Precise automatic multi-axis numerical control electric spark forming machine data processing method and system

Abstract

The invention discloses a data processing method and system for a precise automatic multi-axis numerical control electric spark forming machine, and the method comprises the steps: mapping preprocessed data into a digital twin model for fusion analysis, and extracting fusion features including instantaneous energy density, an effective machining angle, a thermal error compensation amount and a chip removal efficiency evaluation value; based on the fused feature data, performing real-time state evaluation and future discharge trend prediction by using a LightGBM neural network model, and outputting a processing prediction result; and a multi-target optimization function is established, an improved NSGA-II algorithm is utilized to carry out rapid optimization search by taking the current processing parameters as an initial population, a multi-axis cooperative motion instruction is generated, and the multi-axis cooperative motion instruction is transmitted to a multi-axis motion control module and a pulse power supply module to be executed. And the processing stability and reliability are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of numerical control machining, in particular to a data processing method and system for a precision automatic multi-axis numerical control EDM machine. BACKGROUND

[0002] The traditional method relies on offline experience library and manual judgment, lacks real-time dynamic analysis capability, and needs to wait for the completion of processing to obtain the surface quality, size accuracy and other results through detection equipment to judge whether the processing is qualified. If an abnormality occurs, it is easy to cause the workpiece to be scrapped or the electrode to be damaged, causing state evaluation to lag behind. It is impossible to predict the future discharge state based on the current processing data, such as being unable to identify potential risks in advance, and only being able to deal with passive shutdown after a fault occurs, resulting in reduced processing efficiency. SUMMARY

[0003] The present application relates to the technical field of numerical control machining, in particular to a data processing method and system for a precision automatic multi-axis numerical control EDM machine.

[0004] To achieve the above-mentioned purpose, the technical scheme of the present application is as follows: further, in the above-mentioned data processing method for a precision automatic multi-axis numerical control EDM machine, the data processing method comprises the following steps: Real-time collection of discharge state, motion axis, machine tool state and environmental data and standardized processing to obtain pre-processing data; Mapping the pre-processing data to the digital twin model for fusion analysis, extracting fusion features including instantaneous energy density, effective processing angle, thermal error compensation amount and chip removal efficiency evaluation value to obtain fusion feature data; Based on the fusion feature data, using a LightGBM neural network model for real-time state evaluation and future discharge trend prediction, and outputting a processing prediction result; Establishing a multi-objective optimization function and using an improved NSGA-II algorithm to perform fast optimization search with the current processing parameters as the initial population to generate multi-axis coordinated motion instructions, wherein the crossover rate and mutation rate of the improved NSGA-II algorithm are dynamically adjusted according to the processing prediction result; Transmitting the multi-axis coordinated motion instructions to the multi-axis motion control module and the pulse power supply module for execution.

[0005] Further, in the above-mentioned data processing method for a precision automatic multi-axis numerical control EDM machine, the real-time collection of discharge state, motion axis, machine tool state and environmental data and standardized processing to obtain pre-processing data comprises: Real-time acquisition of discharge state, motion axis, machine tool state and environmental data, linear interpolation method is used to fill in short-time missing data; long-time missing data is replaced by calling historical similar working condition data, and cleaning data is obtained; The data of different orders of magnitude is standardized by Z-score, and the standardized data is converted into JSON format to obtain preprocessed data.

[0006] Further, in the above-mentioned precision automatic multi-axis numerical control electric spark forming machine data processing method, the preprocessed data is mapped into the digital twin model for fusion analysis, and fusion features including instantaneous energy density, effective machining angle, thermal error compensation amount and chip removal efficiency evaluation value are extracted to obtain fusion feature data, including: Start the digital twin system corresponding to the machine tool, match the actual size of the machine tool with the geometric model, integrate the electric spark machining energy transmission formula and thermal deformation formula into the physical model, and preset the multi-axis motion coordination logic in the behavior model; Establish the correspondence between the preprocessed data and the twin model parameters to generate a mapping table and store it in the model configuration library; The twin model is based on injected data for real-time simulation, simulating the current machining scene, calculating the instantaneous energy density based on the three-dimensional coordinates of the discharge area in the twin model, and extracting the effective machining angle through the angle between the electrode axis and the normal of the workpiece machining surface in the twin model combined with the motion axis attitude data.

