavelet decomposition to identify and remove the high-frequency noise and unstable components introduced by compensation. An 8th-order Butterworth low-pass filter is designed, and the cutoff frequency is set to 1.5 times the main frequency component of the signal. The denoised compensation signal is filtered to eliminate residual high-frequency interference. The stability of the compensated and filtered signals is evaluated by calculating the variance ratio and signal-to-noise ratio improvement of the signals before and after filtering. If the variance is reduced by more than 50% and the signal-to-noise ratio is improved by more than 6dB, it is determined that a stable vibration signal is obtained. In the vibration analysis of an industrial equipment, the nonlinear damping characteristic curve of the shock absorber is measured by the MTS831.50 dynamic stiffness tester. The curve is divided into 5 linear intervals, and the local linear function is fitted in each interval using the least squares method. The corrected vibration displacement peak is 0.5mm, and the velocity peak is 0.05m / s. Substituting into the corresponding interval function, the actual damping force is 850N. The original vibration signal with a sampling rate of 10kHz was converted to the frequency domain using 8192-point FFT, and a frequency domain compensation function was constructed. The amplitude compensation coefficient decreased from 1.2 to 1.0 in the range of 0-100Hz, and the phase compensation decreased from 15° to 0° in the range of 0-100Hz. The compensation function was smoothed using the Hanning window function, and the window length was 5% of the compensation function length. The compensated frequency domain signal was converted back to the time domain using IFFT to obtain a preliminary compensated vibration signal. The Matlab wavelet toolbox was used to perform a three-layer db4 wavelet decomposition, and the coefficients with an absolute value less than 1.5 times the mean in the highest frequency band were removed. An 8th-order Butterworth low-pass filter was designed with a cutoff frequency of 150Hz, and the filtfilt function was used for zero-phase filtering. The variance ratio of the signal before and after filtering was calculated to be 0.45, and the signal-to-noise ratio was improved by 8.3dB, and it was determined that a stable vibration signal was obtained. The peak-to-peak value of the stable vibration signal finally obtained was 1.2mm, and the amplitude of the main frequency of 100Hz was 0.4mm.

[0042] Step S106, an abnormal analysis is performed on the stabilized vibration signal to identify abnormal vibration patterns, and an equipment fault warning model is established based on the identification results. Based on the vibration abnormality information in the fault warning model and the position of the shock absorber in the three-dimensional point cloud model of the environment, the warning information and on-site vibration data are transmitted to the remote expert terminal in real time.

[0043] Obtain the stabilized vibration signal, perform time-frequency analysis on the vibration signal, use db4 wavelet to perform 3-layer wavelet packet transform, and obtain the time-frequency characteristics of the vibration signal; calculate the energy percentage and entropy value of 8 sub-bands according to the time-frequency characteristics, and construct a 16-dimensional feature vector; reduce the dimension of the feature vector through principal component analysis to obtain low-dimensional features that characterize the vibration mode; use the isolation forest algorithm to perform anomaly detection on the low-dimensional features and identify abnormal vibration modes; match the abnormal vibration mode with a pre-established fault type library to determine the potential fault type and severity; based on the abnormal vibration mode and fault type, construct a multi-classification fault warning model of a support vector machine based on the RBF kernel; extract the precise position coordinates of the shock absorber from the three-dimensional point cloud model of the environment; perform multi-source data fusion of the position coordinates and the output result of the fault warning model through a Kalman filter to obtain fused warning information; use OPC The UA protocol transmits the fused warning information and real-time vibration data to the remote expert terminal through the industrial Ethernet; if the remote expert terminal receives the warning information and real-time vibration data, the LZ4 algorithm is used to compress the warning information and real-time vibration data in real time; the AES-256 encryption algorithm is used to encrypt the compressed data; and the encrypted data is transmitted through the UDP protocol.

