Grinding machine tool machining precision error intelligent correction method and system

Abstract

The invention discloses a grinding machine tool machining precision error intelligent correction method and system, and relates to the technical field of grinding control. Target sound data and target workpiece data are obtained, the target sound data are input into the target model to obtain target grinding head data and the target workpiece data, and target grinding head point cloud data and target grinding head abrasion data are determined according to the target grinding head data; inputting the target workpiece data into the target model to obtain workpiece wear data, and determining the wear cycle of the grinding head according to the target grinding head wear data and the workpiece wear data; feed amount compensation is determined according to the wear cycle and the grinding head point cloud data, and the feed amount compensation is fed back to the machine tool so that the machine tool can correct the machining parameters; sound and workpiece data are obtained in a non-contact mode, rapid matching is conducted through a target model, precise evaluation of the abrasion state of the grinding head and the workpiece is achieved, feed amount compensation is determined in combination with the abrasion period and point cloud data, the feed amount compensation is fed back to a machine tool to correct machining parameters, and the machining precision and efficiency are improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of grinding control, and particularly relates to a grinding machine tool machining precision error intelligent correction method and system. BACKGROUND

[0002] As the core equipment of precision manufacturing, the development of grinding machine tools is closely related to the innovation of industrial technology. Early grinding processes rely on manual operation, and the efficiency and precision are limited. With the breakthroughs in numerical control technology, high-precision sensors and intelligent control systems, modern grinding machine tools gradually realize automation and digitization, and can stably complete micron-level or even nanometer-level processing. In recent years, the processing demand for complex curved surfaces and super-hard materials has increased rapidly in the fields of aerospace, optical devices and new energy, which has promoted the evolution of grinding technology towards high speed, combination and intelligence.

[0003] Patent No. CN118720864A discloses a high-efficiency and high-precision grinding control method and control device. The output power and acoustic emission signals of the grinding wheel motor are obtained during the grinding process. The grinding condition of the workpiece is determined based on the output power. The grinding speed of the grinding wheel on the workpiece is controlled in real time based on the grinding condition. The frequency spectrum change graph is generated based on the acoustic emission signal. It is determined whether the workpiece has a burn condition based on the frequency spectrum change graph. When the workpiece has a burn condition, a first burn frequency spectrum change graph is collected from the frequency spectrum change graph. Different grinding parameters are adjusted separately based on the burn frequency spectrum. The second burn frequency spectrum change graph corresponding to each grinding parameter after adjustment is obtained. The sensitivity of different grinding parameters to the burn condition of the workpiece is distinguished based on the second burn frequency spectrum change graph. The grinding efficiency is improved by changing the grinding parameters with low sensitivity. Online burn detection is realized, and the severity of the burn can be reflected.

[0004] During the machining process of the grinding machine tool, the traditional contact type data acquisition method is easily disturbed by factors such as machining vibration and cutting force, resulting in inaccurate data, which further affects the accuracy of subsequent wear state evaluation and machining parameter adjustment. There is a lack of fast and effective data matching and analysis model, it is difficult to determine the grinding head wear period in real time and accurately calculate the feed amount compensation, which affects the machining efficiency and the service life of the grinding head. SUMMARY

[0005] The purpose of the present application is to solve the problem that the traditional contact type data acquisition method is easily disturbed by factors such as machining vibration and cutting force, resulting in inaccurate data, which further affects the accuracy of subsequent wear state evaluation and machining parameter adjustment. There is a lack of fast and effective data matching and analysis model, it is difficult to determine the grinding head wear period in real time and accurately calculate the feed amount compensation, which affects the machining efficiency and the service life of the grinding head. A grinding machine tool machining precision error intelligent correction method and system are proposed.

[0006] In the first aspect of the embodiment of the present application, an intelligent correction method for grinding machine processing precision error is first proposed, and the method comprises: Target sound data and target workpiece data are acquired, the target sound data is input into a target model to obtain target grinding head data and target workpiece data, and target grinding head point cloud data and target grinding head wear data are determined according to the target grinding head data; The target workpiece data is input into the target model to obtain workpiece wear data, and the wear period of the grinding head is determined according to the target grinding head wear data and the workpiece wear data; the wear period includes an initial period, an intermediate period and a final period; The feed compensation is determined according to the wear period and the grinding head point cloud data, and the feed compensation is fed back to the machine tool to correct the machining parameters of the machine tool; the machining parameters include feed speed, cutting depth, grinding head rotating speed and workpiece rotating speed.

[0007] Optionally, the target sound data is input into a target model to obtain target grinding head data and target workpiece data, and the training process of the target model comprises: Image data and sound data of the workpiece during machining are acquired in real time, grinding head data and workpiece data are obtained by improving a YOLOv5 model to extract features from the image data; the image data and the sound data correspond to each other in a time period; the grinding head data includes grinding head point cloud data and grinding head wear data; A wear model sequence set is obtained by three-dimensional modeling of the grinding head according to the grinding head point cloud data, a wear label sequence set is determined according to the grinding head wear data, and the wear model sequence set, the wear label sequence set, the workpiece data and the sound data are used as training data; the wear model sequence set and the wear label sequence set correspond to each other in a time sequence; the training data further includes grinding head rotating speed, feed speed and machining material; The training data is input into a preset model for training, the output of the preset model is used as a model update parameter, the model parameters of the preset model are updated according to the model update parameter to obtain a target model; the output of the target model is grinding head data and workpiece wear data.

[0008] Optionally, the wear model sequence set is obtained by three-dimensional modeling of the grinding head according to the grinding head point cloud data, and comprises: A rectangular coordinate system is constructed according to the grinding head point cloud data, a feature point set in the point cloud data is determined, coordinates of feature points A, feature points B and feature points C are determined according to the feature point set, and coordinates of feature points E are calculated according to the feature points A, the feature points B and the feature points C; The calculation formula of the feature points E is: , wherein (x A ,y A ) is the coordinate of point A, (x B ,y B ) is the coordinate of point B, (x E ,y E ) is the coordinate of point E, and (x C ,y C ) is the coordinate of point C; obtaining the radius of the grinding head, calculating the coordinate of the center O point according to the radius, the coordinate of the feature point A, and the coordinate of the feature point E, calculating the slope k according to the feature point A and the feature point B, and calculating the grinding head center axis equation according to the slope k and the coordinate of the center O point; rotating the coordinate system of the grinding head center axis equation according to a preset angle to obtain a target coordinate system, calculating the distance d i of each feature point in the feature point set from the center line, i rotating one circle to obtain a three-dimensional wear model; i is 1, 2, …, n; the three-dimensional wear model is any one of the wear model sequence set; coordinate system conversion formula: , wherein (x S ,y S ) is the current coordinate system, and (x' S ,y' S ) is the target coordinate system.