[0007] Further, in the above-mentioned precision automatic multi-axis numerical control electric spark forming machine data processing method, characterized in that, based on the fusion feature data, a LightGBM neural network model is used for real-time state evaluation and future discharge trend prediction, and a machining prediction result is output, including: Extract the fusion feature data of the last 10 batches from the fusion feature data as the input vector of the LightGBM model; Load the pre-trained LightGBM gradient boosting tree model, set the learning rate of the model to 0.05, the number of trees to 100, and the number of leaf nodes to 31; Input the input vector into the LightGBM model, and output the real-time machining state evaluation result, including machining stability level, surface quality prediction value and abnormal risk probability.

[0008] Further, in the above-mentioned precision automatic multi-axis numerical control electric spark forming machine data processing method, characterized in that, based on the fusion feature data, a LightGBM neural network model is used for real-time state evaluation and future discharge trend prediction, and a machining prediction result is output, including: Extract 50 batches of fusion feature data, construct time series input sequence, and use LightGBM time series prediction mode to predict the discharge trend of the next 3 batches, including at least discharge frequency trend, discharge energy fluctuation range and abnormal time window.

[0009] Further, in the above-mentioned precision automatic multi-axis numerical control electric spark forming machine data processing method, the multi-objective optimization function is established, and the improved NSGA-II algorithm is used to perform fast optimization search with the current machining parameter as the initial population to generate multi-axis cooperative motion instructions, including: Initial population generation, taking the current machining parameter as the reference, using Latin hypercube sampling method to generate 100 initial population individuals; Set the basic parameters of the improved NSGA-II, including 50 iterations, population size 100, crossover mode simulated binary crossover, mutation mode polynomial mutation, initialize non-dominated sorting counter and crowding degree calculation matrix; After iteration, extract the non-dominated individuals of level 1 to generate a set of Pareto optimal solutions, convert the optimal parameters into motion control instructions, and the multi-axis motion parameters include the target position, motion speed and acceleration of each axis, and the pulse power parameters include the discharge voltage, peak current and pulse width.

[0010] Further, in the above-mentioned precision automatic multi-axis numerical control electric spark forming machine data processing method, the multi-axis cooperative motion instructions are transmitted to the multi-axis motion control module and the pulse power module for execution, including: After receiving the instructions, analyze the axis motion parameters, drive the servo motors of each axis to move according to the target trajectory, analyze the pulse parameters, and adjust the power output.

[0011] Further, in a precision automatic multi-axis numerical control electric spark forming machine data processing system, the precision automatic multi-axis numerical control electric spark forming machine data processing system includes the following modules: Data acquisition and processing module for real-time acquisition and standardized processing of discharge state, motion axis, machine tool state and environmental data to obtain preprocessed data; Feature analysis and fusion module for mapping preprocessed data to digital twin model for fusion analysis, extracting fusion features including instantaneous energy density, effective machining angle, thermal error compensation amount and chip removal efficiency evaluation value to obtain fusion feature data; Machining result prediction module for real-time state evaluation and future discharge trend prediction based on the fusion feature data using LightGBM neural network model, and outputting the machining prediction result; The motion instruction generation module is configured to establish a multi-objective optimization function, and perform fast optimization search by using an improved NSGA-II algorithm with the current machining parameter as an initial population to generate a multi-axis collaborative motion instruction, wherein the crossover rate and the mutation rate of the improved NSGA-II algorithm are dynamically adjusted according to the machining prediction result. The multi-axis collaborative control module is configured to transmit the multi-axis collaborative motion instruction to a multi-axis motion control module and a pulse power supply module for execution.

[0012] Further, in the data processing system of the precision automatic multi-axis NC electric spark forming machine, the machining result prediction module comprises the following sub-modules: The extraction sub-module is configured to extract 50 batches of fusion feature data, construct a time sequence input sequence, and predict the discharge trend of the next 3 batches by using a time sequence prediction mode of LightGBM, at least including a discharge frequency trend, a discharge energy fluctuation range, and an abnormal occurrence time window.

[0013] Further, in the data processing system of the precision automatic multi-axis NC electric spark forming machine, the multi-axis collaborative control module comprises the following sub-modules: The execution sub-module is configured to receive the instruction, analyze the axis motion parameters, drive the servo motors of each axis to move according to the target trajectory, analyze the pulse parameters, and adjust the output of the power supply.

