A multi-modal vibration suppression method and system for vertical grinding machine based on digital twinning

Abstract

The application discloses a kind of based on digital twinning vertical grinder processing multimode vibration suppression method and system, it is related to vertical grinder processing technical field, including: the multimodal signal data is carried out synchronous digital twinning mapping processing, forms digital twinning simulation environment;Multimodal signal data is input to vibration source diagnostic model, and output vibration source fingerprint;Based on digital twinning simulation environment, vibration source fingerprint and real-time processing condition are input to vibration suppression strategy generation model, generate optimal vibration suppression strategy and drive;Based on two-way game evolution mechanism, vibration source diagnostic model and vibration suppression strategy generation model are iteratively updated in coordination;Using vibration suppression decision knowledge graph, vibration source diagnostic model and vibration suppression strategy generation model after collaborative iterative update, vibration suppression decision is carried out to the multimodal signal data acquired in real time, generates executable vibration suppression strategy, and drives vibration suppression execution unit to execute.The application improves vertical grinder processing quality, equipment stability and production efficiency.

Description

Technical Field

[0001] This invention relates to the field of vertical grinding machine processing technology, and more specifically, to a multimodal vibration suppression method and system for vertical grinding machine processing based on digital twins. Background Technology

[0002] Vertical mills are core grinding equipment in industries such as cement, metallurgy, and power. The machining quality of their core wear-resistant components (such as liners and roller sleeves) directly determines the overall lifespan and energy consumption level of the machine. These parts are characterized by their large size (diameter Φ2-5m), complex shape (curved / conical surface), and hard material (high-chromium cast iron, HRC60 and above). In their final finishing process, traditional mechanical clamping methods are prone to workpiece deformation, and the tool-workpiece-machine tool process system is prone to severe chatter under cutting forces, leading to deterioration of machining quality, soaring production costs, and low production efficiency.

[0003] Current traditional vibration suppression methods mainly rely on offline spectrum analysis or online monitoring based on fixed thresholds, making it difficult to achieve accurate source tracing and real-time intervention. While existing digital twin-based vibration suppression technologies can achieve state synchronization and simulation, they primarily use frequency or time domain characteristics at the diagnostic level, failing to adequately reveal the complex vibration mechanisms involving nonlinearity and multi-physics coupling, resulting in delayed early warnings. At the decision-making level, vibration suppression strategies often rely on expert experience or pre-set rule bases, lacking the ability to perform multi-objective verification and autonomous creation in a virtual environment, making it difficult to adapt to complex and changing processing conditions. Furthermore, most existing systems lack self-evolution and knowledge accumulation mechanisms, failing to effectively transform historical vibration suppression experience into reusable and transferable decision-making knowledge, leading to long debugging cycles and unstable vibration suppression effects when facing new workpieces and materials.

[0004] No effective solutions have yet been proposed to address the problems in the relevant technologies. Summary of the Invention

[0005] In response to the problems in related technologies, this invention proposes a method and system for multimodal vibration suppression in vertical grinding machine processing based on digital twins, so as to overcome the above-mentioned technical problems existing in the existing related technologies.

[0006] Therefore, the specific technical solution adopted by the present invention is as follows:

[0007] According to one aspect of the present invention, a method for multimodal vibration suppression in vertical grinding machine machining based on digital twins is provided, the method comprising:

[0008] S1. Acquire multimodal signal data of the vertical grinding machine and perform synchronous digital twin mapping processing on the multimodal signal data to form a digital twin simulation environment;

[0009] S2. Based on high-dimensional phase space reconstruction technology and cross-media feature fusion technology, construct a vibration source diagnostic model; input multimodal signal data into the vibration source diagnostic model and output vibration source fingerprint;

[0010] S3. Construct a vibration suppression strategy generation model; Based on the digital twin simulation environment, input the vibration source fingerprint and real-time processing conditions into the vibration suppression strategy generation model to generate the optimal vibration suppression strategy and drive the vibration suppression execution unit to execute it.

[0011] S4. Based on the bidirectional game evolution mechanism, the vibration source diagnosis model and the vibration suppression strategy generation model are updated collaboratively and iteratively using the multimodal signal data after the vibration suppression execution unit is executed; the vibration characteristics, strategy parameters and performance indicators obtained throughout the process are stored in a structured manner to construct a vibration suppression decision knowledge graph.

[0012] S5. Using the vibration suppression decision knowledge graph, the vibration source diagnosis model and the vibration suppression strategy generation model updated through collaborative iteration, vibration suppression decisions are made on the real-time acquired multimodal signal data, generating an executable vibration suppression strategy that adapts to the current processing conditions, and driving the vibration suppression execution unit to execute it.

[0013] Furthermore, acquiring multimodal signal data from the vertical grinding machine and performing synchronous digital twin mapping processing on the multimodal signal data to form a digital twin simulation environment includes the following steps:

[0014] S11. Obtain the geometric structure, motion relationship and dynamic characteristics of the vertical grinding machine to form its physical properties;

[0015] S12. Using multimodal sensors arranged in key parts of the vertical grinding machine, vibration signals, thermal imaging signals and acoustic signals are acquired in real time to form multimodal signal data;

[0016] S13. Perform real-time synchronous digital twin mapping processing on multimodal signal data and physical characteristics to form a real-time synchronous digital twin simulation environment.

[0017] Furthermore, based on high-dimensional phase space reconstruction technology and cross-media feature fusion technology, a vibration source diagnostic model is constructed; inputting multimodal signal data into the vibration source diagnostic model and outputting the vibration source fingerprint includes the following steps:

[0018] S21. Extract vibration signals from multimodal signal data, and use high-dimensional phase space reconstruction technology to perform phase space mapping on the vibration signals to obtain vibration dynamic characteristics;

[0019] S22. Extract thermal imaging signals and acoustic signals from multimodal signal data, perform time-series analysis and frequency domain transformation respectively, and obtain thermal evolution characteristics and acoustic spectrum characteristics;

[0020] S23. Based on the attention mechanism, cross-media feature fusion of vibration dynamics features, thermal evolution features and acoustic spectrum features is performed to generate a fused feature vector;

[0021] S24. Match the fused feature vector with the pre-built vibration source physical map library to output a vibration source fingerprint containing the vibration source type, spatial location and energy contribution.

[0022] Furthermore, the vibration signals are extracted from the multimodal signal data, and the vibration signals are mapped into phase space using high-dimensional phase space reconstruction technology to obtain the vibration dynamic characteristics, including the following steps:

[0023] S211. Extract vibration signals from multimodal signal data, and perform noise reduction and normalization processing on the vibration signals in sequence to obtain preprocessed vibration signals.

[0024] S212. Using the delayed coordinate embedding method, the preprocessed vibration signal is reconstructed in phase space to obtain the trajectory point set in high-dimensional phase space.

[0025] S213. Perform attractor morphology analysis on the trajectory point set, extract attractor geometric features, and calculate the Lyapunov exponent of the trajectory point set.

[0026] S214. The attractor geometric features are combined with the Lyapunov index to obtain the vibration dynamics features.

