A multi-surface special-shaped magnet one-time clamping linkage grinding process

Abstract

The application discloses a one-time clamping linkage grinding process of a multi-surface special-shaped magnet and relates to the technical field of precision manufacturing and automatic processing. The application discloses a one-time clamping linkage grinding process of a multi-surface special-shaped magnet and relates to the technical field of precision manufacturing and automatic processing. The application discloses a one-time clamping linkage grinding process of a multi-surface special-shaped magnet and relates to the technical field of precision manufacturing and automatic processing. The application discloses a one-time clamping linkage grinding process of a multi-surface special-shaped magnet and relates to the technical field of precision manufacturing and automatic processing. The application discloses a one-time clamping linkage grinding process of a multi-surface special-shaped magnet and relates to the technical field of precision manufacturing and automatic processing. The application discloses a one-time clamping linkage grinding process of a multi-surface special-shaped magnet and relates to the technical field of precision manufacturing and automatic processing. The application discloses a one-time clamping linkage grinding process of a multi-surface special-shaped magnet and relates to the technical field of precision manufacturing and automatic processing. The application discloses a one-time clamping linkage grinding process of a multi-surface special-shaped magnet and relates to the technical field of precision manufacturing and automatic processing. The application discloses a one-time clamping linkage grinding process of a multi-surface special-shaped magnet and relates to the technical field of precision manufacturing and automatic processing. The application discloses a one-time clamping linkage grinding process of a multi-

Description

Technical Field

[0001] This invention relates to the technical field, specifically to a one-time clamping and linkage grinding process for multifaceted irregularly shaped magnets. Background Technology

[0002] For multi-faceted irregular magnets, when the workpiece size is too large, the structure is too complex, or the existing five-axis equipment is limited, problems such as insufficient machine tool travel, mutual interference, or overly complex fixture design may prevent all surfaces of the workpiece from being machined in one clamping. In this case, sequential machining is required, that is, the workpiece needs to be clamped or rotated multiple times. However, the traditional sequential machining process has the following problems. 1) Precision failure caused by thermo-mechanical coupling: The local high heat (instantaneous temperature can reach hundreds of degrees Celsius) and mechanical stress generated during the grinding process will change the micro-stress state of the magnet, causing unpredictable local deformation. When the other side is processed in the subsequent process, the stress released or the accumulated thermal deformation of the previously processed surface will be "transferred" to the new clamping state, resulting in the final cumulative deformation exceeding the tolerance. This kind of error is difficult to predict and compensate for by traditional offline measurement. 2) Fixed and suboptimal processing sequence: Traditional processes rely on experience to determine the processing sequence (e.g., "large surfaces first, then small surfaces"). However, for complex irregular parts, different processing sequences have vastly different effects on the overall rigidity, cumulative thermal deformation, and final accuracy of the workpiece, lacking a scientific and adaptive decision-making method; 3) Black box of process status: After each clamping, the key physical states (such as grinding force, spindle power, and local temperature of the workpiece) are not effectively monitored and correlated with the processing results (dimensions, chipping) in real time. The adjustment of the process window is lagging and blind. 4) Uncontrollable propagation of chipping damage: chipping on one surface may become a new stress concentration point during clamping in subsequent processes, or it may expand due to the force applied during flipping clamping. Damage control lacks a global perspective. Summary of the Invention

[0003] The purpose of this invention is to provide a one-time clamping and grinding process for multifaceted irregular magnets, in order to solve the problems mentioned in the background art, such as unstable final accuracy, high risk of edge breakage, and long process debugging cycle caused by the accumulation of thermal-force deformation, solidification of processing sequence, and unknown process state when multifaceted irregular magnets need to be clamped or rotated multiple times.

[0004] To achieve the above objectives, the present invention provides the following technical solution: a one-time clamping and linkage grinding process for multifaceted irregularly shaped magnets, comprising the following steps: S1: Initial data acquisition: Acquire the physical parameters of the 3D CAD model of the multifaceted irregular magnet to be processed and the specific structural model of the one-time clamping fixture, and establish an accurate geometric expression; S2: Establish a digital twin simulation model: Extract the key physical properties and initial stress state of the multifaceted irregular magnet to be processed based on physical parameters, and establish a grinding force model, a grinding heat model and a cooling model. The key physical properties include elastic modulus, Poisson's ratio, coefficient of thermal expansion, specific heat capacity and thermal conductivity. S3: Process pre-planning: Based on digital twin simulation, a set of candidate process routes containing multiple processing sequences and clamping schemes is generated; S4: First-order processing and processing data acquisition: The first process is executed according to the candidate process route set. During the processing, the temperature is adjusted by the active temperature control module to generate a temperature control strategy log, the process signal monitoring module collects the process signal timing data, and the processing result is detected after the first process is completed. The actual process parameters, timing data, temperature control strategy log and processing result are associated to realize the data acquisition of the processing process and form a structured and traceable "process fingerprint" of the process. S5: Adaptive Decision Loop: Before each subsequent process begins, based on the actual workpiece state represented by all previous "process fingerprints" and combined with the current initial data acquisition of the multifaceted irregular magnet and the digital twin simulation model, dynamically optimize the current process's processing route, process parameters, and temperature control strategy, then execute the processing and form a new "process fingerprint". S6: Iterative learning: Based on the "process fingerprint" dataset of multiple batch processing, the process model and benchmark parameters are optimized through machine learning to form an adaptive process knowledge package.

[0005] In a preferred embodiment, the present invention can be further configured as follows: "Process fingerprint" is a structured, associative, and traceable data object containing input data, process status data, and output result data, which fully encapsulates the entire chain of information from "input" to "output" of a single grinding process. The method for generating the candidate process route set includes simulating the workpiece deformation and stress distribution under different processing sequences through finite element analysis, and selecting multiple non-dominant solutions to form a candidate process route set with multiple objectives of minimizing cumulative deformation and maximizing processing stability.

[0006] In a preferred embodiment, the present invention may be further configured such that the process signal monitoring module includes at least a force sensor for monitoring triaxial grinding force, a temperature sensor for monitoring the local temperature field of the workpiece, and a power sensor for monitoring the spindle load.