[0014] The beneficial effects are that the deep fusion and collaborative intelligence break the information silos between subsystems through digital twin technology, realize the global, multi-dimensional and visual perception of the machining process, and provide a data basis for intelligent decision-making. The forward-looking prediction and active control utilize the time series machine learning model to realize the advanced prediction of the machining abnormality and state trend, change the control mode from lagging response to forward-looking suppression, and significantly improve the machining stability and reliability. The dynamic optimization and adaptability utilize the adaptive evolutionary algorithm dynamically coupled with the machining state for online multi-objective optimization, so that the control parameters can be automatically and optimally adjusted with the change of the machining conditions, and the machining is always kept near the Pareto optimal frontier. The continuous evolution capability enables the learning model of the system to continuously update and optimize itself by using production data, so that the adaptability of the machine tool to different workpieces and electrode materials is continuously enhanced. The overall improvement of the machining performance ultimately reflects the effective improvement of the machining efficiency, the reduction of the electrode loss, the guarantee of the surface quality consistency, and the reduction of the excessive dependence on the experience of the operators. BRIEF DESCRIPTION OF DRAWINGS

[0015] Various other advantages and benefits will become apparent to those of ordinary skill in the art upon reading the following detailed description of the preferred embodiments with reference made to the accompanying drawings. The drawings are for purposes of illustration only and are not considered as limiting the application.

[0016] Figure 1 Fig. 1 is a schematic diagram of a first embodiment of a data processing method of a precision automatic multi-axis NC electric spark forming machine according to the present application; Figure 2 Fig. 2 is a schematic diagram of a second embodiment of a data processing method of a precision automatic multi-axis NC electric spark forming machine according to the present application; Figure 3 Fig. 3 is a schematic diagram of a first embodiment of a data processing system of a precision automatic multi-axis NC electric spark forming machine according to the present application. DETAILED DESCRIPTION

[0017] In order to make the objects, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application.

[0018] Those skilled in the art can understand that, unless specifically stated otherwise, the singular forms "one", "a", and "the" used herein can also include the plural forms. It should be further understood that the use of the term "comprise" in the specification of the present application means that there are features, integers, steps, operations, elements and / or groups, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, groups and / or their combinations.

[0019] The present application will be described in detail below with reference to the accompanying drawings, Figure 1 As shown in the drawings, a data processing method of a precision automatic multi-axis NC electric spark forming machine includes the following steps, Step 101, real-time acquisition of discharge state, motion axis, machine tool state and environmental data and standardized processing to obtain pre-processing data; Specifically, in the present embodiment, the discharge state, motion axis, machine tool state and environmental data are collected in real time, the linear interpolation method is used to fill in the short-time missing data, the historical similar working condition data is called to replace the long-time missing data to obtain cleaning data, the Z-score standardization is used for different orders of magnitude data, and the standardized data is converted into JSON format to obtain pre-processing data.

[0020] Multi-dimensional data acquisition preparation, deployment of acquisition hardware: install voltage / current sensors on the discharge circuit with a sampling frequency of 1kHz~5kHz to collect discharge state data such as discharge voltage, peak current, discharge frequency, and gap voltage; install grating encoders on each motion axis X / Y / Z / U / V with a resolution of ≤1μm to collect axis position, speed, and acceleration data; install temperature sensors with an accuracy of ±0.1℃ and vibration sensors with a range of 0~500Hz on the machine tool bed, spindle, and worktable to collect machine tool state data; install temperature and humidity sensors with an accuracy of ±5%RH / ±0.5℃ and air pressure sensors in the machining cabin to collect environmental data. Synchronous time reference: establish a unified timestamp with an accuracy of 1ms through a PLC programmable logic controller to ensure that all sensor data is time-aligned. Real-time acquisition of raw data, start acquisition program: after the machine tool is turned on, trigger the data acquisition module based on LabVIEW or C++ development to read each sensor data in real time at the set frequency, and store it to the cache database such as Redis to avoid data loss. Abnormal interception: if there is no data feedback or the data exceeds the range such as discharge current > rated value 120%, immediately trigger temporary storage backup and send a sensor fault warning signal.

[0021] Data cleaning, missing value filling, for short-term missing data ≤3 consecutive sampling points, use linear interpolation method to fill in; for long-term missing data >3 sampling points, mark as invalid segment, and call similar historical data of the same period as temporary replacement. Outlier rejection: use the 3σ principle to identify abnormal data such as sudden voltage spikes and axis position jumps, and replace them with a sliding average window size of 5 sampling points after rejection. Data standardization conversion, dimension unification: use Z-score standardization to unify different orders of magnitude data to the [-1,1] interval. Data format regularization: convert the standardized data to JSON format, store it in the timestamp-data type-value structure, and facilitate subsequent digital twin model calling. Preprocessed data output, generate dataset: divide the preprocessed data into 100ms / batches, each batch contains all dimensions of standardized data within that period, store it in a real-time database such as InfluxDB, and mark the data integrity such as complete / with completion / with invalid segment.

