Grinding parameter optimization method of full-automatic coarse and fine grinding all-in-one machine based on particle swarm optimization

Abstract

The invention relates to the technical field of control parameter optimization, in particular to a particle swarm optimization-based grinding parameter optimization method for a full-automatic coarse and fine grinding all-in-one machine. The method comprises the following steps: acquiring equipment operation multi-modal data, and performing multi-thread aggregation to obtain structured grinding field sensing data; performing grinding parameter semantic abstraction on the structured grinding field sensing data to obtain a grinding parameter intention nested mapping table; performing rough solution space boundary evolution on the grinding parameter intention nesting mapping table to obtain a coarse-grained grinding parameter track group; soft clustering biased learning is carried out on the coarse-grained grinding parameter trajectory group, and a high-dimensional grinding parameter candidate map is obtained; and parameter tensor loopback optimization is conducted according to the high-dimensional grinding parameter candidate atlas, and a stable grinding parameter set is obtained. According to the invention, the precision and the production efficiency of the equipment are improved, the energy consumption is reduced, and the abrasion and the maintenance cost of the equipment are reduced, so that higher benefits and sustainable development are realized in industrial production.

Description

Technical Field

[0001] The present invention relates to the technical field of control parameter optimization, and in particular to a grinding parameter optimization method for a fully automatic coarse and fine grinding machine based on particle swarm optimization. Background Art

[0002] As an intelligent equipment that integrates rough grinding and fine grinding processes, the fully automatic coarse and fine grinding machine has the advantages of high efficiency, high precision and process integration. In actual application, the reasonable configuration of grinding process parameters (including grinding speed, feed speed, grinding depth, grinding wheel linear speed, etc.) has a direct impact on indicators such as workpiece surface quality, processing efficiency, equipment stability and energy consumption control. In the existing technology, the selection of grinding parameters mostly relies on manual experience or rule-based decision-making systems, which is difficult to effectively deal with complex working condition control problems under multi-objective and multi-constraint conditions. Some technical solutions introduce traditional optimization methods such as genetic algorithms and simulated annealing to adjust parameters, but because the grinding process exhibits highly nonlinear, dynamic coupling and uncertainty characteristics, traditional optimization methods have problems such as low search efficiency, easy to fall into local optimality, and sensitivity to initial population, which makes it difficult to meet the global optimization requirements in high-dimensional parameter space. Summary of the Invention

[0003] Based on this, it is necessary for the present invention to provide a grinding parameter optimization method for a fully automatic coarse and fine grinding machine based on particle swarm optimization to solve at least one of the above technical problems.

[0004] To achieve the above objectives, a grinding parameter optimization method for a fully automatic coarse and fine grinding machine based on particle swarm optimization is proposed, comprising the following steps: Step S1: Acquire multimodal data of equipment operation and perform preliminary edge-level processing on the multimodal data to extract implicit physical disturbance features; perform multi-threaded aggregation on the implicit physical disturbance features to obtain structured grinding field perception data; Step S2: performing multimodal grinding behavior feature analysis on the structured grinding field perception data to obtain grinding behavior pattern data; performing grinding parameter semantic abstraction based on the grinding behavior pattern data to obtain a grinding parameter intention nested mapping table; Step S3: performing an initialization basic particle swarm calculation on the grinding parameter intention nested mapping table to obtain an initialization basic particle swarm; performing a rough solution space boundary evolution based on the initialization basic particle swarm to obtain a coarse-grained grinding parameter trajectory family; Step S4: Perform subgraph nesting reconstruction on the coarse-grained grinding parameter trajectory family and construct a heterogeneous trajectory graph encoding vector set; perform soft clustering bias learning on the heterogeneous trajectory graph encoding vector set to obtain a high-dimensional grinding parameter candidate map; Step S5: performing a multi-dimensional working condition simulation based on the high-dimensional grinding parameter candidate map, and performing parameter tensor loop optimization based on the multi-dimensional working condition simulation results to obtain a stable grinding parameter set; Step S6: The stable grinding parameter set is transmitted to the management platform of the fully automatic rough and fine grinding machine to execute the control parameters, and iterative precision compensation mapping is performed according to the actual grinding deviation feedback data obtained in real time to obtain the optimized grinding parameter set.

[0005] The present invention effectively solves the problems of instability and low efficiency in the traditional grinding process through precise data acquisition, disturbance identification and optimization adjustment. First, in the raw data acquisition stage, the collaborative work of multimodal sensors comprehensively captures relevant information on the equipment, workpiece and environmental status, providing a solid data foundation for subsequent analysis. By reasonably setting the time window and standard deviation threshold, the cleanliness and reliability of the data are guaranteed, noise interference is eliminated, and the accuracy of subsequent processing is ensured. In the disturbance identification stage, the precise disturbance detection capability enables the early identification of abnormal fluctuations that may occur in the grinding process, provides a basis for subsequent precise adjustment, and avoids errors caused by external factors or equipment instability. In the process of behavioral modeling and map generation, by adopting center transfer learning and fuzzy clustering technology, the key features of the grinding process can be effectively extracted and high-quality behavioral maps can be generated. These behavioral maps provide detailed references for each stage of the grinding process, effectively guide the optimization adjustment of parameters, and ensure the stability and accuracy of the grinding process. The set clustering parameters and clustering thresholds help the system accurately classify different operating conditions, making the map more realistic and enabling rapid response to operating condition changes and real-time adjustments, thus ensuring process adaptability and accuracy. During the grinding parameter optimization phase, a particle swarm search algorithm intelligently adjusts the grinding parameters to find the optimal parameters for different operating conditions, avoiding the subjectivity and uncertainty inherent in manual adjustments in traditional methods. This optimization method not only improves grinding efficiency and reduces energy consumption, but also effectively reduces product rejection rates and enhances production quality. Furthermore, through multi-threaded aggregation and feedback compensation mechanisms, the optimization system can adapt to changing operating conditions in real time, ensuring that the grinding process remains optimal. In summary, through the combined application of these technical approaches, optimization technology has significantly enhanced the stability, efficiency, and intelligence of the fully automatic coarse and fine grinding machine. Reasonable parameter settings and precise operating condition feedback mechanisms not only improve equipment precision and production efficiency, but also reduce energy consumption, equipment wear, and maintenance costs, thereby achieving higher efficiency and sustainable development in industrial production. BRIEF DESCRIPTION OF THE DRAWINGS

[0006] Other features, objects and advantages of the present invention will become more apparent upon reading the detailed description of non-limiting embodiments thereof made with reference to the following drawings: Figure 1 Schematic diagram of the steps of the grinding parameter optimization method of the fully automatic coarse and fine grinding machine based on particle swarm optimization of the present invention; Figure 2 Detailed step flow diagram of step S1 in the present invention; Figure 3 Detailed step flow diagram of step S3 in the present invention; The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings and in conjunction with the embodiments. DETAILED DESCRIPTION

[0007] The following is a clear and complete description of the technical method of the present invention in conjunction with the accompanying drawings. It is obvious that the embodiments described are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making any creative efforts are within the scope of protection of the present invention.

[0008] In addition, the accompanying drawings are merely schematic illustrations of the present invention and are not necessarily drawn to scale. Identical reference numerals in the figures denote identical or similar parts, and thus repetitive descriptions thereof will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities that do not necessarily correspond to physically or logically separate entities. These functional entities may be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor and / or microcontroller approaches.

[0009] It should be understood that although the terms "first," "second," and the like may be used herein to describe various elements, these elements should not be limited by these terms. These terms are used solely to distinguish one element from another. For example, a first element may be referred to as a second element, and similarly, a second element may be referred to as a first element, without departing from the scope of the exemplary embodiments. The term "and / or" as used herein includes any and all combinations of one or more of the listed associated items.

[0010] To achieve this, please refer to Figures 1 to 3 The present invention provides a grinding parameter optimization method for a fully automatic coarse and fine grinding machine based on particle swarm optimization, the method comprising the following steps: Step S1: Acquire multimodal data of equipment operation and perform preliminary edge-level processing on the multimodal data to extract implicit physical disturbance features; perform multi-threaded aggregation on the implicit physical disturbance features to obtain structured grinding field perception data; In this embodiment, the real-time operating status of the fully automatic coarse and fine grinding machine is analyzed, and multi-source data on heat flow, current, vibration, and sound spectrum generated during a continuous 20-minute grinding cycle are collected. The sampling frequencies are set to 10 Hz for heat flow, 100 Hz for current, 1 kHz for vibration, and 8 kHz for sound spectrum, respectively. First-order noise reduction and time series normalization are performed on the data of each channel. After preliminary processing, a parallel multi-threaded structure is used to extract a sequence of feature points that are highly correlated with changes in the contact state of the grinding wheel surface, including heat flow surge intervals, current fluctuation segments, vibration peak slope mutation points, and abnormal sound spectrum frequency bands. The above feature point sequences are aligned and fused according to the time clues to form structured grinding field perception data based on the grinding wheel linear speed, which facilitates the subsequent analysis of behavioral dynamic characteristics.

[0011] Step S2: performing multimodal grinding behavior feature analysis on the structured grinding field perception data to obtain grinding behavior pattern data; performing grinding parameter semantic abstraction based on the grinding behavior pattern data to obtain a grinding parameter intention nested mapping table; In this embodiment, structured grinding field perception data is imported into the time series recognition process. Each set of data is sliced ​​and annotated with grinding behavior. Continuous behavior frames are segmented into 4-second time windows, and a critical change threshold is set to identify contact arc disturbance characteristics within the grinding zone. Using the grinding wheel linear speed as the main axis reference, each modal disturbance response is mapped onto a unified behavioral timeline, and high-impact behavior segments, including abrasive particle breakage, shedding, and localized thermal burns in the contact zone, are identified. Based on this, a semantic response mapping relationship is established between behaviors and parameter groups, and a nested mapping table of grinding parameter intentions is output. Each behavior segment is associated with at least three adjustable parameter dimensions, and a response weight is generated based on the frequency and amplitude change intensity.

[0012] Step S3: performing an initialization basic particle swarm calculation on the grinding parameter intention nested mapping table to obtain an initialization basic particle swarm; performing a rough solution space boundary evolution based on the initialization basic particle swarm to obtain a coarse-grained grinding parameter trajectory family; In this example, based on the grinding parameter intention nested mapping table, the initial particle size of the particle swarm was set to 30, with the particle dimensions corresponding one-to-one to the valid adjustable dimensions in the grinding parameter set (such as linear speed, feed rate, and grinding depth). The parameter swarm was initialized using the behavioral response weight as the initial position offset. During operation, the boundary relaxation coefficient for each evolution round was set to 0.1, limiting each parameter update offset to no more than 20% of the local optimum in the previous round. The densest trajectory community was extracted from 10 preliminary evolution rounds to form a coarse-grained grinding parameter trajectory cluster. All of these trajectories exhibited dynamic characteristics with a behavioral response significance threshold exceeding 0.8, which served as guidance for the subsequent trajectory structure construction and optimization.

