Automobile spoiler manipulator polishing system and polishing method

Abstract

The invention relates to the field of grinding and polishing, in particular to an automobile spoiler mechanical arm grinding system and method.The automobile spoiler mechanical arm grinding system comprises a path planning module, a state monitoring module, a parameter compensation module and a waste collecting module which are in communication connection, and the path planning module is used for scanning a spoiler in real time through a visual sensor and dynamically adjusting a grinding path; the visual sensor is a 3D structured light sensor or a stereoscopic visual camera; the state monitoring module is used for monitoring the vibration state of the grinding head in real time through the vibration sensor and obtaining vibration abnormal data. The parameter compensation module is used for automatically adjusting grinding parameters according to the abnormal vibration data and adjusting a grinding head for grinding; the waste collecting module is used for collecting and treating dust and waste generated in the polishing process in a centralized mode. The polishing path can be precisely planned, the polishing quality is improved, the production efficiency is improved, the cost is reduced, and dust and waste materials are effectively treated.

Description

Technical Field

[0001] The invention relates to the field of grinding and polishing, and in particular to a car tail wing manipulator grinding system and a grinding method. Background Art

[0002] With the development of the automobile industry, the production quality requirements of automobile tail wings, as important components to improve automobile performance and aesthetics, are increasing day by day. In the manufacturing process of automobile tail wings, polishing is one of the key processes. In the prior art, the polishing path planning of the manipulator is not accurate enough, and it is often a preset path, which cannot be adaptively adjusted according to the actual shape change of the tail wing, which can easily lead to uneven polishing, omissions or over-polishing. Moreover, during the polishing process, the vibration state of the polishing head cannot be effectively monitored. When abnormal vibration occurs, such as uneven tail wing material, hard spots and other problems, the polishing parameters cannot be adjusted in time, which not only affects the polishing quality, but also may damage the polishing tools and tail wing. In addition, the imperfect collection and treatment of waste and dust can easily cause environmental pollution and affect the working environment and the health of operators. Summary of the invention

[0003] In order to solve the above problems, the present invention provides a car tail wing manipulator grinding system and a grinding method, which can accurately plan the grinding path to improve the grinding quality, improve production efficiency, reduce costs and effectively handle dust and waste.

[0004] To achieve the above object, the technical solution adopted by the present invention is:

[0005] A car tail wing manipulator polishing system includes a path planning module, a state monitoring module, a parameter compensation module and a waste collection module connected in communication, wherein:

[0006] The path planning module is used to scan the tail wing in real time through a visual sensor and dynamically adjust the polishing path; the visual sensor is a 3D structured light sensor or a stereo vision camera;

[0007] The state monitoring module is used to monitor the vibration state of the grinding head in real time through the vibration sensor and obtain vibration abnormality data;

[0008] The parameter compensation module is used to automatically adjust the grinding parameters according to the abnormal vibration data and adjust the grinding head for grinding;

[0009] The waste collection module is used to centrally collect and process the dust and waste generated during the grinding process.

[0010] Furthermore, the real-time scanning of the tail wing by a visual sensor and the dynamic adjustment of the grinding path include the following steps:

[0011] The visual sensor is fixedly connected to the top or side of the manipulator through a bracket and a shock-absorbing device;

[0012] Connect the data acquisition computer and the manipulator control system via a two-way communication link;

[0013] The polishing path is planned and dynamically adjusted through the path planning model;

[0014] During the grinding process, the visual sensor parameters are continuously monitored by the monitoring program in the data acquisition computer and the manipulator control system. When the sensor parameters exceed the set threshold, an alarm is issued and the grinding is suspended.

[0015] Furthermore, the path planning model is implemented by the A* algorithm and the feature-based model matching algorithm, wherein:

[0016] The A* algorithm is used to search from the starting point to the target point in path planning, generate the optimal path based on the polishing cost, and provide a reference path for subsequent dynamic adjustments;

[0017] The feature-based model matching algorithm is used to quantify the deviation between the tail wing model and the initial model after the real-time feature alignment with the initial model, and dynamically adjust the decision basis of the polishing path.

[0018] Furthermore, the A* algorithm comprises the following steps:

[0019] Start the visual sensor to perform an all-round scan of the tail wing placed and fixed on the workbench, and perform three-dimensional reconstruction of each frame of the image using a three-dimensional reconstruction algorithm based on stereo vision to obtain a three-dimensional point cloud model of the tail wing;

[0020] The three-dimensional space of the tail wing is divided into grids of uniform size, with the starting position of the grinding tool as the starting point and the set of target positions on the tail wing surface that need to be polished as the target point set. The A* algorithm searches for the path by maintaining an open list and a closed list.

[0021] In each iteration, the node with the smallest cost function value is selected from the open list to determine the optimal path. The algorithm ends when the target point is added to the closed list. The initial polishing path is obtained by backtracing the parent node from the target point to the starting point and serves as the input of the feature-based model matching algorithm.

[0022] Furthermore, the feature-based model matching algorithm comprises the following steps:

[0023] During the polishing process, the visual sensor continuously scans the tail wing at a set frequency, converts the real-time image frame into a real-time tail wing 3D point cloud model through a 3D reconstruction algorithm, and extracts features from the corresponding model of the initial polishing path and the real-time model;

[0024] Corner features are extracted using the 3D Harris corner detection algorithm, and edge features are obtained by calculating the gradient direction change of the point cloud. The extracted feature point sets are the initial model feature point set and the real-time model feature point set.

[0025] The nearest neighbor matching algorithm is used to calculate the distance between each feature point in the real-time model feature point set and the closest feature point in the initial model feature point set. The transformation relationship that minimizes the sum of distances between all matching feature points is found through iterative optimization. The transformation relationship includes: translation vector, rotation matrix and scaling factor.

[0026] According to the transformation relationship obtained by model matching, the deviation between the real-time tail surface and the initial grinding path is analyzed, and the rotation angle deviation and scaling factor deviation are analyzed to determine the grinding path that needs to be adjusted;

[0027] The grinding path is modified according to the path adjustment decision, and the adjusted grinding path is fed back to the robot control system again.