[0027] Furthermore, the vibration suppression strategy generation model includes: a strategy generator, a twin simulation discriminator, and a strategy parsing distributor;

[0028] The strategy generator is used to generate a variety of vibration suppression strategy combinations in a digital twin simulation environment based on the vibration source fingerprint and real-time processing conditions.

[0029] The twin simulation discriminator is used to perform multi-objective performance simulation evaluation of multiple vibration suppression strategy combinations based on a digital twin simulation environment, and output the optimal vibration suppression strategy.

[0030] The strategy parser and distributor is used to parse the optimal vibration suppression strategy into multi-channel collaborative control commands and send them to the vibration suppression execution unit to perform vibration suppression operations.

[0031] Furthermore, based on a two-way game evolution mechanism, the vibration source diagnosis model and the vibration suppression strategy generation model are collaboratively iteratively updated using the multimodal signal data after the vibration suppression execution unit is executed, including the following steps:

[0032] Acquire multimodal signal data after the vibration suppression execution unit is executed, and construct a closed-loop feedback dataset containing vibration suppression strategy parameters and processing conditions;

[0033] The vibration suppression effect was evaluated based on the closed-loop feedback dataset, and performance deviation features were extracted.

[0034] Based on the performance deviation characteristics, a diagnostic correction objective function for optimizing the vibration source diagnostic model and a strategy optimization objective function for optimizing the vibration suppression strategy generation model are constructed respectively, forming a two-way game optimization mechanism.

[0035] Based on a two-way game optimization mechanism, the vibration source diagnosis model and the vibration suppression strategy generation model are updated collaboratively and iteratively.

[0036] Furthermore, based on the performance deviation characteristics, a diagnostic correction objective function for optimizing the vibration source diagnostic model and a strategy optimization objective function for optimizing the vibration suppression strategy generation model are constructed respectively, forming a two-way game optimization mechanism including the following steps:

[0037] Based on performance deviation characteristics, the vibration source identification error vector is extracted;

[0038] Using the vibration source identification error vector as input, a diagnostic correction objective function for the vibration source diagnostic model is constructed.

[0039] Based on multi-objective performance indicators including vibration residual energy, processing quality and control energy consumption, and combined with the consistency requirements of simulation and actual execution, a strategy optimization objective function for the vibration suppression strategy generation model is constructed.

[0040] The diagnostic correction objective function is coupled with the strategy optimization objective function, and the vibration source diagnostic model and the vibration suppression strategy generation model are driven to co-evolve in an adversarial manner through alternating optimization until an equilibrium state is reached.

[0041] Furthermore, the vibration characteristics, strategy parameters, and performance indicators acquired throughout the process are stored in a structured manner, and a vibration suppression decision knowledge graph is constructed, including the following steps:

[0042] The vibration source fingerprint, processing conditions, vibration suppression strategy parameters, and corresponding multi-dimensional performance indicators generated throughout the entire process from vibration source diagnosis and strategy generation to the execution of the vibration suppression execution unit are obtained to form vibration suppression event data.

[0043] The vibration suppression event data are sequentially formatted and feature aligned, and causal and triggering relationships are established between the vibration suppression event data.

[0044] Based on the causal and triggering relationships between vibration suppression event data, a vibration suppression decision knowledge graph is constructed with processing conditions, vibration source type, vibration suppression strategy, and performance results as the core quadruple.

[0045] In subsequent collaborative iterative updates, newly generated vibration suppression event data will be injected into the vibration suppression decision knowledge graph.

[0046] Furthermore, by utilizing a vibration suppression decision knowledge graph, a collaboratively iteratively updated vibration source diagnostic model, and a vibration suppression strategy generation model, vibration suppression decisions are made on the real-time acquired multimodal signal data. This generates an executable vibration suppression strategy adapted to the current processing conditions and drives the vibration suppression execution unit to execute the following steps:

[0047] S51. Using the collaboratively iteratively updated vibration source diagnostic model, the real-time acquired multimodal signal data is diagnosed to obtain the real-time vibration source fingerprint;

[0048] S52. Vectorize the real-time vibration source fingerprint and the current processing condition to form a multi-dimensional semantic retrieval key; based on the multi-dimensional semantic retrieval key, search the vibration suppression decision knowledge graph to obtain historical vibration suppression strategies.

[0049] S53. Based on the differences between the current processing conditions and historical processing conditions, the historical vibration suppression strategy is adaptively adjusted to generate an initial candidate strategy set;

[0050] S54. In the digital twin simulation environment, the initial candidate strategy is used as a priori guide and input into the vibration suppression strategy generation model after collaborative iterative update. The executable vibration suppression strategy is then selected and driven to execute by the vibration suppression execution unit.

[0051] According to another aspect of the present invention, a multimodal vibration suppression system for vertical grinding machine machining based on digital twin is also provided. The system includes: a twin synchronization and simulation module, a vibration source diagnosis module, a strategy generation and execution module, a game evolution and knowledge construction module, and an adaptive execution module.

[0052] The twin synchronization and simulation module is used to acquire multimodal signal data of the vertical grinding machine and perform synchronous digital twin mapping processing on the multimodal signal data to form a digital twin simulation environment;

[0053] The vibration source diagnostic module is used to construct a vibration source diagnostic model based on high-dimensional phase space reconstruction technology and cross-media feature fusion technology; it inputs multimodal signal data into the vibration source diagnostic model and outputs the vibration source fingerprint.

[0054] The strategy generation and execution module is used to construct a vibration suppression strategy generation model. Based on the digital twin simulation environment, the vibration source fingerprint and real-time processing conditions are input into the vibration suppression strategy generation model to generate the optimal vibration suppression strategy and drive the vibration suppression execution unit to execute it.

[0055] The Game Evolution and Knowledge Construction Module is used to collaboratively iteratively update the vibration source diagnosis model and the vibration suppression strategy generation model based on the bidirectional game evolution mechanism and the multimodal signal data after the vibration suppression execution unit is executed; it also stores the vibration characteristics, strategy parameters and performance indicators obtained throughout the process in a structured manner to construct a vibration suppression decision knowledge graph.

[0056] The adaptive execution module is used to make vibration suppression decisions on real-time acquired multimodal signal data by utilizing the vibration suppression decision knowledge graph, the collaboratively iteratively updated vibration source diagnosis model and vibration suppression strategy generation model, generate executable vibration suppression strategies that are adapted to the current processing conditions, and drive the vibration suppression execution unit to execute them.

[0057] The beneficial effects of this invention are as follows:

[0058] 1. This invention introduces a high-dimensional phase space reconstruction and cross-media feature fusion mechanism to construct a structured vibration source fingerprint, which can identify the root cause of vibration from the essential level of nonlinear dynamics. It can accurately capture the beginning of flutter before traditional spectral indicators show significant anomalies, and significantly improve the foresight and sensitivity of vibration source diagnosis.

[0059] 2. This invention proposes a vibration suppression strategy generation model architecture, which places the strategy generator and the twin simulation discriminator in a digital twin environment for adversarial game, so that the vibration suppression strategy is executed after multi-objective verification in virtual space, which not only ensures physical safety, but also avoids the processing interruption and tool wear caused by traditional trial and error.