[0007] In a preferred embodiment, the present invention can be further configured as follows: the active temperature control module is a multi-channel independent temperature control jet system, which, based on process instructions, performs fixed-point, fixed-temperature, and fixed-flow impact cooling or heating on specific local areas of the workpiece to regulate the overall temperature field distribution during the workpiece processing.

[0008] In a preferred embodiment, the present invention may be further configured such that the "process fingerprint" includes at least the actual process parameters used in the process, the real-time timing data collected by the process signal monitoring module, the execution strategy log of the active temperature control module, and the measured dimensions / form and position errors and surface damage status information of the workpiece after the process is completed.

[0009] In a preferred embodiment, the present invention can be further configured as follows: the dynamic optimization selection method includes taking the final measured workpiece state of the previous process as input condition, performing rapid deformation simulation on the remaining candidate process routes, selecting the route with the smallest predicted cumulative deformation for execution, and adaptively adjusting the cutting parameters and cooling intensity of the current process according to the signal characteristics in the previous process fingerprint.

[0010] In a preferred embodiment, the present invention may be further configured such that S4 also includes a "transition surface special treatment" process, which automatically uses a smaller depth of cut, a higher grinding wheel linear speed, and stronger cooling protection for the edge region adjacent to the machined surface in the current process, compared to other regions, to prevent secondary damage to the machined edge.

[0011] In a preferred embodiment, the present invention can be further configured such that: the adaptive decision loop includes automatically reducing the depth of cut or feed rate in the current process if the preceding "process fingerprint" shows that the grinding force or spindle power is consistently higher than the reference range; and attempting to make positive efficiency fine-tuning on the reference parameters for similar features in the current process if the preceding temperature control is good and there is no damage.

[0012] In a preferred embodiment, the present invention can be further configured as follows: iterative learning is based on using regression analysis, neural networks or reinforcement learning algorithms to mine a historical "process fingerprint" database, establish a predictive model from process signal features to processing quality results, and use the model to optimize the initial benchmark parameter set and candidate route generation logic for subsequent processing.

[0013] A CNC grinding system, comprising: The main body of the CNC grinding machine includes a process decision and control module, a multi-sensor fusion monitoring module, and an active temperature-controlled jet execution module; The process decision and control module includes a digital twin simulation unit, which performs initial data acquisition (S1) and establishes a digital twin simulation model (S2); a process planning and optimization unit, which generates a set of candidate process routes based on the digital twin simulation to achieve process pre-planning (S3); an adaptive decision unit, which dynamically optimizes the processing route, process parameters, and temperature control strategy based on the preceding "process fingerprint" during processing, and performs an adaptive decision loop (S5); and an iterative learning and knowledge management unit, which optimizes the process model and parameters based on multiple batches of "process fingerprint" data, achieves iterative learning (S6), and forms an adaptive process knowledge base. The multi-sensor fusion monitoring module is used to synchronously collect process signal timing data during the machining process, providing process status data for the formation of the first-order machining and machining process data acquisition S4. It includes a three-dimensional grinding force monitoring unit integrated on the CNC grinding machine body, used to collect grinding force timing signals in real time; a temperature field monitoring unit, used to collect local temperature field data of the workpiece in real time; and a spindle load monitoring unit, used to collect spindle power or current signals in real time. The active temperature control jet execution module is used to execute active temperature control strategies during processing and dynamically adjust temperature control actions based on real-time data in the adaptive decision loop S5. It includes a multi-channel independent temperature control jet unit integrated on the CNC grinding machine body, with jet nozzles that can independently adjust temperature, flow rate and direction; and a jet control and drive unit, which is used to receive process instructions and drive the jet nozzles to perform fixed-point, fixed-temperature and fixed-flow cooling or heating on specific areas of the workpiece.

[0014] Compared with the prior art, the beneficial effects of the present invention are: This invention achieves dynamic prediction and active control of workpiece thermal-mechanical deformation by integrating digital twin simulation with "process fingerprint" data acquisition in actual processing. Before each process begins, the system performs rapid finite element simulation based on the actual processing state of the previous process, and re-plans the subsequent processing route and process parameters, thereby cutting off the error transmission chain caused by multiple clamping and heat accumulation. At the same time, the active temperature control module can perform fixed-point, fixed-temperature cooling on local areas of the workpiece, effectively suppressing grinding heat accumulation, reducing stress release and deformation caused by temperature gradient, and improving the cumulative accuracy stability after multi-sequence processing. This invention employs a candidate process route set generation method based on finite element analysis and multi-objective optimization. In the process pre-planning stage, it simulates the deformation and stress distribution under various processing sequences and selects multiple non-dominant solutions. During actual processing, the system can dynamically select the current optimal processing route based on the actual workpiece state reflected by the previous "process fingerprint" to achieve adaptive adjustment of the processing sequence. This decision-making mechanism based on real-time data breaks the limitations of traditional reliance on fixed experience sequences and can flexibly optimize the processing sequence for different workpiece states, thereby maximizing process efficiency while ensuring processing stability. This invention achieves full-chain transparency and traceability of the processing process by integrating multi-sensor monitoring modules and constructing a "process fingerprint" data system. The input parameters, process signals and output results of each process are structured, linked and recorded to form analyzable data objects. This transforms the adjustment of the process window from lagging and blind experience-based debugging to precise response based on real-time data. The system can instantly adapt decision-making cycles and cooling strategies based on process signal characteristics, thereby improving process controllability and consistency. This invention introduces a "transition surface special treatment" strategy in the adaptive decision-making loop. It uses a smaller cutting depth, a higher grinding wheel linear speed, and stronger cooling protection to refine the machined edge areas, effectively preventing secondary damage. At the same time, the system continuously monitors the edge chipping status through "process fingerprinting" and dynamically adjusts the clamping and processing strategies based on historical damage data in subsequent processes. This suppresses the propagation and expansion of damage from a global perspective. Combined with an iterative learning mechanism, the system can continuously optimize process parameters and route selection, gradually forming a damage-resistant process knowledge package for specific irregular magnets, thereby systematically reducing the risk of edge chipping. Attached Figure Description

[0015] Figure 1 This is a flowchart of a one-time clamping and linkage grinding process for a multi-faceted irregularly shaped magnet according to the present invention; Figure 2 This is a flowchart of the process pre-planning of the present invention; Figure 3 This is a schematic diagram of the "process fingerprint" module of the present invention. Detailed Implementation