[0022] Step 102, map the preprocessed data to the digital twin model for fusion analysis, extract fusion features including instantaneous energy density, effective machining angle, thermal error compensation amount, and chip removal efficiency evaluation value, and obtain fusion feature data; Specifically, in the embodiment, the digital twin system corresponding to the machine tool is started, the geometric model matches the actual size of the machine tool, the physical model integrates the energy transmission formula of electric spark machining and the thermal deformation formula, and the behavior model presets the multi-axis motion coordination logic. A mapping table is generated by establishing a correspondence between preprocessed data and twin model parameters and stored in a model configuration library. The twin model is simulated in real time based on the injected data, simulates the current machining scene, calculates the instantaneous energy density based on the three-dimensional coordinates of the discharge area in the twin model, and extracts the effective machining angle by the angle between the electrode axis and the normal of the workpiece machining surface in the twin model combined with the motion axis attitude data.

[0023] The digital twin model initializes the model loading, starts the digital twin system corresponding to the machine tool, and contains the geometric model, the physical model, and the behavior model. The geometric model matches the actual size error of the machine tool ≤0.01mm, the physical model integrates the energy transmission formula of electric spark machining and the thermal deformation formula, and the behavior model presets the multi-axis motion coordination logic. Data mapping configuration, establish a correspondence between preprocessed data and twin model parameters such as discharge voltage to model discharge energy input shaft position to model axis coordinate environmental temperature to model thermal boundary condition, generate a mapping table and store it in a model configuration library.

[0024] Real-time data and twin model fusion, data injection: preprocessed data is injected into the twin model in batches at a rate of 50ms / time, and real-time parameters in the model are updated, such as synchronizing the position of each axis and the discharge state once every 50ms. Dynamic simulation operation: the twin model is simulated in real time based on the injected data, simulates the current machining scene such as the relative position of the electrode and the workpiece, the temperature distribution of the discharge area, and the motion trajectory of the debris, and the simulation step is consistent with the data injection frequency 50ms / step.

[0025] Fusion feature extraction, instantaneous energy density calculation: based on the three-dimensional coordinates of the discharge area in the twin model derived from the axis position data, combined with the discharge power voltage x current calculation, the sampling interval is 10ms. Effective machining angle extraction: through the angle between the electrode axis and the normal of the workpiece machining surface in the twin model, combined with the motion axis attitude data U / V axis swing angle, the angle value within the effective machining range such as 0°~60° is selected, and the average value in this period is taken as the effective machining angle. Thermal error compensation amount calculation: based on the temperature field distribution of the bed and the spindle in the thermal simulation results of the twin model, the thermal deformation formula is substituted, and the thermal deformation amount of each motion axis is calculated as the thermal error compensation amount. The calculation of the evaluation value of the chip removal efficiency: combined with the debris concentration distribution in the twin model derived from the discharge gap data and the cooling liquid flow rate data, and whether the chip removal channel is blocked is judged by the vibration sensor data, the fuzzy comprehensive evaluation method is adopted, the weight is: debris concentration 60%, channel smoothness 40%, and the output is the chip removal efficiency evaluation value of 0~100.

[0026] Fusion feature data verification and output, feature validity check: calculate the correlation between each fusion feature, such as the Pearson correlation coefficient of instantaneous energy density and chip removal efficiency, and eliminate redundant features with a correlation of >0.9; compare the feature value with the feature range of the historical optimal machining condition, and if it exceeds the range, mark it for review. Feature data output, store the verified fusion feature instantaneous energy density, effective machining angle, thermal error compensation amount, and chip removal efficiency evaluation value in the feature database such as PostgreSQL in the format of timestamp-feature name-value-validity mark for subsequent model calling.

[0027] Step 103, based on the fusion feature data, use the LightGBM neural network model to perform real-time state evaluation and future discharge trend prediction, and output the machining prediction result; Specifically, in this embodiment, the fusion feature data of the last 10 batches is extracted from the fusion feature data as the input vector of the LightGBM model; the pre-trained LightGBM gradient boosting tree model is loaded, the learning rate of the model is set to 0.05, the number of trees is 100, and the number of leaf nodes is 31; the input vector is input into the LightGBM model, and the real-time machining state evaluation result is output, including the machining stability level, the surface quality prediction value and the abnormal risk probability. Extract 50 batches of fusion feature data to construct a time series input sequence, and use the time series prediction mode of LightGBM to predict the discharge trend of the next 3 batches, including at least the discharge frequency trend, the discharge energy fluctuation range and the abnormal time window.