[0028] Furthermore, the real-time monitoring of the vibration state of the grinding head by the vibration sensor to obtain abnormal vibration data includes the following steps:

[0029] A vibration sensor is connected between the grinding head and the manipulator, and the vibration sensor is connected to a data acquisition card via a wired method. The data acquisition card continuously collects output signals of the vibration sensor during the grinding process and performs signal preprocessing.

[0030] The DC component of the vibration signal is obtained by calculating the signal mean, the discrete degree of the signal is obtained by calculating the standard deviation of the signal, the peak factor of the signal is extracted to determine the presence of impact vibration, the signal spectrum is obtained by performing fast Fourier transform on the signal, and the spectrum amplitude is calculated. The main frequency component of the vibration is determined by the frequency point with a larger amplitude;

[0031] According to the anomaly detection algorithm, thresholds are determined for the signal mean, standard deviation, peak factor and main frequency amplitude. When multiple features exceed the threshold at the same time, it is judged as abnormal vibration. By performing statistical analysis on a large amount of vibration feature data under normal grinding conditions, the mean and standard deviation of the distribution are estimated, and the probability of the newly collected vibration feature data under the assumed distribution is calculated. When it is lower than the probability threshold, it is judged as abnormal vibration and the vibration feature data is recorded.

[0032] Furthermore, after the abnormal vibration is determined, abnormal vibration detection is performed through a support vector machine or a multi-layer perceptron to identify abnormal vibrations in complex patterns.

[0033] Furthermore, the automatic adjustment of grinding parameters according to the abnormal vibration data and the adjustment of the grinding head for grinding include the following steps:

[0034] The abnormal vibration data is fuzzified into fuzzy language variables, and the fuzzy set in each fuzzy language variable corresponds to a membership function; according to the fuzzy rule base, the fuzzified vibration characteristics are associated with the grinding parameter adjustment amount; according to the current fuzzified vibration characteristics, the fuzzy reasoning algorithm is used to infer in the rule base to obtain the fuzzified grinding parameter adjustment amount; the fuzzified adjustment amount is converted into an accurate parameter adjustment value;

[0035] The calculated grinding parameter adjustment value is sent to the robot control system to perform the parameter adjustment operation. After the parameter adjustment, the vibration of the grinding head is continuously monitored to determine whether it has returned to normal. If it is still abnormal, repeat the above steps to adjust the grinding parameters again, and then use machine learning to assist in the adjustment.

[0036] Furthermore, the dust and waste generated during the grinding process are collected and processed in a centralized manner, including the following steps:

[0037] A dust suction port is arranged under the tail wing, and each dust suction port is connected to a central dust collection device through a pipeline;

[0038] Start the dust collection system before grinding, and control the air volume and speed of the dust collection system by adjusting the fan speed and valve opening. Install a pressure difference sensor in the pipeline system to monitor the pressure difference at both ends of the pipeline to determine whether the pipeline is blocked and perform flow balance control.

[0039] Waste separation and classification are carried out through cyclone separation technology. During the waste treatment process, the operating parameters of the treatment equipment are monitored, and waste information is recorded to establish a waste treatment file.

[0040] A method for polishing a car rear wing robot comprises the following steps:

[0041] Scan the tail wing in real time through the visual sensor and dynamically adjust the polishing path;

[0042] The vibration sensor is used to monitor the vibration state of the grinding head in real time and obtain abnormal vibration data;

[0043] Automatically adjust grinding parameters according to abnormal vibration data and adjust the grinding head for grinding;

[0044] The dust and waste generated during the grinding process are collected and treated centrally.

[0045] The beneficial effects of the present invention are:

[0046] 1. Through the combination of visual sensors, A* algorithms and feature-based model matching algorithms, the grinding path can be accurately planned and dynamically adjusted according to the actual shape of the tail. Whether it is a complex surface, edge or different geometric shapes, it can ensure that the grinding tool covers every part that needs to be processed, avoiding omissions and uneven grinding, making the surface quality of the tail higher and meeting high-quality production standards. The vibration sensor cooperates with the advanced anomaly detection algorithm to detect abnormal vibration of the grinding head in time. The grinding parameters are automatically adjusted through the parameter adjustment algorithm based on fuzzy logic. This effectively reduces the grinding defects caused by abnormal vibration and further improves the grinding quality.

[0047] 2. Dynamic grinding path planning does not require frequent manual intervention. During the grinding process, the system can automatically adjust the path according to the change of the tail shape. Compared with the traditional preset fixed path grinding method, it reduces the time for re-grinding or adjustment due to poor grinding effect, and improves the overall grinding efficiency. The timely vibration abnormality handling mechanism ensures the stability of the grinding process, reduces the number and time of downtime caused by faults such as grinding head damage and tail damage, and enables grinding work to be carried out continuously and efficiently.

[0048] 3. The precise grinding path and stable grinding process avoid over-grinding, reduce the waste of tail wing materials, and effectively reduce the cost of raw materials. By effectively controlling the grinding parameters, the damage to the grinding head and manipulator caused by abnormal vibration is avoided, and the cost of equipment maintenance and replacement is reduced. The perfect waste collection system can efficiently collect the dust and waste generated during the grinding process, which not only reduces the pollution to the environment and protects the health of the operators, but also complies with the requirements of environmental protection regulations and avoids the potential costs and risks caused by environmental problems. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 It is a module diagram of the automobile rear wing robot grinding system.

[0050] Figure 2 It is a flow chart of the method for polishing the rear wing of an automobile by a robot. DETAILED DESCRIPTION

[0051] See also Figure 1 and Figure 2 As shown, the present invention relates to a car tail wing manipulator polishing system, including a path planning module, a state monitoring module, a parameter compensation module and a waste collection module connected in communication, wherein:

[0052] The path planning module is used to scan the tail wing in real time through a visual sensor and dynamically adjust the polishing path; the visual sensor is a 3D structured light sensor or a stereo vision camera;

[0053] The state monitoring module is used to monitor the vibration state of the grinding head in real time through the vibration sensor and obtain vibration abnormality data;

[0054] The parameter compensation module is used to automatically adjust the grinding parameters according to the abnormal vibration data and adjust the grinding head for grinding;

[0055] The waste collection module is used to centrally collect and process the dust and waste generated during the grinding process.