[0009] Optionally, the improved YOLOv5 model is used to extract features of the image data to obtain the grinding head data and the workpiece data, the C3 module in the YOLOv5 model is replaced by an improved C3 module to obtain an improved YOLOv5 model, which comprises: working principle of the improved YOLOv5 model: the output of the third layer improved C3 module is convoluted by a 1×1 CBS module to obtain a first feature map, the output of the fifth layer improved C3 module is input into the 1×1 CBS module to be convoluted to obtain a second feature map, the output of the seventh layer improved C3 module is input into the 1×1 CBS module to be convoluted to obtain a third feature map, and the output of the tenth layer SPPF module is input into the 1×1 CBS module to be convoluted to obtain a fourth feature map; the fourth feature map is up-sampled to obtain a target fourth feature map, the second feature map is input into the 1×1 CBS module to be convoluted to obtain a target second feature map, and the target fourth feature map, the target second feature map, and the third feature map are spliced to obtain a first fusion feature map; inputting the first fusion feature map into a 1*1 CBS module for convolution to obtain a target fusion feature map, inputting the first feature map into a 1*1 CBS module for convolution to obtain a target first feature map, splicing the target first feature map, the second feature map and the target fusion feature map to obtain a second fusion feature map; inputting the second fusion feature map into a 1*1 CBS module for convolution to obtain a third fusion feature map, taking the third fusion feature map as the output of a first detection head, splicing the third fusion feature map and the target fusion feature map to obtain a fourth feature fusion map, taking the fourth feature fusion map as the output of a second detection head, inputting the fourth feature fusion map into a 1*1 CBS module for convolution to obtain a fifth feature fusion map, splicing the fifth feature fusion map and the fourth feature map to obtain a sixth feature fusion map, and taking the sixth feature fusion map as the output of a third detection head.

[0010] Optionally, the improved C3 module comprises: replacing BottleNeck*N in the C3 module with DenseBlock*N, and adding an NAM module at the end of the C module to obtain the improved C3 module; The working principle of the improved C3 module is as follows: inputting the first processing graph into the improved C3 module, inputting the first processing graph into a 1*1 CBS module for convolution to obtain a second processing graph, inputting the second processing graph into DenseBlock*N for dense connection to obtain a third processing graph, inputting the third processing graph into an NAM module for weighting to obtain a fourth processing graph, and taking the fourth processing graph as the output of the improved C3 module.

[0011] In a second aspect of the embodiment of the present application, an intelligent correction system for grinding machine tool machining precision error is provided, comprising a data acquisition module, a cycle determination module and a data feedback module: The data acquisition module is configured to acquire target sound data and target workpiece data, input the target sound data into a target model to obtain target grinding head data and target workpiece data, and determine target grinding head point cloud data and target grinding head wear data according to the target grinding head data. The cycle determination module is configured to input the target workpiece data into a target model to obtain workpiece wear data, and determine a wear cycle of the grinding head according to the target grinding head wear data and the workpiece wear data; the wear cycle comprises an initial stage, an intermediate stage and a final stage. The data feedback module is configured to determine a feed amount compensation according to the wear period and the grinding head point cloud data, and feed the feed amount compensation to a machine tool to make the machine tool correct machining parameters; the machining parameters include a feed speed, a cutting depth, a grinding head rotating speed, and a workpiece rotating speed.

[0012] Optionally, the data acquisition module comprises a training data acquisition module, a training data processing module, and a model updating module. The training data acquisition module is configured to acquire image data and sound data of a workpiece in real time during machining, and extract features of the image data by improving a YOLOv5 model to obtain grinding head data and workpiece data; the image data and the sound data correspond to each other in a time period; the grinding head data comprises grinding head point cloud data and grinding head wear data; The training data processing module is configured to perform three-dimensional modeling on the grinding head according to the grinding head point cloud data to obtain a wear model sequence set, determine a wear label sequence set according to the grinding head wear data, and use the wear model sequence set, the wear label sequence set, the workpiece data, and the sound data as training data; the wear model sequence set and the wear label sequence set correspond to each other in a time sequence; the training data further comprises a grinding head rotating speed, a feed speed, and a machining material; The model updating module is configured to input the training data into a preset model for training, use an output of the preset model as a model updating parameter, update model parameters of the preset model according to the model updating parameter to obtain a target model; the output of the target model is grinding head data and workpiece wear data.

[0013] Optionally, the training data processing module comprises a feature point determination module, an equation calculation module, and a three-dimensional model generation module. The feature point determination module is configured to construct a rectangular coordinate system according to the grinding head point cloud data, determine a feature point set in the point cloud data, determine coordinates of a feature point A, a feature point B, and a feature point C according to the feature point set, and calculate coordinates of a feature point E according to the feature point A, the feature point B, and the feature point C; The calculation formula of the feature point E is as follows: , wherein, (x A ,y A ) is the coordinates of the point A, (x B ,y B ) is the coordinates of the point B, (x E ,y E ) is the coordinates of the point E, and (x C ,y C ) is the coordinates of the point C. The equation calculation module is configured to obtain a radius of the grinding head, calculate a center O point coordinate according to the radius, a feature point A coordinate and a feature point E coordinate, calculate a slope k according to the feature point A and the feature point B, and calculate a grinding head center axis equation according to the slope k and the center O point coordinate; The three-dimensional model generation module is configured to rotate a coordinate system of the grinding head center axis equation according to a preset angle to obtain a target coordinate system, and calculate distances d of each feature point in the feature point set from a center line i The distances d of the feature point set are calculated i The distances d of the feature point set are calculated The three-dimensional model generation module is configured to rotate a coordinate system of the grinding head center axis equation according to a preset angle to obtain a target coordinate system, and calculate distances d of each feature point in the feature point set from a center line , wherein (x S ,y S ) is a current coordinate system, and (x' S ,y' S ) is a target coordinate system.

[0014] Optionally, the working principle of the improved YOLOv5 model includes: The output of the third layer improved C3 module is convoluted through a 1*1 CBS module to obtain a first feature map, the output of the fifth layer improved C3 module is input into the 1*1 CBS module to be convoluted to obtain a second feature map, the output of the seventh layer improved C3 module is input into the 1*1 CBS module to be convoluted to obtain a third feature map, and the output of the tenth layer SPPF module is input into the 1*1 CBS module to be convoluted to obtain a fourth feature map. The fourth feature map is up-sampled to obtain a target fourth feature map, the second feature map is input into the 1*1 CBS module to be convoluted to obtain a target second feature map, and the target fourth feature map, the target second feature map and the third feature map are spliced to obtain a first fusion feature map. The first fusion feature map is input into the 1*1 CBS module to be convoluted to obtain a target fusion feature map, the first feature map is input into the 1*1 CBS module to be convoluted to obtain a target first feature map, and the target first feature map, the second feature map and the target fusion feature map are spliced to obtain a second fusion feature map. The second fusion feature map is input into a 1*1 CBS module for convolution to obtain a third fusion feature map, the third fusion feature map is taken as the output of a first detection head, the third fusion feature map and the target fusion feature map are spliced to obtain a fourth feature fusion map, the fourth feature fusion map is taken as the output of a second detection head, the fourth feature fusion map is input into a 1*1 CBS module for convolution to obtain a fifth feature fusion map, the fifth feature fusion map is spliced with the fourth feature map to obtain a sixth feature fusion map, and the sixth feature fusion map is taken as the output of a third detection head.