[0013] Step S4: Perform subgraph nesting reconstruction on the coarse-grained grinding parameter trajectory family and construct a heterogeneous trajectory graph encoding vector set; perform soft clustering bias learning on the heterogeneous trajectory graph encoding vector set to obtain a high-dimensional grinding parameter candidate map; In this embodiment, the coarse-grained grinding parameter trajectory family is constructed into a nested subgraph structure based on the trajectory continuity and parameter change coupling characteristics. Each trajectory subgraph constructs an edge set based on the parameter change sequence, and only connecting edges with adjacent behavioral response strengths greater than 0.05 are included in the graph. Nodes represent parameter states at discrete time points, and the maximum number of nodes in a single graph is set to no more than 300. The resulting multi-subgraph set represents the behavioral response network under different parameter evolution paths. On this basis, a window sampling method is used to perform structural encoding on each subgraph, extracting its joint statistical features in the three domains of time evolution, intensity response, and frequency fluctuation, and forming a unified dimensional original encoding vector. To eliminate the uneven impact of different feature dimensions, a zero-mean unit variance normalization method is used for normalization: the sample mean and standard deviation of each feature dimension are first calculated, and then the original values ​​are mapped to standardized values ​​so that all feature dimensions obey a zero mean and 1 standard deviation distribution, thereby generating a unified scale encoding vector set. A behavioral bias vector constructed based on a historical operating condition model was introduced as an initial membership guide to guide the fuzzy membership process toward true behavioral discrimination characteristics. During the clustering process, the target number of clusters was set to 10, and the fuzzy weighting index was set to 2.1 to enhance the boundary fuzzy control capability of soft clustering. An iterative center adjustment mechanism was employed, with each update round limiting the maximum offset of the cluster center to no more than 0.05 to ensure a stable clustering convergence path. After clustering, the clustering results were analyzed for stability. The variance of the center vector of each cluster was calculated, and unstable clusters with variances greater than 0.08 were eliminated. Finally, the retained cluster centers were used as representatives of the grinding parameter semantic clusters. By tracking the attribution of each encoded vector to multiple centers, a multi-membership structure map was constructed. This map, based on multiple associations between vectors, presents a highly confident semantic space for grinding parameters.

[0014] Step S5: performing a multi-dimensional working condition simulation based on the high-dimensional grinding parameter candidate map, and performing parameter tensor loop optimization based on the multi-dimensional working condition simulation results to obtain a stable grinding parameter set; In this example, a high-dimensional grinding parameter candidate map was imported into a multidimensional working condition response simulation framework. Using a library of typical working conditions (e.g., silicon carbide workpieces, high-strength steel workpieces, and ceramic materials) from a fully automatic roughing and finishing grinding machine, and setting a material matching factor of 0.9, the grinding response under different parameter combinations was simulated and reconstructed, outputting four metrics: grinding efficiency, heat load, energy consumption, and surface roughness. A tensor structure was constructed for the simulation output, with the response error serving as the loop update condition and an error convergence threshold set to 0.02. Three rounds of feedback optimization were performed to ultimately generate a stable grinding parameter set, and its feasibility and consistency were verified in a real-world environment.

[0015] Step S6: The stable grinding parameter set is transmitted to the management platform of the fully automatic rough and fine grinding machine to execute the control parameters, and iterative precision compensation mapping is performed according to the actual grinding deviation feedback data obtained in real time to obtain the optimized grinding parameter set.

[0016] In this embodiment, a stable grinding parameter set is transmitted via an industrial Ethernet interface to the equipment management controller of a fully automatic roughing and finishing grinding machine. The controller then coordinates the equipment to execute the grinding task. The equipment collects real-time data on execution current, voltage fluctuations, machining accuracy, and thermal load, and calculates the deviation from the preset operating conditions. A deviation feedback iteration cycle is set to 5 minutes. Each cycle identifies the type and magnitude of deviation in the current parameter execution, and fine-grained adjustments are made through a bidirectional parameter mapping mechanism. This outputs an optimized grinding parameter set after deviation compensation, thereby continuously improving equipment operational stability and machining quality consistency. The bidirectional parameter mapping mechanism is a parameter update strategy that combines forward and reverse adjustment paths. Its core purpose is to accurately compensate for grinding parameter deviations caused by operating condition fluctuations while maintaining the stability of the equipment's current operating state. Specifically, it involves two data loops: 1) Forward mapping, starting from the original stable parameter set, the controller maps these parameters into executable machine control instructions based on the preset operating condition model and grinding task requirements. This process corresponds to the path by which parameters are "distributed" or "driven" to the equipment. 2) Reverse mapping is to collect actual feedback data (such as current, voltage, processing deviation, thermal load, etc.) during equipment operation, determine the deviation between this feedback and the desired target, and then infer the direction and amplitude of the current parameter adjustment based on the deviation type and amplitude. This is a feedback-based adaptive compensation path.

[0017] Optionally, step S1 specifically includes: Step S11: Acquiring multimodal data of equipment operation, wherein the multimodal data of equipment operation includes a grinding parameter group, a high-frequency vibration signal, a spindle current fluctuation record, a temperature rise thermal flow field image, and an accompanying acoustic spectrum signal; In this embodiment, a multi-type sensor array deployed on a fully automatic roughing and finishing grinding machine captures multimodal data during its operation. This data includes a grinding parameter set (covering grinding depth, linear speed, and feed rate), a three-axis acceleration vibration signal with a high-speed sampling frequency of 20kHz, a real-time current fluctuation curve of the spindle drive system, a dynamic image sequence of the temperature rise and thermal flow field acquired using infrared thermal imaging (with a frame rate of no less than 30 frames per second), and an acoustic spectrum signal in the 0–10kHz range acquired by a coupled acoustic sensor. All acquisition channels are synchronized with unified timestamps via an embedded data acquisition unit to ensure temporal consistency. The data is then encoded and stored in a unified data packet format for subsequent decoupling and analysis.

[0018] Step S12: performing channel-by-channel decoupling processing on the multimodal data of the device operation, and performing physical field disturbance trend identification on the decoupling processing results to obtain physical field disturbance trend data; In this embodiment, the acquired multimodal data is divided based on the data channel and decoupled in the physical domain. The vibration signal channel uses short-time Fourier transform to extract the excitation energy within the frequency band; the current signal channel uses a sliding window method to calculate the root mean square of the current fluctuation amplitude; the thermal flow image channel uses the rate of change of the local temperature gradient as the time domain feature; and the spectrum channel obtains the sound pressure power value of each frequency band through 1 / 3 octave decomposition. Subsequently, on each channel feature sequence, the position points where the continuous change gradient in the time series is greater than the threshold (the threshold is set at 0.15g / s in the vibration domain, 0.12A / s in the current domain, and 1.2°C / s in the thermal flow domain) are determined as the key points of the disturbance trend, and after integration, cross-channel physical field disturbance trend data is formed.

[0019] Step S13: extracting implicit physical disturbance segments based on the physical field disturbance trend data to obtain implicit physical disturbance primary annotation data; In this embodiment, based on the constructed physical field disturbance trend data, the disturbance threshold criteria are set in combination with the characteristics of different physical channels, and the implicit physical disturbance segment extraction is carried out in sequence. Specifically, the interval in which the temperature rise change rate is greater than 1.5°C / s for more than three consecutive time points and the vibration change amplitude reaches 0.2g is judged as a potential thermal-mechanical coupling disturbance; if the current peak change rate exceeds 0.15A / s and the difference in the mean square increment of sound pressure in two adjacent windows exceeds 10%, it is further marked as a composite disturbance segment. All identified disturbance segments are recorded according to the start and end time, channel attributes and disturbance intensity, and the disturbance level is marked to form a primary disturbance annotation data set.

[0020] Step S14: Thread the implicit physical disturbance primary annotation data, construct time segment indexes for the heat flow, vibration, current, and sound spectrum signal segments respectively, and set the aggregation window step size to 0.5s to perform thread merging operations to obtain a disturbance event structure vector set; In this embodiment, the extracted primary disturbance annotation data is thread-divided according to the data channel type, and time segment indexes are established for the thermal flow image frame sequence, vibration acceleration record, current fluctuation record and sound spectrum density function curve respectively. Each type of segment is intercepted as a segment window according to the range of 1.5 seconds before and after the time point of the disturbance, and the aggregation window step is set to 0.5 seconds to perform the thread merging operation. During the thread merging process, if multiple channels have interference intervals in the same time period, the one with the largest disturbance level is taken as the main event, and a composite structure is established with the help of other channel segments, and finally a disturbance event set in a unified structural vector format is output. Each disturbance event structure contains the disturbance source, the active channel, the maximum change amplitude, the duration and the channel coordination mark.

[0021] Step S15: perform structural consistency encoding on the disturbance event structure vector set, and fuse the acquired equipment operation timestamp, working condition identification and location index information to obtain structured grinding field perception data.

[0022] In this embodiment, the constructed disturbance event structure vector set is processed for structural consistency. Each vector is first sorted and classified by channel sequence and event intensity. Then, using a template nesting approach, the vectors are unified into a coded form containing six fixed fields: channel type identifier, disturbance start time, disturbance duration, maximum disturbance intensity value, impact level score, and interference frequency range. The processing task number, machine working part number, and a unified timestamp synchronized with the main control PLC system are simultaneously obtained from the equipment operation control system at the time of the event. These three information are integrated and embedded into the vector structure, completing the structural reconstruction of the data at the temporal, spatial, and semantic levels. Ultimately, structured grinding field perception data is generated for subsequent behavioral analysis and grinding behavior feature modeling.

[0023] Optionally, step S13 is specifically as follows: Step S131: setting the window length to 2s and the step length to 0.5s, performing segmented sliding window segmentation on the physical field disturbance trend data to obtain sliding segment disturbance trend data; In this embodiment, the sliding window length is set to 2 seconds and the step size is 0.5 seconds, and the physical field disturbance trend data obtained during the operation of the equipment are segmented. Each sliding window scrolls on the time axis at intervals of 0.5 seconds, and the data within the corresponding 2-second window is extracted each time it scrolls. Specifically, multi-channel data such as heat flow, vibration, current and sound spectrum are distinguished by synchronized time stamps. The data of each sliding segment will be analyzed as independent disturbance trend data. These data contain the changes in physical quantities and corresponding disturbance performance of each channel during the time period, ensuring that further disturbance analysis and feature extraction can be performed for each segment in subsequent steps.

[0024] Step S132: Calculate the disturbance intensity differential value for each sliding segment in the sliding segment disturbance trend data, and set the disturbance intensity mutation threshold to 1.75 to identify the initial segment of potential disturbance anomaly, thereby generating a disturbance segment preliminary screening data set; In this embodiment, in each sliding segment of the obtained sliding segment disturbance trend data, the disturbance intensity of each channel is differentially calculated, with special attention paid to its rate of change. The disturbance intensity mutation threshold is set to 1.75, and the disturbance intensity fluctuation of each segment is detected based on this standard. Once the disturbance intensity change mutation in the segment exceeds the threshold, it is considered that there is potential abnormal disturbance behavior in the segment and it is marked as a preliminary abnormal fragment. In specific implementation, by calculating the rate of change of the disturbance intensity in each time step (for example, the rate of change of the current amplitude and the temperature gradient), once it exceeds the set threshold, it is considered that a significant disturbance change has occurred in the area. All segments that meet this condition will be compiled into a preliminary screening data set for subsequent disturbance behavior analysis and pattern recognition.

[0025] Step S133: Group the disturbance segment preliminary screening data set by modal channel type, construct vibration frequency-energy diagrams, current fluctuation spectrum diagrams, heat flux intensity spectrum diagrams, and acoustic spectrum density diagrams, and perform frequency domain entropy increment analysis on each group of spectrums to screen segments that meet the preset multi-channel synchronous entropy fluctuation conditions and generate candidate multi-channel coupled disturbance segments; In this embodiment, the screened disturbance segments are further subdivided into different channel types for grouping. For vibration signals, a vibration frequency-energy diagram is constructed, and the energy distribution of each frequency segment is analyzed using FFT; for current fluctuations, a current fluctuation spectrum is constructed to evaluate the frequency components of current fluctuations; for heat flow data, a heat flow intensity spectrum is generated to reflect the intensity of local temperature changes; for sound spectrum data, a sound spectrum density map is constructed to measure the distribution of sound wave signals in different frequency bands. Then, a frequency domain entropy increment analysis is performed on each spectrum to calculate the entropy fluctuations in each segment. The preset multi-channel synchronous entropy fluctuation condition is set as follows: when the entropy changes of multiple channels exceed a specific fluctuation threshold at the same time, it is considered that the segment has significant disturbance behavior. All segments that meet this condition are summarized as multi-channel coupled disturbance candidate segments for further analysis.