[0016] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0017] See Figure 1 This invention provides an embodiment of a one-time clamping linkage grinding process for multi-faceted irregularly shaped magnets. The workpiece to be processed is a cuboid-based NdFeB magnet with multiple features. All features to be processed after one clamping include: a top surface (surface A), a bottom surface (surface B), four side surfaces (surfaces C1, C2, C3, C4), an irregularly shaped groove penetrating the side surfaces (feature D), and a curved boss on the top surface (feature E). All features must be processed in one clamping operation by linkage control of the machine tool spindle and... The worktable / rotary table completes all grinding operations. The machining requirements are: dimensional accuracy of each surface ±0.003mm, perpendicularity of adjacent surfaces ≤0.005mm, and no visible chipping on any edges ≤0.01mm. The machining system uses a CNC precision grinding machine with multi-axis linkage, equipped with a tilting rotary worktable (B / C axes) or a similar composite rotary table. The workpiece is clamped onto this rotary worktable in one go using a special fixture. The multi-axis linkage of the machine tool ensures the tool's accessibility to all surfaces to be machined, including the following steps: S1: Initial Data Acquisition: Obtain the precise 3D CAD model of the magnet blank and fixture, and complete the virtual clamping in the software. Clarify the fixed position and attitude of the workpiece relative to the machine tool coordinate system under this clamping, as well as the geometric description and spatial orientation of all features (AEs) to be processed. Extract the physical property parameters of the workpiece material: elastic modulus E=160GPa, Poisson's ratio ν=0.25, coefficient of thermal expansion α=6.0e-6 / K, specific heat capacity c=460J / (kg·K), thermal conductivity k=8.5W / (m·K). Obtain the initial residual stress distribution data of the blank through non-destructive testing. S2: Establish a digital twin simulation model: Based on the data from step S1, construct a digital twin simulation environment that includes the workpiece-fixture-machine tool motion chain, and integrate the models. Kinematic model: Describes the motion relationships of each axis of the machine tool and verifies that there is no collision or interference in all machining paths; Multi-faceted grinding force / thermal model: Establish corresponding grinding force and grinding heat generation models for different feature surfaces, such as planes, curved surfaces, and groove walls, as moving loads and heat sources; Structure-thermal coupling model: simulates the stress redistribution and cumulative deformation inside the workpiece due to sequential heat input and material removal during a single clamping and continuous machining of different surfaces; like Figure 2 S3: Process Pre-planning: The core task is to plan the optimal machining sequence, machining path, and corresponding machine tool linkage trajectory for all surfaces to be machined (AE) under a single clamping constraint, and to perform virtual machining simulation using a digital twin model. 1. Generate a candidate processing sequence set: Based on geometric reachability and technological constraints, automatically generate multiple feasible total order processing sequence solutions, for example: Sequence a1: Rough grinding of top surface A → Rough grinding of bottom surface B → Fine grinding of top surface A → Machining of four sides C1-C4 → Machining of groove D → Fine grinding of curved surface boss E; Sequence a2: Rough grinding of bottom surface B (establishing datum) → Machining of four sides C1-C4 → Machining of groove D → Rough and fine grinding of top surface A and boss E; Sequence a3: Rough grinding of top surface A → Machining of two side surfaces C1, C2 → Machining of groove D → Machining of the other two side surfaces C3, C4 → Fine grinding of top surface A and boss E → Fine grinding of bottom surface B; The candidate machining sequence set is generated using a graph search algorithm based on geometric constraints, employing a framework that combines constraint methods with connectivity graph methods. This algorithm has been successfully applied in the field of robot assembly sequence planning because it can explicitly handle geometric interference constraints and connectivity constraints, making it highly suitable for the automatic generation of machining sequences. 2. Multi-objective simulation and optimization: Perform a complete thermo-mechanical coupled temporal simulation for each candidate sequence. Evaluation metrics include: the final cumulative deformation of the workpiece after the entire machining process, the change in the overall stiffness of the workpiece during machining, the peak temperature and its distribution, and the stress concentration risk at critical edges. 3. Screening of Candidate Process Routes: Using a multi-objective decision-making method, 3-4 "non-dominant solution" sequences that demonstrate balanced performance in suppressing cumulative deformation, controlling thermal effects, and maintaining stability are selected to form the initial candidate process route set. Each candidate route includes detailed processing steps, baseline process parameters for each step, and theoretical temperature control strategies. S4: Initial Machining and Process Data Acquisition: Select a sequence from the candidate process route set, for example, sequence a1, to begin actual machining. The machine tool runs according to the first step of a1, rough grinding the top surface A program. The multi-sensor fusion monitoring module works throughout the process: a triaxial force sensor monitors the grinding force, an infrared thermal imager monitors the workpiece surface, especially the area to be machined, the temperature field, and a power sensor monitors the spindle load. During the rough grinding of surface A, the active temperature control jet system, according to instructions, performs point cooling on the side area of ​​the workpiece at each pass interval, pre-controlling the substrate for subsequent side machining. After the rough grinding of surface A is completed, without changing the clamp, the flatness and key dimensions of surface A are quickly measured on the machine using the machine tool probe. At the same time, the vision system checks the condition of the machined edges. All data—including actual process parameters, process force / heat / power signals, temperature control logs, and measured geometric results of surface A—are linked and packaged to form the first "process fingerprint FP1". FP1 not only records the machining results of surface A, but more importantly, it characterizes the actual physical state of the workpiece after the first process, including initial stress release, minor deformation, and temperature distribution. Specifically, during this process, real-time process signal timing data is collected synchronously, including three-dimensional grinding force (Fx, Fy, Fz), workpiece monitoring point temperature T(t), and spindle power P(t). Simultaneously, the active jet temperature control module is used to actively pre-cool or maintain a constant temperature in the adjacent area of ​​the recently machined surface or the parent area of ​​the next feature to be machined during the intervals or non-cutting movement segments of the grinding path to suppress heat accumulation. After machining, the actual machining dimensions and form and position errors Δ1 of the process are measured on the machine tool using a contact probe or vision system, and the chipping state is detected. A "process fingerprint" file for the process is established: {actual process parameters, process signal timing data, active temperature control strategy, and final machining result (Δ1, chipping)} are associated and bound. S5: Adaptive Decision Loop: Before executing the second process, such as the rough grinding of the bottom surface B in a1, the system initiates online adaptive decision-making: First, a status assessment is performed. The "process fingerprint FP1" is read. The system learns that after rough grinding of surface A, the flatness is good but there is a measurable micro-deformation trend (such as a 0.004mm central depression). The local temperature during processing has reached 90% of the predetermined threshold. The grinding force spectrum shows that the grinding wheel is in normal condition. Then, dynamic sequence and parameter correction are performed. Taking the actual workpiece state reflected in FP1 as a new starting point, the digital twin model is quickly updated. The system performs rapid resimulation on the remaining candidate machining sequences (the subsequent steps of a1 and other candidate sequences) to predict the final result of continuing to process different sequences from the current state. For example, the simulation may show that continuing to process the bottom surface B according to the original a1 sequence may aggravate the concave deformation of surface A; while switching to the next step of sequence a3 (processing the two sides first) is beneficial to use symmetrical material removal to balance stress and improve the overall deformation state. The system dynamically adjusts the subsequent machining sequence accordingly, and automatically fine-tunes the process parameters of the next process according to the process signal in FP1. For example, because the temperature is close to the threshold, the system decides to slightly reduce the feed rate in the next machining and advance and strengthen the active cooling intensity of the corresponding area. For the edges adjacent to the already machined surface A in the next step, the "transition surface special treatment" strategy is automatically applied, with smaller depth of cut, higher linear speed, and enhanced protective cooling. Finally, execution and iteration are carried out. The next processing step is executed according to the new instructions generated by dynamic decision. After completion, the machine is also measured and a new "process fingerprint FP2" is generated. This closed loop of "processing-measurement-evaluation-decision-reprocessing" continues within a single clamping and runs through the entire process from the second process to the last process. Each "process fingerprint" is a snapshot of the real-time status of the workpiece, driving the continuous optimization of subsequent processing strategies. S6: Iterative Learning and Process Knowledge Package Generation: After completing one clamping and processing of a batch (e.g., 15 pieces) of workpieces, the system collects hundreds of "process fingerprints" covering different processes and workpiece states. The system uses machine learning algorithms to analyze this data. For example, by analyzing the fingerprints of all workpieces in the "machining groove D" process, it is found that when the workpiece body temperature is below a certain critical value and a specific grinding wheel cutting angle is used, the groove edge breakage rate is significantly reduced. Such patterns are transformed into optimization rules and updated in the process knowledge base. For example, the trigger condition for updating "machining groove D" is updated: it can only be executed when the overall average temperature of the workpiece is detected to be below T_crit. Otherwise, the system will automatically insert an additional "global cooling waiting period" or adjust the preceding processing parameters, and finally generate an "adaptive process knowledge package" for one clamping and processing of the specific part.