[0015] Optionally, the improved C3 module comprises: replacing BottleNeck*N in the C3 module with DenseBlock*N, and adding an NAM module at the end of the C module to obtain the improved C3 module. The working principle of the improved C3 module is as follows: The first processing map is input into the improved C3 module, the first processing map is input into a 1*1 CBS module for convolution to obtain a second processing map, the second processing map is input into DenseBlock*N for dense connection to obtain a third processing map, the third processing map is input into an NAM module for weighting to obtain a fourth processing map, and the fourth processing map is taken as the output of the improved C3 module.

[0016] The beneficial effects of the present application are as follows: The present application provides an intelligent correction method for grinding machine tool machining precision error, by obtaining target sound data and target workpiece data, inputting the target sound data into the target model to obtain target grinding head data and target workpiece data, determining the target grinding head point cloud data and target grinding head wear data according to the target grinding head data; inputting the target workpiece data into the target model to obtain the workpiece wear data, determining the wear period of the grinding head according to the target grinding head wear data and the workpiece wear data; determining the feed compensation according to the wear period and the grinding head point cloud data, and feeding back the feed compensation to the machine tool to correct the machining parameters; non-contact acquisition of sound and workpiece data, and rapid matching by using the target model to realize accurate evaluation of the grinding head and workpiece wear state, and determination of the feed compensation according to the wear period and the point cloud data, and feedback to the machine tool to correct the machining parameters, thereby improving the machining precision and efficiency. BRIEF DESCRIPTION OF DRAWINGS

[0017] The present application will be further described below in conjunction with the drawings.

[0018] Figure 1 A flowchart of an intelligent correction method for grinding machine tool machining precision error provided by the embodiment of the present application is shown in the figure. Figure 2A three-dimensional structural schematic diagram of an intelligent correction method for grinding machine tool machining precision error is provided for an embodiment of the present application. Figure 3 A model structure schematic diagram of an intelligent correction method for grinding machine tool machining precision error is provided for an embodiment of the present application. Figure 4 A framework diagram of an intelligent correction system for grinding machine tool machining precision error is provided for an embodiment of the present application. DETAILED DESCRIPTION

[0019] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. The term “and / or” in this document is only used to describe the association relationship of the associated objects, and can represent three relationships, for example, A and B can represent three cases of A alone, A and B together, and B alone. In addition, the description of “first”, “second” and the like in the present application is only for the purpose of description, and cannot be understood as indicating or implying the relative importance of the indicated technical features or implicitly indicating the number of the indicated technical features. Therefore, the features limited by “first”, “second” can explicitly or implicitly include at least one of the features. In addition, the technical solutions of various embodiments can be combined with each other, but it must be based on the fact that a person skilled in the art can realize it. When the combination of technical solutions appears contradictory or unachievable, it should be considered that the combination of technical solutions does not exist, and is not within the protection scope required by the present application.

[0020] Based on the embodiments in the present application, all other embodiments obtained by a person skilled in the art without creative labor are within the scope of protection of the present application.

[0021] An intelligent correction method for grinding machine tool machining precision error is provided in the embodiments of the present application. Referring to Figure 1 , Figure 1 A flowchart of an intelligent correction method for grinding machine tool machining precision error is provided for an embodiment of the present application. The method comprises the following steps: S101, target sound data and target workpiece data are acquired, the target sound data is input into a target model to obtain target grinding head data and target workpiece data, and target grinding head point cloud data and target grinding head wear data are determined according to the target grinding head data; S102, the target workpiece data is input into the target model to obtain workpiece wear data, and the wear period of the grinding head is determined according to the target grinding head wear data and the workpiece wear data; S103, the feed compensation is determined according to the wear period and the grinding head point cloud data, and the feed compensation is fed back to the machine tool, so that the machine tool corrects the machining parameters. The wear cycle includes: initial, intermediate and final; the machining parameters include feed speed, cutting depth, grinding head rotation speed and workpiece rotation speed.

[0022] Based on the grinding machine processing precision error intelligent correction method provided by the embodiment of the application, the sound and workpiece data are obtained by non-contact, and the target model is quickly matched, the wear state of the grinding head and the workpiece is accurately evaluated, the feed compensation is determined by combining the wear cycle and the point cloud data, and the machining parameters are corrected by the feedback to the machine tool, the machining precision and efficiency are improved, and the service life of the grinding head is prolonged.

[0023] In an implementation manner, target sound data (acquired by a microphone array) and target workpiece data (workpiece surface images and grinding head surface images acquired by laser scanning) are obtained by a non-contact manner, and the data is input into a target model to quickly obtain target grinding head data and target workpiece data. The non-contact data acquisition manner not only avoids direct interference to the machining process, ensures the continuity and stability of the machining, but also can acquire data in real time and efficiently. The target model can quickly match and process the data, and timely output target grinding head point cloud data and target grinding head wear data, to provide timely and accurate information for subsequent wear state evaluation, and greatly shorten the time from data acquisition to wear evaluation.

[0024] In an implementation manner, the workpiece wear data is further obtained by using the target workpiece data, and the wear cycle (initial, intermediate and final) of the grinding head is determined by combining the target grinding head wear data. By comprehensively analyzing the wear data of the grinding head and the workpiece, the fine evaluation of the wear state of the grinding head is realized. The fine evaluation can more accurately reflect the actual wear condition of the grinding head in different machining stages, and provide more targeted basis for subsequent feed compensation. Compared with the traditional single wear evaluation method, the evaluation method based on the wear cycle can better adapt to the needs of different machining stages, and improve the accuracy and reliability of the wear evaluation.

[0025] In an implementation, the calculation of the feed rate compensation is mainly based on the wear cycle of the grinding head and the grinding head point cloud data; determining the wear cycle: according to the target grinding head wear data and the workpiece wear data, it is judged whether the grinding head is currently in the early stage, the middle stage or the late stage of the wear cycle. The wear characteristics of each stage are different, and the corresponding compensation strategy also has some differences; analyzing the grinding head point cloud data: a three-dimensional wear model is constructed through the grinding head point cloud data, and the depth and range of the grinding head surface wear are calculated. The point cloud data provides the macro morphology of the grinding head wear, which is used to quantify the wear degree; calculating the compensation amount: according to the wear cycle and the wear degree, the feed rate compensation value is determined in combination with the processing process requirements. For example, in the early stage of wear, the compensation amount may be small to maintain the processing accuracy; while in the late stage of wear, the compensation amount may be large to avoid excessive wear leading to processing errors (and burning of the workpiece); feedback to the machine tool: the calculated feed rate compensation value is fed back to the machine tool control system, and the machine tool adjusts the processing parameters such as feed speed, cutting depth, grinding head speed and workpiece speed according to the compensation value, thereby realizing dynamic correction of the processing process and ensuring the stability and consistency of the processing quality.