[0060] 3. This invention designs a two-way game evolution mechanism and a vibration suppression decision knowledge graph, enabling the vibration source diagnosis model and the strategy generation model to continuously optimize in actual operation, and structuring and precipitating effective experience into a quadruple of operating condition-vibration source-strategy-performance; the vibration suppression decision knowledge graph supports intelligent retrieval and migration of historical strategies, effectively solving the cold start problem under new operating conditions.

[0061] 4. This invention effectively solves the technical problem of the difficulty in coordinating the suppression of multi-source composite vibrations in high-precision, heavy-load, and variable-condition machining of vertical grinding machines by deeply integrating digital twins, nonlinear dynamics, cross-media fusion, generative adversarial learning, causal knowledge graphs and multi-objective optimization, and significantly improves the surface quality, equipment stability and production efficiency of the machined surface. Attached Figure Description

[0062] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0063] Figure 1 This is a flowchart of a multimodal vibration suppression method for vertical grinding machine based on digital twin according to an embodiment of the present invention;

[0064] Figure 2 This is a schematic diagram of a multimodal vibration suppression system for vertical grinding machines based on digital twins, according to an embodiment of the present invention.

[0065] In the picture:

[0066] 1. Twin synchronization and simulation module; 2. Vibration source diagnosis module; 3. Strategy generation and execution module; 4. Game evolution and knowledge construction module; 5. Adaptive execution module. Detailed Implementation

[0067] To further illustrate the various embodiments, the present invention provides accompanying drawings, which are part of the disclosure of the present invention. These drawings are mainly used to illustrate the embodiments and can be used in conjunction with the relevant descriptions in the specification to explain the operating principles of the embodiments. With reference to these drawings, those skilled in the art should be able to understand other possible implementation methods and the advantages of the present invention. The components in the drawings are not drawn to scale, and similar component symbols are generally used to represent similar components.

[0068] According to an embodiment of the present invention, a method and system for multimodal vibration suppression in vertical grinding machine processing based on digital twin is provided.

[0069] The present invention will now be further described in conjunction with the accompanying drawings and specific embodiments, such as... Figure 1 As shown, according to an embodiment of the present invention, a multimodal vibration suppression method for vertical grinding machine machining based on digital twins includes:

[0070] S1. Acquire multimodal signal data of the vertical grinding machine and perform synchronous digital twin mapping processing on the multimodal signal data to form a digital twin simulation environment.

[0071] It should be explained that by using multi-dimensional modeling of geometry, physics, and behavior, and real-time mapping of multi-modal data, a virtual simulation platform with high fidelity to the vertical grinding machine is constructed, providing a foundation for subsequent vibration source diagnosis and strategy verification.

[0072] In this optional embodiment, acquiring multimodal signal data of a vertical grinding machine and performing synchronous digital twin mapping processing on the multimodal signal data to form a digital twin simulation environment includes the following steps:

[0073] S11. Obtain the geometric structure, motion relationship and dynamic characteristics of the vertical grinding machine to form its physical properties;

[0074] Specifically, based on the actual structure of the vertical grinding machine, high-precision point cloud data is acquired, and then a three-dimensional geometric model (such as STEP or IGES format) containing key components such as the bed, column, worktable, and spindle box is reconstructed, clarifying the assembly relationships and spatial constraints between the components. On this basis, based on the CNC system parameters and mechanical transmission chain of the machine tool, the kinematic relationships of each motion axis are defined on the three-dimensional geometric model, including the stroke range, speed and acceleration limits, and the linkage logic between axes. Through experimental modal analysis, with the machine tool stationary, a hammer or vibrator is used to excite the key structure, while response signals are collected by arranged acceleration sensors to identify the first few natural frequencies, damping ratios, and mode shapes of the structure. The acquired geometric structure, kinematic relationships, and experimentally measured dynamic parameters together constitute the physical characteristics used to construct the virtual carrier.

[0075] S12. Using multimodal sensors arranged in key parts of the vertical grinding machine, vibration signals, thermal imaging signals and acoustic signals are acquired in real time to form multimodal signal data;

[0076] Specifically, the sensor arrangement is as follows: accelerometers (sampling rate ≥ 10kHz) are installed at both ends of the spindle and at key stress points on the grinding disc to ensure comprehensive acquisition of vibration signals; an infrared thermal imager (resolution 640×480, frame rate 30Hz) is used to target the core joint area between the spindle and the grinding disc to capture temperature field changes in real time and form thermal imaging signals; acoustic sensors (frequency band 100–500kHz) are installed on the side of the frame near the vibration transmission path to collect acoustic signals during the processing.

[0077] Data acquisition involves simultaneously receiving three signals via a multi-channel data acquisition card and performing preprocessing operations, including mean removal and power frequency filtering, to form standardized multimodal data frames at fixed time intervals, ensuring data timeliness and consistency.

[0078] S13. Perform real-time synchronous digital twin mapping processing on multimodal signal data and physical characteristics to form a real-time synchronous digital twin simulation environment.

[0079] It needs to be explained that a set of explicit mapping rules are established in advance. For example, the vibration acceleration data frame of the spindle bearing housing is specified to drive the corresponding degree of freedom of the spindle assembly in the 3D geometric model to move; the average temperature of a specific pixel area in the thermal imaging data is specified to update the temperature attribute of the finite element mesh node in that area in the 3D geometric model; and the RMS (root mean square) value of the acoustic emission signal is specified to trigger a visualization event of an energy level in the virtual environment.

[0080] Multimodal signal data is transmitted in real time via a high-speed communication interface (such as TCP / IP protocol or shared memory) to a simulation platform running a virtual carrier (i.e., a 3D visualization scene and dynamic calculation kernel built based on physical characteristics). The data interface and driver module in the simulation platform parse the multimodal signal data in real time according to the aforementioned mapping rules, and call the simulation engine API to drive the corresponding state variables (such as displacement, velocity, temperature, and energy) in the 3D geometric model to be updated instantaneously.

[0081] Through the continuous, high-frequency (e.g., 10Hz or higher) data injection and status updates, the dynamic response of the virtual carrier is synchronized with the actual operating status of the physical machine tool at the millisecond level, thus forming a digital twin simulation environment that can be used for online analysis, prediction, and simulation.

[0082] S2. Based on high-dimensional phase space reconstruction technology and cross-media feature fusion technology, construct a vibration source diagnostic model; input multi-modal signal data into the vibration source diagnostic model and output vibration source fingerprint.

[0083] It should be explained that this step, through in-depth analysis and feature fusion of multimodal signals, overcomes the limitations of single-signal diagnosis, achieves accurate identification and quantitative characterization of vibration sources, and provides a targeted basis for the generation of subsequent vibration suppression strategies.