[0018] The "Adaptive Process Knowledge Package" includes an optimized standard machining sequence, an adaptive process parameter adjustment rule library for each process, a heat-deformation prediction model trained based on historical data, and an optimal active temperature control strategy mapping table. When machining a new batch of the same type of parts, the system uses this knowledge package as a starting point, making its pre-planning more accurate and its adaptive adjustment faster, thereby achieving iterative improvement in machining accuracy and stability in a single setup.

[0019] like Figure 3 The specific components of a "process fingerprint": "Process fingerprint" is a structured, associative, and traceable data object defined in this process. It completely encapsulates the entire chain of information from "input" to "output" of a single grinding process. Its data structure can be compared to a detailed "medical record," containing the following three main categories of data: 1. Input data: Process parameter set: All parameters actually called by the CNC program when executing this process, for example: Cutting parameters: grinding wheel linear speed Vs, table feed rate Vw, depth of cut Ap, grinding strategy; Cooling parameters: temperature, flow rate, spray angle, and timing of coolant in each channel; Clamping information: clamp identification, clamping force setting.

[0020] 2. Process status data: Multi-source sensor timing signals: These are the core feature values ​​of the "fingerprint," recorded synchronously with timestamps; Force signals: waveforms of triaxial grinding forces Fx(t), Fy(t), and Fz(t), including their mean, peak value, variance, and spectral characteristics. For example, the amplitude at a specific frequency can reflect the sharpness of the grinding wheel, the uniformity of the workpiece material, and the risk of chatter. Thermal signal: The temperature T(t) curve of key monitoring points on the workpiece can be used to calculate the maximum temperature, temperature rise rate, and high temperature duration to assess the risk of thermal damage; Power / current signal: The spindle motor power P(t) or current I(t) curve is a macroscopic representation of the grinding load and is corroborated by the grinding force; Active intervention log: Records when, where, and with what parameters the active temperature control system intervened, such as "t=12.3s, apply 5°C jet to area A for 2 seconds".

[0021] 3. Output results data: Geometric quality: After the process is completed, the actual dimensional error Δ and geometric tolerances of the machined surface obtained by online or offline measurement, such as flatness and parallelism.

[0022] Surface / subsurface quality: The results of inspection on the machined surface, such as: Surface roughness Ra, Rz; Statistics on chipped edges size and quantity; Indirect assessment or sampling results of the subsurface damage layer depth.

[0023] A complete "process fingerprint" data package associates and binds the above three categories of data through a unified process ID, timestamp, and workpiece unique code, forming the following logical relationship: "When input parameters are used and processing is performed in the process state, the output result is obtained." The underlying logic of "process fingerprinting": Its underlying logic is a closed-loop intelligent system of "perception-modeling-decision-evolution", which can be divided into four levels: First layer: Data perception and association Logic: In traditional processing, operators can only see the input (program) and the final output (good or bad products), while the intermediate process is a black box. This invention establishes a precise causal relationship between the three by synchronously collecting input, process, and output data. To achieve precise correlation, the system adopts a synchronization mechanism based on the IEEE 1588 Precision Time Protocol (PTP). All sensor (force, temperature, power) data acquisition cards support PTP and are synchronized with the CNC system clock via industrial Ethernet to ensure that the sampling timestamp error of each channel is less than 1 microsecond. The data acquisition software synchronously reads the buffers of each sensor at a fixed frequency of 10kHz to generate timestamp-aligned multidimensional time-series data frames, thereby ensuring the precise correspondence of force, heat, and power signals on the time axis. Example: The system can not only record "edge chipping", but also accurately trace back to the moment when the edge chipping occurred, the Fy force suddenly increased by 50N, the local temperature reached 120°C, and the high-feed grooving action being executed. This transforms vague "experience" into analyzable "evidence".

