A method and system for monitoring the honing process

Abstract

This invention provides a method and system for monitoring the honing process. The method includes: acquiring gear tooth surface roughness, honing wheel meshing torque, and honing wheel surface wear texture data using a line laser profilometer, a dynamic torque sensor, and an industrial camera, respectively; generating a fused feature vector of the processing state through time synchronization and principal component analysis for dimensionality reduction; training a pre-set convolutional neural network using the fused feature vector as input, combined with historical data from multiple machine tools and virtual trial cutting data from a digital twin, to output the corresponding anomaly level and cause; associating the corresponding processing information through the gear's RDID identifier, and locating the corresponding abnormal workpiece by combining the anomaly timestamp, thus locking the corresponding target honing wheel; matching the corresponding adjustment processing parameters in a process knowledge base according to the anomaly level and cause, and adjusting the processing parameters to adjust the actual working state of the target honing wheel. This invention can effectively improve gear production efficiency.

Description

Technical Field

[0001] This invention relates to the field of honing technology, and in particular to a method and system for monitoring the honing process. Background Technology

[0002] Honing is the core process of gear finishing and is widely used in the manufacturing of key gears in the automotive, construction machinery and other fields. Through the meshing action of the honing wheel and the gear, the surface accuracy (from IT8~IT9 grade to IT6~IT7 grade) and surface quality (Ra value as low as 0.2~0.8μm) are improved, which directly determines the gear transmission performance and service life. Therefore, quality control in the machining process is of utmost importance.

[0003] Currently, the industry generally adopts a honing quality control model of "specific frequency sampling inspection + post-processing traceability and isolation": Sampling intervals are preset based on production experience (e.g., 3-5 pieces per hour after processing 50 pieces), and quality is judged by detecting tooth profile error, tooth direction accuracy, etc.; if defective products are found, products before and after the sampled batch are traced and isolated, and the causes of abnormalities such as honing wheel wear and parameter drift are analyzed offline. This model infers overall quality through sampling, balancing cost and efficiency.

[0004] However, this model has obvious drawbacks: the sampling interval makes it impossible to identify processing abnormalities such as honing wheel failure and cooling contamination in real time, resulting in the continuous generation of non-conforming products between two sampling inspections; after the problem is discovered in the sampling inspection, it is difficult to accurately define the scope of traceability and isolation, which can easily lead to the omission of non-conforming products or the over-isolation of qualified products. At the same time, the non-conforming products that have been generated will cause waste of raw materials and labor time, increase rework costs and production delay risks, especially in mass production, the economic losses are significant, and it cannot meet the real-time quality control needs of modern manufacturing industry. Summary of the Invention

[0005] Therefore, the purpose of this invention is to provide a honing process monitoring method and system to solve the problem that the existing technology cannot adjust the machining accuracy of the honing wheel in a timely manner, which easily leads to a large number of defective products.

[0006] The first aspect of the present invention proposes: A method for monitoring the honing process specifically includes the following steps: The tooth surface roughness of the gear, the meshing torque of the honing wheel, and the wear texture data of the honing wheel surface are collected by a line laser profilometer, a dynamic torque sensor, and an industrial camera, respectively. After time synchronization and dimensionality reduction by principal component analysis, a fused feature vector of the processing state is generated. Using the fused feature vector as input, a preset convolutional neural network is trained by combining historical data from multiple machine tools and virtual trial cutting data from a digital twin, so as to output the corresponding anomaly level and anomaly cause. By associating the corresponding processing information with the RDID identifier of the gear, and combining the abnormal timestamp, the corresponding abnormal workpiece can be located, so as to lock the corresponding target honing wheel. Based on the anomaly level and cause, the corresponding adjustment processing parameters are matched in the process knowledge base, and the actual working state of the target honing wheel is adjusted accordingly using the adjustment processing parameters.

[0007] The beneficial effects of this invention are as follows: By collecting specific processing parameters of the gear, a fused feature vector for subsequent analysis can be generated. Based on this, to facilitate rapid subsequent analysis, the current fused feature vector is combined with multi-level bed historical data and virtual trial cutting data from a digital twin to train a convolutional neural network, so as to accurately output the level and cause of abnormality encountered by the gear during processing. Based on this, while collecting the corresponding processing information, the corresponding target honing wheel can be determined. Based on this, the actual working state of the current target honing wheel is adjusted according to the generated adjustment processing parameters, thereby effectively detecting abnormalities and avoiding the generation of a large number of defective products, thus improving the production efficiency of the gear.