[0026] In an implementation, by converting the workpiece wear data (workpiece surface image) into a heat map, the depth of the workpiece surface ravines, i.e. the wear amount of the workpiece by the grinding head, can be intuitively reflected. Combined with the target grinding head wear data (grinding head wear condition), the wear cycle of the grinding head can be quantitatively determined. Specifically, the color depth in the heat map is proportional to the ravine depth, and by setting a threshold, the wear depth is divided into early stage (light color area, shallow wear), middle stage (medium color area, moderate wear) and late stage (dark color area, deep wear). At the same time, combined with the key indicators in the grinding head wear data (such as the degree of particle wear on the grinding head surface, the size of the wear area in the grinding head point cloud data, etc.), the judgment of the wear cycle is further refined. For example, when the particle wear on the grinding head surface reaches a certain proportion and the proportion of the dark color area in the heat map increases, it can be determined that the wear is in the late stage. This method of combining heat map and grinding head wear data not only can intuitively and quantitatively determine the wear cycle, but also can provide more accurate basis for subsequent feed rate compensation.

[0027] In one implementation, the feed rate compensation is determined according to the wear cycle and the grinding head point cloud data, and the compensation is fed back to the machine tool to realize dynamic correction of the machining parameters. This process uses the refined evaluation results of the wear cycle and the macro wear pattern of the grinding head point cloud data to calculate the appropriate feed rate compensation value. By feeding back the compensation value to the machine tool, the machine tool can adjust the machining parameters such as feed speed and cutting depth in real time, thereby ensuring the stability of the machining process and the consistency of the machining quality. This data-driven dynamic compensation mechanism not only improves the machining accuracy, but also prolongs the service life of the grinding head and improves the intelligent level and efficiency of the entire machining process; each correction will feed back to the machine tool, and the feed rate adjusted each time is equivalent to fine tuning, which improves the machining accuracy through multiple fine tuning.

[0028] In one embodiment, target sound data is input into a target model to obtain target grinding head data and target workpiece data, and the training process of the target model includes: Real-time acquisition of image data and sound data of the workpiece during machining, feature extraction of the image data by improving the YOLOv5 model to obtain grinding head data and workpiece data; the image data and the sound data correspond one-to-one in time period; the grinding head data includes grinding head point cloud data and grinding head wear data; According to the grinding head point cloud data, a three-dimensional model of the grinding head is established to obtain a wear model sequence set, and according to the grinding head wear data, a wear label sequence set is determined, and the wear model sequence set, the wear label sequence set, the workpiece data and the sound data are used as training data; the wear model sequence set and the wear label sequence set correspond one-to-one in time sequence; the training data further includes grinding head speed, feed speed and machining material; The training data is input into a preset model for training, the output of the preset model is used as a model update parameter, the model parameters of the preset model are updated according to the model update parameter to obtain a target model; the output of the target model is the grinding head data and the workpiece wear data.

[0029] In one implementation, the preset model is, for example: 1. GA-BP model (genetic algorithm optimized BP neural network model), 2. Model based on deep learning acoustic scene classification, 3. Model based on optimized SVM (support vector machine), and 4. LSTM long short-term memory network model, etc.; the image data includes images of the grinding head and the workpiece, wherein the grinding head image is an image during machining (obtained by laser scanning to prevent image motion blur), and the workpiece image is an image after the grinding head has just processed.

[0030] In an implementation, during the grinding machine processing, image data and sound data during workpiece processing are acquired in real time, and an improved YOLOv5 model is used to extract features of the image data to obtain grinding head data and workpiece data. The non-contact data acquisition mode avoids the interference on the processing process caused by the traditional contact detection, and the introduction of sound data enables the processing state to be reflected according to the sound in subsequent analysis. The image data and the sound data correspond to each other in the time period, ensuring the synchronization and correlation of the data, providing accurate input for subsequent model training. Compared with the traditional complex data acquisition and analysis method, the combination of image and sound data reduces the calculation amount during real-time acquisition and data analysis, improves the data processing efficiency, and can more quickly provide a basis for subsequent wear state judgment and compensation.

[0031] In an implementation, grinding head point cloud data is used to perform three-dimensional modeling on the grinding head to obtain a wear model sequence set, and grinding head wear data is used to determine a wear label sequence set. The grinding head point cloud data can construct the three-dimensional shape of the grinding head from a macroscopic perspective, reflecting the overall wear of the grinding head. The grinding head wear data focuses on the wear of the particles on the surface of the grinding head, which is a microscopic wear analysis. The grinding head wear data can also be comprehensively judged in combination with the time of use of the grinding head. The wear data of the macroscopic and microscopic perspectives are combined and formed into a sequence set, so that the description of the wear state of the grinding head is more comprehensive and detailed. This multi-dimensional modeling and analysis method can more accurately grasp the wear degree of the grinding head at different processing stages, provides rich and accurate training data for subsequent model training, and helps to improve the accuracy of the model in judging the wear state of the grinding head.

[0032] In an implementation, the wear model sequence set, the wear label sequence set, the workpiece data, and the sound data are input into a preset model as training data for training, and the model parameters of the preset model are updated according to the output of the model as an update parameter to finally obtain a target model. This process trains the model by using a large amount of data related to the wear state of the grinding head (training by using historical data), so that the model can learn the complex mapping relationship between different data features and the wear state of the grinding head. The continuous updating of the model parameters further optimizes the performance of the model and improves the accuracy and reliability of the model in judging the wear state of the grinding head.

[0033] In one implementation, after the target model is trained, only the target sound data needs to be obtained and input into the target model to obtain the target wear data, and then the compensation is corrected according to the target wear data to determine the feed compensation of the grinding machine. This mode simplifies the process of wear condition judgment and compensation, avoids the collection and processing of other complex data, and in the actual processing process, the sound data is relatively easy to obtain and is not disturbed by too many factors such as processing environment. Through the analysis of the sound data by the target model, the grinding head wear condition under the same working configuration (such as grinding head speed, feed speed, and processed material) can be quickly and accurately found. Based on the rapid feedback mechanism of the sound data, the feed of the grinding machine can be adjusted in time to ensure the stability of the processing process and the processing quality, and the intelligent level and processing efficiency of the grinding machine are improved.