[0024] Second layer: Status assessment and prediction Logic: The system has a built-in rule base and model based on physical models and statistical learning, which is used to interpret the "process fingerprint" in real time; Real-time diagnosis: When the energy of the force signal spectrum rises abnormally in a certain high-frequency band, the model judges that "micro-blockage of the grinding wheel" or "early stage of flutter" may have occurred, and immediately issues an early warning, or even automatically reduces the speed or triggers cleaning and cooling. Result prediction: By analyzing the process signals in this "fingerprint", such as the stability of force and temperature peak, the system can predict whether the surface roughness or edge chipping probability of this processing is within the acceptable range without waiting for the measurement results. Subsequent impact prediction: In S4, the system takes the measured results from the previous "fingerprint", such as the deformation amount Δ, as new initial conditions and inputs them into the modified finite element model. The simulation predicts how the cumulative deformation of the workpiece will develop if the next surface is processed according to scheme A or scheme B. This changes the decision-making from being based on "ideal drawings" to being based on "the real state of the physical entity".

[0025] Third layer: Adaptive decision-making and control Logic: This layer is where the value of the "fingerprint" is realized. The system dynamically adjusts subsequent actions based on the evaluation of the current and previous "fingerprints". Parameter adaptation: If the current "fingerprint" shows that the grinding force is too large but stable, the next similar process may automatically fine-tune and reduce Ap; if the force value is small and stable, it may try to slightly increase Vw to improve efficiency. Path / sequence adaptation: Based on the prediction results of S4, the system selects the route that best suits the current state of the specific workpiece from multiple predefined candidate process routes. For example, if it is found that the workpiece has warped in a specific direction, the machining surface and clamping method that can partially correct the warping are selected as the next step. Active intervention trigger: The temperature field data recorded in the "fingerprint" is directly used to optimize the active temperature control strategy for the next process, such as strengthening cooling in advance in heat-sensitive areas.

[0026] Fourth layer: Iterative learning and knowledge evolution Logic: This is the ultimate manifestation of system intelligence. A single "fingerprint" is a sample, while a batch of "fingerprints" constitutes a database; Machine learning mining: The system runs analysis algorithms periodically (e.g., after processing 100 pieces) to search for hidden patterns that are difficult for humans to discover in massive "fingerprint" data. For example, cluster analysis may reveal that "when the force signal variance is lower than X and the coolant impact angle is Y, the edge breakage rate approaches zero regardless of the size of Ap".

[0027] The specific methods used are as follows: 1. Data labeling: The system automatically associates the final processing quality, such as dimensional error and edge chipping level, with each "process fingerprint" as a label. For example, if the final inspection of the workpiece is qualified, the "process fingerprints" of all intermediate processes in that batch will be labeled as "good"; if the final inspection is unqualified, the fingerprints of the key processes that cause the deviation will be labeled as "poor" based on the in-machine measurement results. 2. Dataset partitioning: The system sorts all historical "process fingerprint" data in chronological order, with the first 70% used as the training set, the middle 15% as the validation set, and the last 15% as the test set, to simulate the model's ability to predict future unknown data and prevent overfitting. 3. Model Design and Training: Regression analysis: Using multiple linear regression or support vector regression, a quantitative relationship model is established between "process signal characteristics, such as peak force and temperature integral" and "final dimensional error". Neural Network: Construct a fully connected feedforward network containing an input layer (signal features), 2-3 hidden layers (128 neurons per layer, using the ReLU activation function), and an output layer (predicted quality indicators). Supervised learning is performed using the training set data. The optimizer is Adam, and the loss function is mean squared error (MSE). Reinforcement learning: Modeling process parameter adjustment as a Markov decision process. The state is the current "process fingerprint" feature, the action is the parameter fine-tuning amount for the next process, and the reward is the improvement in post-processing quality. Q-learning or policy gradient algorithm is used to train the agent so that it can autonomously learn the optimal parameter adjustment strategy. 4. Model Deployment: The trained prediction model is packaged into a dynamic link library and deployed in the process decision and control module for real-time assessment of process risks and recommendation of parameters; Knowledge base update: These newly discovered patterns are transformed into new process rules or optimized baseline parameters and updated into the system's knowledge base. When processing a new batch for the next time, the system's initial candidate routes and parameters will already be better.

[0028] Process package formation: Finally, for a specific part, the system will generate a unique, data-driven "adaptive process knowledge package". This package not only contains CNC G-code, but also a complete set of "fingerprint" interpretation rules, adaptive strategies and prediction models that match the part, realizing the assetization and inheritance of the process.

[0029] The underlying logic of "process fingerprinting" is essentially to equip the manufacturing system with "sensory nerves" and "cerebral cortex". Sensory nerves, used for data acquisition: perceiving and processing the entire process; The cerebral cortex is the logical layer: through association, diagnosis, prediction, decision-making, and learning, it transforms perceptual information into optimized actions.

[0030] The ultimate goal is to achieve the precision of "one-time clamping" to the greatest extent possible through digital intelligence, even under the physical constraints of sequential processing, and to realize continuous self-optimization of the processing process, thus upgrading the manufacturing paradigm from "digital program control" to "physical information fusion intelligent decision-making".

[0031] The method for generating a candidate process route set specifically includes simulating workpiece deformation and stress distribution under different processing sequences through finite element analysis, and selecting multiple non-dominated solutions to form a candidate process route set with the goal of minimizing cumulative deformation and maximizing processing stability. The generation of the candidate process route set relies on digital twin technology and finite element analysis. First, an accurate three-dimensional model of the workpiece is established, including material properties and initial stress state. The thermo-mechanical coupling deformation and residual stress distribution under different processing sequences are simulated through finite element analysis. With the goal of minimizing cumulative deformation and maximizing processing stability, a multi-objective optimization algorithm, such as the Pareto front, is used to select multiple "non-dominated solutions" to form a candidate process route set. Virtual processing simulation is performed on each candidate solution to evaluate its feasibility and risks under ideal conditions. Force sensors are used to verify the grinding force assumptions in the simulation, temperature sensors are used to verify the accuracy of the thermal model, and power sensors are used to verify the consistency between the spindle load and the simulation.