[0084] In this optional embodiment, a vibration source diagnostic model is constructed based on high-dimensional phase space reconstruction technology and cross-media feature fusion technology; inputting multimodal signal data into the vibration source diagnostic model and outputting the vibration source fingerprint includes the following steps:

[0085] S21. Extract vibration signals from multimodal signal data, and use high-dimensional phase space reconstruction technology to perform phase space mapping on the vibration signals to obtain vibration dynamic characteristics;

[0086] S22. Extract thermal imaging signals and acoustic signals from multimodal signal data, perform time-series analysis and frequency domain transformation respectively, and obtain thermal evolution characteristics and acoustic spectrum characteristics;

[0087] Specifically, thermal evolution feature extraction involves inter-frame difference processing of thermal imaging signals to analyze the temporal characteristics of core regions (such as spindle bearings and grinding disc contact surfaces), such as temperature change rate and temperature gradient, reflecting the changing law of frictional heating caused by vibration; acoustic spectrum feature extraction uses fast Fourier transform to convert acoustic signals from the time domain to the frequency domain, extracting frequency domain features such as characteristic frequency, power spectral density, and spectral peak energy. Different vibration sources will correspond to specific acoustic spectrum peak distributions.

[0088] S23. Based on the attention mechanism, cross-media feature fusion of vibration dynamics features, thermal evolution features and acoustic spectrum features is performed to generate a fused feature vector;

[0089] Specifically, the feature dimension uniformly adopts the feature mapping method to transform the features of the three different modes into the same dimensional space, ensuring the feasibility of fusion; the attention mechanism is introduced by constructing an attention weight allocation model to analyze the correlation between different modal features and vibration source identification, assigning higher basic weights to vibration dynamic features, and dynamically adjusting the weights of each mode according to the real-time signal quality (such as increasing the weight of acoustic spectrum features when the vibration signal is disturbed); the fusion execution adopts a weighted summation method to fuse the features of each mode, generating a fused feature vector that combines the advantages of each mode, strengthening the representation of key information of the vibration source.

[0090] S24. Match the fused feature vector with the pre-built vibration source physical map library to output a vibration source fingerprint containing the vibration source type, spatial location and energy contribution.

[0091] It should be explained that the vibration source physical map library is established through historical data learning or simulation analysis, storing typical feature vector templates corresponding to different types of vibration sources and their possible positional distribution in the machine tool coordinate system. The matching process uses nearest neighbor search or support vector machine classification algorithms to find the template most similar to the fused feature vector, and outputs a vibration source fingerprint containing the corresponding vibration source type, the most likely spatial location, and the estimated contribution of the vibration source to the overall vibration energy.

[0092] In this optional embodiment, extracting vibration signals from multimodal signal data and mapping the vibration signals into phase space using high-dimensional phase space reconstruction technology to obtain vibration dynamic characteristics includes the following steps:

[0093] S211. Extract vibration signals from multimodal signal data, and perform noise reduction and normalization processing on the vibration signals in sequence to obtain preprocessed vibration signals.

[0094] S212. Using the delayed coordinate embedding method, the preprocessed vibration signal is reconstructed in phase space to obtain the trajectory point set in high-dimensional phase space.

[0095] S213. Perform attractor morphology analysis on the trajectory point set, extract attractor geometric features, and calculate the Lyapunov exponent of the trajectory point set.

[0096] S214. The attractor geometric features are combined with the Lyapunov index to obtain the vibration dynamics features.

[0097] It should be explained that vibration signals, as the core data for vibration source diagnosis, can be reconstructed in high-dimensional phase space to uncover their nonlinear dynamic nature and capture subtle vibration characteristics that are difficult to detect using traditional analysis methods. The specific implementation steps for vibration dynamic feature extraction are as follows:

[0098] Step 1: Extract the triaxial vibration acceleration signal (sampling frequency 10kHz) from the synchronously acquired multimodal data stream at the spindle or grinding wheel frame. Use wavelet threshold denoising (such as db4 wavelet, 5-level decomposition) to suppress high-frequency noise and eliminate dimensional differences through Z-score normalization to obtain the preprocessed single-channel vibration time sequence signal.

[0099] Step 2: The preprocessed vibration signal is reconstructed into a high-dimensional phase space using the delayed coordinate embedding method. The time delay is determined by the first minimum value of the mutual information function (usually 1 / 4 to 1 / 10 of the main period of the signal), and the embedding dimension is determined by the spurious nearest neighbor method (usually 3 to 10). The one-dimensional time series is reconstructed into a set of trajectory points in the high-dimensional phase space. Each state point contains the characteristic information of the signal at different times, realizing a multi-dimensional representation of the vibration behavior.

[0100] Step 3: Perform nonlinear dynamic analysis on the reconstructed trajectory point set. On one hand, calculate the geometric characteristics of the attractors in phase space, such as correlation dimension, Kolmogorov entropy, or the ratio of the principal axis lengths of the attractors. These characteristics quantify the complexity and degrees of freedom of the system's motion. On the other hand, use the Wolf algorithm or the small data method to calculate the maximum Lyapunov exponent of the trajectory point set. A positive exponent indicates that the system has chaotic characteristics, and its magnitude directly reflects the system's sensitivity to initial conditions (i.e., the intensity of chaos). These two characteristics together constitute a description of the essential dynamic behavior of the vibration system.

[0101] Step 4: The extracted attractor geometric features (such as fractal dimension and correlation dimension) are integrated with the calculated Lyapunov exponent using a feature concatenation method to form a feature vector containing the essence of vibration nonlinear dynamics; the differences in the dimensions of different features are eliminated through feature normalization processing, and finally standardized vibration dynamic features are obtained.

[0102] S3. Construct a vibration suppression strategy generation model; Based on the digital twin simulation environment, input the vibration source fingerprint and real-time processing conditions into the vibration suppression strategy generation model to generate the optimal vibration suppression strategy and drive the vibration suppression execution unit to execute it.

[0103] It should be explained that this step achieves intelligent generation and verification of vibration suppression strategies through an adversarial network architecture, avoids the risk of physical trial and error by relying on a digital twin simulation environment, and ensures the accuracy of strategy execution by combining multi-component collaboration, thus forming a closed-loop link of generation-verification-execution.

[0104] In this optional embodiment, the vibration suppression strategy generation model includes: a strategy generator, a twin simulation discriminator, and a strategy parsing distributor;

[0105] The strategy generator is used to generate a variety of vibration suppression strategy combinations in a digital twin simulation environment based on the vibration source fingerprint and real-time processing conditions.

[0106] The twin simulation discriminator is used to perform multi-objective performance simulation evaluation of multiple vibration suppression strategy combinations based on a digital twin simulation environment, and output the optimal vibration suppression strategy.

[0107] The strategy parser and distributor is used to parse the optimal vibration suppression strategy into multi-channel collaborative control commands and send them to the vibration suppression execution unit to perform vibration suppression operations.

[0108] It should be explained that the vibration suppression strategy generation model consists of three core components: a strategy generator, a twin simulation discriminator, and a strategy parser and distributor, which work together on the digital twin simulation environment.

[0109] The policy generator is a conditional generative network (like the generator part of a conditional generative adversarial network cGAN). The policy generator takes as input a concatenated vector of vibration source fingerprints (including type, location, and energy contribution) and real-time machining conditions (such as spindle speed, feed rate, and depth of cut). In one specific implementation, the policy generator employs a multi-layer fully connected neural network structure, outputting a multi-dimensional vector through forward propagation. This multi-dimensional vector encodes a complete set of potentially effective vibration suppression strategy parameters for the current vibration source and machining conditions.