[0034] In one embodiment, a grinding head is three-dimensionally modeled according to grinding head point cloud data to obtain a wear model sequence set, comprising: A rectangular coordinate system is constructed according to the grinding head point cloud data, a feature point set in the point cloud data is determined, coordinates of feature points A, B and C are determined according to the feature point set, and coordinates of feature point E are calculated according to the feature points A, B and C. The calculation formula of the feature point E is: , Wherein, (x A ,y A ) is the coordinate of point A, (x B ,y B ) is the coordinate of point B, (x E ,y E ) is the coordinate of point E, and (x C ,y C ) is the coordinate of point C.

[0035] The radius of the grinding head is obtained, the coordinates of the center O point are calculated according to the radius, the coordinates of the feature point A and the coordinates of the feature point E, the slope k is calculated according to the feature point A and the feature point B, and the grinding head center axis equation is calculated according to the slope k and the coordinates of the center O point. The coordinate system of the grinding head center axis equation is rotated according to a preset angle to obtain a target coordinate system, the distance d i of each feature point in the feature point set from the center line is calculated, the distance d i of the feature point set is rotated for one revolution to obtain a three-dimensional wear model; the value of i is 1, 2, …, n; and the three-dimensional wear model is any one of the wear model sequence set. The coordinate system conversion formula is: , Wherein, (x S ,y S) is the current coordinate system, (x' S ,y' S ) is the target coordinate system.

[0036] In an implementation, referring to Figure 2 , Figure 2 is a three-dimensional structural schematic diagram of an intelligent correction method for grinding machine tool processing precision error provided by an embodiment of the application, and the dashed lines of AE and EC are wear areas; by constructing a rectangular coordinate system and determining the feature point set in the point cloud data, the coordinates of the feature points A, B and C can be accurately calculated, and the coordinates of the feature point E can be calculated by means of a given formula. This process realizes accurate conversion from point cloud data to key feature point coordinates, and provides an accurate point basis for subsequent grinding head wear modeling. Accurate calculation of feature points ensures the reliability of subsequent model construction, avoids modeling deviation caused by point error, and thus improves the accuracy of the entire wear model.

[0037] In an implementation, based on the grinding head radius and the known feature point coordinates, the grinding head center O point coordinates and the center axis equation are calculated, realizing the derivation from local features to the overall geometric relationship of the grinding head. By rotating the coordinate system, the distance of the feature point from the center line is expanded to a three-dimensional wear model. This process not only completely restores the wear pattern of the grinding head, but also considers the wear of the grinding head at different angles, making the wear model more comprehensive and accurate. The conversion from two-dimensional to three-dimensional enhances the model's ability to describe the wear state and provides strong support for the quantitative analysis of grinding head wear.

[0038] In an implementation, accurate feature point calculation provides a reliable basis for subsequent modeling, accurate solution of the grinding head center axis and coordinate system rotation operation enable the wear model to accurately reflect the actual wear condition of the grinding head. This modeling method based on point cloud data not only avoids complex manual measurement, but also improves the construction efficiency of the wear model, and overall realizes rapid and accurate modeling of the grinding head wear state, providing a high-quality data basis for subsequent wear analysis and prediction, and improving the intelligent level of grinding head wear monitoring and maintenance.

[0039] In one embodiment, the improved YOLOv5 model is used to extract features from image data to obtain grinding head data and workpiece data, the C3 module in the YOLOv5 model is replaced with an improved C3 module to obtain an improved YOLOv5 model, which includes: Working principle of the improved YOLOv5 model: The output of the third layer improved C3 module is convoluted by a 1x1 CBS module to obtain a first feature map, the output of the fifth layer improved C3 module is input into a 1x1 CBS module to obtain a second feature map, the output of the seventh layer improved C3 module is input into a 1x1 CBS module to obtain a third feature map, and the output of the tenth layer SPPF module is input into a 1x1 CBS module to obtain a fourth feature map; The fourth feature map is up-sampled to obtain a target fourth feature map, the second feature map is input into a 1x1 CBS module to obtain a target second feature map, and the target fourth feature map, the target second feature map and the third feature map are spliced to obtain a first fusion feature map; The first fusion feature map is input into a 1x1 CBS module to obtain a target fusion feature map, the first feature map is input into a 1x1 CBS module to obtain a target first feature map, and the target first feature map, the second feature map and the target fusion feature map are spliced to obtain a second fusion feature map; The second fusion feature map is input into a 1x1 CBS module to obtain a third fusion feature map, the third fusion feature map is taken as the output of the first detection head, the third fusion feature map and the target fusion feature map are spliced to obtain a fourth feature fusion map, the fourth feature fusion map is taken as the output of the second detection head, the fourth feature fusion map is input into a 1x1 CBS module to obtain a fifth feature fusion map, the fifth feature fusion map and the fourth feature map are spliced to obtain a sixth feature fusion map, and the sixth feature fusion map is taken as the output of the third detection head.

[0040] In an implementation manner, referring to Figure 3 , Figure 3 A model structure schematic diagram of a grinding machine tool machining precision error intelligent correction method provided by the embodiment of the application; a first feature map: the output (resolution is 80x80x256) of the third layer improved C3 module is output as a feature map with a resolution of 80x80x128 through a 1x1 CBS module (channel compression); a second feature map: the output (40x40x512) of the fifth layer improved C3 module is output as a feature map with a resolution of 40x40x256 through a 1x1 CBS module; a third feature map: the output (20x20x1024) of the seventh layer improved C3 module is output as a feature map with a resolution of 20x20x512 through a 1x1 CBS module; a fourth feature map: the output (20x20x1024) of the tenth layer SPPF module is output as a feature map with a resolution of 20x20x512 through a 1x1 CBS module; the redundant channels are compressed through 1x1 convolution, the subsequent calculation amount is reduced (for example, the channel number of the first feature map is compressed from 256 to 128), and the key features are retained.

[0041] In one implementation, the target fourth feature map: the fourth feature map (20x20x512) is upsampled to 40x40x512; the target second feature map: the second feature map (40x40x256) is adjusted to 40x40x512 in channel by a 1x1 CBS module; the splicing operation: the target fourth feature map (40x40x512), the target second feature map (40x40x512) and the third feature map (20x20x512) are upsampled to 40x40x512, and the first fusion feature map (40x40x1536) is generated by splicing; the high-resolution information after upsampling, the deep-layer semantics (20x20) and the middle-layer feature (40x40) are fused, so as to improve the recognition ability of the workpiece surface feature; and the channel number is unified by the 1x1 CBS, so as to avoid feature conflict in cross-resolution splicing.