[0056] Furthermore, the real-time scanning of the tail wing by a visual sensor and the dynamic adjustment of the grinding path include the following steps:

[0057] The visual sensor is fixedly connected to the top or side of the manipulator through a bracket and a shock-absorbing device;

[0058] Connect the data acquisition computer and the manipulator control system via a two-way communication link;

[0059] The polishing path is planned and dynamically adjusted through the path planning model;

[0060] During the grinding process, the visual sensor parameters are continuously monitored by the monitoring program in the data acquisition computer and the manipulator control system. When the sensor parameters exceed the set threshold, an alarm is issued and the grinding is suspended.

[0061] Specifically, 3D structured light sensors acquire three-dimensional information by projecting specific light patterns onto the surface of an object and analyzing the reflected light, with high accuracy and resolution; stereo vision cameras, based on the principle of triangulation, reconstruct three-dimensional models by photographing objects from different angles with two or more cameras. The installation of the visual sensor must ensure that its field of view is not blocked and that it is firmly installed to reduce the impact of vibration and displacement on image acquisition. During the polishing process, the visual sensor acquires images of the surface of the tail wing at a set frequency. For 3D structured light sensors, images containing light pattern information are acquired; for stereo vision cameras, images from multiple perspectives are acquired simultaneously. The acquired images are preprocessed, including denoising, contrast enhancement, and other operations, such as removing random noise in the image through Gaussian filtering, enhancing the image contrast through histogram equalization, and improving image quality for subsequent processing.

[0062] The data acquisition computer is a computer device specially used for data acquisition. It has the function of collecting various data, can obtain information from different data sources, and store, process and analyze these data. The manipulator control system refers to a system used to control the operation of the manipulator. This system realizes precise control of various actions of the manipulator, such as grasping, moving, placing and other operations. The data acquisition computer is connected to the manipulator control system to establish a two-way communication link so that the data processing results collected by the visual sensor can be fed back to the control system to realize real-time adjustment of the grinding path.

[0063] During the entire polishing process, by setting up monitoring programs in the data acquisition computer and control system, the working temperature of the visual sensor, the packet loss rate of data transmission, the running time of the algorithm and other parameters are continuously monitored. If the sensor temperature is found to be too high (exceeding the set temperature threshold), an alarm will be issued and the polishing will be suspended to check the heat dissipation of the sensor; if the packet loss rate is higher than the set threshold, check the data transmission line; if the algorithm running time exceeds a certain proportion of the interval time of each scan, it may cause real-time problems, and it is necessary to optimize the algorithm or upgrade the hardware.

[0064] Furthermore, the path planning model is implemented by the A* algorithm and the feature-based model matching algorithm, wherein:

[0065] The A* algorithm is used to search from the starting point to the target point in path planning, generate the optimal path based on the polishing cost, and provide a reference path for subsequent dynamic adjustments;

[0066] The feature-based model matching algorithm is used to quantify the deviation between the tail wing model and the initial model after the real-time feature alignment with the initial model, and dynamically adjust the decision basis of the polishing path.

[0067] Specifically, in the application scenario of automobile tail wing polishing, the main task of the A* algorithm is to search for paths to all target positions that need to be polished (such as the edge of the tail wing, different curved surface areas, etc.) in the three-dimensional space of the tail wing starting from the starting position (starting point) of the polishing tool. Due to the complex shape of the tail wing, with multiple curved surfaces and irregular geometric shapes, the A* algorithm can effectively perform path exploration in such a three-dimensional space. The A* algorithm defines a cost function to measure the "cost" of passing each grid point. This cost function comprehensively considers multiple factors, such as the distance from the grid point to the polishing starting point, whether the grid point is on the tail wing surface, and the curvature change near the grid point. By minimizing the total cost of the path, the initial polishing path generated by the A* algorithm is optimal to a certain extent, enabling the polishing tool to traverse the tail wing surface in a relatively efficient and reasonable manner. The initial polishing path output by the A* algorithm serves as the input of the subsequent feature-based model matching algorithm and is the basis of the entire dynamic polishing path adjustment process. It provides the system with an expected polishing path reference. During the polishing process, when the tail surface shape changes, the polishing path is optimized in real time by comparing and adjusting with this reference path.

[0068] During the polishing process, the real-time three-dimensional model of the tail is obtained through the visual sensor. The first thing that the feature-based model matching algorithm needs to do is to align the real-time model with the model corresponding to the initial polishing path generated by the A* algorithm. By accurately finding the correspondence between the feature points in the two models, the actual position change of the tail surface during the polishing process can be determined. After completing the feature alignment, the algorithm can quantify the deviation between the real-time tail model and the initial model. By calculating these deviations, it is possible to accurately locate which areas of the tail surface have undergone shape changes and the extent of the changes. This is critical for determining the specific location and adjustment range of the polishing path that needs to be adjusted. If the displacement of the feature points in a certain area exceeds the set threshold, or the rotation angle changes greatly, it means that the polishing of the area is not in line with expectations, which may be caused by uneven tail material, wear of the polishing tool, or other factors. Based on this deviation information, the system can make targeted adjustments to the polishing path to ensure the uniformity and integrity of the polishing, so that the surface quality of the polished tail meets the requirements.

[0069] Furthermore, the A* algorithm comprises the following steps:

[0070] Start the visual sensor to perform an all-round scan of the tail wing placed and fixed on the workbench, and perform three-dimensional reconstruction of each frame of the image using a three-dimensional reconstruction algorithm based on stereo vision to obtain a three-dimensional point cloud model of the tail wing;

[0071] The three-dimensional space of the tail wing is divided into grids of uniform size, with the starting position of the grinding tool as the starting point and the set of target positions on the tail wing surface that need to be polished as the target point set. The A* algorithm searches for the path by maintaining an open list and a closed list.

[0072] In each iteration, the node with the smallest cost function value is selected from the open list to determine the optimal path. The algorithm ends when the target point is added to the closed list. The initial polishing path is obtained by backtracing the parent node from the target point to the starting point and serves as the input of the feature-based model matching algorithm.