[0028] When extracting the fusion feature data of the last 10 batches from the fusion feature database, the batch division rule and data filtering logic need to be determined first. According to the previous step setting, the fusion feature data is stored at 100ms / batch, each batch contains 4 core features of instantaneous energy density, effective machining angle, thermal error compensation amount, and chip removal efficiency evaluation value, and is accompanied by timestamp-validity mark. During the extraction process, first, filter the latest 10 continuous batches according to the timestamp in descending order, covering the time range of the last 1s, and check the validity mark of each batch - only select batches with complete or incomplete data ratio ≤30%, if a batch is marked as containing invalid segments and the invalid data ratio >30%, skip this batch and trace back, until 10 batches that meet the requirements are selected. Then, arrange the 4 features of the 10 batches in time order-feature dimension to construct an input matrix with a dimension of 10x4, where the row represents the batch number from early to late, and the column corresponds to the 4 fusion features, finally forming an input vector recognizable by the LightGBM model, ensuring that the model can perform state evaluation based on the continuous feature data of the last 1s.

[0029] LightGBM model loading and parameter configuration Before loading the pre-trained LightGBM gradient boosting tree model, the model inference environment needs to be initialized first, usually based on the Python Scikit-learn framework, and the LightGBM library and model dependent numpy, pandas and other toolkits are imported to ensure that the environment version is consistent with the model training, such as LightGBM version >= 3.3.2, to avoid model loading failure due to version compatibility problems. In the model loading stage, the pre-trained model file format is.pkl or.txt, which is read from the specified path. This model has been trained on 1000+ sets of fused feature data under different processing conditions, and the parameters have been optimized through 5-fold cross-validation during training. After loading, set the core parameters as required: set the learning rate to 0.05 to control the weight update range of each tree to avoid model overfitting, set the number of trees to 100 to balance model accuracy and inference speed, 100 trees can ensure evaluation accuracy while controlling inference time within 50ms, set the number of leaf nodes to 31 to limit the complexity of a single tree to prevent overlearning of noise in the training data, and in addition, other parameters such as the target function for multi-classification and regression mixed target and the split rule for gradient-based single-side sampling during model training need to be fixed to ensure that the model inference logic is consistent with the training logic.

[0030] LightGBM model inference and real-time processing state evaluation After inputting the constructed 10x4 input vector into the LightGBM model, the model will output real-time processing state evaluation results through the collaborative inference of multiple gradient boosting trees. For processing stability level, the model calculates the standard deviation of instantaneous energy density within 10 batches based on the fluctuation amplitude of instantaneous energy density in the input vector, and combines the fluctuation amplitude-stability level mapping relationship established during training to output evaluation results from 1 to 5, 1 corresponding to standard deviation <= 5%, representing the most stable processing; 5 corresponding to standard deviation > 20%, representing extremely unstable processing; the inference of surface quality prediction value Ra value integrates effective processing angle and thermal error compensation, and through the nonlinear mapping relationship learned by the model, the numerical values of these two features are converted into corresponding surface roughness prediction value unit μm, the prediction error can be controlled within ± 0.1 μm; the calculation of abnormal risk probability takes the chip removal efficiency evaluation value as the core input, if the chip removal efficiency evaluation value of 10 batches is continuously lower than 60, the model will combine the correlation between chip removal efficiency and abnormal occurrence in historical fault data to output the risk probability range 0~100% of specific abnormal types such as arc discharge and chip removal blockage.

[0031] The 50 batch fusion feature data is used to construct the time sequence input sequence, and the batch screening rule consistent with the real-time state evaluation is followed, that is, the latest 50 continuous effective batches covering the processing process in the last 5s are selected, and the 4 features of each batch need to pass the effectiveness check. When constructing the time sequence, the feature data of the 50 batches is arranged in chronological order according to the timestamp, forming a time sequence matrix with a dimension of 50x4, wherein the row represents a time step of 100ms, the column corresponds to the 4 fusion features, and the timestamp information of each batch is also retained to provide a basis for subsequent time window labeling of trend prediction.