[0042] In one implementation, the target fusion feature map: the first fusion feature map (40x40x1536) is compressed to 40x40x256 by a 1x1 CBS module; the target first feature map: the first feature map (80x80x128) is adjusted to 80x80x256 by a 1x1 CBS module; the splicing operation: the target first feature map (80x80x256), the second feature map (40x40x256) and the target fusion feature map (40x40x256) are upsampled to 80x80x256, and the second fusion feature map (80x80x768) is generated by splicing; small target detection enhancement: the 80x80 high-resolution feature map retains the texture details of the workpiece surface after grinding; and the calculation amount is reduced while the multi-scale information is maintained by multiple channel compression (1536→256→768).

[0043] In one implementation, the third fusion feature map: the second fusion feature map (80x80x768) is compressed to 80x80x255 by a 1x1 CBS module, and is output as a first detection head (detecting small targets); the fourth feature fusion map: the third fusion feature map (80x80x255) and the target fusion feature map (40x40x256) are upsampled to 80x80x256 and spliced to generate 80x80x511, which is output as a second detection head; the sixth feature fusion map: the fifth feature fusion map (generated by the fourth feature fusion map by a 1x1 CBS) and the fourth feature map (20x20x512) are upsampled to 40x40x512 and spliced to generate 40x40x1024, which is output as a third detection head; the first detection head (80x80x255): focusing on micro defects (grinding marks at the end of the grinding head); the second detection head (80x80x511): balancing small and medium defect detection (grinding marks at the middle stage of the grinding head); and the third detection head (40x40x1024): processing large-scale defects (grinding marks at the initial stage of the grinding head).

[0044] In one embodiment, the improved C3 module comprises: Replace BottleNeck*N in C3 module with DenseBlock*N and add NAM module at the end of C module to obtain improved C3 module; Working principle of improved C3 module: Input the first processing graph into the improved C3 module, input the first processing graph into the 1*1 CBS module for convolution to obtain a second processing graph, input the second processing graph into DenseBlock*N for dense connection to obtain a third processing graph, input the third processing graph into NAM for weighting to obtain a fourth processing graph, and the fourth processing graph is taken as the output of the improved C3 module.

[0045] In an implementation manner, BottleNeck*N is replaced with DenseBlock*N, through cross-layer dense connection (splicing of output of each layer with input of all subsequent layers), reuse of shallow texture features (such as shallow scratches on the surface of a workpiece) and deep semantic features (deep scratches) is realized, positioning accuracy of micro defects is improved, and dense connection of DenseBlock alleviates gradient vanishing problem of a deep network and accelerates convergence speed.

[0046] In an implementation manner, NAM (Normalization-based Attention Module) dynamically adjusts response intensity of different regions in a feature map through channel normalization and spatial attention weight distribution, suppresses irrelevant background noise (such as debris on the surface of a workpiece), and at the same time enhances feature contrast of a defect region (such as a grinding deep region); core mechanisms include: 1. Channel dimension normalization: balancing feature distribution of each channel, alleviating influence of uneven illumination on detection; 2. Spatial attention weighting: highlighting a defect region and weakening background interference through learnable parameters; overall effect: in the improved C3 module, NAM is taken as an end feature optimization component, and the network's ability to capture key information is improved; through enhancement of feature expression of a defect region and suppression of noise, detection robustness of the model in a complex industrial scene is improved, and false detection rate is reduced, so that the entire network realizes a balance between higher accuracy and efficiency.

[0047] Based on the same inventive concept, the embodiment of the present application also provides a grinding machine tool machining precision error intelligent correction system. Referring to Figure 4 , Figure 4 A structure diagram of a grinding machine tool machining precision error intelligent correction system provided by the embodiment of the present application comprises a data acquisition module, a period determination module and a data feedback module: The data acquisition module is used for acquiring target sound data and target workpiece data, inputting the target sound data into a target model to obtain target grinding head data and target workpiece data, and determining target grinding head point cloud data and target grinding head wear data according to the target grinding head data; A cycle determination module is configured to input target workpiece data into a target model to obtain workpiece wear data, and determine a wear cycle of the grinding head according to the target grinding head wear data and the workpiece wear data; the wear cycle includes an initial stage, an intermediate stage and a final stage. A data feedback module is configured to determine a feed compensation according to the wear cycle and the grinding head point cloud data, and feed the feed compensation back to the machine tool to correct the machining parameters; the machining parameters include a feed speed, a cutting depth, a grinding head rotating speed and a workpiece rotating speed.

[0048] The grinding machine machining precision error intelligent correction system provided by the embodiment of the present application can realize accurate evaluation of the wear state of the grinding head and the workpiece by non-contact acquisition of sound and workpiece data and rapid matching using a target model. The feed compensation is determined according to the wear cycle and the point cloud data, and is fed back to the machine tool to correct the machining parameters, thereby improving the machining precision and efficiency and prolonging the service life of the grinding head.

[0049] In one embodiment, the data acquisition module includes a training data acquisition module, a training data processing module and a model updating module. The training data acquisition module is configured to acquire image data and sound data of the workpiece in real time during machining, and extract features of the image data by improving a YOLOv5 model to obtain grinding head data and workpiece data; the image data and the sound data correspond to each other in a time cycle; the grinding head data includes grinding head point cloud data and grinding head wear data; The training data processing module is configured to model the grinding head in three dimensions according to the grinding head point cloud data to obtain a wear model sequence set, and determine a wear label sequence set according to the grinding head wear data; the wear model sequence set and the wear label sequence set correspond to each other in a time sequence; the training data further includes a grinding head rotating speed, a feed speed and a machining material; The model updating module is configured to input the training data into a preset model for training, take the output of the preset model as a model updating parameter, update the model parameters of the preset model according to the model updating parameter to obtain a target model; the output of the target model is the grinding head data and the workpiece wear data.

[0050] In one embodiment, the training data processing module includes a feature point determination module, an equation calculation module and a three-dimensional model generation module. The feature point determination module is configured to construct a rectangular coordinate system according to the grinding head point cloud data, determine a feature point set in the point cloud data, determine coordinates of a feature point A, a feature point B and a feature point C according to the feature point set, and calculate coordinates of a feature point E according to the feature point A, the feature point B and the feature point C; The calculation formula of the feature point E is as follows: , Wherein, (x A ,y A ) is the coordinate of point A, (x B ,y B ) is the coordinate of point B, (x E ,y E ) is the coordinate of point E, and (x C ,y C ) is the coordinate of point C.

[0051] An equation calculation module is configured to obtain the radius of the grinding head, calculate the coordinates of the center O point according to the radius, the coordinates of the feature point A and the coordinates of the feature point E, calculate the slope k according to the feature point A and the feature point B, and calculate the grinding head center axis equation according to the slope k and the coordinates of the center O point. A three-dimensional model generation module is configured to rotate the coordinate system of the grinding head center axis equation according to a preset angle to obtain a target coordinate system, calculate the distance d i of each feature point in the feature point set from the center line, and rotate the distance d i of the feature point set for one revolution to obtain a three-dimensional wear model; i is 1, 2, …, n; and the three-dimensional wear model is any one of the wear model sequence set. Coordinate system conversion formula: , Wherein, (x S ,y S ) is the current coordinate system, and (x' S ,y' S ) is the target coordinate system.