[0044] Specifically, the time-frequency analysis of the stabilized vibration signal is carried out, and the db4 wavelet is used to perform a three-layer wavelet packet transform to extract the time-frequency characteristics of the signal. The energy percentage and entropy value of the 8 sub-bands are calculated to construct a 16-dimensional feature vector. The feature dimension is reduced by principal component analysis, and the principal components with a cumulative contribution rate of 95% are retained to obtain low-dimensional features that characterize the vibration mode. The isolation forest algorithm is used to detect anomalies in the reduced-dimensional features. 100 trees are set, the subsampling size is 256, and the anomaly threshold is -0.5. The abnormal vibration mode is identified and matched with the pre-established fault type library to determine the potential fault type and severity, and generate labeled data for fault warning model training. Based on the identified abnormal vibration mode and fault type, a multi-classification fault warning model of a support vector machine based on the RBF kernel is constructed. The kernel parameter γ and the penalty coefficient C are optimized through grid search and 5-fold cross validation to improve the generalization ability and prediction accuracy of the model. The precise position coordinates of the shock absorber are extracted from the three-dimensional point cloud model of the environment. The position information and the output results of the fault warning model are fused through the Kalman filter for multi-source data. The fused warning information and real-time vibration data are transmitted to the remote expert terminal through the industrial Ethernet using the OPCUA protocol. The transmission data is compressed in real time using the LZ4 algorithm, the AES-256 encryption algorithm is used to ensure the security of data transmission, and low-latency real-time data transmission is achieved through the UDP protocol. In a vibration monitoring system of an industrial equipment, a stable vibration signal with a sampling rate of 10kHz is analyzed. The db4 wavelet 3-layer wavelet packet transform is implemented using the PyWavelets library, and 8 frequency bands are obtained, 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 to construct a 16-dimensional feature vector. The PCA algorithm of the scikit-learn library was used to reduce the dimension, retaining 4 principal components with a cumulative contribution rate of 96.8%. Anomaly detection was performed using an isolation forest model with 100 trees and a subsampling size of 256. The anomaly threshold was set to -0.5, and 3 abnormal vibration modes were identified. The abnormal mode was compared with a library containing 10 fault types and was determined to be a bearing inner race fault with a moderate severity. The RBF kernel SVM multi-classification model was implemented using scikit-learn's SVC, and parameter optimization was performed using GridSearchCV. The optimal parameters were C=10, γ=0.01, and the 5-fold cross validation accuracy reached 92%. The coordinates of the 4 shock absorbers were extracted from the point cloud model with sub-millimeter accuracy. The Kalman filter was implemented using the FilterPy library to fuse the position information and fault warning results. The OPCUA server was implemented using the python-opcua library, and the data update frequency was 100Hz.The lz4 library is used to compress the transmission data, with a compression ratio of 1:5. The AES-256 algorithm of the pycryptodome library is used to encrypt the data. The asyncio library is used to implement asynchronous transmission based on UDP, with an average delay of less than 10ms. The remote expert terminal receives 100kb of compressed and encrypted data per second, including complete vibration waveforms and fault warning information.

[0045] Step S107, decompress and restore the received warning information and on-site vibration data, and remotely monitor and diagnose the equipment through the vibration status and fault warning information of the equipment. If data transmission abnormality or equipment failure occurs, an alarm prompt will be issued according to the abnormality handling mechanism, and the equipment will be shut down or maintained urgently, realizing product positioning guidance of industrial vision.

[0046] Receive the compressed and encrypted warning information and vibration data, decrypt the warning information and vibration data using the AES-256 decryption algorithm, and restore the original data using the LZ4 decompression algorithm. Store the decompressed original data in the Redis high-performance cache database. Extract the latest equipment vibration status and fault warning information from the Redis cache database, and use the InfluxDB time series database to store historical data. Use a 60-second sliding time window to calculate the root mean square value and peak factor of the vibration signal, and build an equipment health index in combination with the warning information. Judge the equipment operation status according to the equipment health index. If data transmission abnormality or equipment failure is detected, trigger the exception handling mechanism. Automatically select the processing strategy according to the preset decision tree algorithm, and the decision tree contains data integrity, vibration amplitude, and frequency feature nodes. Communicate with the on-site PLC through the OPCUA protocol to perform corresponding processing operations. Based on the real-time image collected by the industrial camera, use the YOLOv5 target detection algorithm to identify the product location, and combine the equipment three-dimensional model and vibration data to calculate the precise coordinates of the product relative to the equipment. Use WebGL technology to overlay and display product positioning information and equipment status indication on the remote monitoring interface.