[0026] Step S134: performing co-occurrence pattern recognition on the candidate segments of multi-channel coupled disturbance, screening out isolated segments, and setting the modal consistency evaluation index to be greater than 0.6 to select segments with high synergy of interference components as primary annotated segments of implicit physical disturbance; In this embodiment, co-occurrence pattern recognition is performed based on the obtained multi-channel coupled disturbance candidate segments. By analyzing the temporal synchronization of disturbance events in different channels, segments with highly coordinated interference are identified, and segments without sufficient coordination or isolated disturbances are eliminated. In order to select segments with high coordination of interference components, the modal consistency evaluation index is set to be greater than 0.6, that is, only segments with highly consistent disturbance characteristics of vibration, heat flow, current and spectral signals are identified as implicit physical disturbances. Here, the consistency evaluation is completed by calculating the similarity between the signals of each channel. All qualified segments will be marked as highly coordinated disturbance segments and further used as primary annotation data for implicit physical disturbances.

[0027] Step S135: Aggregate the implicit physical disturbance primary annotation segments to obtain implicit physical disturbance primary annotation data.

[0028] In this example, the implicit physical disturbance segments marked as highly cooperative disturbances are chronologically aggregated. In practice, these segments are aggregated into a disturbance event set, with each disturbance event including information such as its start time, duration, interference intensity, and channel cooperation. Then, based on the physical characteristics and cooperative properties of each disturbance event, all segments are summarized into primary annotated data for implicit physical disturbances. This dataset serves as the basis for subsequent analysis and optimization, enabling further disturbance pattern recognition, operating condition prediction, and process control.

[0029] Optionally, step S14 is specifically as follows: Step S141: grouping and labeling the implicit physical disturbance primary annotation data according to the modal channel type to obtain thermal flow field segments, vibration frequency band segments, current disturbance segments, and sound spectrum fluctuation segments; In this embodiment, data is grouped according to the modal channel type in the implicit physical disturbance primary annotation data. Specifically, all disturbance segment data are classified according to the modal signal (heat flow, vibration, current, sound spectrum) from which they come. The thermal flow field segment includes the temperature change data recorded during the operation of the equipment, the vibration frequency band segment covers the vibration signal generated by the equipment, the current disturbance segment involves the current fluctuation of the equipment, and the sound spectrum fluctuation segment refers to the change in the sound wave spectrum generated when the equipment is running. In this way, the data of each modal signal is divided into independent groups, which facilitates the subsequent separate analysis and processing of each channel. During specific operations, a dedicated processing channel is set for each type of data to facilitate subsequent parallel processing and synchronization.

[0030] Step S142: assigning the thermal flow field segment, vibration frequency band segment, current disturbance segment, and sound spectrum fluctuation segment to corresponding preset thread pools, and activating four types of parallel processing threads to generate an initial channel set for a multi-threaded data stream; In this embodiment, for the thermal flow field segments, vibration frequency band segments, current disturbance segments, and acoustic spectrum fluctuation segments obtained in step S141, each type of data is assigned to a corresponding preset thread pool. Each thread pool contains multiple worker threads, and the number of worker threads in each thread pool is set to four based on actual needs. For example, the thermal flow signal is processed in parallel by a thread pool containing four threads, the vibration signal is processed by another thread pool containing four threads, and so on. During initialization, each thread pool specifies a maximum number of threads of four, and its task queue uses a blocking queue (such as a LinkedBlockingQueue) to ensure orderly execution of tasks and prevent resource overload caused by excessive task accumulation. To avoid thread contention, each thread in the thread pool is responsible for processing a specific subset of data, and data is distributed to different threads using a predefined strategy. Each data type is processed by a separate thread pool, which are activated and run in parallel. Specifically, the thermal flow signal is processed by one thread pool, the vibration signal by another thread pool, the current signal by a third thread pool, and the acoustic spectrum signal by a fourth thread pool. Each worker thread in the thread pool processes its assigned data fragments in parallel, generating the initial set of channels for the multi-threaded data flow. The thread pool is sized to handle four threads per signal type to ensure efficient concurrent processing and avoid resource contention.

[0031] Step S143: Construct a standard time segment index for the disturbed segment data in each thread of the initial channel set of the multi-threaded data stream, set the single index time window to 0.5s, and perform sliding alignment on the disturbed segment data in each thread, eliminating segments with cross-channel time drift greater than 200ms to generate a synchronized time index segment set; In this embodiment, the disturbance segment data in each thread in the initial channel set of the multi-threaded data stream is subjected to standard time segment indexing processing. Specifically, the data in each thread will be segmented within a time window of 0.5 seconds to generate a time segment index. The length of each sliding window is 0.5 seconds, and the sliding step is 0.5 seconds, ensuring that the data in each time period is fully analyzed and synchronized. When aligning the time segments, if it is found that the time drift across channels exceeds 200 milliseconds, the data is considered inconsistent and needs to be eliminated. This is intended to ensure that the signals of different channels can be accurately synchronized, thereby ensuring the accuracy and reliability of subsequent analysis. Finally, after this processing step, a synchronized time index segment set is obtained.

[0032] Step S144: setting the aggregation window step size to 0.5s, performing thread aggregation operations on the synchronized time index segment set within each aggregation window, extracting the cross-channel collaborative perturbation structure, and generating a perturbation thread fusion node set; In this embodiment, the aggregation window step size is set to 0.5 seconds to ensure sufficient temporal overlap between the disturbance segments within each time window. Within each aggregation window, all synchronized time segment data from different channels is aggregated, and cross-channel collaborative disturbance extraction begins. Cross-modal feature correlation analysis techniques are employed during thread aggregation. Within each aggregation window, the disturbance data in the synchronized time segments is first analyzed in the time and frequency domains for each channel to extract key features in time, frequency, and amplitude, such as the frequency content of the vibration signal and the fluctuation amplitude of the current signal. Subsequently, collaborative analysis techniques are employed to identify cross-channel disturbance structures by calculating the similarity and correlation between different modal signals. For example, methods such as the Pearson correlation coefficient or mutual information can be used to measure the degree of correlation between different signal channels. If, within a given time window, the correlation between the vibration signal and the current signal is high and their disturbance intensities both reach a preset threshold, cross-channel collaborative disturbance is considered to exist within that time period. After the collaborative disturbance analysis is completed, the cross-channel collaborative disturbance structure is extracted based on the collaborative relationships between the different channels. Using multimodal coupling analysis, the disturbance features of different signals are fused within the same time window to form a fusion node of a multimodal disturbance structure. The fusion node is marked according to features such as signal correlation and synergy strength, and a set of disturbance thread fusion nodes is generated. In order to further ensure that the extracted collaborative disturbance structure has a high degree of credibility, each structural node is screened during the extraction process. Specifically, screening conditions are set. For example, the disturbance structure is only considered valid when the synergy degree of cross-channel disturbance is greater than 0.6. At the same time, the threshold requirements of the disturbance intensity must be met to ensure that only disturbance nodes with high influence are retained.

[0033] Step S145: Serialize and integrate the disturbance thread fusion node set, and assign a unique structure label to each disturbance thread fusion node to obtain a disturbance event structure vector set.

[0034] In this embodiment, the obtained disturbance thread fusion node set is serialized and integrated, and a unique structural label is assigned to each disturbance thread fusion node. These fusion nodes are representations of the collaborative disturbances of various modal signals during the operation of the equipment, reflecting the relationship and influence between different signals. During specific operations, each node will generate a unique label based on its position in the time series, disturbance intensity, cross-channel coordination and other characteristics. For example, different disturbance modes are marked by combining equipment status, operating condition information and disturbance intensity. Finally, all disturbance thread fusion nodes will be integrated into a disturbance event structure vector set for further analysis and decision support. Each disturbance event will contain its collaborative characteristics under multimodal signals and provide structured data support for subsequent prediction and optimization.

[0035] Optionally, the multimodal grinding behavior feature analysis in step S2 is specifically as follows: The structured grinding field perception data is expanded in the time domain, and the time series frames are synchronously reconstructed according to the grinding wheel linear speed in the grinding parameter group to obtain the grinding field perception time series frame set; In this embodiment, time domain expansion processing is performed based on the collected structured grinding field perception data. This process uses the "grinding wheel linear speed" in the grinding parameter group as the synchronization reference benchmark, and performs linear recalibration processing on the time axis of multimodal perception data such as heat flow, current, vibration, and sound spectrum. To ensure timing consistency, the sampling period is unified to 1 ms, and the data points with sampling errors greater than 5 ms are time-aligned by interpolation correction. Subsequently, the continuous processing sections in the grinding process are segmented and intercepted, and the effective perception sequence dominated by the grinding wheel linear speed in each section is extracted. Based on the synchronous frame reconstruction method, the different modal data are organized into a unified frame structure to generate a grinding field perception time frame set with continuous time sequence numbers and consistent data format.

[0036] Acquire abrasive structure data, and perform grinding ratio change trend analysis based on the abrasive structure data and the grinding parameter group to obtain cutting area grinding ratio change trend data; In this embodiment, the process of obtaining abrasive structure data includes reading the abrasive parameter file preset in the system, and simultaneously parsing the currently used grinding wheel brand and abrasive grain size grade. The abrasive structure data must include the abrasive morphology factor and the abrasive grain arrangement density (unit: g / cm³). This data is combined with the real-time grinding parameter group for analysis, and the grinding ratio (G ratio) change trend of the grinding wheel in the cutting section is calculated using a timed rolling window method (window length: 10 seconds, step length: 2 seconds). In the trend analysis, the grinding depth threshold is set to 0.03mm, and the cutting speed fluctuation rate is limited to within the range of ±5% to eliminate error interference under abnormal working conditions. The final output result is a trend curve of the evolution of the grinding ratio of the cutting area over time, which is used to indicate the abrasive consumption stage and the characteristics of the abrasive grain state change.

[0037] The grinding field perception time series frame set is combined with the grinding ratio change trend data of the cutting area to perform multimodal time synchronization alignment of heat flow, current, vibration and sound spectrum to generate a modal reconstruction sequence group; In this embodiment, the grinding field perception time series frame set is associated and matched with the extracted cutting area grinding ratio change trend data. The synchronous alignment operation is based on the grinding ratio mutation point as the anchor point, combined with the peak response in the same time period in the multimodal perception signal, to complete the reconstruction of the main time axis of the data frame. The maximum alignment offset is set to ±20ms, and the data segments that do not meet the alignment conditions are smoothly interpolated. The aligned modal data form a unified modal reconstruction structure in each frame, and are organized into modal reconstruction sequence groups according to the time number. The sequence group has fields such as heat flux density, instantaneous current value, vibration frequency band power spectrum density and sound spectrum energy distribution, which provide a time series basis with consistent resolution for subsequent disturbance identification.

[0038] The contact arc disturbance feature is identified for the modal reconstruction sequence group, and the transient components and energy mutation points of the modal disturbance are extracted; In this embodiment, a contact arc zone disturbance identification process is performed based on the modal reconstruction sequence group. Initially, candidate disturbances are identified based on a heat flux change rate greater than 40W / cm² / s, a current surge rate exceeding 1.5A / ms, and an instantaneous power increase exceeding 8dB in the 2-5kHz frequency band of the acoustic spectrum. Based on this, transient components that meet the disturbance characteristic criteria are extracted from each modal signal, and energy mutation points are marked using wavelet analysis.