[0032] In the aforementioned process, "minimizing cumulative deformation" and "maximizing processing stability" are the core indicators for multi-objective optimization, and their specific quantitative definitions are as follows: 1. Quantitative indicators of "cumulative deformation" Cumulative deformation refers to the deviation between the final geometry of a workpiece and the ideal design after multiple processing steps due to the thermo-mechanical coupling effect. In finite element simulation and optimization screening, the following three quantitative indicators are used for comprehensive evaluation: Maximum deformation: refers to the maximum displacement of each node after the entire machining process is completed. It is usually expressed in millimeters. This indicator directly reflects whether the workpiece has experienced local out-of-tolerance deformation. Average deformation: refers to the average displacement of all nodes on the key feature surfaces of the workpiece, such as the datum surface and mating surface. This indicator reflects the overall deformation trend and is used to evaluate the clamping stability of subsequent processes. Deformation field uniformity index: By calculating the standard deviation or gradient of the deformation field, the uniformity of the workpiece deformation distribution is evaluated. The larger the deformation gradient, the more severe the local stress concentration, and the higher the risk of edge chipping in subsequent processing. During the optimization process, the system performs thermo-mechanical coupled time-series simulation on each candidate process route, extracts the above three indicators, and forms a "cumulative deformation comprehensive evaluation value" by weighted summation, which serves as one of the optimization objectives.

[0033] 2. Quantitative definition of "processing stability" Combining the general definition in the manufacturing field with the process characteristics of this invention, "processing stability" is quantified from the following two dimensions: (1) Process stability – refers to the degree of fluctuation in the state of the process system throughout the entire processing process. Quantitative indicators include: Grinding force fluctuation coefficient: The ratio of the standard deviation to the mean of the time-series signal collected by the force sensor, the coefficient of variation, CV. The smaller the value, the smoother the grinding process and the more stable the grinding wheel condition and material removal process. Temperature field uniformity index: The temperature gradient in key areas is monitored by temperature sensors. The larger the temperature gradient, the more severe the thermal stress concentration, the higher the risk of thermal deformation of the workpiece, and the worse the process stability. Spindle power fluctuation rate: reflects the stability of the cutting load and indirectly determines whether there are abnormal conditions such as chatter or material inhomogeneity.

[0034] (2) Production stability – refers to the degree of consistency in processing quality between different batches and different workpieces. Quantitative indicators include: Process capability index: Statistical process control of critical dimensions to evaluate the ability of a process to maintain accuracy in multiple repeated processing steps; Standard deviation of pass rate: Calculates the range of fluctuation of pass rate in multiple batches of processing. The smaller the fluctuation, the higher the production stability.

[0035] In the S3 process pre-planning stage, the system simulates the process state of each candidate route through finite element simulation, extracts the above-mentioned quantitative indicators, and transforms "maximizing processing stability" into specific objective functions such as "minimizing grinding force fluctuation coefficient", "minimizing temperature gradient" and "maximizing process capability index prediction value", which are then incorporated into a multi-objective optimization model for solution.

[0036] The process signal monitoring module includes at least a force sensor for monitoring triaxial grinding force, a temperature sensor for monitoring the local temperature field of the workpiece, and a power / current sensor for monitoring the spindle load. The triaxial grinding force sensor is usually piezoelectric and is mounted on the spindle or grinding wheel head. The temperature sensor is generally an infrared thermal imager or a distributed thermocouple to monitor the temperature field in key areas of the workpiece. The power / current sensor is used to monitor the spindle motor load. The force sensor is based on the piezoelectric effect to convert mechanical force into an electrical signal. The temperature sensor is based on infrared radiation or the thermoelectric effect of thermocouples to collect temperature data in real time. The power sensor calculates real-time power by measuring current and voltage. During operation, the timestamps of all sensors are synchronized to form a multi-dimensional time-series data stream. The force, temperature, and power signals are fused to form a "process fingerprint". The process status is judged in real time by signal characteristics such as force peak and temperature gradient.

[0037] The active temperature control module is a multi-channel independent temperature-controlled jet system. Based on process instructions, it performs point-to-point, temperature-to-flow-rate impact cooling or heating on specific local areas of the workpiece to regulate the overall temperature field distribution during workpiece processing. The active temperature control module consists of a multi-channel independent temperature-controlled jet system, including multiple independently controlled nozzles. The coolant temperature is adjustable from -10°C to 25°C, and the flow rate and spray angle are adjustable. Point-to-point spraying is achieved by auxiliary axis drive. The process decision system generates temperature control instructions based on the temperature distribution, processing path, and material characteristics in the "process fingerprint". During grinding intervals or non-cutting sections, it performs pre-cooling or constant temperature control on areas to be processed or already processed, and adjusts the coolant temperature, flow rate, and spray timing based on real-time temperature data.

[0038] To achieve -10°C low-temperature cooling, the system employs a cascade refrigeration principle. The first-stage refrigeration cycle uses R404A refrigerant to cool an intermediate medium, such as an ethylene glycol aqueous solution, to approximately -15°C. The second-stage refrigeration cycle uses R23 refrigerant, which, based on the pre-cooling in the first stage, exchanges heat with the coolant through an evaporator, further lowering its temperature to -10°C. Heating is achieved through a PTC ceramic heater integrated in the liquid receiver, which can raise the coolant temperature to 25°C. The entire temperature control process is controlled in a closed-loop manner using a PID algorithm, achieving a temperature control accuracy of ±0.5°C. The system's response time, from 25°C to -10°C, is less than 5 minutes, ensuring timely process adjustments. The active temperature control module includes two sets of underlying working logic: a thermal management strategy and an adaptive adjustment. Thermal management is achieved by controlling the local temperature gradient to suppress thermal deformation and thermal damage. Adaptive adjustment optimizes the cooling strategy of the current process based on the temperature history of the preceding process.

[0039] The "process fingerprint" should include at least the actual process parameters used in the process, the real-time timing data collected by the process signal monitoring module, the execution strategy log of the active temperature control module, and the measured dimensions / form and position errors and surface damage status information of the workpiece after the process is completed.