[0047] Specifically, the compressed and encrypted warning information and on-site vibration data are received, the AES-256 decryption algorithm is used to decrypt the data, the LZ4 decompression algorithm is used to restore the original data, the CRC32 checksum algorithm is used to verify the data integrity, and the decompressed data is stored in the Redis high-performance cache database. The latest equipment vibration status and fault warning information are extracted from the Redis cache, and the 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 a 50% overlap rate. The equipment health index is constructed in combination with the warning information, and the equipment operation status is determined by threshold judgment. If data transmission anomalies or equipment failures are detected, the exception handling mechanism is triggered, and the processing strategy is automatically selected according to the preset decision tree algorithm. The decision tree contains data integrity, vibration amplitude, and frequency feature nodes. The branch conditions cover three levels: normal, warning, and severe. The alarm prompt is sent, maintenance suggestions or emergency shutdown instructions are generated, and the OPCUA protocol is used to communicate with the on-site PLC to perform shutdown operations. At the same time, the abnormal events are recorded in the MongoDB log database. Based on the real-time images collected by industrial cameras, the YOLOv5 target detection algorithm is used to identify the product location. Combined with the equipment 3D model and vibration data, the precise coordinates of the product relative to the equipment are calculated. The product positioning information and equipment status indication are superimposed and displayed on the remote monitoring interface through WebGL technology, realizing the product positioning guidance and equipment status visualization of industrial vision. In a remote monitoring system of industrial equipment, 100KB of compressed encrypted data is received per second, and the AES-256-CBC mode of the OpenSSL library is used for decryption, with a key length of 256 bits. The LZ4 algorithm is decompressed with a 4KB block size, and the average decompression speed reaches 5GB / s. The CRC32 check uses the IEEE802.3 standard polynomial 0x04C11DB7 to check 32-bit data blocks. The decompressed data is stored in the Redis6.0 cluster, using 3 master nodes, 3 slave nodes, and a total memory of 24GB. InfluxDB2.0 stores 30 days of historical data, using the TSM engine, with a compression ratio of 10:1. The vibration signal is collected at a sampling rate of 10kHz, and the RMS value, peak factor, skewness, and kurtosis are calculated in a 60-second window. The equipment health index is in the range of 0-100, and the thresholds are set to warning = 70 and severe = 40. The decision tree has a depth of 4 and contains 7 leaf nodes. It is trained using the CART algorithm with an accuracy of 95%. It communicates with Siemens S7-1500 PLC via the OPC UA protocol, with a response time of <100ms. The YOLOv5 model runs on an NVIDIA T4 GPU with a frame rate of 30fps and mAP@0.5 of 0.89. WebGL uses the Three.js library to render 3D scenes, and displays product location and equipment status smoothly at 60fps. The average delay of the entire remote monitoring system is <500ms, with a reliability of 99.99%, and supports simultaneous online monitoring of 100 devices.

[0048] The above only lists some preferred embodiments of the present invention, but the present invention is not limited thereto, and many improvements and changes can be made. As long as the improvements and changes are made on the basis of the basic principles of the present invention, they should be regarded as falling within the protection scope of the present invention.