[0008] Furthermore, the step of generating a fused feature vector of the processing state through time synchronization and principal component analysis dimensionality reduction includes: Using the rotational speed signal of the honing spindle as the time reference, the time anchor points of each sensor data are extracted, and the timestamps are corrected through a deviation compensation algorithm to generate the corresponding standard dataset. The standard dataset is converted into the corresponding original feature matrix based on the LOF algorithm and Fourier transform. The target principal components are extracted from the original feature matrix using principal component analysis algorithm, and then the target principal components are transformed into the fused feature vector.

[0009] Furthermore, the step of extracting the target principal components from the original feature matrix using principal component analysis algorithm, and then converting the target principal components into the fused feature vector, includes: The mutual information entropy between the original feature matrix and the gear processing quality index is calculated, and the mutual information entropy is weighted and fused with the original feature matrix to generate the corresponding weighted feature matrix. Principal component analysis is performed on the weighted feature matrix to select initial principal components, and an attention mechanism is introduced to calculate the contribution score of the initial principal components to the identification of processing anomalies. Key principal components are selected based on the contribution scores, and fusion feature calculations are performed on the key principal components to generate the fusion feature vector.

[0010] Furthermore, the step of training a preset convolutional neural network using the fused feature vector as input, combined with historical data from multiple machine tools and virtual trial cutting data from a digital twin, to output the corresponding anomaly level and cause of the anomaly includes: The historical data of the multi-machine tool is subjected to oversampling of abnormal samples and undersampling of normal samples. A balanced dataset is generated by the SMOTE algorithm. The virtual trial cutting data is divided into a training set and a transfer validation set. The dual-source data are aligned by the domain adaptation algorithm to generate a corresponding collaborative training dataset. The balanced dataset and the co-training dataset are fused and temporally correlated using the GRU algorithm to generate the corresponding target training set. The preset convolutional neural network is trained using the target training set and a preset cross-entropy loss function to output the anomaly level and the cause of the anomaly.

[0011] Furthermore, the step of training the preset convolutional neural network using the target training set and a preset cross-entropy loss function to output the anomaly level and the cause of the anomaly includes: Extract the processing condition labels from the target training set, and perform dynamic enhancement processing on the target training set based on the processing condition labels to construct the corresponding multi-condition enhanced training set; A dual-branch convolutional neural network with a shared stratigraphic feature extraction layer and an independent fully connected layer is constructed, and a corresponding anomaly detection model is built by combining the multi-condition enhanced training set. The anomaly detection model outputs the anomaly level and cause corresponding to the gear.

[0012] Furthermore, the step of matching the corresponding adjustment processing parameters in the process knowledge base based on the anomaly level and cause includes: Construct a three-level index of anomaly level, anomaly cause, and parameter category to match the corresponding target processing case in the process knowledge base; The physical constraint rules for honing are extracted from the target machining case, and the machining effect is simulated using digital twins to output the corresponding quality deviation. When the minimum quality deviation is detected, the corresponding target parameter is extracted and used as the adjustment processing parameter.

[0013] Furthermore, the step of extracting the physical constraint rules for honing in the target machining case and simulating the machining effect through digital twin to output the corresponding quality deviation includes: Based on the actual machining conditions of the gear, the physical constraint rules of the honing process are dynamically adjusted using a fuzzy clustering algorithm to generate a suitable set of constraint rules. The constraint rule set is embedded into the digital twin model, and case parameters and actual processing data are input to simulate the meshing process and output multi-dimensional processing results. Differential weights are assigned to the multi-dimensional processing results to calculate the corresponding comprehensive deviation, and the deviation difference of the historical best case is combined to output the corresponding quality deviation.

[0014] The second aspect of the present invention proposes: A honing process monitoring system, wherein the system includes: The acquisition module is used to acquire gear tooth surface roughness, honing wheel meshing torque, and honing wheel surface wear texture data through a line laser profilometer, dynamic torque sensor, and industrial camera, respectively. After time synchronization and principal component analysis for dimensionality reduction, a fused feature vector of the processing state is generated. The training module is used to train a preset convolutional neural network by taking the fused feature vector as input and combining historical data of multiple machine tools with virtual trial cutting data of digital twins, so as to output the corresponding abnormality level and abnormality cause. The positioning module is used to associate the corresponding processing information with the RDID identifier of the gear, and locate the corresponding abnormal workpiece by combining the abnormal timestamp, so as to lock the corresponding target honing wheel. The adjustment module is used to match the corresponding adjustment processing parameters in the process knowledge base according to the abnormality level and the cause of the abnormality, and adjust the actual working state of the target honing wheel accordingly through the adjustment processing parameters.

[0015] Furthermore, the acquisition module is specifically used for: Using the rotational speed signal of the honing spindle as the time reference, the time anchor points of each sensor data are extracted, and the timestamps are corrected through a deviation compensation algorithm to generate the corresponding standard dataset. The standard dataset is converted into the corresponding original feature matrix based on the LOF algorithm and Fourier transform. The target principal components are extracted from the original feature matrix using principal component analysis algorithm, and then the target principal components are transformed into the fused feature vector.