[0032] In step 104, a multi-objective optimization function is established, and an improved NSGA-II algorithm is used to perform fast optimization search with the current processing parameters as the initial population to generate multi-axis collaborative motion instructions, wherein the crossover rate and mutation rate of the improved NSGA-II algorithm are dynamically adjusted according to the processing prediction result. Specifically, in the initial population generation in the embodiment, the Latin hypercube sampling method is used to generate 100 initial population individuals based on the current processing parameters; the basic parameters of the improved NSGA-II are set, including 50 iterations, a population size of 100, a simulated binary crossover for the crossover method, a polynomial mutation for the mutation method, and initialization of the non-dominated sorting counter and the crowdedness calculation matrix; after iteration, the non-dominated individuals of level 1 are extracted to generate a Pareto optimal solution set, the optimal parameters are converted into motion control instructions, and the multi-axis motion parameters include the target position, motion speed and acceleration of each axis, and the pulse power parameters include the discharge voltage, peak current and pulse width.

[0033] The multi-objective optimization function is constructed, and three core objectives are defined: the maximum material removal rate, the minimum surface roughness, and the minimum electrode loss rate. The maximum material removal rate is the material removal rate, unit mm³; t is the processing time, unit s, based on the material removal simulation formula of the twin model; the minimum surface roughness is the predicted surface quality value, unit μm, which converts the minimization problem into a maximization problem; the minimum electrode loss rate is the electrode loss rate, unit %, based on the discharge energy density and thermal error compensation amount calculation.

[0034] Constraint setting: discharge parameter constraint: discharge voltage 50~300V, peak current 5~50A, pulse width 1~100μs; motion axis constraint: each axis stroke in the rated range such as X axis 0~500mm, speed≤10mm / s, acceleration≤5mm / s²; environment constraint: processing cabin temperature 20~28℃, humidity 40%~60%RH. Improve the initialization of the NSGA-II algorithm, and generate the initial population: based on the current processing parameters of discharge voltage, current, pulse parameters, and axis motion speed / acceleration, use the Latin hypercube sampling method to generate 100 initial population individuals, each individual corresponds to a complete set of processing parameters, and ensure that the population covers the parameter space within the constraint conditions. Initialize the algorithm parameters: set the basic parameters of the improved NSGA-II to 50 iterations, population size 100, crossover mode simulated binary crossover, mutation mode polynomial mutation, and initialize the non-dominated sorting counter and crowding degree calculation matrix.

[0035] Dynamic adjustment of crossover rate and mutation rate, and the adjustment rules are based on the processing prediction results in step three. If the predicted processing stability is level 1~2, and the abnormal risk probability is <30%, the crossover rate is increased to 0.8~0.9, the existing optimal solution is iterated, the mutation rate is reduced to 0.01~0.02 to reduce random disturbance; if the predicted processing is unstable, the stability level is 4~5, and the abnormal risk probability is >60%, the crossover rate is reduced to 0.4~0.5 to avoid the transmission of inferior solutions, and the mutation rate is increased to 0.05~0.08 to explore new parameter space; if there is an abnormal trend such as possible arc in the future 300ms, temporarily increase the mutation rate to 0.1 to force search to avoid abnormal parameters. Real-time adjustment execution: before each iteration, read the latest processing prediction results, adjust the crossover rate and mutation rate according to the rules, and update the algorithm parameter configuration.

[0036] Fast optimization search, non-dominated sorting: sort the 100 individuals in the current population according to the advantages and disadvantages of the 3 objective functions, and divide them into different non-dominated levels. Level 1 is the optimal solution, and level 2 is the second. Crowding degree calculation: calculate the crowding degree of individuals in the same level in the target space to reflect the diversity of individuals. The larger the crowding degree, the more unique the individual is in the parameter space. Selection and iteration: select 50 optimal individuals into the next generation population according to the non-dominated level priority + crowding degree sorting rule; generate 50 new individuals through crossover and mutation operations to supplement the population to maintain the size of 100; repeat the iteration until the set number of 50 times is reached.

[0037] Optimization result screening and instruction generation, Pareto optimal solution set extraction: After iteration, extract the non-dominated individuals of level 1 to form a Pareto optimal solution set, usually containing 5-10 optimal parameter combinations. Optimal parameter selection: based on actual processing requirements such as priority efficiency, select the individual with the maximum \(f_1\), priority surface quality, select the individual with the maximum \(f_2\), and screen 1 set of optimal processing parameters. Multi-axis coordinated motion instruction generation: convert the optimal parameters into motion control instructions, including: multi-axis motion parameters: target position, motion speed, acceleration of each axis X / Y / Z / U / V, motion speed, acceleration format conforms to G code standard, such as G01X100Y50Z-20F1000; pulse power parameters: discharge voltage, peak current, pulse width / interval format conforms to M code standard, such as M03P50I10T20, P for voltage, I for current, T for pulse width.