Claims

1. A product positioning guidance method for industrial vision, characterized in that: The method comprises: According to the vibration frequency characteristics of precision equipment in industrial environments, select the shock absorber type and damping coefficient, and match the spring stiffness to form a shockproof structure that meets the shockproof requirements of precision equipment; An industrial camera is placed according to the anti-vibration structure that meets the anti-vibration requirements of precision equipment. The image of the vibrating equipment on site is collected by the industrial camera. The image is pre-processed according to the characteristics of the anti-vibration material. The texture and shape features of the image are extracted through the convolutional neural network. At the same time, the network is fine-tuned by fine-tuning the large model to adapt to the characteristic changes of the vibrating equipment in the current industrial environment and extract the image feature information. Based on the extracted image feature information and the layout of the anti-vibration device, a three-dimensional point cloud model of the environment including the location of the shock absorber is constructed, and the shock absorber in the three-dimensional point cloud model of the environment is identified and marked; Based on the markers of the shock absorbers in the three-dimensional point cloud model of the environment, the vibration acceleration data of the equipment is obtained, the vibration response of the shockproof device is estimated, and the corrected vibration displacement and velocity are obtained; According to the damping characteristics of the shock absorber, the corrected vibration displacement and velocity, the original vibration signal is accurately compensated, and the compensated signal is filtered through a digital filter to obtain a stable vibration signal; Perform abnormal analysis on the stabilized vibration signal to identify abnormal vibration patterns, establish an equipment fault warning model based on the identification results, and transmit the warning information and on-site vibration data to the remote expert terminal in real time based on the vibration abnormality information in the fault warning model and the position of the shock absorber in the environmental 3D point cloud model; Decompress and restore the received warning information and on-site vibration data, and remotely monitor and diagnose the equipment through the equipment's vibration status and fault warning information. If data transmission abnormalities or equipment failures occur, alarm prompts are given according to the exception handling mechanism, and the equipment is urgently shut down or maintained, thus realizing product positioning guidance with industrial vision.

2. The method according to claim 1, wherein: According to the vibration frequency characteristics of precision equipment in the industrial environment, the shock absorber type and damping coefficient are selected, and the spring stiffness is matched to form a shockproof structure that meets the shockproof requirements of precision equipment, including: Receive a data collection instruction carrying a unique identification number of the precision equipment, the data collection instruction including vibration frequency characteristic data, equipment weight data and installation space data, and simultaneously record the temperature, humidity and corrosive gas of the industrial environment; Calculating a main frequency range according to the vibration frequency characteristic data, wherein the main frequency range is combined with equipment weight data, installation space data, and environmental characteristic parameters to select a shock absorber type; For the selected shock absorber type, a genetic algorithm is used to optimize and calculate the damping coefficient, and the optimization objective function of the genetic algorithm is set to minimize the system vibration transmissibility; Acquiring a damping coefficient obtained by the genetic algorithm optimization, wherein the damping coefficient is combined with shock absorber type data to establish a finite element model of a shock absorption system; The spring stiffness is optimized by using a binary search method, wherein the convergence condition of the binary search method is set as that the difference between the natural frequency of the system and the vibration frequency of the equipment is less than a preset threshold value; Constructing a mathematical model of the seismic-proof structure from the shock absorber type and damping coefficient and the optimal spring stiffness; A time domain response analysis is implemented in the mathematical model of the earthquake-proof structure. The time domain response analysis selects a random vibration excitation signal that meets the characteristics of the industrial environment to calculate the displacement response and acceleration response of the structure.

3. The method according to claim 1, wherein: The industrial camera is placed according to the anti-vibration structure that meets the anti-vibration requirements of precision equipment, and the image of the vibrating equipment on site is collected by the industrial camera. The image is pre-processed according to the characteristics of the anti-vibration material, and the texture and shape features of the image are extracted by the convolutional neural network. At the same time, the network is fine-tuned by fine-tuning the large model to adapt to the characteristic changes of the vibrating equipment in the current industrial environment, and the image feature information is extracted, including: Calculate the optimal shooting distance and angle of the industrial camera according to the frequency response characteristics of the shockproof structure, adjust the installation position of the industrial camera, and obtain a sequence of images of the vibrating device; Using a median filter algorithm to perform noise reduction processing on the image sequence to obtain a first image; Performing contrast enhancement on the first image using an adaptive histogram equalization method to obtain a second image; Construct a multi-layer convolutional neural network structure and use the ResNet50 pre-trained model to import initial weights; Fine-tune the convolutional neural network using the second image to obtain a fine-tuned convolutional neural network; Extracting a multi-scale feature map from the fine-tuned convolutional neural network; Calculating a feature covariance matrix of the multi-scale feature map; Performing eigenvalue decomposition on the characteristic covariance matrix; Select the principal component whose contribution rate exceeds the preset threshold value to obtain the feature vector after dimension reduction; Grouping the feature vectors by a k-means clustering algorithm; If the result of the feature vector grouping meets a preset condition, it is determined that the feature vector is a low-dimensional feature representation that characterizes the feature change of the vibration equipment in the current industrial environment.