[0073] Specifically, the visual sensor is started to perform a full-scale scan of the tail wing placed on the workbench and fixed. Let the scanning time be T, and the number of image frames collected during this period be N = fT. For each frame of the image, a three-dimensional reconstruction algorithm based on stereo vision is used for three-dimensional reconstruction. Specifically, let the optical center coordinates of the two cameras be O 1 (x 1 ,y 1 , z 1 ) and O 2 (x 2 ,y 2 , z 2 ), for any point P(x, y) in the image, its coordinates (X, Y, Z) in three-dimensional space can be calculated using the triangulation formula:

[0074]

[0075] Among them, C x , C y is the principal point coordinate of the camera, f is the focal length of the camera, and Z is calculated by matching the same-name points in the two camera images using the parallax principle. By processing the points in all image frames, the three-dimensional point cloud model M of the tail is obtained. 0 .

[0076] The three-dimensional space of the tail is divided into grids of uniform size. Let the grid side length be a. For each grid point G(i, j, k) (where i, j, k are the indices of the grid point in the three-dimensional space), define its cost function C(G): C(G) = ω 1 d(G)+ω 2 s(G)+ω 3 c(G); where d(G) is the Euclidean distance from the grid point G to the grinding starting point, calculated as: Where (x 0 ,y 0 , z 0 ) is the coordinate of the starting point of grinding, (x G ,y G , z G) is the coordinate of the grid point G; s(G) indicates whether the grid point is on the tail surface. If it is on the tail surface, s(G) = 0, otherwise s(G) = 1; c(G) is the curvature change near the grid point, which can be approximated by calculating the second-order derivative of the point cloud model near the point, ω 1 ,ω 2 ,ω 3 It is a weight coefficient determined based on experience.

[0077] The starting position of the grinding tool is taken as the starting point S, and the set of target positions on the tail surface that need to be polished is taken as the target point set {T n The A* algorithm searches for paths by maintaining two lists: an open list and a closed list. In each iteration, the node n with the smallest cost function value is selected from the open list. The calculation formula is F(n) = G(n) + H(n), where G(n) is the actual cost from the starting point to node n (that is, the sum of the cost function values ​​of all grid points passed), and H(n) is the heuristic estimated cost from node n to the target point using the Euclidean distance as the heuristic function, that is:

[0078]

[0079] When the target point is added to the closed list, the algorithm ends and the initial polishing path P is obtained by backtracking from the target point to the parent node of the starting point. 0 , which is represented by a series of grid point sequences P 0 =(G 1 , G 2 , …, G m ), this path serves as the input to the feature-based model matching algorithm.

[0080] Furthermore, the feature-based model matching algorithm comprises the following steps:

[0081] During the polishing process, the visual sensor continuously scans the tail wing at a set frequency, converts the real-time image frame into a real-time tail wing 3D point cloud model through a 3D reconstruction algorithm, and extracts features from the corresponding model of the initial polishing path and the real-time model;

[0082] Corner features are extracted using the 3D Harris corner detection algorithm, and edge features are obtained by calculating the gradient direction change of the point cloud. The extracted feature point sets are the initial model feature point set and the real-time model feature point set.

[0083] The nearest neighbor matching algorithm is used to calculate the distance between each feature point in the real-time model feature point set and the closest feature point in the initial model feature point set. The transformation relationship that minimizes the sum of distances between all matching feature points is found through iterative optimization. The transformation relationship includes: translation vector, rotation matrix and scaling factor.

[0084] According to the transformation relationship obtained by model matching, the deviation between the real-time tail surface and the initial grinding path is analyzed, and the rotation angle deviation and scaling factor deviation are analyzed to determine the grinding path that needs to be adjusted;

[0085] The grinding path is modified according to the path adjustment decision, and the adjusted grinding path is fed back to the robot control system again.

[0086] Specifically, during the polishing process, the visual sensor continuously scans the tail wing at a set frequency, and also uses a 3D reconstruction algorithm to convert the real-time image frame into a real-time tail wing 3D point cloud model M t . From the initial grinding path P 0 The corresponding model M 0 and real-time model M t Extract features. For corner feature extraction, the 3D Harris corner detection algorithm is used. Let the neighborhood point set of point p (x, y, z) in the 3D point cloud model be N (p), and its autocorrelation matrix H (p) be:

[0087]

[0088] Among them, I x ,I y ,I z is the gradient of the point cloud in the x, y, and z directions, and ω is a weighting function (such as a Gaussian function). Calculate the eigenvalue λ of H(p) 1 , 2 , 3 , if λ is satisfied 1 >λ th And 2 >λ th , where λ th is the set threshold, then point p is a corner feature. For edge features, we detect by calculating the gradient direction change of the point cloud. When the gradient direction change exceeds a certain threshold, it is considered an edge point. The extracted feature point sets are F 0 (initial model feature point set) and F t (Real-time model feature point set).

[0089] The nearest neighbor matching algorithm is based on the feature-based model matching algorithm. For the real-time model feature point set F t Each feature point F in t，i , in the initial model feature point set F 0 Find the closest feature point f 0，j , calculate the distance d(f t，i , f 0，j). Through iterative optimization, find the transformation relationship that minimizes the sum of distances between all matching feature point pairs. This transformation relationship includes the translation vector The rotation matrix R and the scaling factor s. The transformation parameters can be solved by minimizing the objective function as follows:

[0090]

[0091] According to the transformation relationship obtained by model matching, the deviation between the real-time tail surface and the initial grinding path is analyzed. The average displacement of the feature points is calculated. Where n is the number of matching feature points. where d th If the displacement threshold is less than the set value, it is considered that there is a significant deviation. At the same time, the rotation angle deviation and scaling factor deviation are analyzed to determine which parts of the grinding path need to be adjusted. For example, if the displacement and rotation angle deviation of the feature points in a local area are large, it means that the grinding in this area is uneven or the shape of the tail has changed, and the grinding path of this area needs to be adjusted.

[0092] According to the path adjustment decision, the grinding path is modified. Let the point on the original grinding path be P 0 =(G 1 , G 2 , ..., G m ), for point G in the area to be adjusted i , calculate its adjusted position G′ based on the deviation i If there is a translational deviation, If there is a rotational deviation, G′ i =RG i ; If there is a scaling bias, G′ i =sG i Adjusted grinding path P t =(G′ 1 , G′ 2 , ..., G′ m ) is fed back to the robot control system to achieve real-time dynamic adjustment of the grinding process to ensure the uniformity and integrity of the grinding.