[0016] Furthermore, the acquisition module is specifically used for: The mutual information entropy between the original feature matrix and the gear processing quality index is calculated, and the mutual information entropy is weighted and fused with the original feature matrix to generate the corresponding weighted feature matrix. Principal component analysis is performed on the weighted feature matrix to select initial principal components, and an attention mechanism is introduced to calculate the contribution score of the initial principal components to the identification of processing anomalies. Key principal components are selected based on the contribution scores, and fusion feature calculations are performed on the key principal components to generate the fusion feature vector.

[0017] Furthermore, the training module is specifically used for: The historical data of the multi-machine tool is subjected to oversampling of abnormal samples and undersampling of normal samples. A balanced dataset is generated by the SMOTE algorithm. The virtual trial cutting data is divided into a training set and a transfer validation set. The dual-source data are aligned by the domain adaptation algorithm to generate a corresponding collaborative training dataset. The balanced dataset and the co-training dataset are fused and temporally correlated using the GRU algorithm to generate the corresponding target training set. The preset convolutional neural network is trained using the target training set and a preset cross-entropy loss function to output the anomaly level and the cause of the anomaly.

[0018] Furthermore, the training module is specifically used for: Extract the processing condition labels from the target training set, and perform dynamic enhancement processing on the target training set based on the processing condition labels to construct the corresponding multi-condition enhanced training set; A dual-branch convolutional neural network with a shared stratigraphic feature extraction layer and an independent fully connected layer is constructed, and a corresponding anomaly detection model is built by combining the multi-condition enhanced training set. The anomaly detection model outputs the anomaly level and cause corresponding to the gear.

[0019] Furthermore, the adjustment module is specifically used for: Construct a three-level index of anomaly level, anomaly cause, and parameter category to match the corresponding target processing case in the process knowledge base; The physical constraint rules for honing are extracted from the target machining case, and the machining effect is simulated using digital twins to output the corresponding quality deviation. When the minimum quality deviation is detected, the corresponding target parameter is extracted and used as the adjustment processing parameter.

[0020] Furthermore, the adjustment module is specifically used for: Based on the actual machining conditions of the gear, the physical constraint rules of the honing process are dynamically adjusted using a fuzzy clustering algorithm to generate a suitable set of constraint rules. The constraint rule set is embedded into the digital twin model, and case parameters and actual processing data are input to simulate the meshing process and output multi-dimensional processing results. Differential weights are assigned to the multi-dimensional processing results to calculate the corresponding comprehensive deviation, and the deviation difference of the historical best case is combined to output the corresponding quality deviation.

[0021] The third aspect of the present invention proposes: A computer includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the honing process monitoring method as described above.

[0022] The fourth aspect of the present invention proposes: A readable storage medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the honing process monitoring method as described above.

[0023] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0024] Figure 1 A flowchart of the honing process monitoring method provided in the first embodiment of the present invention; Figure 2 This is a structural block diagram of the honing process monitoring system provided in the third embodiment of the present invention.

[0025] The following detailed description, in conjunction with the accompanying drawings, will further illustrate the present invention. Detailed Implementation

[0026] To facilitate understanding of the present invention, a more complete description will be given below with reference to the accompanying drawings. Several embodiments of the invention are illustrated in the drawings. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete.

[0027] It should be noted that when a component is said to be "fixed to" another component, it can be directly on the other component or there may be an intervening component. When a component is said to be "connected to" another component, it can be directly connected to the other component or there may be an intervening component. The terms "vertical," "horizontal," "left," "right," and similar expressions used in this document are for illustrative purposes only.

[0028] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0029] Please see Figure 1 The image shows a honing process monitoring method provided in the first embodiment of the present invention. The honing process monitoring method provided in this embodiment can quickly identify abnormalities in gear processing, thereby avoiding the generation of a large number of defective products and improving production efficiency.

[0030] Specifically, this embodiment provides: A method for monitoring the honing process specifically includes the following steps: Step S10: The tooth surface roughness of the gear, the meshing torque of the honing wheel, and the wear texture data of the honing wheel surface are collected by a line laser profilometer, a dynamic torque sensor, and an industrial camera, respectively. After time synchronization and principal component analysis for dimensionality reduction, a fused feature vector of the processing state is generated. It should be noted that the line laser profilometer, with its high-precision profile scanning capability, can directly capture the microscopic undulations of the gear tooth surface, thereby quantifying the tooth surface roughness; the dynamic torque sensor senses the torque fluctuations during the meshing process of the honing wheel and gear in real time, reflecting the meshing stability; and the industrial camera captures details such as the morphology and density of the wear texture on the honing wheel surface through high-definition imaging. These three devices collect data from three key dimensions: machining quality results, machining process force signals, and tool wear status. Since the sampling frequency and triggering timing of different devices differ, time synchronization is used to eliminate data time deviations. Then, principal component analysis is used for dimensionality reduction to remove redundant information and retain core features. The generated fused feature vector can comprehensively characterize the machining status, facilitating subsequent processing.