[0039] The disturbance significance threshold is set to 0.85, and the characteristics of abrasive particle breakage, shedding, and local burning during the grinding process are identified based on transient components and energy mutation points, thereby forming a grinding modal response significance map. In this embodiment, for the identification of mutation points, the local energy change rate threshold is set to 150%, and the effective disturbance duration is stipulated to be no less than 5ms. The transient components of each modal data such as heat flow, current, vibration and sound spectrum in the modal reconstruction sequence group are extracted. With 10ms as the sliding time window, the instantaneous energy change rate curve of each modal signal is calculated, and the mutation segment that exceeds 150% of the benchmark and lasts for more than 5ms is marked as a benchmark based on its average change rate. Subsequently, the disturbance type is identified in combination with the typical response characteristics of each mode. 1) For abrasive crushing type disturbances, the following three conditions must be met at the same time: first, the power spectrum density in the 13kHz frequency band is detected to increase by more than 15dB in the vibration mode and last for no less than 7ms; second, there is a rapid current drop slope of more than -1.0A / ms in the current mode; third, a high-frequency response with an energy jump of more than 12dB appears in the 57kHz frequency band of the sound spectrum signal. If the above three items appear in conjunction, it is determined to be an abrasive crushing type disturbance frame. 2) The identification criteria for abrasive shedding type disturbances include: the instantaneous energy increase in the 12kHz frequency band in the vibration mode is between 810dB and lasts for more than 10ms; the current decline slope is between -0.3 and -0.6A / ms; the heat flux density drops by more than 20% within 50ms, and the energy average of the 13kHz frequency band in the sound spectrum shows a 36dB background noise increase. Frames that meet the above conditions are classified as abrasive shedding type disturbance. 3) When identifying local burn type disturbances, heat flux data is used as the leading indicator. The judgment conditions are that the heat flux density surges to more than 80W / cm² within 30ms, the increase gradient reaches 2.5W / cm² / ms, and there are peaks of more than 1.8A in the current mode, and the second-order derivative of the spindle current is greater than 10A / s². The acoustic spectrum shows a high-frequency broadband surge of more than 10dB in the 68kHz band, accompanied by a sudden drop in energy in the low-frequency vibration band (0.5-1.5kHz). Frames meeting these combined conditions are labeled as localized burn-type disturbances. Finally, all identified disturbance frames are organized chronologically and labeled with their disturbance type and modal response intensity to generate a grinding modal response significance map.

[0040] Based on the structured grinding field sensing data, the cutting energy characteristics of each mode are extracted to obtain the cutting force fluctuation frequency, the grinding interval sound pressure surge rate and the second-order derivative of the spindle current. In this example, after extracting the significance map of the grinding modal response, the structured grinding field perception data was further subjected to feature extraction in the cutting energy dimension. Based on the aligned modal data, a 5-second sliding window (with a step size of 1 second) was used to calculate the cutting force fluctuation frequency (based on the difference between the mean and standard deviation of the vibration frequency, in Hz), the grinding interval sound pressure surge rate (in dB / s), and the second-order derivative of the spindle current (in A / s²) within that window. Periods with a cutting force fluctuation frequency greater than 60 Hz, a sound pressure surge rate exceeding 12 dB / s, and an absolute value of the second-order derivative of the current exceeding 20 A / s² were recorded as high-energy disturbance segments. These features constitute key indicators reflecting process stability and abrasive stress response, providing a high-precision energy description for further constructing an energy synergy structure.

[0041] The cross-modal energy synergy tensor is constructed by integrating the cutting force fluctuation frequency, the grinding zone acoustic pressure surge rate, and the second derivative of the spindle current, and combining it with the grinding modal response significance map. In this embodiment, the energy feature data of each dimension are fused to construct a unified cross-modal energy synergy tensor. The construction of this tensor is based on a time window length of 10 seconds, integrating three modal features: cutting force fluctuation frequency, grinding sound pressure surge rate, and spindle current second-order derivative. To ensure the numerical stability of the tensor, Z-score normalization is first performed, and the dynamic feature channel mutual information screening method is used to screen out low-contribution channels (channels with mutual information less than 0.1 will be discarded). After the tensor is constructed, it is divided into sub-tensor blocks with time continuity to facilitate behavior clustering and change boundary identification analysis. The tensor structure is expressed in three dimensions: time frame × modal feature × state value, which fully covers the timing feature fluctuations and synergistic response behavior of the entire grinding process.

[0042] Based on the cross-modal energy synergy tensor, behavior pattern aggregation and boundary identification are performed to obtain a set of sequential grinding behavior fragments. In this embodiment, the behavioral pattern aggregation and boundary identification tasks are carried out on the constructed cross-modal energy synergy tensor. The specific method includes dividing the tensor into sliding blocks of 5 seconds, and identifying the behavioral state boundary points by jointly modeling the relative amplitude and frequency of the changes in the eigenvalues ​​of each mode in each block. The behavioral change threshold is set to the amplitude of the eigenvalue change exceeding 2 times the standard deviation of the mean, and the number of mutation dimensions is not less than two modes, that is, it is judged as a state transition boundary. Continuous tensor segments without significant boundaries are defined as the same behavioral segment, and finally a complete set of time-series grinding behavior segments is output. Each behavioral segment has a clear start and end frame, a corresponding modal change trend and an energy fluctuation structure, which is convenient for subsequent behavioral pattern recognition and tracing.

[0043] The set of sequential grinding behavior segments is mapped to the pre-connected grinding condition behavior library, and the morphological similarity of the behavior sequences is calculated to obtain the grinding behavior pattern data.

[0044] In this embodiment, the identified set of sequential grinding behavior fragments is mapped to a pre-connected grinding condition behavior library. This behavior library presets structural feature templates for various standard working condition behaviors, including behavioral units such as grinding stabilization period, initial shedding of abrasive particles, and precursors to severe burns. During the mapping process, the morphological similarity with the behavior templates in the library is calculated one by one based on the energy fluctuation spectrum, modal collaborative structure, and state duration of the behavior fragment (calculated using DTW distance, and a similarity greater than 0.85 is considered a match). Finally, the standard working condition type matched by each behavior fragment is output, along with the corresponding similarity score, to form the grinding behavior pattern data. This data can be used for further grinding process backtracking, quality prediction, and parameter tuning.

[0045] It is particularly important to construct the cross-modal energy synergy tensor as follows: A multimodal energy characteristic matrix is ​​constructed based on the cutting force fluctuation frequency, the acoustic pressure surge rate in the grinding zone, and the second-order derivative of the spindle current. In this embodiment, the cutting force fluctuation frequency, the sound pressure surge rate in the grinding interval, and the second-order derivative of the spindle current are synchronized and aligned on the time axis and then normalized to form an energy feature matrix with a matrix structure of N×3, where N is the number of time sampling frames, the matrix rows represent time series segments, and the columns are three types of energy indicators: the cutting force fluctuation frequency, the sound pressure surge rate in the grinding interval, and the second-order derivative of the spindle current.

[0046] The multimodal energy feature matrix was screened for responsiveness based on the grinding modal response significance map, and energy features with corresponding significance weights lower than the significance threshold of 0.85 were eliminated to obtain the significant energy response set. In this embodiment, the above-mentioned energy feature matrix is ​​mapped one-to-one with the time axis in the significance map of the grinding modal response. In the specific operation, each row of feature vectors corresponds to the saliency map labeled frame according to its timestamp, and its maximum response value in the disturbance saliency layer is extracted. The significance weight is normalized in the range of 0 to 1, and the response threshold is set to 0.85. All energy feature rows below this threshold are eliminated and no longer participate in subsequent feature analysis. This processing method ensures that the filtered feature matrix is ​​composed only of energy fragments that are representative in the actual disturbance response, avoiding fuzzy interference in subsequent modeling due to low response sections. The matrix retained after screening is defined as a significant energy response set, the size of which is determined by the actual number of disturbances, usually accounting for 15% to 25% of the total number of sampling frames.

[0047] The modal coupling criterion is set to 0.7, and bilateral correlation analysis between modes is performed in the significant energy response set. The synergy weight matrix between different modal feature pairs is calculated, and the modal pairs with coupling greater than the modal coupling criterion are screened and retained to obtain high-coupling modal pairs. In this embodiment, in order to identify the coupling behavior characteristics between multiple modes, a bilateral correlation analysis is required between any two modal features in the significant energy response set. Specifically, the cutting force fluctuation frequency and the sound pressure surge rate, the sound pressure surge rate and the second-order derivative of the spindle current, and the cutting force fluctuation frequency and the second-order derivative of the spindle current are respectively composed of modal feature pairs, and the Pearson correlation coefficient is calculated for each pair of features within a time window length of 200ms, and the time window overlap rate is set to 50%. The calculation results form a 3×1 coupling coefficient vector, which represents the average coupling strength between the three pairs of modes. The modal coupling criterion is set to 0.7. If the coupling strength is greater than the threshold, the modal combination is retained as a high-coupling modal pair. This analysis process can effectively identify the feature combination with a synchronous and coordinated change trend between multiple modes during the grinding disturbance process, providing a reliable modal structure foundation for subsequent tensor construction.

[0048] A third-order tensor structure is constructed based on highly coupled modal pairs to obtain a cross-modal energy synergy tensor.

[0049] In this embodiment, based on the screened high-coupling mode pairs, a cross-modal energy synergy tensor is constructed to express its synergy characteristics. The construction method is to use the time step as the first dimension, the mode pair combination as the second dimension, and the coupling strength, synergy energy jump amplitude and transient peak delay between each pair of modes as the third dimension. The overall structure is a three-order tensor structure in the form of T×P×F, where T is the time series length, P is the number of high-coupling mode pairs, and F is the number of synergy features of each pair of modes. In the specific implementation, for each high-coupling mode pair, three types of indicators are extracted: joint peak alignment error, energy growth rate, and resonant frequency band synchronization ratio. All features are written into the tensor after being processed in a standardized interval. The tensor structure finally constructed can be used as an input carrier for subsequent behavior aggregation and grinding condition identification, support the process of multi-dimensional pattern recognition, and provide a structural basis for cross-modal coupling for the analysis of grinding behavior mechanisms.

[0050] Optionally, the semantic abstraction of the grinding parameters in step S2 is specifically: The grinding behavior pattern data were preliminarily screened for behavior-parameter mapping, and a potential correlation matrix between behavior segments and grinding parameters was constructed; In this embodiment, during the initial screening of the grinding behavior-parameter mapping, the behavioral pattern data obtained through perturbation analysis and energy coupling modeling is first decomposed into a set of behavioral segments by time segment, and the processing parameter records within the corresponding time period are extracted. Processing parameters include but are not limited to feed rate, spindle speed, coolant flow rate, grinding wheel linear speed, and grinding wheel dressing interval. For each behavioral segment, a sliding window method is used to calculate the variation range and trend of the corresponding parameter dimension. The parameter distribution is expressed using a five-number summary method (minimum value, lower quartile, median, upper quartile, and maximum value). When constructing the latent association matrix, the behavioral segments are divided into three categories according to type: perturbation type, overload type, and stable type. The co-occurrence frequency, interval overlap, and trend alignment between each behavior class and parameter group are used as the three factors of the matrix elements. Finally, a three-dimensional latent association matrix of size M×N×3 is constructed, where M is the number of behavioral segment classes, N is the number of processing parameters, and 3 is the characteristic factor dimension.

[0051] The local sensitivity threshold was set to 0.75, and the maximum mutual information criterion analysis was performed on the potential correlation matrix to identify the significant behavioral response areas, thereby generating a parameter candidate cross-table; In this embodiment, in the identification link of significant behavioral response areas, the above-mentioned potential correlation matrix needs to be subjected to mutual information quantification processing to determine the correlation strength between each behavior type and different parameters in the local segment. During the operation, the local sensitivity threshold is set to 0.75, and for each parameter and behavior segment combination in the matrix, the mutual information value is calculated by sliding within a time window of 0.5s, and the mutual information value is normalized. All combinations greater than the set threshold are marked as significant response areas, and candidate association pairs are established. In order to improve the screening accuracy, the candidate combinations are filtered in combination with the disturbance density index, and only high-intensity parameter response pairs that appear simultaneously in more than two behavior segments are retained, and they are organized into a parameter candidate cross-table. This cross-table serves as the initial structure for subsequent parameter semantic modeling, in which each row corresponds to a behavior pattern and each column corresponds to a potential response parameter.