4. The method according to claim 1, wherein: The method of constructing an environmental three-dimensional point cloud model including the location of the shock absorber based on the extracted image feature information and in combination with the layout of the shockproof device, and identifying and marking the shock absorber in the environmental three-dimensional point cloud model includes: Acquire a multi-frame environment depth map, where the multi-frame environment depth map is collected by a structured light scanner; Generate a dense point cloud data sequence according to the multi-frame environment depth map, wherein the dense point cloud data sequence is obtained by mapping two-dimensional image pixels and their features to three-dimensional space coordinates through a back-projection algorithm; Performing registration processing on the dense point cloud data sequence to obtain a three-dimensional point cloud model of the environment, wherein the registration processing adopts an iterative closest point algorithm, and if the number of iterations reaches a preset maximum number of iterations or the convergence error is less than a preset convergence threshold, the iteration is stopped; Extracting local geometric features and texture features from the three-dimensional point cloud model of the environment, wherein the local geometric features include normal vectors and curvatures, and the texture features include spin graphs and fast point feature histograms; Using a support vector machine classifier to classify each point in the three-dimensional point cloud model of the environment, the support vector machine classifier uses a radial basis kernel function and is trained using a labeled shock absorber point cloud data set; Performing regional growth on the points classified as shock absorbers in the three-dimensional point cloud model of the environment to obtain a complete shock absorber entity, and if the distance between adjacent points is less than a preset growth threshold, adding the adjacent points to the shock absorber entity; Performing spatial marking on the shock absorber entity, wherein the spatial marking is implemented by using a minimum bounding box method; It also includes: stitching the depth image features from different perspectives to obtain a complete three-dimensional point cloud model of the shock absorber's surrounding environment; searching for areas similar to the CAD model in the environmental point cloud based on a pre-established three-dimensional CAD model library of the shock absorber; preliminarily determining the rough position of the shock absorber; extracting a point cloud subset of the rough position of the shock absorber; further refining the precise boundary and posture information of the shock absorber through a point cloud segmentation algorithm; and marking the shock absorber as identified.

5. The method according to claim 4, wherein: The depth image features of different perspectives are stitched together to obtain a complete 3D point cloud model of the surrounding environment of the shock absorber. According to the pre-established 3D CAD model library of the shock absorber, an area similar to the CAD model is searched in the environmental point cloud, the rough position of the shock absorber is preliminarily determined, and a point cloud subset of the rough position of the shock absorber is extracted. The precise boundary and posture information of the shock absorber are further refined through a point cloud segmentation algorithm, and the shock absorber is marked as recognized, including: Acquire a depth image of the environment around the shock absorber collected by a multi-view depth camera, apply bilateral filtering to the depth image to remove noise, and obtain a filtered depth image; According to the filtered depth image, SIFT feature point matching and RANSAC algorithm are used to calculate the transformation matrix between adjacent perspectives, and an iterative closest point algorithm is used to align and fuse multiple depth images to obtain a three-dimensional point cloud model of the environment; Extracting key geometric features from a pre-established three-dimensional CAD model library of a shock absorber, wherein the key geometric features include surface normal vector distribution, curvature distribution, and local shape descriptors, and constructing a feature description vector; Searching for an area similar to the feature description vector in the three-dimensional point cloud model of the environment through a sliding window, and if a high-similarity area is identified, performing a cluster analysis on the high-similarity area; The point cloud subset obtained by the cluster analysis is finely segmented using a region growing algorithm, and the precise boundary of the shock absorber is obtained through iterative expansion. The principal component analysis method is used to calculate the principal axis direction and center position of the shock absorber to determine the precise position and posture of the shock absorber.