[0093] Furthermore, the real-time monitoring of the vibration state of the grinding head by the vibration sensor to obtain abnormal vibration data includes the following steps:

[0094] A vibration sensor is connected between the grinding head and the manipulator, and the vibration sensor is connected to a data acquisition card via a wired method. The data acquisition card continuously collects output signals of the vibration sensor during the grinding process and performs signal preprocessing.

[0095] The DC component of the vibration signal is obtained by calculating the signal mean, the discrete degree of the signal is obtained by calculating the standard deviation of the signal, the peak factor of the signal is extracted to determine the presence of impact vibration, the signal spectrum is obtained by performing fast Fourier transform on the signal, and the spectrum amplitude is calculated. The main frequency component of the vibration is determined by the frequency point with a larger amplitude;

[0096] According to the anomaly detection algorithm, thresholds are determined for the signal mean, standard deviation, peak factor and main frequency amplitude. When multiple features exceed the threshold at the same time, it is judged as abnormal vibration. By performing statistical analysis on a large amount of vibration feature data under normal grinding conditions, the mean and standard deviation of the distribution are estimated, and the probability of the newly collected vibration feature data under the assumed distribution is calculated. When it is lower than the probability threshold, it is judged as abnormal vibration and the vibration feature data is recorded.

[0097] Specifically, install the vibration sensor at a position close to the grinding head, such as directly on the outer shell of the grinding head or on the mechanical arm part that connects the grinding head to the manipulator. Ensure that the sensor can accurately sense the vibration of the grinding head and that the installation method will not affect the normal operation of the grinding head and the movement of the manipulator. Calibrate the sensor according to the standard vibration source. Use a vibration generator with known frequency and amplitude to excite vibration at different frequencies and amplitudes within the operating frequency range of the sensor, and record the output signal of the sensor. Based on the recorded data, establish a calibration curve or calibration equation between the sensor output signal and the actual vibration parameters (frequency, amplitude and phase). Assume that there is a linear relationship between the sensor output voltage signal V and the vibration frequency f and amplitude A (the actual situation may be more complicated), and use the least squares method to fit to obtain a shape such as V=k 1 f+k 2 The calibration equation of A+b, where k 1 k 2 and b are the calibration coefficients.

[0098] Connect the vibration sensor to the data acquisition card in the data acquisition device, the sampling frequency f s It should be determined based on the highest possible frequency of the grinding head vibration. According to the Nyquist-Shannon sampling theorem, f is usually selected. s ≥2f max , where f max is the expected maximum vibration frequency. During the grinding process, the data acquisition system continuously collects the output signal x(n) of the vibration sensor according to the set sampling frequency, where n represents the sampling point number x(n) is a discrete time series signal. Since the collected vibration signal may contain noise and interference components, such as electromagnetic interference, other vibrations transmitted by the mechanical structure, etc., the signal needs to be filtered. In order to facilitate subsequent data analysis and comparison, the filtered signal is normalized.

[0099] Calculate the mean of a signal Where N is the length of the signal (number of sampling points), the mean value can reflect the DC component of the vibration signal, and an abnormal DC offset may indicate abnormal vibration of the grinding head. Calculate the standard deviation of the signal The standard deviation is a measure of the dispersion of the signal. A larger standard deviation may indicate unstable vibration. where z peak is the peak value of the signal. The crest factor can be used to detect whether there is impact vibration in the signal, such as when the grinding head encounters a hard spot, a higher crest factor will be generated. Perform a fast Fourier transform on the normalized signal z(n) to obtain the signal spectrum Z(k), where k represents the frequency index. Calculate the amplitude of the spectrum |Z(k)|. Extract the amplitude and frequency of the main frequency component, and determine the main frequency component of the vibration by finding the frequency point with a larger amplitude.

[0100] In the anomaly detection algorithm, the threshold judgment method is used: based on experience or experimental data, thresholds are set for the extracted features (such as mean, standard deviation, peak factor, main frequency amplitude, etc.). When multiple features exceed their respective thresholds at the same time, the credibility of the anomaly judgment is increased. For example, if not only the standard deviation exceeds the threshold, but the peak factor also exceeds the normal range, then it can be more certain that the grinding head is in an abnormal vibration state. Through the statistical model method: assume that the characteristic data of normal vibration obeys a certain probability distribution, such as normal distribution. By performing statistical analysis on a large amount of vibration characteristic data under normal grinding conditions, the parameters of the distribution (such as mean and standard deviation) are estimated. For the newly collected vibration characteristic data, its probability under the assumed distribution is calculated. If the probability is lower than the set probability threshold, it is judged as abnormal vibration.

[0101] When abnormal vibration data is detected, the relevant vibration characteristic data (such as timestamp, characteristic value, spectrum, etc.) is recorded for subsequent analysis of the abnormal cause. This data can be stored in a local database or file. The alarm system is triggered, such as by sound, light or sending a message to the operator to remind the staff of abnormal vibration of the grinding head. The alarm information can include the type of abnormality (such as excessive vibration, abnormal frequency, etc.) and possible location (based on the position information of the grinding head on the tail).

[0102] Furthermore, after the abnormal vibration is determined, abnormal vibration detection is performed through a support vector machine or a multi-layer perceptron to identify abnormal vibrations in complex patterns.