[0052] The dynamic characteristics of parameters related to grinding behavior are extracted based on the parameter candidate cross table, and the parameter information density distribution of the dynamic characteristics of parameters related to grinding behavior is evaluated to obtain the semantic weight map of grinding parameters; In this embodiment, based on the parameter candidate cross-table, in order to further refine the behavioral association characteristics of the parameters, the dynamic changes of the parameters contained therein in all significant behavioral segments are trend-classified. Taking the spindle speed as an example, the standard deviation statistics of its amplitude changes in the contact instability segment are performed, and its change amplitude and average gradient in each disturbance cycle are extracted, and its average information density is calculated. The information density is defined as the number of effective value jumps per unit time. Repeat this process to complete the dynamic trend feature extraction of all parameters. Subsequently, a three-factor semantic scoring system is constructed using the frequency of parameters appearing in different behavioral patterns, the information density peak and the distribution diffusion. The scoring results are aggregated to form a semantic weight map of grinding parameters. The map is presented in the form of a heat map, with the X-axis being the behavior type, the Y-axis being the parameter type, and the heat value representing its representative intensity at the semantic layer.

[0053] Perform behavioral semantic projection on the parameter abnormality patterns that frequently appear in the grinding parameter semantic weight map, and transform the parameter abnormality patterns into semantic behavioral labels including contact instability segments, local overload segments, and thermal drift risk segments. In this embodiment, in order to land the key behavior semantics on identifiable labels, it is necessary to perform behavioral semantic projection on the parameter anomaly patterns with high weights in the semantic weight map. In the specific operation, parameter behavior combinations with semantic strength values ​​more than twice the global mean of the map are selected to extract their dynamic profiles under different behavior types. Taking the sudden drop in grinding wheel linear speed as an example, if it appears synchronously with the sharp increase in sound pressure and the fluctuation of spindle current in multiple disturbance segments, and the co-occurrence rate is higher than 80%, it will be marked as the semantic feature of "contact instability segment". Similarly, if the slope of the spindle current continues to rise and the heat flux mutation signal appears, it is classified as a "local overload segment"; if the temperature curve drifts after multiple cycles of processing and is superimposed on the slow rise of other parameters, it is marked as a "thermal drift risk segment". After all behavior-parameter combinations are semantically named, a semantic behavior label set is formed, which serves as the core basis for the next step of parameter intent analysis.

[0054] Construct a multi-nested index structure between semantic behavior labels and grinding parameter groups to generate grinding parameter behavior intention mapping clusters; In this embodiment, when constructing the mapping structure between semantic behavior labels and parameter groups, it is necessary to establish a clear hierarchical relationship based on the process significance and risk level embodied by the semantic behavior labels. First, a first-level classification is performed according to the behavior type, such as "contact instability", "local overload", "thermal drift", etc., and each category is assigned a specific label, such as "micro-impact instability", "short-cycle load peak", "thermal accumulation drift", etc., as the index primary key. For each label, the grinding process parameter data associated with the traceable behavior fragment is used to extract the parameter combination with the highest frequency of occurrence in multiple samples of the label to preliminarily form a parameter list. In order to enhance the semantic expression ability of the list, the parameters need to be functionally divided. The judgment criteria are set as follows: if the disturbance change of the parameter significantly precedes within 5 seconds before the start of the semantic behavior label, it is defined as a core parameter; if it fluctuates closely following the core parameter during the behavior period, it is defined as a secondary parameter; if the parameter change is relatively stable but its setting affects the fluctuation amplitude of the core or secondary parameter, it is defined as a background parameter. For example, in the "local overload segment" tag, the spindle current peak typically occurs 1-2 seconds before the start of the behavior segment, making it a core parameter. The grinding wheel dressing interval significantly affects the current fluctuation rhythm, but the change lags slightly, making it a secondary parameter. The coolant flow rate is stable, but the adjustment level affects the thermal effect of the current, making it a background parameter. Once constructed, a three-level nested structure of behavior and parameters is formed, with each layer possessing parseable process implications. Using the behavior tag as the primary index key, the corresponding core, secondary, and background parameters are hierarchically aggregated to construct a parameter behavior intention mapping cluster. In actual process flows, this mapping cluster can be directly connected to the grinding process monitoring system to implement behavior recognition-driven parameter linkage adjustment. For example, when the "unstable contact" tag is triggered, the feed rate and grinding wheel linear speed can be prioritized according to the mapping cluster's recommendations as part of the response strategy.

[0055] The grinding parameter behavior intention mapping cluster is converted into a multi-dimensional nested structure to perform parameter dimension semantic compression and upper and lower encoding, and the grinding parameter intention nested mapping table is obtained.

[0056] In this embodiment, multiple mapping clusters are semantically hierarchically merged and encoded with hierarchical relationships. In this process, the parameter sets involved in all mapping clusters must first be vectorized and normalized, and the parameter values ​​must be linearly normalized according to the standard interval (such as [0,1]) to avoid the accuracy of structure construction being affected by parameter dimension differences. The mapping clusters are hierarchically divided according to the behavioral risk level reflected by the semantic label. The risk level assessment standard refers to the parameter response intensity, behavior duration and the impact of the behavior on grinding stability. For example, the "thermal drift" label is classified as low risk, "contact instability" is set to medium risk, and "local overload" is classified as medium-high risk. Based on this structure, semantic compression is performed on large-scale mapping clusters with the same behavioral attribution. The specific method is to merge the corresponding multiple behavioral labels into a unified semantic node, and extract the intersection of their core parameter sets to form a "high semantic density" compressed node to avoid semantic duplication. After completing the semantic merging, a multi-dimensional nested mapping table is constructed. The mapping table adopts a tree-like structure, with risk level as the first-level index node, and is divided into a semantic label layer, a parameter type layer (core / secondary / background), and a specific parameter item layer. Each path represents a mapping chain from high-level behavioral semantics to specific parameters. By introducing a nested structure, parameter commonality mining across semantic labels can be achieved. For example, under the "medium-risk behavior" node, both the "contact instability segment" and the "short-cycle instability segment" can call the two parameters of grinding wheel linear speed and spindle current, facilitating the merging of strategies and unification of processes. The final generated grinding parameter intention nested mapping table supports not only upward deduction (identifying risky behaviors from parameter fluctuations) but also downward analysis (deriving key monitoring parameters from risk warnings), greatly improving the engineering practicality and response efficiency of the semantic parameter system.

[0057] Optionally, step S3 specifically includes: Step S31: performing tensor expansion on the grinding parameter intention nested mapping table to construct an initial parameter tensor field; In this embodiment, to address the high-dimensional data processing requirements of the nested intent mapping table, a tensor expansion operation is performed on the nested intent mapping table for grinding parameters. Specifically, each "semantic behavior label-parameter combination path" in the mapping table is converted into a tensor element, with the semantic label encoding as the first dimension, the parameter category (core, secondary, background) as the second dimension, and the parameter value range as the third dimension. To enhance structural expressiveness, the parameter value range is divided into five levels using the quintile method, and the upper and lower bounds of the range are annotated. The tensor is designed as a three-dimensional tensor of the form L, C, V, where L is the number of semantic labels, C is the number of parameter categories, and V is the number of interval divisions for each parameter category. Taking the "local overload segment" label as an example, the corresponding tensor unit has a high-load range of 0.9 to 1.2 A for the spindle current, which is calibrated as the fifth level range. The resulting initial parameter tensor field has a unified data format, facilitating subsequent structural perturbation and path encoding processing.

[0058] Step S32: constructing a constraint space boundary set based on the upper and lower boundary combinations of each parameter dimension in the initial parameter tensor field, and performing non-uniform perturbations on the constraint space boundary set to obtain a coarse solution spatial structure grid; In this embodiment, to model the parameter optimization path, a constraint space boundary set is constructed by combining the upper and lower bounds of each parameter dimension based on the initial parameter tensor field. Specifically, the upper and lower bounds of each parameter in the tensor field under each semantic label are extracted, combined to form constraint boundary tuples, and then uniformly classified to form a multi-parameter boundary set. To improve the adaptability of the search space, a non-uniform perturbation strategy is further introduced. This strategy applies unequally spaced perturbation points within each boundary interval, with the perturbation point spacing set within a range of 3% to 10% of the original interval. A Gaussian offset (with the mean set to the interval center and the standard deviation set to 0.2 times the interval width) is used to control the perturbation point density. In the "thermal drift risk section," if the spindle temperature rise is set to 50°C ± 5°C, the generated perturbation points will cover a unequal density distribution between 46.5°C and 53.5°C. This method generates a coarse solution spatial structure grid, laying the foundation for the preliminary spatial structure of the subsequent parameter search.

[0059] Step S33: performing particle swarm spatial initialization on the rough solution spatial structure grid, and performing initial parameter path encoding on each particle in the particle swarm to obtain an initialized basic particle swarm; In this embodiment, after obtaining the coarse solution structure grid, the group structure in space needs to be initialized. The particle swarm initialization method can be used to construct the group starting point structure of the parameter behavior path. First, the number of group particles is set to P = 80, and each particle represents a set of parameter path combinations to be optimized. During initialization, each particle randomly selects a set of parameter upper and lower limit combinations from the coarse solution grid as the initial position, and performs a three-dimensional path encoding on it based on the tensor path structure, where the semantic label index, parameter category index and interval encoding are input as encoding vectors. To enhance the coverage of the search space, the initial distribution between particles is kept as uniform as possible in each dimension. The parameter initialization path length is fixed to 6 steps, and each step encoding combination contains 1 core parameter, 1 secondary parameter and 1 background parameter. The initial distribution state of this particle structure in space is the basic particle swarm, which has strong diversity and structural analysis capabilities.

[0060] Step S34: performing coarse-grained search evolution on the initialized basic particle group to generate a coarse-grained grinding parameter trajectory family.

[0061] In this embodiment, after completing the construction of the basic particle swarm, it is necessary to carry out preliminary path evolution exploration to obtain a coarse-grained trajectory family of grinding parameters. A coarse-grained search evolution operation is performed on the initialized group. The specific method is that in each round of iteration, each particle adjusts its forward direction according to the adaptability of its current parameter combination under the semantic behavior target. The adjustment rule consists of two parts: one is to evaluate the correlation strength of the current parameter combination based on the historical mapping weight of the semantic label, and the other is to adjust the parameter jump amplitude according to the path smoothness of the current combination in the tensor structure. In actual operation, the maximum parameter adjustment step size per round is set to 15% of the initial interval width, the number of search rounds is 50 rounds, and the particles with the top 20% fitness in each round are retained as the trajectory retention group, and the rest are randomly perturbed and regenerated. When "contact instability segment" is used as the target semantic label, most of the retained particles show a clear convergence trend in the feed rate and normal force direction, forming a stable coarse-grained parameter trajectory family, which provides a sample reference for the subsequent fine tuning stage.