[0052] In one embodiment, the working principle of improving the YOLOv5 model includes: The output of the third layer improved C3 module is convoluted through a 1*1 CBS module to obtain a first feature map, the output of the fifth layer improved C3 module is input into the 1*1 CBS module for convolution to obtain a second feature map, the output of the seventh layer improved C3 module is input into the 1*1 CBS module for convolution to obtain a third feature map, and the output of the tenth layer SPPF module is input into the 1*1 CBS module for convolution to obtain a fourth feature map. The fourth feature map is up-sampled to obtain a target fourth feature map, the second feature map is input into the 1*1 CBS module for convolution to obtain a target second feature map, and the target fourth feature map, the target second feature map and the third feature map are spliced to obtain a first fusion feature map. The first fusion feature map is input into a 1*1 CBS module for convolution to obtain a target fusion feature map, the first feature map is input into a 1*1 CBS module for convolution to obtain a target first feature map, the target first feature map, the second feature map and the target fusion feature map are spliced to obtain a second fusion feature map; The second fusion feature map is input into a 1*1 CBS module for convolution to obtain a third fusion feature map, the third fusion feature map is taken as the output of the first detection head, the third fusion feature map and the target fusion feature map are spliced to obtain a fourth feature fusion map, the fourth feature fusion map is taken as the output of the second detection head, the fourth feature fusion map is input into a 1*1 CBS module for convolution to obtain a fifth feature fusion map, the fifth feature fusion map and the fourth feature map are spliced to obtain a sixth feature fusion map, and the sixth feature fusion map is taken as the output of the third detection head.

[0053] In one embodiment, the improved C3 module comprises: replacing BottleNeck*N in the C3 module with DenseBlock*N, and adding an NAM module at the end of the C module to obtain the improved C3 module. Working principle of the improved C3 module: The first processing map is input into the improved C3 module, the first processing map is input into a 1*1 CBS module for convolution to obtain a second processing map, the second processing map is input into a DenseBlock*N for dense connection to obtain a third processing map, the third processing map is input into an NAM module for weighting to obtain a fourth processing map, and the fourth processing map is taken as the output of the improved C3 module.

[0054] The above has described one embodiment of the present application in detail, but the content is only the preferred embodiment of the present application, and cannot be considered as used for limiting the implementation range of the present application. Any equivalent change and improvement made according to the scope of the present application application should still belong to the patent coverage range of the present application.

Claims

1. A method for intelligent correction of grinding machine accuracy errors, characterized in that, The method comprises: Obtaining target sound data and target workpiece data, inputting the target sound data into a target model to obtain target grinding head data and target workpiece data, and determining target grinding head point cloud data and target grinding head wear data according to the target grinding head data; Inputting the target workpiece data into the target model to obtain workpiece wear data, and determining the wear period of the grinding head according to the target grinding head wear data and the workpiece wear data; the wear period comprises an initial period, an intermediate period and a final period; According to the wear period and the grinding head point cloud data, the feed compensation is determined, and the feed compensation is fed back to the machine tool to correct the machining parameters; the machining parameters include feed speed, cutting depth, grinding head speed and workpiece speed.

2. The intelligent correction method for machining precision error of a grinding machine according to claim 1, characterized in that, Inputting the target sound data into the target model to obtain target grinding head data and target workpiece data, and the training process of the target model comprises: Real-time acquisition of image data and sound data of the workpiece during machining, feature extraction of the image data by improving the YOLOv5 model to obtain grinding head data and workpiece data; the image data and the sound data correspond to each other in time period; the grinding head data includes grinding head point cloud data and grinding head wear data; According to the grinding head point cloud data, a wear model sequence set is obtained by three-dimensional modeling of the grinding head, and a wear label sequence set is determined according to the grinding head wear data; the wear model sequence set, the wear label sequence set, the workpiece data and the sound data are used as training data; the wear model sequence set and the wear label sequence set correspond to each other in time sequence; the training data further includes grinding head speed, feed speed and machining material; The training data is input into a preset model for training, the output of the preset model is used as a model update parameter, the model parameters of the preset model are updated according to the model update parameter to obtain a target model; the output of the target model is grinding head data and workpiece wear data.

3. The intelligent correction method for machining accuracy error of a grinding machine according to claim 2, characterized in that, According to the grinding head point cloud data, a wear model sequence set is obtained by three-dimensional modeling of the grinding head, and a wear label sequence set is determined according to the grinding head wear data; the wear model sequence set, the wear label sequence set, the workpiece data and the sound data are used as training data; the wear model sequence set and the wear label sequence set correspond to each other in time sequence; the training data further includes grinding head speed, feed speed and machining material; The training data is input into a preset model for training, the output of the preset model is used as a model update parameter, the model parameters of the preset model are updated according to the model update parameter to obtain a target model; the output of the target model is grinding head data and workpiece wear data. According to the grinding head point cloud data, a wear model sequence set is obtained by three-dimensional modeling of the grinding head, and a wear label sequence set is determined according to the grinding head wear data; the wear model sequence set, the wear label sequence set, the workpiece data and the sound data are used as training data; the wear model sequence set and the wear label sequence set correspond to each other in time sequence; the training data further includes grinding head speed, feed speed and machining material; ， wherein is the coordinate of point A, is the coordinate of point B, is the coordinate of point E, is the coordinate of point C; The working principle of the improved YOLOv5 model: Rotating the coordinate system of the center axis equation of the grinding head according to a preset angle to obtain a target coordinate system, and calculating distances d of each feature point in the feature point set from the center line i , the distance d of the feature point set i Rotating one circle to obtain a three-dimensional wear model; i is 1, 2, …, n; the three-dimensional wear model is any one of the wear model sequence set ​ ， wherein, is the current coordinate system, is the target coordinate system.

4. The intelligent correction method for machining precision error of a grinding machine according to claim 2, characterized in that, ​ ​ The output of the third layer improved C3 module is convoluted by a 1*1 CBS module to obtain a first feature map, the output of the fifth layer improved C3 module is input into a 1*1 CBS module to obtain a second feature map, the output of the seventh layer improved C3 module is input into a 1*1 CBS module to obtain a third feature map, and the output of the tenth layer SPPF module is input into a 1*1 CBS module to obtain a fourth feature map; The fourth feature map is up-sampled to obtain a target fourth feature map, the second feature map is input into a 1*1 CBS module to obtain a target second feature map, and the target fourth feature map, the target second feature map and the third feature map are spliced to obtain a first fusion feature map; The first fusion feature map is input into a 1*1 CBS module to obtain a target fusion feature map, the first feature map is input into a 1*1 CBS module to obtain a target first feature map, and the target first feature map, the second feature map and the target fusion feature map are spliced to obtain a second fusion feature map; The second fusion feature map is input into a 1*1 CBS module to obtain a third fusion feature map, the third fusion feature map is taken as the output of a first detection head, the fourth feature map is taken as the output of a second detection head, the fourth feature map is input into a 1*1 CBS module to obtain a fifth feature fusion map, the fifth feature fusion map and the fourth feature map are spliced to obtain a sixth feature fusion map, and the sixth feature fusion map is taken as the output of a third detection head.