[0031] Step S20: Using the fused feature vector as input, train a preset convolutional neural network by combining historical data of multiple machine tools and virtual trial cutting data of digital twins to output the corresponding abnormality level and abnormality cause. It's important to note that using fused feature vectors as model input leverages the advantages of combining historical data from multiple machine tools with digital twin virtual trial cutting data. Specifically, the former provides accumulated data on real-world operating conditions, while the latter addresses the lack of real-world data under extreme or new conditions. By training a pre-defined convolutional neural network, the model can accurately learn the mapping relationship between machining states and anomalies, thereby outputting the anomaly level (e.g., mild, moderate, severe) and specific causes (e.g., honing wheel wear, excessive torque fluctuations), facilitating subsequent processing.

[0032] Step S30: The corresponding processing information is associated with the RDID identifier of the gear, and the corresponding abnormal workpiece is located by combining the abnormal timestamp, so as to lock the corresponding target honing wheel. It's important to note that the unique "identity code" of the gear's RDID can be used to quickly link all processing information, including the processing equipment, time, and process. Combined with the timestamp of the anomaly, the specific workpiece causing the anomaly can be accurately located. Furthermore, workpieces processed consecutively by the same honing wheel are correlated, allowing the identification of the target honing wheel causing the anomaly. This facilitates subsequent processing.

[0033] Step S40: Match the corresponding adjustment processing parameters in the process knowledge base according to the abnormality level and abnormality cause, and adjust the actual working state of the target honing wheel according to the adjustment processing parameters.

[0034] It should be noted that the process knowledge base integrates core data such as historical processing experience and theoretical parameter ranges. It adjusts parameters based on the anomaly level and cause, and then applies this adjustment to the actual working state of the target honing wheel, achieving precise intervention in the processing and completing a closed loop of "anomaly detection - problem location - problem resolution." This facilitates subsequent processing.

[0035] Second Embodiment Furthermore, the step of generating a fused feature vector of the processing state through time synchronization and principal component analysis dimensionality reduction includes: Using the rotational speed signal of the honing spindle as the time reference, the time anchor points of each sensor data are extracted, and the timestamps are corrected through a deviation compensation algorithm to generate the corresponding standard dataset. The standard dataset is converted into the corresponding original feature matrix based on the LOF algorithm and Fourier transform. The target principal components are extracted from the original feature matrix using principal component analysis algorithm, and then the target principal components are transformed into the fused feature vector.

[0036] It should be noted that the spindle speed signal is used as the time reference because the spindle speed is the core reference parameter for honing. Its signal has high stability and is completely synchronized with the machining process. By extracting the time anchor points of each sensor data, the correspondence between different data sources and the spindle rotation cycle can be established. Then, the deviation compensation algorithm is used to correct the timestamp deviation caused by equipment response delay, signal transmission time, etc., to ensure that all data accurately correspond to the same machining moment. The generated standard dataset lays a consistent foundation for subsequent feature extraction.

[0037] The LOF algorithm (Local Outlier Factor algorithm) can effectively identify and process abnormal data points in standard datasets caused by sensor failures, transient interference, etc. The Fourier transform converts sensor data in the time domain (such as the torque fluctuation curve over time) into frequency domain features, and mines hidden periodic patterns in the data (such as abnormal meshing frequency). The combination of the two transforms the standard dataset into an original feature matrix containing time domain, frequency domain and basic statistical features, comprehensively covering the data information.

[0038] Principal component analysis (PCA) is used to select target principal components with high variance contributions from the original feature matrix. These principal components are linear combinations of multiple original features, which can retain the core information of the original data while significantly reducing the feature dimensionality, avoiding the "curse of dimensionality" in subsequent model training. Finally, the target principal components are transformed into fused feature vectors that are both representative and concise, facilitating subsequent processing.

[0039] Furthermore, the step of extracting the target principal components from the original feature matrix using principal component analysis algorithm, and then converting the target principal components into the fused feature vector, includes: The mutual information entropy between the original feature matrix and the gear processing quality index is calculated, and the mutual information entropy is weighted and fused with the original feature matrix to generate the corresponding weighted feature matrix. Principal component analysis is performed on the weighted feature matrix to select initial principal components, and an attention mechanism is introduced to calculate the contribution score of the initial principal components to the identification of processing anomalies. Key principal components are selected based on the contribution scores, and fusion feature calculations are performed on the key principal components to generate the fusion feature vector.