[0103] Specifically, after the anomaly detection algorithm, the supervised learning algorithm, support vector machine (SVM) or neural network (such as multi-layer perceptron) is used to detect abnormal vibration. First, a large amount of feature data under normal and abnormal vibration states is collected as a training set. The training set is labeled (normal is 0, abnormal is 1), and then the model is trained using the training set. In the test phase, the newly extracted vibration feature data is input into the trained model, and the model outputs a probability or classification result, indicating whether the vibration is abnormal. For example, SVM divides normal and abnormal data by finding an optimal hyperplane, and neural networks perform classification by learning the complex relationship between feature data. When the abnormal pattern of the vibration signal is complex and is affected by multiple factors (such as uneven tail material, grinding head wear, different grinding process parameters, etc.), the machine learning algorithm can automatically mine the potential relationship between these complex factors and vibration anomalies by learning a large amount of sample data. Once the training is completed, it has a certain degree of adaptability and generalization ability. It can adapt to the vibration changes under different tail shapes, grinding tools, grinding speeds, etc., without manually adjusting the threshold or statistical model for each case. When there is enough high-quality data, the machine learning algorithm can combine multiple features for comprehensive judgment, avoiding the misjudgment that may occur when judging by a single feature threshold. By learning the boundaries between normal and abnormal vibration states, the vibration state can be classified more accurately, thereby improving the accuracy and reliability of vibration anomaly detection and better ensuring the stability of the grinding process and product quality.

[0104] Furthermore, the automatic adjustment of grinding parameters according to the abnormal vibration data and the adjustment of the grinding head for grinding include the following steps:

[0105] The abnormal vibration data is fuzzified into fuzzy language variables, and the fuzzy set in each fuzzy language variable corresponds to a membership function; according to the fuzzy rule base, the fuzzified vibration characteristics are associated with the grinding parameter adjustment amount; according to the current fuzzified vibration characteristics, the fuzzy reasoning algorithm is used to infer in the rule base to obtain the fuzzified grinding parameter adjustment amount; the fuzzified adjustment amount is converted into an accurate parameter adjustment value;

[0106] The calculated grinding parameter adjustment value is sent to the robot control system to perform the parameter adjustment operation. After the parameter adjustment, the vibration of the grinding head is continuously monitored to determine whether it has returned to normal. If it is still abnormal, repeat the above steps to adjust the grinding parameters again, and then use machine learning to assist in the adjustment.

[0107] Specifically, the detected abnormal vibration data is classified. According to the abnormal conditions of the vibration characteristics such as frequency, amplitude, and peak factor, the abnormal vibration is divided into different types, such as high-frequency abnormal vibration, large-amplitude abnormal vibration, and impact-type abnormal vibration. The possible causes of each type of abnormal vibration are analyzed: high-frequency abnormal vibration may be caused by the high-frequency resonance of the grinding head and the tiny uneven structure on the surface of the tail; large-amplitude abnormal vibration may be caused by the grinding head being subjected to greater resistance, such as a sudden change in the local hardness of the tail material; impact-type abnormal vibration may be caused by the grinding head hitting a hard spot or impurity on the surface of the tail. Determine the corresponding grinding parameter adjustment strategy according to the type of abnormal vibration. For high-frequency abnormal vibration, adjust the grinding head speed or change the smoothness of the grinding path; for large-amplitude abnormal vibration, it is necessary to reduce the grinding pressure or adjust the feed speed of the grinding head; for impact-type abnormal vibration, pause the grinding head to move instantly or change the grinding direction to avoid hard spots. Construct a mapping relationship table between adjustment strategies and abnormal vibration types based on empirical data, experimental results, or theoretical analysis. For example, when high-frequency abnormal vibration is detected and the frequency is within a certain range, the corresponding adjustment strategy is to reduce the grinding head speed by 10%-20% and increase the smoothness coefficient of the grinding path.

[0108] Abnormal vibration data (such as frequency deviation, amplitude deviation, etc.) are fuzzified into fuzzy linguistic variables. For example, the amplitude deviation is divided into fuzzy sets such as "positive large", "positive middle", "positive small", "zero", "negative small", "negative middle", and "negative large", and each fuzzy set corresponds to a membership function. Commonly used membership functions include triangular membership function and trapezoidal membership function. According to the fuzzy rule base, the fuzzified vibration characteristics are associated with the grinding parameter adjustment amount. For example, the rule is "If the amplitude deviation is positive large and the frequency deviation is positive small, then the grinding pressure is reduced a lot and the grinding head speed is reduced a little." These rules are established based on expert experience and experimental data. According to the current fuzzified vibration characteristics, the Mamdan i reasoning method or Sugeno reasoning method in the fuzzy reasoning algorithm is used to reason in the rule base to obtain the fuzzified grinding parameter adjustment amount. The fuzzified adjustment amount is converted into an accurate parameter adjustment value, such as the centroid method or the maximum membership method in the defuzzification method. For example, the specific adjustment values ​​of the grinding pressure and the grinding head speed are calculated by the centroid method.

[0109] Send the calculated grinding parameter adjustment values ​​to the robot control system and the grinding head drive system to perform parameter adjustment operations. For example, change the grinding head speed by controlling the motor driver, and adjust the grinding pressure through the hydraulic or pneumatic system. After the parameter adjustment, continue to monitor the vibration of the grinding head. Obtain new vibration data through the same vibration sensor and signal processing method as the abnormal vibration detection. Determine whether the vibration has returned to the normal range. If it is still abnormal, repeat the above steps to adjust the grinding parameters again; if the vibration is normal, continue the grinding process and keep monitoring the vibration to deal with abnormal situations that may occur again.

[0110] After the polishing parameter adjustment is executed and monitored, machine learning can be used to assist in the adjustment. If a machine learning method is used, a reinforcement learning algorithm is used. The state of the polishing system (including abnormal vibration data, current polishing parameters, etc.) is used as the state space, the polishing parameter adjustment action is used as the action space, and the polishing quality and vibration stability are used as reward signals. By continuously testing different parameter adjustment actions in the actual polishing process, the optimal parameter adjustment strategy is learned based on the reward signal. For example, through the deep Q-network (DQN) algorithm, a neural network is trained to approximate the optimal Q-value function, so as to select the best polishing parameter adjustment action according to the current state. This method can continuously optimize the parameter adjustment strategy during the long-term polishing process to adapt to different tail wings and polishing conditions.

[0111] Furthermore, the dust and waste generated during the grinding process are collected and processed in a centralized manner, including the following steps:

[0112] A dust suction port is arranged under the tail wing, and each dust suction port is connected to a central dust collection device through a pipeline;

[0113] Start the dust collection system before grinding, and control the air volume and speed of the dust collection system by adjusting the fan speed and valve opening. Install a pressure difference sensor in the pipeline system to monitor the pressure difference at both ends of the pipeline to determine whether the pipeline is blocked and perform flow balance control.