[0110] Specifically, the vibration suppression strategy combination parameter vector defines the suggested control quantities for multiple vibration suppression execution units (such as hydraulic leveling modules, magnetorheological damping modules, and piezoelectric actuation modules) within a future control cycle. These include the target displacement of the leveling cylinder, the target current value of the damper coil, and the amplitude and phase of the driving voltage of the piezoelectric ceramic. The multidimensional vector is designed to simultaneously output multiple different parameter vectors, forming a candidate strategy set.

[0111] The twin simulation discriminator is an intelligent agent with multi-objective evaluation capabilities based on a digital twin simulation environment. Its core function is not to distinguish between genuine and fake policies as in traditional GANs, but rather to evaluate the merits of different policies. The twin simulation discriminator receives each set of candidate policy parameters output by the policy generator. For each set of parameters, the discriminator inputs it as a control command into the digital twin simulation environment, driving the virtual vibration damping execution unit to perform actions. Through rapid numerical simulation, it predicts the dynamic response of the virtual machine bed over several control cycles after executing the policy.

[0112] Based on simulation results, the twin simulation discriminator calculates and integrates multiple performance indicators, including but not limited to: 1) Vibration attenuation rate: the percentage decrease in vibration energy at key measuring points; 2) Control energy consumption: the total power consumed by each execution unit; 3) Actuator load: assessing whether the amplitude and rate of change of control commands are within safe limits; 4) Impact on machined surface quality: the predicted surface roughness change based on simulated vibration data. The twin simulation discriminator uses a pre-trained evaluation network (such as a deep Q-network or a multi-objective evaluation network) to perform weighted or Pareto front analysis on these indicators, assigning a comprehensive evaluation score to each candidate strategy. Finally, the twin simulation discriminator outputs the strategy with the highest score as the optimal vibration suppression strategy.

[0113] The strategy parser and distributor is a module responsible for instruction conversion and synchronous distribution. It receives the optimal vibration suppression strategy selected by the twin simulation discriminator. Internally, the parser and distributor stores control protocol mapping tables for each execution unit. Its workflow is as follows: First, it parses the abstract strategy parameters into low-level control instruction frames that conform to the communication protocols of each execution unit's controller and have precise timing requirements.

[0114] For example, the command "magnetorheological damping current value 1.5A" is parsed into a CAN message containing the target address, data length, current setpoint, and checksum. Then, via a real-time industrial network, these multi-channel control commands are precisely and simultaneously transmitted to the corresponding vibration damping actuators (hydraulic leveler, magnetorheological damper, piezoelectric actuator) on the physical machine tool in a time-synchronized manner. Upon receiving the commands, the vibration damping actuators immediately activate, thereby achieving vibration suppression in the physical world.

[0115] In addition, the entire adversarial network is optimized through adversarial training in the offline phase: the discriminator provides gradient feedback to guide the generator to improve the effectiveness of the policy; in the online phase, the network weights are fixed and only forward inference and simulation evaluation are performed to ensure real-time performance (end-to-end latency <50ms).

[0116] S4. Based on the bidirectional game evolution mechanism, the multimodal signal data after the vibration suppression execution unit is executed is used to collaboratively iteratively update the vibration source diagnosis model and the vibration suppression strategy generation model; the vibration characteristics, strategy parameters and performance indicators obtained throughout the process are structured and stored to construct a vibration suppression decision knowledge graph.

[0117] It should be explained that this step achieves collaborative optimization of the two models through a two-way game evolution mechanism, and at the same time relies on knowledge graphs to complete the structured accumulation and reuse of data throughout the process, forming a long-term evolutionary link of execution feedback - model optimization - knowledge accumulation.

[0118] In this optional embodiment, based on a two-way game evolution mechanism, the collaborative iterative update of the vibration source diagnosis model and the vibration suppression strategy generation model using the multimodal signal data after the vibration suppression execution unit is executed includes the following steps:

[0119] Acquire multimodal signal data after the vibration suppression execution unit is executed, and construct a closed-loop feedback dataset containing vibration suppression strategy parameters and processing conditions;

[0120] The vibration suppression effect was evaluated based on the closed-loop feedback dataset, and performance deviation features were extracted.

[0121] Based on the performance deviation characteristics, a diagnostic correction objective function for optimizing the vibration source diagnostic model and a strategy optimization objective function for optimizing the vibration suppression strategy generation model are constructed respectively, forming a two-way game optimization mechanism.

[0122] Based on a two-way game optimization mechanism, the vibration source diagnosis model and the vibration suppression strategy generation model are updated collaboratively and iteratively.

[0123] It should be explained that the implementation steps of the dual-model collaborative iterative update based on the two-way game evolution mechanism are as follows:

[0124] Step 1: Constructing the closed-loop feedback dataset:

[0125] 1) Data Acquisition. Vibration signals, thermal imaging signals and acoustic signals after execution are collected synchronously through multimodal sensors. At the same time, the vibration suppression strategy parameters (such as control parameters of each execution unit and execution sequence) and real-time processing condition data (such as material hardness and spindle load changes during execution) are collected through the equipment control system.

[0126] 2) Data integration. The "multimodal signals after execution - vibration suppression strategy parameters - processing conditions" are bound by timestamps to form a single complete vibration suppression closed-loop data record.

[0127] 3) Feedback dataset construction. Data records with execution errors are removed, ultimately forming a closed-loop feedback dataset containing multiple batches and operating conditions.

[0128] Step 2: Vibration Suppression Effect Evaluation and Performance Deviation Feature Extraction:

[0129] 1) Establish a multi-dimensional evaluation index system. Establish core indicators (such as vibration attenuation rate and machining surface accuracy) and auxiliary indicators (such as execution response delay, energy consumption value, and execution unit temperature rise); calculate the vibration attenuation rate by comparing the vibration amplitude before and after execution, obtain the machining surface accuracy through a surface roughness measuring instrument, and obtain auxiliary index data in conjunction with a data acquisition system.

[0130] 2) Deviation Feature Extraction. The difference between the actual value of the evaluation index and the preset target value is calculated to obtain the deviation of a single index; for the vibration source diagnosis dimension, the difference between the vibration source fingerprint after re-diagnosis and the vibration source fingerprint before execution is calculated to extract the vibration source identification error feature; for the strategy effect dimension, performance deviation features such as the amplitude of vibration attenuation rate failure, the amount of machining accuracy deviation, and the energy consumption excess rate are extracted; through feature normalization processing, various deviation features are integrated into a standardized performance deviation feature vector.

[0131] Step 3: Implement the game mechanism through objective function construction and coupling:

[0132] 1) Source identification error vector extraction. Based on the source diagnosis-related deviations in the performance deviation characteristics, a feature decomposition method is used to extract three core components: source type identification error, spatial location positioning error, and energy contribution calculation error. A three-dimensional source identification error vector is constructed. Each component in the vector is obtained after standardization of the deviation value, which directly reflects the accuracy shortcomings of the source diagnosis model.