[0040] It should be noted that the mutual information entropy between the original feature matrix and gear machining quality indicators (such as tooth surface accuracy, tooth thickness deviation, etc.) is calculated. The mutual information entropy quantifies the nonlinear correlation between features and quality indicators. By weighting and fusing it with the original feature matrix, features that are strongly correlated with machining quality can be given higher weights. The resulting weighted feature matrix focuses more on the core information that has a direct impact on machining quality and reduces the interference of irrelevant features.

[0041] First, initial principal components are selected through principal component analysis. Although these principal components retain most of the information, their contribution to anomaly identification varies. After introducing an attention mechanism, the model can autonomously learn the importance of different initial principal components in the anomaly identification task and calculate the contribution score (e.g., the wear texture related principal components contribute more to the "honing wheel wear" anomaly).

[0042] Key principal components are selected based on contribution scores, and redundant principal components with extremely low contribution are removed. Then, a fusion feature vector is generated through fusion feature calculation (such as feature concatenation, weighted summation, etc.). This vector can accurately match the needs of subsequent anomaly identification tasks and improve the accuracy of model diagnosis.

[0043] Furthermore, the step of training a preset convolutional neural network using the fused feature vector as input, combined with historical data from multiple machine tools and virtual trial cutting data from a digital twin, to output the corresponding anomaly level and cause of the anomaly includes: The historical data of the multi-machine tool is subjected to oversampling of abnormal samples and undersampling of normal samples. A balanced dataset is generated by the SMOTE algorithm. The virtual trial cutting data is divided into a training set and a transfer validation set. The dual-source data are aligned by the domain adaptation algorithm to generate a corresponding collaborative training dataset. The balanced dataset and the co-training dataset are fused and temporally correlated using the GRU algorithm to generate the corresponding target training set. The preset convolutional neural network is trained using the target training set and a preset cross-entropy loss function to output the anomaly level and the cause of the anomaly.

[0044] It should be noted that historical data from multiple machine tools often suffers from an imbalance problem, with "more normal samples and fewer abnormal samples." Oversampling of abnormal samples and undersampling of normal samples can initially adjust the sample ratio. The SMOTE algorithm further optimizes the dataset balance by synthesizing new abnormal samples, avoiding missed detections caused by the model biasing towards normal samples. The digital twin virtual trial cutting data is divided into a training set and a transfer validation set, and then the dual-source data is aligned using a domain adaptation algorithm. Specifically, because there are domain differences between real and virtual data (such as the difference in friction coefficient between the virtual and real environments), domain adaptation can eliminate these differences, making the dual-source data form a unified collaborative training dataset and improving data utilization.

[0045] The GRU (Gated Recurrent Unit) algorithm excels at processing time-series data. The honing process has obvious temporal characteristics (such as processing progress and parameter changes over time). By fusing and associating the temporal information in the balanced dataset and the co-training dataset through GRU, the dynamic change patterns of features during the processing can be discovered, and the generated target training set is more in line with the actual temporal characteristics of the processing.

[0046] Using the target training set as input, a convolutional neural network is trained using a pre-defined cross-entropy loss function. The cross-entropy loss function provides more accurate error calculation for classification tasks and effectively guides the model to learn the mapping relationship between processing states and anomaly levels and causes, ultimately achieving accurate output of anomaly levels and causes for subsequent processing.

[0047] Furthermore, the step of training the preset convolutional neural network using the target training set and a preset cross-entropy loss function to output the anomaly level and the cause of the anomaly includes: Extract the processing condition labels from the target training set, and perform dynamic enhancement processing on the target training set based on the processing condition labels to construct the corresponding multi-condition enhanced training set; A dual-branch convolutional neural network with a shared stratigraphic feature extraction layer and an independent fully connected layer is constructed, and a corresponding anomaly detection model is built by combining the multi-condition enhanced training set. The anomaly detection model outputs the anomaly level and cause corresponding to the gear.

[0048] It should be noted that the processing condition labels (such as gear material, module, processing accuracy requirements, etc.) are extracted from the target training set. Based on these labels, the training set is dynamically enhanced (such as adjusting parameters to generate simulated samples under different working conditions). The constructed multi-working-condition enhanced training set can cover a variety of actual processing scenarios, avoiding the insufficient generalization ability caused by the model only adapting to a single working condition.

[0049] A dual-branch convolutional neural network is constructed. The shared stratum feature extraction layer can extract common features under different working conditions (such as the basic variation law of meshing torque), while the independent fully connected layer extracts specific features for different working conditions (such as the difference in tooth surface roughness features of gears of different materials). The anomaly detection model constructed by combining the multi-working-condition enhanced training set can capture common laws and adapt to the differences in working conditions.