[0062] Optionally, step S34 is specifically as follows: Step S341: mapping the initial parameter path of the initialized basic particle swarm to the corresponding constraint space, and calculating the position similarity of the particles in the constraint space, so as to assign the initial velocity and inertia weight to the initialized basic particle swarm and obtain the initial velocity particle swarm; In this embodiment, a structured mapping operation is required for the initial parameter path of the initialized basic particle group. Specifically, assuming that the total number of particles is N = 100, each particle is composed of a grinding parameter combination containing three dimensions: spindle speed, feed rate and grinding depth, which are set to randomly and uniformly generate initial values ​​in the intervals [5000, 8000] rpm, [100, 300] mm / min, and [0.01, 0.05] mm, respectively. To achieve structural constraint mapping, these three-element combinations need to be embedded in a predefined non-uniform perturbation constraint space (coarse solution space structure grid). The tensor structure uses nonlinear scaling in each dimension based on the perturbation density. For example, the spindle speed dimension is divided into logarithmic segmented intervals to increase the perturbation sensitivity in the medium and high intervals. After the mapping is completed, each particle obtains its position vector in the tensor space, expressed as a three-dimensional coordinate form. Taking a particle's parameters (6800 rpm, 200 mm / min, 0.03 mm) as an example, its coordinates in tensor space might be mapped to (0.62, 0.47, 0.55). Subsequently, a 100×100-dimensional position similarity matrix was constructed based on the Euclidean distance formula. A distance threshold was set: for any two particles, if their distance was less than 0.35 (the standardized similarity threshold of 0.65 is equivalent), they were grouped together; otherwise, they were labeled as exhibiting divergent behavior. Furthermore, for each particle, an initial velocity vector was calculated based on the average distance between it and its neighbors. The vector's magnitude was constrained to the interval [0.1, 0.9]. The velocity direction was determined by the distance gradient, with particles farther away from similar particles being assigned a higher initial velocity to encourage structural divergence. The inertia weight was initially set to 0.7, serving as a velocity continuation factor in subsequent jump updates. Ultimately, an initial velocity particle swarm with structural embedding information, velocity vectors, and inertia weights was obtained.

[0063] Step S342: using the constraint space to perform random jump updates on the initial velocity particle swarm to obtain a disturbance response particle swarm; In this embodiment, a jump-type perturbation update operation is performed on the initial velocity particle swarm based on the constrained spatial structure grid. The specific implementation method is as follows: based on the current velocity vector of the particle, a non-continuous jump perturbation is applied to its current position parameter combination. The perturbation scale is generated based on a random offset within 25% of the current dimensional interval width, and the perturbation direction is based on the velocity vector direction and superimposed with a Gaussian offset factor (the standard deviation is 0.2 times the perturbation scale). To control the rationality of the structure, the particle position after the jump must remain within the constraint boundary range. If it crosses the boundary, mirror reflection compensation processing is performed. Taking the spindle speed range of 5000~8000rpm as an example, if a particle jumps and updates to 8100rpm, it is readjusted to 7900rpm according to the boundary symmetry point. After this perturbation process, the particle swarm position completes the first jump-type dispersion in the search space, forming a perturbation-responsive particle swarm, providing diversified samples for subsequent fitness evaluation.

[0064] Step S343: Calculate the fitness value of the current solution for each particle in the disturbance response particle swarm; In this embodiment, a fitness evaluation is performed on each particle in the disturbance response particle swarm to obtain its response quality under the current semantic behavior goal. Taking "contact stability improvement" as the semantic behavior goal, relevant feature tensors are extracted from the structured perception data of the fully automatic coarse and fine grinding machine, such as three types of data: normal force fluctuation, spindle current change rate, and contact vibration frequency. The current parameter combination of each particle is mapped to these feature tensor dimensions and similarity matching is performed with historical behavior label data through a sliding window comparison method. The matching degree is calculated using dynamic time warping (DTW) distance weighted superposition evaluation, and the behavior matching score P_i∈[0,1] is output. Secondly, for the local feature sequence of the matching position, the standard deviation and mean change rate within its fluctuation range are evaluated to measure the stability of its behavioral response. The stability score S_i is also normalized to [0,1]. The final fitness value, F_i, is a weighted sum of the two: F_i = 0.6 × P_i + 0.4 × S_i. For example, if a particle's parameter combination has a matching degree of 0.88 and a stability score of 0.75, its fitness is: F = 0.6 × 0.88 + 0.4 × 0.75 = 0.832. To facilitate classification, a fitness grading standard is set: F ≥ 0.85 indicates a high-matching particle (significant response); 0.6 ≤ F < 0.85 indicates a moderately responsive particle; and F < 0.6 indicates a low-fitness particle (not recommended). All particle fitness values ​​are ultimately stored in a structured fitness matrix, which serves as an important basis for subsequent optimal solution screening and behavior trajectory clustering. It can also be used for process monitoring and performance feedback for parameter trajectory learning.

[0065] Step S344: According to the fitness value of the current solution, a local optimal solution and a global optimal solution are selected for each particle, and the particle parameters are collaboratively updated according to the local optimal solution and the global optimal solution to obtain an updated particle swarm; In this embodiment, after completing the fitness evaluation, the particle swarm needs to be updated based on the target guidance structure. The optimal parameter combination in the historical iteration can be found for each particle as the local optimal solution, and the particle with the highest overall fitness is selected from the current particle swarm as the global optimal solution. The update operation is combined and corrected based on the local and global guidance directions. The new position vector of each particle is composed of the current position superimposed with a set of weighted difference vectors: the local guidance item weight is set to 0.4, the global guidance item weight is set to 0.6, and the step factor range is [0.1, 0.3] and is randomly generated. Taking the spindle current as an example parameter, if the current value is 1.0A, the local optimal value is 1.2A, and the global optimal value is 1.4A, the new value is updated to: 1.0+0.4*(1.2–1.0)+0.6*(1.4–1.0)=1.32A. After the above process, a new round of particle swarm structure is obtained and prepared for subsequent iterations.

[0066] Step S345: Return the updated particle swarm to step S342, and perform 10 rounds of iterations, with steps S342 to S345 as one round of iteration. Perform group collaborative clustering based on the solution of each particle in the particle swarm and its fitness evaluation results, and group the particles to output the coarse-grained grinding parameter trajectory family.

[0067] In this embodiment, after the particle swarm completes an iterative update, a reflux operation is performed on it and multiple rounds of behavioral collaborative evolution are carried out. The updated particle swarm is returned to step S342 to perform 10 rounds of random jumps, fitness evaluation and solution updates. After each round, the historical optimal path and fitness value change trend of each particle are recorded. After the end of 10 rounds, a group collaborative clustering operation is performed on all particles based on their fitness performance and parameter path similarity. Clustering is based on a joint distance measurement of the particle encoding vector on the tensor path structure and its final fitness. The number of clusters is set to 5 categories, and the density distribution and semantic label similarity joint criterion is adopted. Finally, multiple coarse-grained parameter trajectory groups are formed, each group has a combination of consistent behavioral response characteristics and high fitness characteristics, which is suitable for subsequent working condition matching and control strategy generation tasks.

[0068] Optionally, the soft clustering bias learning in step S4 is specifically as follows: Perform feature domain normalization mapping on the heterogeneous trajectory map encoding vector set to generate unified scale parameter space data; In this embodiment, in order to unify the numerical scales of different trajectory map encoding vectors in each feature dimension, all encoding vectors need to be normalized in the feature domain. First, the maximum and minimum values ​​of each parameter dimension in the encoding vector set are extracted, and the Min-Max standard scaling formula is used to map all dimensions: normalized value = (original value - minimum value) / (maximum value - minimum value). In order to ensure the numerical stability of the normalization process, the minimum scaling interval width is set to 0.01. If the actual variation range of a parameter dimension is less than the threshold, the dimension is uniformly assigned a value of 0.5 after normalization to avoid weight imbalance due to dimensional flatness. In the processed unified scale parameter space, each dimension value of all encoding vectors is limited to the interval [0,1], thereby providing a consistency measurement basis for subsequent fuzzy clustering operations.

[0069] Based on the uniform scale parameter space data, biased fuzzy clustering is performed to obtain the biased fuzzy initial membership set; In this embodiment, biased fuzzy attribution modeling is performed within the normalized unified-scale parameter space, guided by the historical operating condition model, to construct an initial membership structure. In the specific implementation, the heterogeneous trajectory map encoding vector set is first normalized to ensure that all feature dimensions have zero mean and unit standard deviation distribution characteristics within the standard-scale space, eliminating numerical offsets caused by differences in feature dimensions. Subsequently, a historical operating condition model is constructed based on a grinding condition behavior library. A sequence of at least 500 typical operating condition samples with time labels is extracted. A sliding window behavior segmentation and trend fitting method are used to construct a set of behavior bias vectors to characterize the distribution tendency of historical operating condition evolution trends within the feature space. Finally, based on the unified-scale parameter space data, the number of fuzzy aggregation clusters is set to 10, the fuzzy weighting index is set to 2.1, the historical bias vector is introduced as a guide for fuzzy membership initialization, a Gaussian decay weighting function (bandwidth parameter σ = 0.25) is used to enhance the directional attribution weight, and the fuzzy membership distribution interval is set to [0.3, 0.95]. Each trajectory vector can be assigned to three candidate classes, with inter-class membership differences of no less than 0.2. A label preference factor of 0.15 is also introduced to enhance the appeal of behavior-related cluster centers. This processing results in a biased fuzzy initial membership set, where each vector has probability weights assigned to multiple fuzzy classes. This satisfies the requirements for fuzzy recognition of the multi-semantic features of grinding behavior and lays the foundation for subsequent center transfer modeling and graph fusion.

[0070] Iterative center transfer learning is performed based on the biased fuzzy initial membership set, and the center drift rate of each iteration is constrained to not exceed 0.05, and a set of class center transfer trajectories is obtained; In this embodiment, the class center convergence process is modeled for the initial membership set. The class center coordinates are updated by weighted averaging in each round of iteration, and the weight coefficient is provided by the degree of attribution matrix. In order to ensure the stability and convergence of the clustering process, it is stipulated that the drift rate of each round of class center update shall not exceed 0.05, and the drift rate is defined as the Euclidean distance between the centers of the previous and next two rounds divided by the maximum distance between the initial classes. If the drift rate of a class center exceeds the limit in a certain iteration, a buffer adjustment is performed through smooth interpolation, and the buffer factor is set to 0.5. The iterative process is executed for a maximum of 30 rounds, or terminated early when the drift rates of all class centers are lower than 0.01. Finally, the dynamic movement path of each class center is recorded, and a class center transfer trajectory set is constructed to provide a historical evolution information basis for subsequent stability judgment.

[0071] The dynamic stability of the class center transfer trajectory set is evaluated, and the classes whose center trajectory variance exceeds the stability threshold of 0.08 are eliminated. The remaining class centers are fitted and modeled to generate the convergent class center space; In this embodiment, the reliability of the class center needs to be judged based on its drift stability in the entire transfer trajectory. Based on the time series of each class center in each dimension, its change variance is calculated, and the stability threshold is set to 0.08. Classes exceeding this threshold are considered unstable classes. For the class center trajectories that are not eliminated, the trajectory curve fitting process is further performed, and the change path of the center in each dimension is modeled using a fifth-order polynomial fitting method. The fitting accuracy requirement requires that the determination coefficient R² must be higher than 0.93. If it is not met, a local fitting replacement is performed. Ultimately, a convergent class center space with stable structure and traceability is formed, providing a robust anchor point for subsequent affiliation back-deduction and weight assignment.

[0072] The multi-class membership distribution of each trajectory vector is calculated inversely based on the convergent class center space, and a multi-membership structure map is constructed. According to the multi-membership structure map, each encoding vector in the heterogeneous trajectory map encoding vector set is assigned a confidence scalar weight, and a high-confidence soft clustering structure of grinding parameters is obtained. In this embodiment, based on the convergent class center structure obtained by fitting, the original coding vector set is reversely calculated. The initial membership of each coding vector is calculated based on the inverse of its distance from each class center, and normalization is used to obtain a complete multi-class attribution probability distribution. The confidence weight distribution function is set as follows: confidence = max (membership distribution) × local density adjustment coefficient. The local density adjustment coefficient is set according to the number of neighbors of the vector within a radius of 0.05 around the class, and the range is 0.6~1.0. Finally, a confidence weight mark is added to each coding vector to obtain a soft clustering structure cluster with multi-class label support, in preparation for graph merging and semantic compression.