[0133] 2) Construction of the Diagnostic Correction Objective Function. With minimizing the vibration source identification error as the core objective, model complexity constraints are introduced to avoid overfitting while also considering diagnostic efficiency requirements. The diagnostic correction objective function uses the magnitude of the vibration source identification error vector as the core optimization term. Regularization terms constrain the complexity of model parameters (such as the embedding dimension of high-dimensional phase space reconstruction and the weight matrix of the attention mechanism), and an upper limit on diagnostic time is set as a constraint. Through the diagnostic correction objective function, the optimization direction of the vibration source diagnosis model is clearly defined as improving accuracy, ensuring efficiency, and maintaining controllable complexity.

[0134] 3) Construction of the Strategy Optimization Objective Function. The core objective is multi-objective optimization, combined with the consistency requirements between simulation and actual execution. The strategy optimization objective function uses residual vibration energy (an inverse indicator of vibration attenuation rate), processing quality deviation, and control energy consumption as core optimization terms, assigning them different weights (dynamically adjusted according to processing requirements, such as increasing the weight of processing quality for high-precision processing). Simultaneously, a deviation penalty term between simulation and actual execution is introduced. When the deviation between the simulation-predicted vibration suppression effect and the actual effect exceeds a set threshold, the objective function value is increased, forcing the strategy generation model to improve the consistency between virtual and real performance. Through this objective function, multi-objective optimization is achieved, resulting in superior vibration suppression, high processing quality, low energy consumption, and good consistency between virtual and real performance.

[0135] 4) Coupling of Objective Functions and Formation of Game Mechanism. The diagnostic correction objective function is coupled with the strategy optimization objective function. The game relationship between the two models is clarified: the output of the vibration source diagnostic model (vibration source fingerprint) is the input of the vibration suppression strategy generation model, and the feedback of the execution effect of the strategy generation model provides a basis for the optimization of the diagnostic model.

[0136] The game evolution is achieved through alternating optimization: first, the model parameters are generated by fixing the strategy, and the diagnostic correction objective function is optimized to improve the diagnostic accuracy; then, based on the output of the optimized diagnostic model, the strategy is optimized to improve the objective function to enhance the vibration suppression effect; this process is repeated until the changes in both objective functions are less than the set threshold, reaching the game equilibrium state, at which point the two models achieve synergistic optimality.

[0137] Step 4: Collaborative Iterative Update of the Two Models

[0138] 1) Parameter adjustment rules are established. For the vibration source diagnosis model, the parameters of high-dimensional phase space reconstruction (such as delay time and embedding dimension) and key parameters of cross-media fusion attention weights are adjusted according to the gradient direction of the diagnostic correction objective function. For the vibration suppression strategy generation model, the network weights of the generator (such as parameters of convolutional and recurrent layers) and the generation range of strategy parameters are adjusted according to the gradient direction of the strategy optimization objective function. An upper limit for the magnitude of single parameter adjustment is set to avoid sudden changes in model performance.

[0139] 2) Iterative process control. A batch iterative approach is adopted, and an iterative update is performed after accumulating a certain amount of closed-loop feedback data (e.g., 10 processing batches). After each iteration, the performance of the dual model is verified using test case samples in the digital twin simulation environment. If the performance improvement meets the set threshold, the updated model parameters are saved. If the performance does not improve or decreases, the optimization direction is adjusted back and the iteration is re-executed.

[0140] 3) Equilibrium state determination. When the performance indicators (diagnostic accuracy and vibration suppression effect compliance rate) of the two models change by less than the set threshold after three consecutive iterations, and the consistency deviation between simulation and actual execution is less than the threshold, the game equilibrium state is determined to be reached, and the current iteration update is paused.

[0141] In this optional embodiment, based on the performance deviation characteristics, a diagnostic correction objective function for optimizing the vibration source diagnostic model and a strategy optimization objective function for optimizing the vibration suppression strategy generation model are constructed respectively, forming a two-way game optimization mechanism including the following steps:

[0142] Based on performance deviation characteristics, the vibration source identification error vector is extracted;

[0143] Using the vibration source identification error vector as input, a diagnostic correction objective function for the vibration source diagnostic model is constructed.

[0144] Based on multi-objective performance indicators including vibration residual energy, processing quality and control energy consumption, and combined with the consistency requirements of simulation and actual execution, a strategy optimization objective function for the vibration suppression strategy generation model is constructed.

[0145] The diagnostic correction objective function is coupled with the strategy optimization objective function, and the vibration source diagnostic model and the vibration suppression strategy generation model are driven to co-evolve in an adversarial manner through alternating optimization until an equilibrium state is reached.

[0146] In this optional embodiment, the vibration characteristics, strategy parameters, and performance indicators acquired throughout the process are stored in a structured manner, and a vibration suppression decision knowledge graph is constructed, including the following steps:

[0147] The vibration source fingerprint, processing conditions, vibration suppression strategy parameters, and corresponding multi-dimensional performance indicators generated throughout the entire process from vibration source diagnosis and strategy generation to the execution of the vibration suppression execution unit are obtained to form vibration suppression event data.

[0148] Specifically, the vibration suppression event data covers the entire process of vibration source diagnosis, strategy generation, and vibration suppression execution, including vibration source fingerprint (type, location, energy contribution), vibration suppression strategy parameters (execution unit type, control parameters, timing), real-time machining conditions (material properties, spindle speed, feed rate, etc.), and multi-dimensional performance indicators after execution (vibration attenuation rate, machining accuracy, energy consumption, etc.).

[0149] The vibration suppression event data are sequentially formatted and feature aligned, and causal and triggering relationships are established between the vibration suppression event data.

[0150] Based on the causal and triggering relationships between vibration suppression event data, a vibration suppression decision knowledge graph is constructed with processing conditions, vibration source type, vibration suppression strategy, and performance results as the core quadruple.

[0151] In subsequent collaborative iterative updates, newly generated vibration suppression event data will be injected into the vibration suppression decision knowledge graph.

[0152] S5. Using the vibration suppression decision knowledge graph, the vibration source diagnosis model and the vibration suppression strategy generation model updated through collaborative iteration, vibration suppression decisions are made on the real-time acquired multimodal signal data, generating an executable vibration suppression strategy that adapts to the current processing conditions, and driving the vibration suppression execution unit to execute it.

[0153] It should be explained that this step integrates the advantages of the evolved dual models and knowledge graph through a full-link decision-making process of real-time diagnosis, knowledge retrieval, strategy adjustment and simulation screening, to achieve accurate adaptation to the current working conditions and efficient vibration suppression, forming a dual guarantee mechanism of knowledge reuse and model optimization.

[0154] In this optional embodiment, a vibration suppression decision knowledge graph, a collaboratively iteratively updated vibration source diagnostic model, and a vibration suppression strategy generation model are used to make vibration suppression decisions on the real-time acquired multimodal signal data, generate an executable vibration suppression strategy adapted to the current processing conditions, and drive the vibration suppression execution unit to execute the following steps:

[0155] S51. Using the collaboratively iteratively updated vibration source diagnostic model, the real-time acquired multimodal signal data is diagnosed to obtain the real-time vibration source fingerprint;

[0156] S52. Vectorize the real-time vibration source fingerprint and the current processing condition to form a multi-dimensional semantic retrieval key; based on the multi-dimensional semantic retrieval key, search the vibration suppression decision knowledge graph to obtain historical vibration suppression strategies.