[0040] The dynamic optimization selection method includes taking the final measured workpiece state of the previous process as input conditions, performing rapid deformation simulation on the remaining candidate process routes, selecting the route with the smallest predicted cumulative deformation for execution, and adaptively adjusting the cutting parameters and cooling intensity of the current process based on the signal characteristics in the previous process fingerprint. The dynamic optimization selection method is as follows: taking the "process fingerprint" of the preceding process, measured dimensions, temperature, and damage status as input, and then based on the modified finite element model, rapidly predicting the deformation of the remaining candidate routes, selecting the route with the smallest predicted cumulative deformation, and finally, adaptively adjusting the cutting parameters and cooling intensity based on the characteristics of the preceding signals, such as force and temperature anomalies. Force sensors will be used to determine the grinding wheel condition and material uniformity, temperature sensors will be used to assess the heat-affected zone and guide the cooling strategy, and power sensors will be used to reflect the macroscopic load and assist in judging the process stability.

[0041] This dynamic optimization selection process is implemented by a set of rule-based decision-making logic in the "adaptive decision unit," and its core steps are as follows: Input parsing: The system first reads the "process fingerprint" of the previous process and extracts key status information, including: the geometric deformation of the workpiece after the process is completed, such as flatness error, the highest temperature monitored during the processing, and the spectral stability characteristics of the grinding force signal, which are obtained by performing a fast Fourier transform on the force signal.

[0042] Rapid Simulation and Evaluation: The system uses the actual workpiece state described above as initial conditions and inputs them into the modified digital twin simulation model. Then, it performs rapid finite element simulation on each of the remaining candidate process routes to predict the cumulative deformation of the workpiece if processing continues along that route. The system compares the predicted deformation of all candidate routes and selects the route with the smallest predicted cumulative deformation as the processing route for the current operation.

[0043] Adaptive Fine-Tuning of Parameters: After selecting the machining path, the system fine-tunes the baseline process parameters of the current process based on the force signal characteristics and temperature characteristics in the previous "process fingerprint". Specific rules are as follows: If the grinding force stability index of the previous process, such as the force fluctuation amplitude, is consistently higher than the preset high threshold, indicating excessive machining load or poor grinding wheel condition, the system will automatically reduce the cutting depth by 20% and the feed rate by 10% in the current process to reduce load and avoid damage. If the grinding force stability index of the previous process is lower than the low threshold and the machining temperature is also lower than the safety threshold, indicating good current conditions, the system will attempt to increase the feed rate by 5% in the current process to improve machining efficiency while ensuring quality.

[0044] Temperature control strategy update: The system automatically generates or adjusts active temperature control strategies based on the temperature field data of the previous process and the characteristics of the current process to be processed. For example, it can activate stronger cooling in advance for heat-sensitive areas or adjust the temperature and flow rate of the coolant.

[0045] Through the above steps, the system achieves closed-loop adaptive decision-making based on real-time data.

[0046] S4 also includes a "transition surface special treatment" process. Based on the edge area adjacent to the machined surface in the current process, the system automatically adopts a smaller depth of cut, a higher grinding wheel linear speed, and stronger cooling protection compared to other areas to prevent secondary damage to the machined edge.

[0047] The adaptive decision loop includes automatically reducing the depth of cut or feed rate in the current process if the preceding "process fingerprint" shows that the grinding force or spindle power is consistently higher than the reference range, and attempting to make positive efficiency fine-tuning on the reference parameters for similar features in the current process if the preceding temperature control is good and there is no damage.

[0048] Iterative learning is based on using regression analysis, neural networks, or reinforcement learning algorithms to mine historical "process fingerprint" databases, establish a predictive model from process signal features to processing quality results, and use this model to optimize the initial benchmark parameter set and candidate route generation logic for subsequent processing. The specific working logic of iterative learning consists of five steps: data accumulation, model training, knowledge update, and continuous evolution. Data accumulation collects "process fingerprint" data from multiple batches and processes. Model training uses algorithms such as regression analysis, neural networks, and reinforcement learning to establish a predictive model of "process parameters + process signals → processing quality". Knowledge update optimizes the benchmark process parameter set, updates the candidate route generation logic, and continuously evolves into an "adaptive process knowledge package" for specific parts, realizing the self-optimization and inheritance of the process.

[0049] The "process fingerprint" data is stored in a structured relational database, with "process fingerprint" as the core table. Its main data fields are designed as follows: Fingerprint unique ID: As the primary key, it consists of "part batch number_process number_time stamp" and is used to uniquely identify a process fingerprint; Workpiece unique code: used as an index to trace the complete processing history of a single workpiece from blank to finished product; Current process step number: Records the sequence of this process step in the overall process route; Input parameter set: Stores all input parameters for this process in JSON format, such as grinding wheel linear speed, feed rate, depth of cut, coolant set temperature, etc. Force signal storage path: The storage path points to a time-series database, such as the path to the stress signal data file in InfluxDB, for fast retrieval; Thermal signal storage path: The path that stores the temperature signal data file in the time series database; Power signal storage path: The path that stores the power signal data file in the timing database; Temperature control execution log: Records the execution details of the active temperature control module in JSON format, including the opening / closing time of each nozzle, set temperature, actual flow rate, etc. Processing result measurement values: The in-machine measurement results after the completion of this process are stored in JSON format, such as flatness error, maximum chipping size, etc. Record generation time: timestamp, which records the time when the fingerprint data was entered into the database.

[0050] Through this structured storage method, the system can efficiently retrieve the fingerprint data of all processes of a workpiece using its unique code. It can also analyze the correlation between specific signal features and the final measurement results through SQL queries, providing a data foundation for iterative learning.