[0073] The high-confidence soft clustering structure of grinding parameters is subjected to similarity graph merging and nested graph compression to output a high-dimensional grinding parameter candidate graph.

[0074] In this embodiment, in order to aggregate the assigned coding vectors into a more compact graph expression, graph fusion processing needs to be performed based on the similarity between vectors. Taking cosine similarity as the core indicator, the graph merging threshold is set to 0.85, and vectors exceeding this threshold are aggregated into a subgraph structure. Furthermore, nested graph compression is performed on subgraphs with overlapping boundaries. The compression strategy adopts a two-stage operation of node merging and edge weight fusion, and the edge weight fusion adopts the maximum confidence selection principle. Finally, a high-dimensional grinding parameter candidate graph with a compact structure and clear semantics is output, which has the ability to support multiple scenarios such as working condition reasoning and strategy formulation.

[0075] What is particularly important is that biased fuzzy clustering is specifically: The feature domain normalization processing is performed on the heterogeneous trajectory map encoding vector set, and each dimension feature is mapped to a standard scale space with zero mean and unit variance to generate unified scale parameter space data; In this embodiment, in order to ensure the comparability of different feature dimensions in the heterogeneous trajectory map encoding vector set in the metric space, it is necessary to perform a standardized mapping process on each dimension of the feature. In the specific implementation process, the full sample mean and standard deviation of each dimension in the vector set are first calculated separately, and then a standardized transformation method is adopted to subtract the original value from the dimension mean and then divide it by the dimension standard deviation, so that each feature dimension presents a distribution state of zero mean and unit standard deviation in the standard scale space. In order to avoid numerical anomalies caused by the denominator approaching zero, the minimum standard deviation threshold is set to 0.02. If the actual standard deviation of a dimension is lower than this value, it is uniformly replaced with the minimum standard deviation threshold and then normalization is performed. After the above transformation, all vectors are mapped to a parameter space of uniform scale, providing basic scale support for subsequent fuzzy processing and behavior modeling processes, and eliminating the risk of numerical deviation caused by inconsistent scales of feature dimensions from different sources. The so-called standard scale space, in this embodiment, refers to a multidimensional vector space in which all dimensional features satisfy a distribution with a mean of zero and a standard deviation of one. Its main technical goal is to eliminate the clustering offset or weight imbalance problems caused by inconsistent numerical scales of different features, and to enhance the stability and adaptability of subsequent processing stages.

[0076] Perform historical working condition time series modeling based on the grinding working condition behavior library to obtain a historical working condition model; In this embodiment, in order to effectively integrate the historical grinding behavior characteristics, it is necessary to extract typical time series evolution samples based on the constructed grinding condition behavior library and perform behavior pattern modeling. In the specific implementation process, first, no less than 500 complete condition sequence samples are screened from the behavior library. The samples must contain three key fields: timestamp, corresponding vector features, and condition labels. A fixed time window sliding method is used to divide the behavior segments into segments every 10 time points in the sequence to form a set of time series segments of equal length. Subsequently, the evolution trends of the key behavior indicators in each time series are extracted by feature statistics, and these evolution trends are fitted into a behavior bias vector model. The bias vector length must be consistent with the trajectory encoding vector dimension, which is used to express the historical distribution tendency of the condition behavior in the feature space. Finally, this model serves as an important reference for the subsequent fuzzy attribution initialization, so that the fuzzy aggregation process has historical behavior guidance, effectively improving the recognition rate of weak sample behaviors in grinding scenarios.

[0077] Based on the unified scale parameter space data, the number of clusters of biased fuzzy clustering is set to 10, the fuzzy weighting index is set to 2.1, and the bias vector of the historical operating condition model is introduced as the membership initialization guide to perform membership initialization, thereby generating a biased fuzzy initial membership set.

[0078] In this embodiment, in order to achieve vector aggregation processing with behavioral orientation in a unified scale parameter space, it is necessary to perform bias-guided fuzzy attribution initialization in combination with the historical operating condition model. In the specific implementation, the number of clusters for fuzzy processing is first set to 10, and the fuzzy weighting index is selected as 2.1 based on experience to control the fuzziness of the membership distribution. At the same time, the constructed historical operating condition model bias vector set is introduced as the initialized behavioral guidance weight, and each bias vector is used as the initial center direction to enhance the attraction effect of the corresponding cluster in the initial stage of attribution allocation. The initial membership matrix is ​​constructed by weighting the distance between the sample point and the bias vector. The distance weight function is set to a Gaussian decay function, and the bandwidth parameter σ is set to 0.25 to enhance the center attraction ability of the bias class. After completing the above initialization, a biased fuzzy initial membership set covering the entire sample set is formed, so that each trajectory encoding vector obtains a set of fuzzy attribution probabilities associated with the historical behavior model, laying a fuzzy structure foundation for subsequent convergence path learning and structure compression.

[0079] Optionally, step S5 is specifically: Step S51: extracting the particle size parameter, cutting depth, linear speed, contact arc length and grinding ratio index corresponding to each candidate parameter structure from the high-dimensional grinding parameter candidate map, and constructing an initial working condition state tensor; In this embodiment, the basic working condition elements associated with each parameter structure need to be extracted one by one from the constructed high-dimensional grinding parameter candidate map, specifically including the particle size parameter (unit: μm), cutting depth (unit: mm), linear speed (unit: m / s), contact arc length (unit: mm), and grinding ratio (dimensionless). During the extraction process, the extraction batch processing unit is set to process 20 groups of candidate parameter structures at a time to ensure parallel construction efficiency. In each group of candidate structures, the five indicators are integrated into a five-dimensional state vector in a fixed order, and the vectors are sequentially superimposed to form a three-dimensional tensor structure. The dimension is defined as [N×5×1], where N is the number of candidate parameters.

[0080] Step S52: Perform multi-physics field simulation based on the initial working condition state tensor, perform 12S continuous simulation for each set of parameters, and evaluate the physical indicators of grinding ratio, surface roughness, heat-affected zone width and grinding energy consumption respectively to obtain working condition simulation response data; In this embodiment, based on the constructed initial working condition state tensor, a virtual simulation evaluation operation of candidate parameters is performed by relying on a simulation environment with thermal-mechanical-grinding multi-field coupling capabilities. The simulation time of each set of parameters is set to 12 seconds, and the sampling frequency during the simulation process is 1000 Hz to ensure that the time evolution curve of each physical indicator has sufficient resolution. The physical evaluation indicators include grinding ratio (measured as the ratio of material removal to grinding wheel loss), surface roughness Ra (unit μm), heat-affected zone width (unit μm) and energy consumption per unit time (unit J / s). The stable segment mean of each physical quantity is recorded by the sampling end as the simulation response output of the group of parameters. In order to improve the simulation efficiency, a multi-threaded parallel simulation framework is adopted, and a simulation batch is run concurrently for every 8 sets of parameters. The average total simulation time is controlled within 30 minutes. The final output is a structured response data table, which contains the corresponding parameter index and the values ​​of four physical indicators as the working condition simulation response data set.

[0081] Step S53: normalizing the working condition simulation response data, setting the index weight group to [0.35, 0.25, 0.2, 0.2], calculating the performance score of each group of candidate parameters, and generating a candidate parameter performance evaluation table; In this embodiment, all working condition simulation response data are normalized using a linear interval normalization method to uniformly map all physical indicators to the [0,1] interval. After normalization, the indicator weight group [0.35, 0.25, 0.2, 0.2] is introduced, corresponding to the grinding ratio, surface roughness, heat-affected zone width and energy consumption, respectively, and surface roughness and heat impact are clearly defined as secondary indicators. The performance score is obtained by weighted summation, and each group of parameters corresponds to a score result. In order to ensure the physical interpretability of the scoring process, the indicators with low values ​​as the best (such as Ra and energy consumption) are taken inversely before normalization. The final result uses the parameter index as the key and the performance score as the value to form a candidate parameter performance evaluation table. This embodiment processes 100 groups of parameters and outputs 100 scoring data to support subsequent parameter screening and judgment.

[0082] Step S54: According to the candidate parameter performance evaluation table, the score threshold is set to 0.82 to screen the stable parameter subset; In this embodiment, in order to achieve the goal of performance stability, it is necessary to perform threshold screening on the scores in the candidate parameter performance evaluation table. Combining historical simulation experiments and empirical data, the evaluation score screening threshold is set to 0.82, with the intention of eliminating parameter combinations with obvious performance disadvantages. During the screening process, the score value of each group of parameters is judged unidirectionally. Those with a score lower than 0.82 are all marked as substandard performance groups and eliminated from the parameter candidate set. This batch processes 100 groups of parameters. After screening, a total of 37 stable subsets with scores higher than the threshold are retained to form a stable parameter index set. This subset will be used for dimensional loop evaluation processing in subsequent steps. The threshold in this step can be flexibly adjusted according to the specific application requirements. The recommended value range is usually between 0.8 and 0.85, which should be determined in combination with the robustness requirements of the target working conditions.

[0083] Step S55: Perform dimensional loop mapping and redundancy removal processing on the stable parameter subset to construct a stable grinding parameter loop tensor set, and perform dimension-by-dimensional convergence evaluation on the stable grinding parameter loop tensor set. Set the stability fluctuation threshold to ±3% to eliminate periodic loop drift parameter combinations and generate a stable grinding parameter set.

[0084] In this embodiment, structured compression processing is further performed. First, a stable grinding parameter loop tensor set is constructed, and each group of parameter vectors is remapped to a ring structure according to the original five-dimensional index order. The loop period is set to 10 groups, forming a tensor set structure with a dimension of [37×5×10]. Subsequently, the mean and maximum fluctuation amplitude of each dimension within the loop period are calculated separately. If the relative fluctuation amplitude of any parameter dimension within the period exceeds the set threshold of ±3%, it is determined that there is a periodic drift, and the group of parameters will be marked as an unstable structure and eliminated. What is finally retained is a parameter structure that is fully stable within the loop period, totaling 24 groups. This stable grinding parameter set will be input into the actual process verification process or further working condition compression stage as a high-confidence configuration scheme to ensure that the subsequent process optimization processing has high reliability and process consistency.

[0085] The present invention is therefore intended to be illustrative and non-restrictive in all respects, with the scope of the invention being defined by the appended claims rather than the foregoing description, and all changes that come within the meaning and range of equivalents of the application documents are intended to be embraced within the present invention.

[0086] The foregoing description is intended only to provide specific embodiments of the present invention, which will enable those skilled in the art to understand and implement the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not intended to be limited to the embodiments shown herein, but is to be construed in the widest possible manner consistent with the principles and novel features disclosed herein.

Claims

1. A grinding parameter optimization method for a fully automatic coarse and fine grinding machine based on particle swarm optimization, characterized in that: The following steps are involved: Step S1: Acquire multimodal data of equipment operation and perform preliminary edge-level processing on the multimodal data to extract implicit physical disturbance features; Multi-threaded aggregation of implicit physical disturbance features is performed to obtain structured grinding field perception data; Step S2: performing multimodal grinding behavior feature analysis on the structured grinding field perception data to obtain grinding behavior pattern data; performing grinding parameter semantic abstraction based on the grinding behavior pattern data to obtain a grinding parameter intention nested mapping table; Step S3: performing an initialization basic particle swarm calculation on the grinding parameter intention nested mapping table to obtain an initialization basic particle swarm; performing a rough solution space boundary evolution based on the initialization basic particle swarm to obtain a coarse-grained grinding parameter trajectory family; Step S4: Perform subgraph nesting reconstruction on the coarse-grained grinding parameter trajectory family and construct a heterogeneous trajectory graph encoding vector set; perform soft clustering bias learning on the heterogeneous trajectory graph encoding vector set to obtain a high-dimensional grinding parameter candidate map; Step S5: performing a multi-dimensional working condition simulation based on the high-dimensional grinding parameter candidate map, and performing parameter tensor loop optimization based on the multi-dimensional working condition simulation results to obtain a stable grinding parameter set; Step S6: The stable grinding parameter set is transmitted to the management platform of the fully automatic rough and fine grinding machine to execute the control parameters, and iterative precision compensation mapping is performed according to the actual grinding deviation feedback data obtained in real time to obtain the optimized grinding parameter set.