[0157] Specifically, the knowledge graph retrieval calls the vibration suppression decision knowledge graph and adopts a composite retrieval strategy of similarity matching and association reasoning. First, it performs preliminary screening based on the similarity between the search key and the processing condition-vibration source fingerprint entity in the graph, retaining the top 5 related entries with the highest similarity. Then, it performs reasoning verification through the causal relationship in the graph (such as the association link between a certain vibration source type and the adaptation strategy), eliminates logically inconsistent entries, and obtains the historical vibration suppression strategy.

[0158] S53. Based on the differences between the current processing conditions and historical processing conditions, the historical vibration suppression strategy is adaptively adjusted to generate an initial candidate strategy set;

[0159] S54. In the digital twin simulation environment, the initial candidate strategy is used as a priori guide and input into the vibration suppression strategy generation model after collaborative iterative update. The executable vibration suppression strategy is then selected and driven to execute by the vibration suppression execution unit.

[0160] According to one embodiment of the present invention, such as Figure 2 As shown, a multimodal vibration suppression system for vertical grinding machine machining based on digital twin is also provided. The system includes: a twin synchronization and simulation module 1, a vibration source diagnosis module 2, a strategy generation and execution module 3, a game evolution and knowledge construction module 4, and an adaptive execution module 5.

[0161] The twin synchronization and simulation module 1 is used to acquire multimodal signal data of the vertical grinding machine and perform synchronous digital twin mapping processing on the multimodal signal data to form a digital twin simulation environment;

[0162] Vibration source diagnostic module 2 is used to construct a vibration source diagnostic model based on high-dimensional phase space reconstruction technology and cross-media feature fusion technology; input multimodal signal data into the vibration source diagnostic model and output vibration source fingerprint;

[0163] Strategy generation and execution module 3 is used to construct a vibration suppression strategy generation model. Based on the digital twin simulation environment, the vibration source fingerprint and real-time processing conditions are input into the vibration suppression strategy generation model to generate the optimal vibration suppression strategy and drive the vibration suppression execution unit to execute it.

[0164] The Game Evolution and Knowledge Construction Module 4 is used to collaboratively iteratively update the vibration source diagnosis model and the vibration suppression strategy generation model based on the bidirectional game evolution mechanism and the multimodal signal data after the vibration suppression execution unit is executed; the vibration characteristics, strategy parameters and performance indicators obtained throughout the process are structured and stored to construct a vibration suppression decision knowledge graph.

[0165] The adaptive execution module 5 is used to make vibration suppression decisions on real-time acquired multimodal signal data by utilizing the vibration suppression decision knowledge graph, the collaboratively iteratively updated vibration source diagnosis model and vibration suppression strategy generation model, generate executable vibration suppression strategies that are adapted to the current processing conditions, and drive the vibration suppression execution unit to execute them.

[0166] In summary, by utilizing the above-mentioned technical solutions of this invention, the present invention constructs a structured vibration source fingerprint by introducing a high-dimensional phase space reconstruction and cross-media feature fusion mechanism. This enables the identification of vibration root causes from the essential level of nonlinear dynamics, accurately capturing the nascent flutter before traditional spectral indicators show significant anomalies, and significantly improving the foresight and sensitivity of vibration source diagnosis. This invention proposes a vibration suppression strategy generation model architecture, placing the strategy generator and the twin simulation discriminator in a digital twin environment for adversarial game theory. This allows the vibration suppression strategy to undergo multi-objective verification in virtual space before being issued for execution, ensuring physical security while avoiding processing interruptions and tool wear caused by traditional trial and error. This invention designs a bidirectional game evolution mechanism and a vibration suppression decision knowledge graph, enabling the vibration source diagnosis model and the strategy generation model to continuously optimize collaboratively during actual operation. Effective experience is structured and precipitated into a quadruple of operating condition-vibration source-strategy-performance. The vibration suppression decision knowledge graph supports intelligent retrieval and migration of historical strategies, effectively solving the cold start problem under new operating conditions. This invention effectively solves the technical challenge of synergistically suppressing multi-source composite vibrations in high-precision, heavy-load, and variable-condition machining of vertical grinding machines by deeply integrating digital twins, nonlinear dynamics, cross-media fusion, generative adversarial learning, causal knowledge graphs, and multi-objective optimization, thereby significantly improving the surface quality, equipment stability, and production efficiency of the machined parts.

[0167] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for multimodal vibration suppression in vertical grinding machine machining based on digital twins, characterized in that, The method includes: S1. Acquire multimodal signal data of the vertical grinding machine and perform synchronous digital twin mapping processing on the multimodal signal data to form a digital twin simulation environment; S2. Based on high-dimensional phase space reconstruction technology and cross-media feature fusion technology, construct a vibration source diagnostic model; input multimodal signal data into the vibration source diagnostic model and output vibration source fingerprint; S3. Construct a vibration suppression strategy generation model; Based on the digital twin simulation environment, input the vibration source fingerprint and real-time processing conditions into the vibration suppression strategy generation model to generate the optimal vibration suppression strategy and drive the vibration suppression execution unit to execute it. S4. Obtain multimodal signal data after the vibration suppression execution unit has executed, and construct a closed-loop feedback dataset containing vibration suppression strategy parameters and processing conditions; evaluate the vibration suppression effect based on the closed-loop feedback dataset and extract performance deviation features; based on the performance deviation features, construct a diagnostic correction objective function for optimizing the vibration source diagnostic model and a strategy optimization objective function for optimizing the vibration suppression strategy generation model, forming a two-way game optimization mechanism; based on the two-way game optimization mechanism, perform collaborative iterative updates on the vibration source diagnostic model and the vibration suppression strategy generation model; and structurally store the vibration features, strategy parameters, and performance indicators obtained throughout the process to construct a vibration suppression decision knowledge graph; wherein, the two-way game optimization mechanism includes: Based on performance deviation characteristics, the vibration source identification error vector is extracted; Using the vibration source identification error vector as input, a diagnostic correction objective function for the vibration source diagnostic model is constructed. Based on multi-objective performance indicators including vibration residual energy, processing quality and control energy consumption, and combined with the consistency requirements of simulation and actual execution, a strategy optimization objective function for the vibration suppression strategy generation model is constructed. The diagnostic correction objective function is coupled with the strategy optimization objective function, and the vibration source diagnostic model and the vibration suppression strategy generation model are driven to co-evolve in the adversarial process through alternating optimization until an equilibrium state is reached. S5. Using the vibration suppression decision knowledge graph, the vibration source diagnosis model and the vibration suppression strategy generation model updated through collaborative iteration, vibration suppression decisions are made on the real-time acquired multimodal signal data, generating an executable vibration suppression strategy that adapts to the current processing conditions, and driving the vibration suppression execution unit to execute it.