[0038] Step 105, transmit the multi-axis coordinated motion instruction to the multi-axis motion control module and the pulse power module for execution.

[0039] Specifically, in the present embodiment, after receiving the instruction, the axis motion parameters are analyzed, the servo motors of each axis are driven to move according to the target trajectory, and the pulse parameters are analyzed to adjust the power output.

[0040] Instruction format verification and conversion, format compliance verification: check whether the motion instruction G / M code conforms to the machine tool control system protocol such as Fanuc FOCAS, Siemens Sinumerik, if there is a syntax error such as axis coordinate exceeding range, parameter missing, return to the optimization algorithm module to regenerate. Instruction conversion if needed: if the multi-axis motion control module supports a dedicated protocol such as EtherCAT, convert the G / M code to an EtherCAT instruction frame containing axis ID, control word, target position, etc. field, to ensure that the instruction can be directly parsed by the module.

[0041] Real-time instruction transmission, transmission channel selection: use EtherCAT bus transmission rate 100Mbps, delay ≤1ms as the main channel, Ethernet TCP / IP protocol as the backup channel, to ensure the real-time and reliability of instruction transmission. Data verification and retransmission: each frame of instruction is attached with CRC check code, which is verified by the motion control module and the pulse power module after receiving. If the verification fails, immediately send a retransmission request, the retransmission number ≤3 times, if more than 3 times, trigger an alarm.

[0042] The instruction execution and real-time feedback, multi-axis motion control module executes: after receiving the instruction, analyzing the axis motion parameters, driving each axis servo motor such as Panasonic A6 series servo motor to move according to the target trajectory, real-time acquisition of the actual position data feedback frequency of the motor encoder is 1kHz, calculate the position deviation if the deviation >5um, start position compensation. Pulse power module execution: analysis of pulse parameters, adjust the power output such as through IGBT module control discharge voltage, current, real-time acquisition of the actual discharge parameters voltage, current, feedback to the power control unit if the actual parameter and instruction deviation >10%, adjust the output circuit.

[0043] The execution state monitoring and closed-loop adjustment, state monitoring: machine tool PLC real-time acquisition of motion axis execution state such as whether to reach the target position, whether there is overload, pulse power execution state such as whether normal discharge, whether there is short circuit, if there is an exception such as axis stuck, power overcurrent, immediately send pause instruction, trigger sound and light alarm. Closed-loop feedback: the actual parameters of the execution axis position, discharge parameters are fed back to the data acquisition module of step one as the original data of the next round of data acquisition, forming a closed-loop control of acquisition-analysis-prediction-optimization-execution-feedback.

[0044] Its beneficial effects are that deep integration and collaborative intelligence, through digital twin technology, it breaks the information island between subsystems, realizes the global, multi-dimensional and visual perception of the processing process, and provides a data basis for intelligent decision-making. Prospective prediction and active control, using time series machine learning model to realize the advanced prediction of processing abnormality and state trend, change the control mode from lag response to prospective suppression, significantly improve the processing stability and reliability. Dynamic optimization and adaptability, using adaptive evolutionary algorithm coupled with processing state for online multi-objective optimization, so that the control parameters can be automatically and optimally adjusted with the change of processing conditions, and the processing is always kept near the Pareto optimal frontier. Continuous evolution ability, the learning model of the system can continuously update and optimize itself using production data, so that the adaptability of the machine tool to different workpieces and electrode materials is continuously enhanced. Comprehensive improvement of processing performance, ultimately reflected in effectively improving the processing efficiency, reducing the electrode loss, ensuring the surface quality consistency, and reducing the excessive dependence on the experience of the operator.

[0045] Please refer to Figure 2 In a precise automatic multi-axis numerical control electric spark forming machine data processing method, the preprocessed data is mapped into a digital twin model for fusion analysis, and fusion features including instantaneous energy density, effective processing angle, thermal error compensation amount and chip removal efficiency evaluation value are extracted to obtain fusion feature data, including the following steps, Step 201, start the digital twin system corresponding to the machine tool, match the geometric model with the actual size of the machine tool, integrate the physical model with the energy transmission formula and thermal deformation formula of the electric spark machining, and preset the multi-axis motion coordination logic in the behavior model; Step 202, establish the corresponding relationship between the pre-processing data and the twin model parameters to generate a mapping table and store it in the model configuration library; Step 203, the twin model performs real-time simulation based on the injected data, simulates the current machining scene, calculates the instantaneous energy density based on the three-dimensional coordinates of the discharge area in the twin model, and extracts the effective machining angle through the angle between the electrode axis and the normal of the workpiece machining surface in the twin model combined with the motion axis posture data.