[0050] The trained anomaly detection model directly outputs the anomaly level and cause corresponding to the currently processed gear, providing a precise basis for subsequent target locking and parameter adjustment, thus facilitating subsequent processing.

[0051] Furthermore, the step of matching the corresponding adjustment processing parameters in the process knowledge base based on the anomaly level and cause includes: Construct a three-level index of anomaly level, anomaly cause, and parameter category to match the corresponding target processing case in the process knowledge base; The physical constraint rules for honing are extracted from the target machining case, and the machining effect is simulated using digital twins to output the corresponding quality deviation. When the minimum quality deviation is detected, the corresponding target parameter is extracted and used as the adjustment processing parameter.

[0052] It should be noted that a three-level index of "abnormality level - abnormality cause - parameter category" is constructed. This index establishes a direct association between abnormal information and process parameters, such as "severe abnormality - honing wheel wear - feed speed, honing wheel speed". Through the index, target processing cases with similar abnormalities can be quickly matched in the process knowledge base, which greatly improves the efficiency of case matching.

[0053] Extract the physical constraints of honing (such as torque limit, feed rate range, etc.) from the target machining case. These rules are the basis for ensuring machining feasibility. Embed them into the digital twin model, input case parameters and current actual machining data, and you can simulate the machining effect under different parameters and output the quality deviation (such as the difference between the actual tooth surface roughness and the standard value).

[0054] By iteratively simulating different parameter combinations, the target parameter corresponding to the minimum detected quality deviation is the optimal parameter that can solve the current anomaly. Using it as the processing parameter adjustment can ensure the effectiveness of parameter adjustment.

[0055] Furthermore, the step of extracting the physical constraint rules for honing in the target machining case and simulating the machining effect through digital twin to output the corresponding quality deviation includes: Based on the actual machining conditions of the gear, the physical constraint rules of the honing process are dynamically adjusted using a fuzzy clustering algorithm to generate a suitable set of constraint rules. The constraint rule set is embedded into the digital twin model, and case parameters and actual processing data are input to simulate the meshing process and output multi-dimensional processing results. Differential weights are assigned to the multi-dimensional processing results to calculate the corresponding comprehensive deviation, and the deviation difference of the historical best case is combined to output the corresponding quality deviation.

[0056] It should be noted that, based on the actual gear machining conditions (such as current machining temperature, lubrication status, etc.), the physical constraint rules for honing in the target case are dynamically adjusted through fuzzy clustering algorithm. For example, when the actual temperature is high, the torque constraint is appropriately relaxed. The generated set of adaptive constraint rules is more in line with the current machining environment, avoiding the disconnect between simulation results and reality caused by the solidification of rules.

[0057] By embedding the adjusted set of constraint rules into the digital twin model and inputting case parameters and actual machining data, the model can accurately simulate the meshing process of the honing wheel and gear, and output multi-dimensional machining results (such as tooth surface roughness, tooth profile accuracy, machining efficiency, etc.).

[0058] Based on the degree of influence of different processing results on the final quality, differentiated weights are assigned (e.g., the weight of tooth surface roughness is higher than that of processing efficiency). A comprehensive deviation is calculated, and then corrected by combining the deviation difference from historical best cases. The output quality deviation more accurately reflects the actual quality effect after parameter adjustment, providing a reliable basis for selecting the optimal parameters and facilitating subsequent processing.

[0059] Please see Figure 2 The third embodiment of the present invention provides: A honing process monitoring system, wherein the system includes: The acquisition module is used to acquire gear tooth surface roughness, honing wheel meshing torque, and honing wheel surface wear texture data through a line laser profilometer, dynamic torque sensor, and industrial camera, respectively. After time synchronization and principal component analysis for dimensionality reduction, a fused feature vector of the processing state is generated. The training module is used to train a preset convolutional neural network by taking the fused feature vector as input and combining historical data of multiple machine tools with virtual trial cutting data of digital twins, so as to output the corresponding abnormality level and abnormality cause. The positioning module is used to associate the corresponding processing information with the RDID identifier of the gear, and locate the corresponding abnormal workpiece by combining the abnormal timestamp, so as to lock the corresponding target honing wheel. The adjustment module is used to match the corresponding adjustment processing parameters in the process knowledge base according to the abnormality level and the cause of the abnormality, and adjust the actual working state of the target honing wheel accordingly through the adjustment processing parameters.

[0060] Furthermore, the acquisition module is specifically used for: Using the rotational speed signal of the honing spindle as the time reference, the time anchor points of each sensor data are extracted, and the timestamps are corrected through a deviation compensation algorithm to generate the corresponding standard dataset. The standard dataset is converted into the corresponding original feature matrix based on the LOF algorithm and Fourier transform. The target principal components are extracted from the original feature matrix using principal component analysis algorithm, and then the target principal components are transformed into the fused feature vector.