2. The method for multimodal vibration suppression in vertical grinding machine machining based on digital twins according to claim 1, characterized in that, The process of acquiring multimodal signal data from a vertical grinding machine and performing synchronous digital twin mapping processing on the multimodal signal data to form a digital twin simulation environment includes the following steps: S11. Obtain the geometric structure, motion relationship and dynamic characteristics of the vertical grinding machine to form its physical properties; S12. Using multimodal sensors arranged in key parts of the vertical grinding machine, vibration signals, thermal imaging signals and acoustic signals are acquired in real time to form multimodal signal data; S13. Perform real-time synchronous digital twin mapping processing on multimodal signal data and physical characteristics to form a real-time synchronous digital twin simulation environment.

3. The method for multimodal vibration suppression in vertical grinding machine machining based on digital twins according to claim 1, characterized in that, The process of constructing a vibration source diagnostic model based on high-dimensional phase space reconstruction technology and cross-media feature fusion technology, and inputting multimodal signal data into the vibration source diagnostic model to output the vibration source fingerprint includes the following steps: S21. Extract vibration signals from multimodal signal data, and use high-dimensional phase space reconstruction technology to perform phase space mapping on the vibration signals to obtain vibration dynamic characteristics; S22. Extract thermal imaging signals and acoustic signals from multimodal signal data, perform time-series analysis and frequency domain transformation respectively, and obtain thermal evolution characteristics and acoustic spectrum characteristics; S23. Based on the attention mechanism, cross-media feature fusion of vibration dynamics features, thermal evolution features and acoustic spectrum features is performed to generate a fused feature vector; S24. Match the fused feature vector with the pre-built vibration source physical map library to output a vibration source fingerprint containing the vibration source type, spatial location and energy contribution.

4. The method for multimodal vibration suppression in vertical grinding machine machining based on digital twins according to claim 3, characterized in that, The process of extracting vibration signals from multimodal signal data and using high-dimensional phase space reconstruction technology to perform phase space mapping on the vibration signals to obtain vibration dynamic characteristics includes the following steps: S211. Extract vibration signals from multimodal signal data, and perform noise reduction and normalization processing on the vibration signals in sequence to obtain preprocessed vibration signals. S212. Using the delayed coordinate embedding method, the preprocessed vibration signal is reconstructed in phase space to obtain the trajectory point set in high-dimensional phase space. S213. Perform attractor morphology analysis on the trajectory point set, extract attractor geometric features, and calculate the Lyapunov exponent of the trajectory point set. S214. The attractor geometric features are combined with the Lyapunov index to obtain the vibration dynamics features.

5. The method for multimodal vibration suppression in vertical grinding machine machining based on digital twins according to claim 1, characterized in that, The vibration suppression strategy generation model includes: a strategy generator, a twin simulation discriminator, and a strategy parsing distributor; The strategy generator is used to generate a variety of vibration suppression strategy combinations in a digital twin simulation environment based on the vibration source fingerprint and real-time processing conditions. The twin simulation discriminator is used to perform multi-objective performance simulation evaluation of multiple vibration suppression strategy combinations based on a digital twin simulation environment, and output the optimal vibration suppression strategy. The strategy parser and distributor is used to parse the optimal vibration suppression strategy into multi-channel collaborative control commands and send them to the vibration suppression execution unit to perform vibration suppression operations.

6. The method for multimodal vibration suppression in vertical grinding machine machining based on digital twins according to claim 1, characterized in that, The process of constructing a vibration suppression decision knowledge graph by structuring and storing the vibration characteristics, strategy parameters, and performance indicators acquired throughout the process includes the following steps: The vibration source fingerprint, processing conditions, vibration suppression strategy parameters, and corresponding multi-dimensional performance indicators generated throughout the entire process from vibration source diagnosis and strategy generation to the execution of the vibration suppression execution unit are obtained to form vibration suppression event data. The vibration suppression event data are sequentially formatted and feature aligned, and causal and triggering relationships are established between the vibration suppression event data. Based on the causal and triggering relationships between vibration suppression event data, a vibration suppression decision knowledge graph is constructed with processing conditions, vibration source type, vibration suppression strategy, and performance results as the core quadruple. In subsequent collaborative iterative updates, newly generated vibration suppression event data will be injected into the vibration suppression decision knowledge graph.

7. The method for multimodal vibration suppression in vertical grinding machine machining based on digital twins according to claim 1, characterized in that, The process of utilizing a vibration suppression decision knowledge graph, a collaboratively iteratively updated vibration source diagnosis model, and a vibration suppression strategy generation model to perform vibration suppression decisions on real-time acquired multimodal signal data, generating an executable vibration suppression strategy adapted to the current processing conditions, and driving the vibration suppression execution unit to execute the following steps: S51. Using the collaboratively iteratively updated vibration source diagnostic model, the real-time acquired multimodal signal data is diagnosed to obtain the real-time vibration source fingerprint; S52. Vectorize the real-time vibration source fingerprint and the current processing condition to form a multi-dimensional semantic retrieval key; based on the multi-dimensional semantic retrieval key, search the vibration suppression decision knowledge graph to obtain historical vibration suppression strategies. S53. Based on the differences between the current processing conditions and historical processing conditions, the historical vibration suppression strategy is adaptively adjusted to generate an initial candidate strategy set; S54. In the digital twin simulation environment, the initial candidate strategy is used as a priori guide and input into the vibration suppression strategy generation model after collaborative iterative update. The executable vibration suppression strategy is then selected and driven to execute by the vibration suppression execution unit.

8. A digital twin-based multimodal vibration suppression system for vertical grinding machine machining, used to implement the digital twin-based multimodal vibration suppression method for vertical grinding machine machining as described in any one of claims 1-7, characterized in that, The system includes: a twin synchronization and simulation module, a vibration source diagnosis module, a strategy generation and execution module, a game evolution and knowledge construction module, and an adaptive execution module; The twin synchronization and simulation module is used to acquire multimodal signal data of the vertical grinding machine and perform synchronous digital twin mapping processing on the multimodal signal data to form a digital twin simulation environment; The vibration source diagnostic module is used to construct a vibration source diagnostic model based on high-dimensional phase space reconstruction technology and cross-media feature fusion technology; it inputs multimodal signal data into the vibration source diagnostic model and outputs the vibration source fingerprint. The strategy generation and execution module is used to construct a vibration suppression strategy generation model. Based on the digital twin simulation environment, the vibration source fingerprint and real-time processing conditions are input into the vibration suppression strategy generation model to generate the optimal vibration suppression strategy and drive the vibration suppression execution unit to execute it. The Game Evolution and Knowledge Construction Module is used to collaboratively iteratively update the vibration source diagnosis model and the vibration suppression strategy generation model based on the bidirectional game evolution mechanism and the multimodal signal data after the vibration suppression execution unit is executed; it also stores the vibration characteristics, strategy parameters and performance indicators obtained throughout the process in a structured manner to construct a vibration suppression decision knowledge graph. The adaptive execution module is used to make vibration suppression decisions on real-time acquired multimodal signal data by utilizing the vibration suppression decision knowledge graph, the collaboratively iteratively updated vibration source diagnosis model and vibration suppression strategy generation model, generate executable vibration suppression strategies that are adapted to the current processing conditions, and drive the vibration suppression execution unit to execute them.

Images