[0114] Waste separation and classification are carried out through cyclone separation technology. During the waste treatment process, the operating parameters of the treatment equipment are monitored, and waste information is recorded to establish a waste treatment file.

[0115] Specifically, a dust suction port is set under the tail wing. The dust suction port should be as close as possible to the source of dust and waste, such as directly below the grinding head and below the edge of the tail wing, so as to improve the collection efficiency. The dust suction port is an opening for sucking dust, debris, etc. The dust suction port is connected to the dust suction pipe, and the surrounding dust is sucked into the equipment by generating suction force for collection and treatment. The shape and size of the dust suction port are designed according to the size of the grinding area and the distribution of dust generation. The dust suction port is generally rectangular or circular, and its size must be guaranteed to cover the main dust diffusion area. Each dust suction port is connected to the central dust collection device through a pipe. The pipe should be made of a material with a smooth inner wall, such as a galvanized steel pipe or a PVC pipe, to reduce the deposition of dust in the pipe. The diameter of the pipe is determined according to the air volume and flow rate requirements of the dust collection system. Minimize the bends and changes in pipe diameter to avoid poor airflow. Use fluid mechanics simulation software, such as Fluent, to simulate and optimize the pipeline system to ensure that the airflow can smoothly transport the dust to the dust collection equipment. Choose the appropriate dust collection equipment according to the amount of dust generated by the grinding system, the size and properties of the dust particles. Common dust collection equipment includes cyclone dust collectors, bag dust collectors, electrostatic dust collectors, etc. Dust particles are large and heavy, so cyclone dust collectors are used for preliminary separation; for the fine dust, bag dust collectors or electrostatic dust collectors are used for efficient collection. Dust collection equipment should be installed in a well-ventilated location to facilitate subsequent maintenance and cleaning.

[0116] Before grinding begins, start the dust collection system to ensure that the dust and waste generated during the grinding process can be collected in time. The startup sequence of the dust collection system is to first turn on the fan of the dust collection equipment, and then open the valves of each dust collection port after the fan reaches the rated speed. The air volume and wind speed of the dust collection system are controlled by adjusting the speed of the fan and the opening of the valve. Wind speed sensors and pressure sensors are installed in the pipeline system to monitor the airflow parameters in real time. According to the grinding process and dust generation, a suitable wind speed range is set to ensure that the dust can be effectively sucked into the dust collection port and transported in the pipeline. A pressure difference sensor is installed in the pipeline system to determine whether the pipeline is blocked by monitoring the pressure difference at both ends of the pipeline. When the pressure difference exceeds the set threshold, it indicates that the pipeline may be blocked. At this time, an alarm is issued through the control system, and corresponding measures are taken, such as turning on the pipeline cleaning device to clear the blockage. The pipeline cleaning device includes a vibrating device or a compressed air purge device. In a system with multiple dust collection ports, in order to ensure that each dust collection port can effectively collect dust, flow balance control is performed. By installing a flow control valve on the pipe of each suction port and combining it with a flow sensor for feedback control, the opening of the flow control valve is adjusted according to the deviation between the actual flow of each suction port and the set flow, so that the suction flow of each suction port is balanced.

[0117] In cyclone separation technology, the dust collection equipment includes a cyclone dust collector, which uses the centrifugal force principle to separate particles of different sizes. When the dust-laden airflow enters the cyclone dust collector, it rotates along the tangent direction of the wall. During the rotation, the larger particles are thrown to the wall due to the centrifugal force, and then fall down along the wall and are collected, while the smaller particles enter the next level of dust removal equipment with the airflow. For recyclable waste materials, such as metal waste, they are washed and dried, and then recycled by compression or packaging. For non-recyclable waste materials, such as plastic dust containing chemical coatings, they are treated in accordance with environmental protection requirements. If the waste contains harmful substances, it is harmlessly treated. During the waste treatment process, the operating parameters of the treatment equipment (such as temperature, pressure, treatment time, etc.) are monitored to avoid secondary pollution. Record the treatment date, treatment volume, treatment method and other information of the waste, and establish a waste treatment file for traceability and meeting environmental protection supervision requirements. These records can be stored in a database for easy query and statistical analysis.

[0118] A method for polishing a car rear wing robot comprises the following steps:

[0119] Scan the tail wing in real time through the visual sensor and dynamically adjust the polishing path;

[0120] The vibration sensor is used to monitor the vibration state of the grinding head in real time and obtain abnormal vibration data;

[0121] Automatically adjust grinding parameters according to abnormal vibration data and adjust the grinding head for grinding;

[0122] The dust and waste generated during the grinding process are collected and treated centrally.

[0123] The above implementation modes are merely descriptions of the preferred implementation modes of the present invention, and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary engineering and technical personnel in the field shall fall within the protection scope determined by the claims of the present invention.

Claims

1. A car tail wing manipulator grinding system, characterized in that: It includes a communication-connected path planning module, a state monitoring module, a parameter compensation module and a waste collection module, wherein: The path planning module is used to scan the tail wing in real time through a visual sensor and dynamically adjust the polishing path; the visual sensor is a 3D structured light sensor or a stereo vision camera; The state monitoring module is used to monitor the vibration state of the grinding head in real time through the vibration sensor and obtain vibration abnormality data; The parameter compensation module is used to automatically adjust the grinding parameters according to the abnormal vibration data and adjust the grinding head for grinding; The waste collection module is used to centrally collect and process the dust and waste generated during the grinding process.

2. The automobile rear wing manipulator grinding system according to claim 1 is characterized in that: The method of scanning the tail wing in real time by a visual sensor and dynamically adjusting the grinding path includes the following steps: The visual sensor is fixedly connected to the top or side of the manipulator through a bracket and a shock-absorbing device; Connect the data acquisition computer and the manipulator control system via a two-way communication link; The polishing path is planned and dynamically adjusted through the path planning model; During the grinding process, the visual sensor parameters are continuously monitored by the monitoring program in the data acquisition computer and the manipulator control system. When the sensor parameters exceed the set threshold, an alarm is issued and the grinding is suspended.