[0051] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A one-time clamping and linkage grinding process for multifaceted irregularly shaped magnets, characterized in that: Includes the following steps: S1: Initial data acquisition: Acquire the physical parameters of the 3D CAD model of the multifaceted irregular magnet to be processed and the specific structural model of the one-time clamping fixture, and establish an accurate geometric expression; S2: Establish a digital twin simulation model: Extract the key physical properties and initial stress state of the multifaceted irregular magnet to be processed based on physical parameters, and establish a grinding force model, a grinding heat model and a cooling model. The key physical properties include elastic modulus, Poisson's ratio, coefficient of thermal expansion, specific heat capacity and thermal conductivity. S3: Process pre-planning: Based on digital twin simulation, a set of candidate process routes containing multiple processing sequences and clamping schemes is generated; S4: First-order processing and processing data acquisition: The first process is executed according to the candidate process route set. During the processing, the temperature is adjusted by the active temperature control module to generate a temperature control strategy log, the process signal monitoring module collects the process signal timing data, and the processing result is detected after the first process is completed. The actual process parameters, timing data, temperature control strategy log and processing result are associated to realize the data acquisition of the processing process and form a structured and traceable "process fingerprint" of the process. S5: Adaptive Decision Loop: Before each subsequent process begins, based on the actual workpiece state represented by all previous "process fingerprints" and combined with the current initial data acquisition of the multifaceted irregular magnet and the digital twin simulation model, dynamically optimize the current process's processing route, process parameters, and temperature control strategy, then execute the processing and form a new "process fingerprint". S6: Iterative learning: Based on the "process fingerprint" dataset of multiple batch processing, the process model and benchmark parameters are optimized through machine learning to form an adaptive process knowledge package.

2. The one-time clamping and linkage grinding process for a multi-faceted irregularly shaped magnet according to claim 1, characterized in that: "Process fingerprint" is a structured, associative, and traceable data object that includes input data, process status data, and output result data. It fully encapsulates the entire chain of information from "input" to "output" of a single grinding process. The method for generating the candidate process route set includes simulating the workpiece deformation and stress distribution under different processing sequences through finite element analysis, and selecting multiple non-dominated solutions to form the candidate process route set with multiple objectives of minimizing cumulative deformation and maximizing processing stability.

3. The one-time clamping and linkage grinding process for a multi-faceted irregularly shaped magnet according to claim 1, characterized in that: The process signal monitoring module includes at least a force sensor for monitoring triaxial grinding force, a temperature sensor for monitoring the local temperature field of the workpiece, and a power sensor for monitoring the spindle load.

4. The one-time clamping and linkage grinding process for a multi-faceted irregularly shaped magnet according to claim 1, characterized in that: The active temperature control module is a multi-channel independent temperature control jet system. Based on process instructions, it performs fixed-point, fixed-temperature, and fixed-flow impact cooling or heating on specific local areas of the workpiece to regulate the overall temperature field distribution during the workpiece processing.

5. The one-time clamping and linkage grinding process for a multi-faceted irregularly shaped magnet according to claim 1, characterized in that: The "process fingerprint" should include at least the actual process parameters used in the process, the real-time timing data collected by the process signal monitoring module, the execution strategy log of the active temperature control module, and the measured dimensions / form and position errors and surface damage status information of the workpiece after the process is completed.

6. The one-time clamping and linkage grinding process for a multi-faceted irregularly shaped magnet according to claim 1, characterized in that: The dynamic optimization selection method includes taking the final measured workpiece state of the previous process as input conditions, performing rapid deformation simulation on the remaining candidate process routes, selecting the route with the smallest predicted cumulative deformation for execution, and adaptively adjusting the cutting parameters and cooling intensity of the current process based on the signal characteristics in the previous process fingerprint.

7. The one-time clamping and linkage grinding process for a multi-faceted irregularly shaped magnet according to claim 1, characterized in that: S4 also includes a "transition surface special treatment" process. Based on the edge area adjacent to the machined surface in the current process, the system automatically adopts a smaller depth of cut, a higher grinding wheel linear speed, and stronger cooling protection compared to other areas to prevent secondary damage to the machined edge.

8. The one-time clamping and linkage grinding process for a multi-faceted irregularly shaped magnet according to claim 1, characterized in that: The adaptive decision loop includes automatically reducing the depth of cut or feed rate in the current process if the previous "process fingerprint" shows that the grinding force or spindle power is consistently higher than the reference range, and attempting to make positive efficiency fine-tuning on the reference parameters for similar features in the current process if the previous temperature control is good and there is no damage.

9. The one-time clamping and linkage grinding process for a multi-faceted irregularly shaped magnet according to claim 1, characterized in that: Iterative learning is based on using regression analysis, neural networks or reinforcement learning algorithms to mine historical "process fingerprint" databases, establish a predictive model from process signal features to processing quality results, and use the model to optimize the initial benchmark parameter set and candidate route generation logic for subsequent processing.

10. A CNC grinding system for implementing a one-time clamping and linkage grinding process for a multi-faceted irregularly shaped magnet as described in any one of claims 1-9, characterized in that, include: The main body of the CNC grinding machine includes a process decision and control module, a multi-sensor fusion monitoring module, and an active temperature-controlled jet execution module; The process decision and control module includes a digital twin simulation unit, which performs initial data acquisition (S1) and establishes a digital twin simulation model (S2); a process planning and optimization unit, which generates a set of candidate process routes based on the digital twin simulation to achieve process pre-planning (S3); an adaptive decision unit, which dynamically optimizes the processing route, process parameters, and temperature control strategy based on the preceding "process fingerprint" during processing, and performs an adaptive decision loop (S5); and an iterative learning and knowledge management unit, which optimizes the process model and parameters based on multiple batches of "process fingerprint" data, achieves iterative learning (S6), and forms an adaptive process knowledge base. The multi-sensor fusion monitoring module is used to synchronously collect process signal timing data during the machining process, providing process status data for the formation of the first-order machining and machining process data acquisition S4. It includes a three-dimensional grinding force monitoring unit integrated on the CNC grinding machine body, used to collect grinding force timing signals in real time; a temperature field monitoring unit, used to collect local temperature field data of the workpiece in real time; and a spindle load monitoring unit, used to collect spindle power or current signals in real time. The active temperature control jet execution module is used to execute active temperature control strategies during processing and dynamically adjust temperature control actions based on real-time data in the adaptive decision loop S5. It includes a multi-channel independent temperature control jet unit integrated on the CNC grinding machine body, with jet nozzles that can independently adjust temperature, flow rate and direction; and a jet control and drive unit, which is used to receive process instructions and drive the jet nozzles to perform fixed-point, fixed-temperature and fixed-flow cooling or heating on specific areas of the workpiece.

Images