[0061] Furthermore, the acquisition module is specifically used for: The mutual information entropy between the original feature matrix and the gear processing quality index is calculated, and the mutual information entropy is weighted and fused with the original feature matrix to generate the corresponding weighted feature matrix. Principal component analysis is performed on the weighted feature matrix to select initial principal components, and an attention mechanism is introduced to calculate the contribution score of the initial principal components to the identification of processing anomalies. Key principal components are selected based on the contribution scores, and fusion feature calculations are performed on the key principal components to generate the fusion feature vector.

[0062] Furthermore, the training module is specifically used for: The historical data of the multi-machine tool is subjected to oversampling of abnormal samples and undersampling of normal samples. A balanced dataset is generated by the SMOTE algorithm. The virtual trial cutting data is divided into a training set and a transfer validation set. The dual-source data are aligned by the domain adaptation algorithm to generate a corresponding collaborative training dataset. The balanced dataset and the co-training dataset are fused and temporally correlated using the GRU algorithm to generate the corresponding target training set. The preset convolutional neural network is trained using the target training set and a preset cross-entropy loss function to output the anomaly level and the cause of the anomaly.

[0063] Furthermore, the training module is specifically used for: Extract the processing condition labels from the target training set, and perform dynamic enhancement processing on the target training set based on the processing condition labels to construct the corresponding multi-condition enhanced training set; A dual-branch convolutional neural network with a shared stratigraphic feature extraction layer and an independent fully connected layer is constructed, and a corresponding anomaly detection model is built by combining the multi-condition enhanced training set. The anomaly detection model outputs the anomaly level and cause corresponding to the gear.

[0064] Furthermore, the adjustment module is specifically used for: Construct a three-level index of anomaly level, anomaly cause, and parameter category to match the corresponding target processing case in the process knowledge base; The physical constraint rules for honing are extracted from the target machining case, and the machining effect is simulated using digital twins to output the corresponding quality deviation. When the minimum quality deviation is detected, the corresponding target parameter is extracted and used as the adjustment processing parameter.

[0065] Furthermore, the adjustment module is specifically used for: Based on the actual machining conditions of the gear, the physical constraint rules of the honing process are dynamically adjusted using a fuzzy clustering algorithm to generate a suitable set of constraint rules. The constraint rule set is embedded into the digital twin model, and case parameters and actual processing data are input to simulate the meshing process and output multi-dimensional processing results. Differential weights are assigned to the multi-dimensional processing results to calculate the corresponding comprehensive deviation, and the deviation difference of the historical best case is combined to output the corresponding quality deviation.

[0066] The fourth embodiment of the present invention provides a computer, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the honing process monitoring method as described above.

[0067] The fifth embodiment of the present invention provides a readable storage medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the honing process monitoring method as described above.

[0068] In summary, the honing process monitoring method and system provided in the above embodiments of the present invention can quickly identify abnormalities in the processing and make rapid adjustments, thereby avoiding the generation of a large number of defective products and improving production efficiency.

[0069] It should be noted that the above modules can be functional modules or program modules, and can be implemented through software or hardware. For modules implemented through hardware, the above modules can reside in the same processor; or the above modules can be located in different processors in any combination.

[0070] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-including system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0071] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0072] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0073] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0074] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the appended claims.

Claims

1. A method for monitoring the honing process, characterized in that, The method includes: The tooth surface roughness of the gear, the meshing torque of the honing wheel, and the wear texture data of the honing wheel surface are collected by a line laser profilometer, a dynamic torque sensor, and an industrial camera, respectively. After time synchronization and dimensionality reduction by principal component analysis, a fused feature vector of the processing state is generated. Using the fused feature vector as input, a preset convolutional neural network is trained by combining historical data from multiple machine tools and virtual trial cutting data from a digital twin, so as to output the corresponding anomaly level and anomaly cause. By associating the corresponding processing information with the RDID identifier of the gear, and combining the abnormal timestamp, the corresponding abnormal workpiece can be located, so as to lock the corresponding target honing wheel. Based on the anomaly level and cause, the corresponding adjustment processing parameters are matched in the process knowledge base, and the actual working state of the target honing wheel is adjusted accordingly through the adjustment processing parameters. The step of training a preset convolutional neural network by using the fused feature vector as input, combining historical data from multiple machine tools with virtual trial cutting data from a digital twin, and outputting the corresponding anomaly level and cause of the anomaly includes: The historical data of the multi-machine tool is subjected to oversampling of abnormal samples and undersampling of normal samples. A balanced dataset is generated by the SMOTE algorithm. The virtual trial cutting data is divided into a training set and a transfer validation set. The dual-source data are aligned by the domain adaptation algorithm to generate a corresponding collaborative training dataset. The balanced dataset and the co-training dataset are fused and temporally correlated using the GRU algorithm to generate the corresponding target training set. The preset convolutional neural network is trained using the target training set and a preset cross-entropy loss function to output the anomaly level and the cause of the anomaly. The step of training the preset convolutional neural network using the target training set and a preset cross-entropy loss function to output the anomaly level and the cause of the anomaly includes: Extract the processing condition labels from the target training set, and perform dynamic enhancement processing on the target training set based on the processing condition labels to construct the corresponding multi-condition enhanced training set; A dual-branch convolutional neural network with a shared stratigraphic feature extraction layer and an independent fully connected layer is constructed, and a corresponding anomaly detection model is built by combining the multi-condition enhanced training set. The anomaly detection model outputs the anomaly level and cause corresponding to the gear.