6. The method according to claim 1, wherein: The method of obtaining equipment vibration acceleration data based on the mark of the shock absorber in the three-dimensional point cloud model of the environment, estimating the vibration response of the shockproof device, and obtaining the corrected vibration displacement and velocity includes: Determine the shock absorber mark position according to the environmental three-dimensional point cloud model, and install a three-axis acceleration sensor on each shock absorber after obtaining the shock absorber mark position; The acceleration signal collected by the three-axis acceleration sensor is subjected to spectrum analysis using a fast Fourier transform algorithm, and the main vibration frequency and amplitude of each shock absorber in the three directions of x, y and z are extracted from the spectrum analysis results to construct a vibration characteristic vector; According to the pre-established mass-spring-damper dynamic model of the shockproof device, the vibration characteristic vector is used as input, and the vibration response of the shockproof device is estimated in real time through the Kalman filter algorithm to obtain the instantaneous displacement and velocity state of each shock absorber; The instantaneous displacement and velocity state are integrated with the shock absorber position information in the three-dimensional point cloud model of the environment by using a Bayesian fusion algorithm, and the instantaneous displacement and velocity state are corrected by a weighted least square method; Using a high-speed camera to collect the motion trajectory of the marking point on the surface of the shock absorber, extracting the position of the marking point from the motion trajectory, comparing the position of the marking point with the corrected instantaneous displacement and velocity state, and judging the accuracy of the corrected instantaneous displacement and velocity state; It also includes: establishing a vibration transfer model based on the different types and locations of shock absorbers, combining the stiffness and damping parameters of the shock absorbers, simulating the propagation and attenuation characteristics of the vibration signal in the equipment structure, so as to obtain a comprehensive equipment vibration response evaluation result.

7. The method according to claim 6, wherein: The vibration transmission model is established according to the shock absorbers of different types and positions, and the propagation and attenuation characteristics of the vibration signal in the equipment structure are simulated by combining the stiffness and damping parameters of the shock absorbers to obtain comprehensive equipment vibration response evaluation results, including: The finite element method is used to construct the geometric model of the equipment structure and the shock absorber is simplified into a spring-damper unit; Measuring the stiffness and damping parameters of the shock absorber using a dynamic stiffness tester to obtain the physical properties of the shock absorber; According to the physical characteristics, assigning corresponding physical parameters to each unit in the geometric model; Using modal analysis methods to calculate the natural frequency and vibration mode of the geometric model to obtain the dynamic characteristics of the equipment structure; Constructing a vibration transfer function, converting the vibration signal at the input end into the response of each key node through the transfer function, and obtaining the vibration propagation characteristics of the equipment structure; Using the response surface method to fit the vibration propagation characteristics, and obtain the vibration response of the equipment structure under different working conditions; The sampling points of the response surface method are determined by using a central composite experimental design method to obtain a mapping relationship between the vibration response and the shock absorber parameters and the vibration input; Monte Carlo simulation samples are generated through Latin hypercube sampling, and statistical analysis is performed on the mapping relationship to obtain a comprehensive evaluation result of the vibration response of the equipment structure.

8. The method according to claim 1, wherein: The method comprises: accurately compensating the original vibration signal according to the damping characteristics of the shock absorber, the corrected vibration displacement and speed, filtering the compensated signal through a digital filter, and obtaining a stable vibration signal, including: Acquire a damping characteristic curve of the shock absorber, and establish a nonlinear mapping relationship between the damping force and the vibration displacement and velocity according to the damping characteristic curve; A piecewise linearization method is used to process nonlinear characteristics, and the damping characteristic curve is divided into a plurality of linear intervals; A local linear damping characteristic function is obtained by fitting the linear interval using the least square method; Substituting the corrected vibration displacement and velocity into the local linear damping characteristic function to calculate the actual damping force; The original vibration signal is converted into the frequency domain by using a fast Fourier transform, and a frequency domain compensation function is constructed according to the actual damping force; The amplitude of the original signal is corrected, and the amplitude compensation coefficient is the ratio of the ideal damping force to the actual damping force; The compensated frequency domain signal is converted back to the time domain by using inverse fast Fourier transform to obtain a preliminary compensated time domain vibration signal; Performing multi-scale analysis on the time-domain vibration signal of the preliminary compensation by wavelet decomposition to identify and remove high-frequency noise and unstable components introduced by the compensation; Design a Butterworth low-pass filter to filter the denoised compensation signal and eliminate residual high-frequency interference; Calculate the variance ratio and signal-to-noise ratio improvement of the signal before and after filtering, and evaluate the stability of the compensated and filtered signal; If the variance decreases by more than a preset threshold and the signal-to-noise ratio increases by more than a preset threshold, it is determined that a stable vibration signal is obtained.