5. The intelligent correction method for machining accuracy error of a grinding machine according to claim 4, characterized in that, The improved C3 module comprises: The BottleNeck*N in the C3 module is replaced by a DenseBlock*N, and an NAM module is added at the end of the C module to obtain an improved C3 module; The working principle of the improved C3 module is as follows: The first processing graph is input into the improved C3 module, the first processing graph is input into a 1*1 CBS module to obtain a second processing graph, the second processing graph is input into a DenseBlock*N to obtain a third processing graph, the third processing graph is input into an NAM module to obtain a fourth processing graph, and the fourth processing graph is taken as the output of the improved C3 module.

6. A grinding machine processing precision error intelligent correction system, characterized in that, The system comprises a data acquisition module, a cycle determination module and a data feedback module: The data acquisition module is configured to acquire target sound data and target workpiece data, input the target sound data into a target model to obtain target grinding head data and target workpiece data, and determine target grinding head point cloud data and target grinding head wear data according to the target grinding head data; The cycle determination module is configured to input the target workpiece data into a target model to obtain workpiece wear data, and determine a wear cycle of a grinding head according to the target grinding head wear data and the workpiece wear data; the wear cycle comprises an initial stage, an intermediate stage and a final stage. The data feedback module is configured to determine a feed amount compensation according to the wear period and the grinding head point cloud data, and feed the feed amount compensation to a machine tool to make the machine tool correct machining parameters, including a feed speed, a cutting depth, a grinding head rotating speed, and a workpiece rotating speed.

7. The grinding machine processing precision error intelligent correction system according to claim 6, characterized in that, The data acquisition module comprises a training data acquisition module, a training data processing module, and a model updating module. The training data acquisition module is configured to acquire image data and sound data of a workpiece in real time during machining, and extract features of the image data by improving a YOLOv5 model to obtain grinding head data and workpiece data; the image data and the sound data correspond to each other in a time period; the grinding head data comprises grinding head point cloud data and grinding head wear data; The training data processing module is configured to model the grinding head in three dimensions according to the grinding head point cloud data to obtain a wear model sequence set, determine a wear label sequence set according to the grinding head wear data, and use the wear model sequence set, the wear label sequence set, the workpiece data, and the sound data as training data; the wear model sequence set and the wear label sequence set correspond to each other in a time sequence; the training data further comprises a grinding head rotating speed, a feed speed, and machining materials; The model updating module is configured to input the training data into a preset model for training, use an output of the preset model as a model updating parameter, update model parameters of the preset model according to the model updating parameter to obtain a target model; the output of the target model is grinding head data and workpiece wear data.

8. The grinding machine processing precision error intelligent correction system according to claim 7, characterized in that, The training data processing module comprises a feature point determination module, an equation calculation module, and a three-dimensional model generation module. The feature point determination module is configured to construct a rectangular coordinate system according to the grinding head point cloud data, determine a feature point set in the point cloud data, determine coordinates of feature points A, B, and C according to the feature point set, and calculate coordinates of feature point E according to the feature points A, B, and C; A calculation formula of the feature point E is as follows: ， wherein, is the coordinate of point A, is the coordinate of point B, is the coordinate of point E, is the coordinate of point C; The equation calculation module is configured to acquire a radius of the grinding head, calculate a center O point coordinate according to the radius, the feature point A coordinate, and the feature point E coordinate, calculate a slope k according to the feature points A and B, and calculate a grinding head center axis equation according to the slope k and the center O point coordinate; The three-dimensional model generation module is configured to rotate a coordinate system of the center axis equation of the grinding head by a preset angle to obtain a target coordinate system, and calculate distances d of each feature point in the feature point set from the center line i The distance d of the feature point set is calculated i A three-dimensional wear model is obtained by rotating one circle; i is an integer of 1, 2, …, n; and the three-dimensional wear model is any one of the wear model sequence set. A coordinate system conversion formula is as follows: ， wherein, is the current coordinate system, is the target coordinate system.

9. The grinding machine processing precision error intelligent correction system according to claim 7, characterized in that, The working principle of the improved YOLOv5 model comprises: The output of the third layer improved C3 module is convoluted through a 1×1 CBS module to obtain a first feature map, the output of the fifth layer improved C3 module is input into the 1×1 CBS module for convolution to obtain a second feature map, the output of the seventh layer improved C3 module is input into the 1×1 CBS module for convolution to obtain a third feature map, and the output of the tenth layer SPPF module is input into the 1×1 CBS module for convolution to obtain a fourth feature map. The fourth feature map is up-sampled to obtain a target fourth feature map, the second feature map is input into a 1*1 CBS module for convolution to obtain a target second feature map, and the target fourth feature map, the target second feature map and the third feature map are spliced to obtain a first fusion feature map; The first fusion feature map is input into a 1*1 CBS module for convolution to obtain a target fusion feature map, the first feature map is input into a 1*1 CBS module for convolution to obtain a target first feature map, and the target first feature map, the second feature map and the target fusion feature map are spliced to obtain a second fusion feature map; The second fusion feature map is input into a 1*1 CBS module for convolution to obtain a third fusion feature map, the third fusion feature map is taken as an output of a first detection head, the third fusion feature map and the target fusion feature map are spliced to obtain a fourth feature fusion map, the fourth feature fusion map is taken as an output of a second detection head, the fourth feature fusion map is input into a 1*1 CBS module for convolution to obtain a fifth feature fusion map, the fifth feature fusion map and the fourth feature map are spliced to obtain a sixth feature fusion map, and the sixth feature fusion map is taken as an output of a third detection head.

10. The grinding machine accuracy error intelligent correction system of claim 9, wherein, The improved C3 module comprises: replacing BottleNeck*N in the C3 module with DenseBlock*N, and adding an NAM module at the end of the C module to obtain the improved C3 module; The working principle of the improved C3 module is as follows: The first processing graph is input into the improved C3 module, the first processing graph is input into a 1*1 CBS module for convolution to obtain a second processing graph, the second processing graph is input into a DenseBlock*N for dense connection to obtain a third processing graph, the third processing graph is input into an NAM module for weighting to obtain a fourth processing graph, and the fourth processing graph is taken as an output of the improved C3 module.

Images