2. The honing process monitoring method according to claim 1, characterized in that, The step of generating a fused feature vector of the processing state through time synchronization and principal component analysis dimensionality reduction includes: Using the rotational speed signal of the honing spindle as the time reference, the time anchor points of each sensor data are extracted, and the timestamps are corrected through a deviation compensation algorithm to generate the corresponding standard dataset. The standard dataset is converted into the corresponding original feature matrix based on the LOF algorithm and Fourier transform. The target principal components are extracted from the original feature matrix using principal component analysis algorithm, and then the target principal components are transformed into the fused feature vector.

3. The honing process monitoring method according to claim 2, characterized in that, The step of extracting the target principal components from the original feature matrix using principal component analysis algorithm, and then converting the target principal components into the fused feature vector, includes: The mutual information entropy between the original feature matrix and the gear processing quality index is calculated, and the mutual information entropy is weighted and fused with the original feature matrix to generate the corresponding weighted feature matrix. Principal component analysis is performed on the weighted feature matrix to select initial principal components, and an attention mechanism is introduced to calculate the contribution score of the initial principal components to the identification of processing anomalies. Key principal components are selected based on the contribution scores, and fusion feature calculations are performed on the key principal components to generate the fusion feature vector.

4. The honing process monitoring method according to claim 1, characterized in that, The step of matching the corresponding adjustment processing parameters in the process knowledge base based on the anomaly level and cause includes: Construct a three-level index of anomaly level, anomaly cause, and parameter category to match the corresponding target processing case in the process knowledge base; The physical constraint rules for honing are extracted from the target machining case, and the machining effect is simulated using digital twins to output the corresponding quality deviation. When the minimum quality deviation is detected, the corresponding target parameter is extracted and used as the adjustment processing parameter.

5. The honing process monitoring method according to claim 4, characterized in that, The steps of extracting the physical constraint rules for honing in the target machining case and simulating the machining effect through digital twin to output the corresponding quality deviation include: Based on the actual machining conditions of the gear, the physical constraint rules of the honing process are dynamically adjusted using a fuzzy clustering algorithm to generate a suitable set of constraint rules. The constraint rule set is embedded into the digital twin model, and case parameters and actual processing data are input to simulate the meshing process and output multi-dimensional processing results. Differential weights are assigned to the multi-dimensional processing results to calculate the corresponding comprehensive deviation, and the deviation difference of the historical best case is combined to output the corresponding quality deviation.

6. A honing process monitoring system, characterized in that, The system is used to implement the honing process monitoring method as described in any one of claims 1 to 5, the system comprising: The acquisition module is used to acquire gear tooth surface roughness, honing wheel meshing torque, and honing wheel surface wear texture data through a line laser profilometer, dynamic torque sensor, and industrial camera, respectively. After time synchronization and principal component analysis for dimensionality reduction, a fused feature vector of the processing state is generated. The training module is used to train a preset convolutional neural network by taking the fused feature vector as input and combining historical data of multiple machine tools with virtual trial cutting data of digital twins, so as to output the corresponding abnormality level and abnormality cause. The positioning module is used to associate the corresponding processing information with the RDID identifier of the gear, and locate the corresponding abnormal workpiece by combining the abnormal timestamp, so as to lock the corresponding target honing wheel. The adjustment module is used to match the corresponding adjustment processing parameters in the process knowledge base according to the abnormality level and the cause of the abnormality, and adjust the actual working state of the target honing wheel accordingly through the adjustment processing parameters.

7. A computer comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the honing process monitoring method as described in any one of claims 1 to 5.

8. A readable storage medium having a computer program stored thereon, characterized in that, When executed by the processor, the program implements the honing process monitoring method as described in any one of claims 1 to 5.

Images