9. The method according to claim 1, wherein: The method includes: performing abnormal analysis on the stabilized vibration signal, identifying abnormal vibration patterns, establishing an equipment fault warning model based on the identification results, and transmitting the warning information and on-site vibration data to the remote expert terminal in real time based on the vibration abnormality information in the fault warning model and the position of the shock absorber in the three-dimensional point cloud model of the environment, including: Acquire a stabilized vibration signal, perform time-frequency analysis on the vibration signal, use db4 wavelet to perform a three-layer wavelet packet transform, and obtain the time-frequency characteristics of the vibration signal; Calculate the energy percentage and entropy value of the eight sub-bands according to the time-frequency characteristics, and construct a 16-dimensional feature vector; Reducing the dimension of the feature vector by principal component analysis to obtain low-dimensional features representing the vibration mode; Using an isolation forest algorithm to perform anomaly detection on the low-dimensional features to identify abnormal vibration patterns; Matching the abnormal vibration pattern with a pre-established fault type library to determine potential fault types and severity; Based on the abnormal vibration mode and fault type, a multi-classification fault warning model of a support vector machine based on an RBF kernel is constructed; Extract the precise position coordinates of the shock absorber from the 3D point cloud model of the environment; The position coordinates and the output result of the fault warning model are subjected to multi-source data fusion through a Kalman filter to obtain fused warning information; The fused warning information and real-time vibration data are transmitted to a remote expert terminal via industrial Ethernet using OPCUA protocol; If the remote expert terminal receives the warning information and the real-time vibration data, the warning information and the real-time vibration data are compressed in real time using the LZ4 algorithm; The compressed data is encrypted using an AES-256 encryption algorithm; The encrypted data is transmitted via the UDP protocol.

10. The method according to claim 1, wherein: The decompression and restoration are performed based on the received warning information and on-site vibration data, and the remote monitoring and diagnosis of the equipment is performed through the vibration status and fault warning information of the equipment. If data transmission abnormality or equipment failure occurs, an alarm prompt is issued according to the abnormality handling mechanism, and the equipment is urgently shut down or maintained, so as to realize the product positioning guidance of industrial vision, including: receiving compressed and encrypted warning information and vibration data, decrypting the warning information and vibration data using an AES-256 decryption algorithm, and restoring the original data using an LZ4 decompression algorithm; The decompressed original data is stored in a Redis high-performance cache database; Extract the latest equipment vibration status and fault warning information from the Redis cache database, and use the InfluxDB time series database to store historical data; The root mean square value and peak factor of the vibration signal are calculated using a 60-second sliding time window, and the equipment health index is constructed in combination with the warning information; The device operating status is determined based on the device health index, and if data transmission anomalies or device failures are detected, an exception handling mechanism is triggered; Automatically select a processing strategy based on a preset decision tree algorithm, wherein the decision tree includes data integrity, vibration amplitude, and frequency feature nodes; Communicate with the on-site PLC through the OPCUA protocol and perform corresponding processing operations; Based on the real-time images collected by industrial cameras, the YOLOv5 target detection algorithm is used to identify the product location, and the precise coordinates of the product relative to the equipment are calculated by combining the equipment 3D model and vibration data; Product location information and equipment status indications are superimposed and displayed on the remote monitoring interface through WebGL technology.

Images