3. The automobile rear wing manipulator grinding system according to claim 2 is characterized in that: The path planning model is implemented by the A* algorithm and the feature-based model matching algorithm, where: The A* algorithm is used to search from the starting point to the target point in path planning, generate the optimal path based on the polishing cost, and provide a reference path for subsequent dynamic adjustments; The feature-based model matching algorithm is used to quantify the deviation between the tail wing model and the initial model after the real-time feature alignment with the initial model, and dynamically adjust the decision basis of the polishing path.

4. The automobile rear wing manipulator grinding system according to claim 3 is characterized in that: The A* algorithm comprises the following steps: Start the visual sensor to perform an all-round scan of the tail wing placed and fixed on the workbench, and perform three-dimensional reconstruction of each frame of the image using a three-dimensional reconstruction algorithm based on stereo vision to obtain a three-dimensional point cloud model of the tail wing; The three-dimensional space of the tail wing is divided into grids of uniform size, with the starting position of the grinding tool as the starting point and the set of target positions on the tail wing surface that need to be polished as the target point set. The A* algorithm searches for the path by maintaining an open list and a closed list. In each iteration, the node with the smallest cost function value is selected from the open list to determine the optimal path. The algorithm ends when the target point is added to the closed list. The initial polishing path is obtained by backtracing the parent node from the target point to the starting point and serves as the input of the feature-based model matching algorithm.

5. The automobile rear wing manipulator grinding system according to claim 4, characterized in that: The feature-based model matching algorithm comprises the following steps: During the polishing process, the visual sensor continuously scans the tail wing at a set frequency, converts the real-time image frame into a real-time tail wing 3D point cloud model through a 3D reconstruction algorithm, and extracts features from the corresponding model of the initial polishing path and the real-time model; Corner features are extracted using the 3D Harris corner detection algorithm, and edge features are obtained by calculating the gradient direction change of the point cloud. The extracted feature point sets are the initial model feature point set and the real-time model feature point set. The nearest neighbor matching algorithm is used to calculate the distance between each feature point in the real-time model feature point set and the closest feature point in the initial model feature point set. The transformation relationship that minimizes the sum of distances between all matching feature points is found through iterative optimization. The transformation relationship includes: translation vector, rotation matrix and scaling factor. According to the transformation relationship obtained by model matching, the deviation between the real-time tail surface and the initial grinding path is analyzed, and the rotation angle deviation and scaling factor deviation are analyzed to determine the grinding path that needs to be adjusted; The grinding path is modified according to the path adjustment decision, and the adjusted grinding path is fed back to the robot control system again.

6. The automobile rear wing manipulator grinding system according to claim 1, characterized in that: The method of monitoring the vibration state of the grinding head in real time by using a vibration sensor to obtain abnormal vibration data includes the following steps: A vibration sensor is connected between the grinding head and the manipulator, and the vibration sensor is connected to a data acquisition card via a wired method. The data acquisition card continuously collects output signals of the vibration sensor during the grinding process and performs signal preprocessing. The DC component of the vibration signal is obtained by calculating the signal mean, the discrete degree of the signal is obtained by calculating the standard deviation of the signal, the peak factor of the signal is extracted to determine the presence of impact vibration, the signal spectrum is obtained by performing fast Fourier transform on the signal, and the spectrum amplitude is calculated. The main frequency component of the vibration is determined by the frequency point with a larger amplitude; According to the anomaly detection algorithm, thresholds are determined for the signal mean, standard deviation, peak factor and main frequency amplitude. When multiple features exceed the threshold at the same time, it is judged as abnormal vibration. By performing statistical analysis on a large amount of vibration feature data under normal grinding conditions, the mean and standard deviation of the distribution are estimated, and the probability of the newly collected vibration feature data under the assumed distribution is calculated. When it is lower than the probability threshold, it is judged as abnormal vibration and the vibration feature data is recorded.

7. The automobile rear wing manipulator grinding system according to claim 6, characterized in that: After the abnormal vibration is determined, abnormal vibration detection is performed through a support vector machine or a multi-layer perceptron to identify abnormal vibration in complex modes.

8. The automobile rear wing manipulator grinding system according to claim 1, characterized in that: The method of automatically adjusting the grinding parameters according to the abnormal vibration data and adjusting the grinding head for grinding comprises the following steps: The abnormal vibration data is fuzzified into fuzzy language variables, and the fuzzy set in each fuzzy language variable corresponds to a membership function; according to the fuzzy rule base, the fuzzified vibration characteristics are associated with the grinding parameter adjustment amount; according to the current fuzzified vibration characteristics, the fuzzy reasoning algorithm is used to infer in the rule base to obtain the fuzzified grinding parameter adjustment amount; the fuzzified adjustment amount is converted into an accurate parameter adjustment value; The calculated grinding parameter adjustment value is sent to the robot control system to perform the parameter adjustment operation. After the parameter adjustment, the vibration of the grinding head is continuously monitored to determine whether it has returned to normal. If it is still abnormal, repeat the above steps to adjust the grinding parameters again, and then use machine learning to assist in the adjustment.

9. The automobile rear wing manipulator grinding system according to claim 1, characterized in that: The dust and waste generated during the grinding process are collected and processed in a centralized manner, including the following steps: A dust suction port is arranged under the tail wing, and each dust suction port is connected to a central dust collection device through a pipeline; Start the dust collection system before grinding, and control the air volume and speed of the dust collection system by adjusting the fan speed and valve opening. Install a pressure difference sensor in the pipeline system to monitor the pressure difference at both ends of the pipeline to determine whether the pipeline is blocked and perform flow balance control. Waste separation and classification are carried out through cyclone separation technology. During the waste treatment process, the operating parameters of the treatment equipment are monitored, and waste information is recorded to establish a waste treatment file.

10. A method for polishing a car's rear wing by a manipulator, applied to the car's rear wing polishing system as claimed in any one of claims 1 to 9, characterized in that: The following steps are involved: Scan the tail wing in real time through the visual sensor and dynamically adjust the polishing path; The vibration sensor is used to monitor the vibration state of the grinding head in real time and obtain abnormal vibration data; Automatically adjust grinding parameters according to abnormal vibration data and adjust the grinding head for grinding; The dust and waste generated during the grinding process are collected and treated centrally.

Images