2. The grinding parameter optimization method of the fully automatic coarse and fine grinding machine based on particle swarm optimization according to claim 1 is characterized in that: Step S1 is specifically as follows: Step S11: Acquiring multimodal data of equipment operation, wherein the multimodal data of equipment operation includes a grinding parameter group, a high-frequency vibration signal, a spindle current fluctuation record, a temperature rise thermal flow field image, and an accompanying acoustic spectrum signal; Step S12: performing channel-by-channel decoupling processing on the multimodal data of the device operation, and performing physical field disturbance trend identification on the decoupling processing results to obtain physical field disturbance trend data; Step S13: extracting implicit physical disturbance segments based on the physical field disturbance trend data to obtain implicit physical disturbance primary annotation data; Step S14: Thread the implicit physical disturbance primary annotation data, construct time segment indexes for the heat flow, vibration, current, and sound spectrum signal segments respectively, and set the aggregation window step size to 0.5s to perform thread merging operations to obtain a disturbance event structure vector set; Step S15: perform structural consistency encoding on the disturbance event structure vector set, and fuse the acquired equipment operation timestamp, working condition identification and location index information to obtain structured grinding field perception data.

3. The grinding parameter optimization method of the fully automatic coarse and fine grinding machine based on particle swarm optimization according to claim 2 is characterized in that: Step S13 is specifically as follows: Step S131: setting the window length to 2s and the step length to 0.5s, performing segmented sliding window segmentation on the physical field disturbance trend data to obtain sliding segment disturbance trend data; Step S132: Calculate the disturbance intensity differential value for each sliding segment in the sliding segment disturbance trend data, and set the disturbance intensity mutation threshold to 1.75 to identify the initial segment of potential disturbance anomaly, thereby generating a disturbance segment preliminary screening data set; Step S133: Group the disturbance segment preliminary screening data set by modal channel type, construct vibration frequency-energy diagrams, current fluctuation spectrum diagrams, heat flux intensity spectrum diagrams, and acoustic spectrum density diagrams, and perform frequency domain entropy increment analysis on each group of spectrums to screen segments that meet the preset multi-channel synchronous entropy fluctuation conditions and generate candidate multi-channel coupled disturbance segments; Step S134: performing co-occurrence pattern recognition on the candidate segments of multi-channel coupled disturbance, screening out isolated segments, and setting the modal consistency evaluation index to be greater than 0.6 to select segments with high synergy of interference components as primary annotated segments of implicit physical disturbance; Step S135: Aggregate the implicit physical disturbance primary annotation segments to obtain implicit physical disturbance primary annotation data.

4. The grinding parameter optimization method of the fully automatic coarse and fine grinding machine based on particle swarm optimization according to claim 2, characterized in that: Step S14 is specifically as follows: Step S141: grouping and labeling the implicit physical disturbance primary annotation data according to the modal channel type to obtain thermal flow field segments, vibration frequency band segments, current disturbance segments, and sound spectrum fluctuation segments; Step S142: assigning the thermal flow field segment, vibration frequency band segment, current disturbance segment, and sound spectrum fluctuation segment to corresponding preset thread pools, and activating four types of parallel processing threads to generate an initial channel set for a multi-threaded data stream; Step S143: Construct a standard time segment index for the disturbed segment data in each thread of the initial channel set of the multi-threaded data stream, set the single index time window to 0.5s, and perform sliding alignment on the disturbed segment data in each thread, eliminating segments with cross-channel time drift greater than 200ms to generate a synchronized time index segment set; Step S144: setting the aggregation window step size to 0.5s, performing thread aggregation operations on the synchronized time index segment set within each aggregation window, extracting the cross-channel collaborative perturbation structure, and generating a perturbation thread fusion node set; Step S145: Serialize and integrate the disturbance thread fusion node set, and assign a unique structure label to each disturbance thread fusion node to obtain a disturbance event structure vector set.

5. The grinding parameter optimization method of the fully automatic coarse and fine grinding machine based on particle swarm optimization according to claim 1 is characterized in that: The multimodal grinding behavior feature analysis in step S2 is as follows: The structured grinding field perception data is expanded in the time domain, and the time series frames are synchronously reconstructed according to the grinding wheel linear speed in the grinding parameter group to obtain the grinding field perception time series frame set; Acquire abrasive structure data, and perform grinding ratio change trend analysis based on the abrasive structure data and the grinding parameter group to obtain cutting area grinding ratio change trend data; The grinding field perception time series frame set is combined with the grinding ratio change trend data of the cutting area to perform multimodal time synchronization alignment of heat flow, current, vibration and sound spectrum to generate a modal reconstruction sequence group; The contact arc disturbance feature is identified for the modal reconstruction sequence group, and the transient components and energy mutation points of the modal disturbance are extracted; The disturbance significance threshold is set to 0.85, and the characteristics of abrasive particle breakage, shedding, and local burning during the grinding process are identified based on transient components and energy mutation points, thereby forming a grinding modal response significance map. Based on the structured grinding field sensing data, the cutting energy characteristics of each mode are extracted to obtain the cutting force fluctuation frequency, the grinding interval sound pressure surge rate and the second-order derivative of the spindle current. The cross-modal energy synergy tensor is constructed by integrating the cutting force fluctuation frequency, the grinding zone acoustic pressure surge rate, and the second derivative of the spindle current, and combining it with the grinding modal response significance map. Based on the cross-modal energy synergy tensor, behavior pattern aggregation and boundary identification are performed to obtain a set of sequential grinding behavior fragments. The set of sequential grinding behavior segments is mapped to the pre-connected grinding condition behavior library, and the morphological similarity of the behavior sequences is calculated to obtain the grinding behavior pattern data.

6. The grinding parameter optimization method of a fully automatic coarse and fine grinding machine based on particle swarm optimization according to claim 1, characterized in that: The semantic abstraction of the grinding parameters in step S2 is specifically: The grinding behavior pattern data were preliminarily screened for behavior-parameter mapping, and a potential correlation matrix between behavior segments and grinding parameters was constructed; The local sensitivity threshold was set to 0.75, and the maximum mutual information criterion analysis was performed on the potential correlation matrix to identify the significant behavioral response areas, thereby generating a parameter candidate cross-table; The dynamic characteristics of parameters related to grinding behavior are extracted based on the parameter candidate cross table, and the parameter information density distribution of the dynamic characteristics of parameters related to grinding behavior is evaluated to obtain the semantic weight map of grinding parameters; Perform behavioral semantic projection on the parameter abnormality patterns that frequently appear in the grinding parameter semantic weight map, and transform the parameter abnormality patterns into semantic behavioral labels including contact instability segments, local overload segments, and thermal drift risk segments. Construct a multi-nested index structure between semantic behavior labels and grinding parameter groups to generate grinding parameter behavior intention mapping clusters; The grinding parameter behavior intention mapping cluster is converted into a multi-dimensional nested structure to perform parameter dimension semantic compression and upper and lower encoding, and the grinding parameter intention nested mapping table is obtained.

7. The grinding parameter optimization method of a fully automatic coarse and fine grinding machine based on particle swarm optimization according to claim 1, characterized in that: Step S3 is specifically as follows: Step S31: performing tensor expansion on the grinding parameter intention nested mapping table to construct an initial parameter tensor field; Step S32: constructing a constraint space boundary set based on the upper and lower boundary combinations of each parameter dimension in the initial parameter tensor field, and performing non-uniform perturbations on the constraint space boundary set to obtain a coarse solution spatial structure grid; Step S33: performing particle swarm spatial initialization on the rough solution spatial structure grid, and performing initial parameter path encoding on each particle in the particle swarm to obtain an initialized basic particle swarm; Step S34: performing coarse-grained search evolution on the initialized basic particle group to generate a coarse-grained grinding parameter trajectory family.

8. The grinding parameter optimization method of the fully automatic coarse and fine grinding machine based on particle swarm optimization according to claim 7 is characterized in that: Step S34 is specifically as follows: Step S341: mapping the initial parameter path of the initialized basic particle swarm to the corresponding constraint space, and calculating the position similarity of the particles in the constraint space, so as to assign the initial velocity and inertia weight to the initialized basic particle swarm and obtain the initial velocity particle swarm; Step S342: using the constraint space to perform random jump updates on the initial velocity particle swarm to obtain a disturbance response particle swarm; Step S343: Calculate the fitness value of the current solution for each particle in the disturbance response particle swarm; Step S344: According to the fitness value of the current solution, a local optimal solution and a global optimal solution are selected for each particle, and the particle parameters are collaboratively updated according to the local optimal solution and the global optimal solution to obtain an updated particle swarm; Step S345: Return the updated particle swarm to step S342, and perform 10 rounds of iterations, with steps S342 to S345 as one round of iteration. Perform group collaborative clustering based on the solution of each particle in the particle swarm and its fitness evaluation results, and group the particles to output the coarse-grained grinding parameter trajectory family.

9. The grinding parameter optimization method of a fully automatic coarse and fine grinding machine based on particle swarm optimization according to claim 1, characterized in that: The soft clustering bias learning in step S4 is specifically as follows: Perform feature domain normalization mapping on the heterogeneous trajectory map encoding vector set to generate unified scale parameter space data; Based on the uniform scale parameter space data, biased fuzzy clustering is performed to obtain the biased fuzzy initial membership set; Iterative center transfer learning is performed based on the biased fuzzy initial membership set, and the center drift rate of each iteration is constrained to not exceed 0.05, and a set of class center transfer trajectories is obtained; The dynamic stability of the class center transfer trajectory set is evaluated, and the classes whose center trajectory variance exceeds the stability threshold of 0.08 are eliminated. The remaining class centers are fitted and modeled to generate the convergent class center space; The multi-class membership distribution of each trajectory vector is calculated inversely based on the convergent class center space, and a multi-membership structure map is constructed. According to the multi-membership structure map, each encoding vector in the heterogeneous trajectory map encoding vector set is assigned a confidence scalar weight, and a high-confidence soft clustering structure of grinding parameters is obtained. The high-confidence soft clustering structure of grinding parameters is subjected to similarity graph merging and nested graph compression to output a high-dimensional grinding parameter candidate graph.

10. The grinding parameter optimization method of a fully automatic coarse and fine grinding machine based on particle swarm optimization according to claim 1, characterized in that: Step S5 is specifically as follows: Step S51: extracting the particle size parameter, cutting depth, linear speed, contact arc length and grinding ratio index corresponding to each candidate parameter structure from the high-dimensional grinding parameter candidate map, and constructing an initial working condition state tensor; Step S52: Perform multi-physics field simulation based on the initial working condition state tensor, perform 12S continuous simulation for each set of parameters, and evaluate the physical indicators of grinding ratio, surface roughness, heat-affected zone width and grinding energy consumption respectively to obtain working condition simulation response data; Step S53: normalizing the working condition simulation response data, setting the index weight group to [0.35, 0.25, 0.2, 0.2], calculating the performance score of each group of candidate parameters, and generating a candidate parameter performance evaluation table; Step S54: According to the candidate parameter performance evaluation table, the score threshold is set to 0.82 to screen the stable parameter subset; Step S55: Perform dimensional loop mapping and redundancy removal processing on the stable parameter subset to construct a stable grinding parameter loop tensor set, and perform dimension-by-dimensional convergence evaluation on the stable grinding parameter loop tensor set. Set the stability fluctuation threshold to ±3% to eliminate periodic loop drift parameter combinations and generate a stable grinding parameter set.

Images