[0046] The above describes an embodiment of the data processing method of the precision automatic multi-axis numerical control electric spark forming machine of the application. Please refer to Figure 3 In a precision automatic multi-axis numerical control electric spark forming machine data processing system, the precision automatic multi-axis numerical control electric spark forming machine data processing system comprises the following modules, A data acquisition and processing module is configured to acquire and process discharge state, motion axis, machine tool state and environment data in real time to obtain pre-processing data; A feature analysis and fusion module is configured to map the pre-processing data to the digital twin model for fusion analysis, extract fusion features including instantaneous energy density, effective machining angle, thermal error compensation amount and chip removal efficiency evaluation value, and obtain fusion feature data; A machining result prediction module is configured to perform real-time state evaluation and future discharge trend prediction based on the fusion feature data using a LightGBM neural network model, and output a machining prediction result; A motion instruction generation module is configured to establish a multi-objective optimization function, and use an improved NSGA-II algorithm to perform fast optimization search with the current machining parameters as the initial population to generate multi-axis coordinated motion instructions, wherein the crossover rate and mutation rate of the improved NSGA-II algorithm are dynamically adjusted according to the machining prediction result; A multi-axis coordinated control module is configured to transmit the multi-axis coordinated motion instructions to the multi-axis motion control module and the pulse power supply module for execution.

[0047] The basic principles, main features and advantages of the application are shown and described above. Those skilled in the art should understand that the application is not limited by the above embodiments, and the above embodiments and descriptions in the specification are only preferred examples of the application and are not intended to limit the application. Without departing from the spirit and scope of the application, various changes and improvements can be made to the application, and these changes and improvements all fall within the scope of the claimed application. The scope of protection of the application is defined by the appended claims and their equivalents.

Claims

1. A data processing method for a precision automatic multi-axis CNC electrical discharge machining (EDM) machine, characterized in that, The data processing method for the precision automatic multi-axis CNC EDM forming machine includes the following steps: Real-time data acquisition of discharge status, motion axis, machine tool status, and environmental data is performed and standardized to obtain preprocessed data. The preprocessed data is mapped to a digital twin model for fusion analysis, and fusion features including instantaneous energy density, effective machining angle, thermal error compensation amount and chip removal efficiency evaluation value are extracted to obtain fusion feature data; Based on the fused feature data, the LightGBM neural network model is used to perform real-time state assessment and future discharge trend prediction, and output the processing prediction results. A multi-objective optimization function is established, and the improved NSGA-II algorithm is used to perform a fast optimization search with the current processing parameters as the initial population to generate multi-axis cooperative motion commands. The crossover rate and mutation rate of the improved NSGA-II algorithm are dynamically adjusted according to the processing prediction results. The multi-axis coordinated motion command is transmitted to the multi-axis motion control module and the pulse power supply module for execution.

2. The data processing method for a precision automatic multi-axis CNC EDM forming machine as described in claim 1, characterized in that, The real-time acquisition and standardization of discharge status, motion axis, machine tool status, and environmental data yields preprocessed data, including: Real-time data acquisition of discharge status, motion axis, machine tool status, and environmental data; short-term missing data is filled using linear interpolation; long-term missing data is replaced by historical data from similar operating conditions to obtain cleaning data. Z-score standardization is applied to data of different magnitudes, and the standardized data is converted into JSON format to obtain preprocessed data.

3. The data processing method for a precision automatic multi-axis CNC EDM forming machine as described in claim 1, characterized in that, The process involves mapping preprocessed data to a digital twin model for fusion analysis, extracting fusion features including instantaneous energy density, effective machining angle, thermal error compensation, and chip removal efficiency assessment values, to obtain fusion feature data, including: Start the digital twin system corresponding to the machine tool, match the geometric model with the actual size of the machine tool, integrate the energy transfer formula and thermal deformation formula of electrical discharge machining into the physical model, and preset the multi-axis motion coordination logic into the behavioral model; Establish a mapping relationship between preprocessed data and twin model parameters, generate a mapping table, and store it in the model configuration library; The twin model performs real-time simulation based on the injected data to simulate the current processing scenario. The instantaneous energy density is calculated based on the three-dimensional coordinates of the discharge area in the twin model. The effective processing angle is extracted by combining the angle between the electrode axis and the normal of the workpiece processing surface in the twin model with the motion axis attitude data.

4. The data processing method for a precision automatic multi-axis CNC EDM forming machine as described in claim 1, characterized in that, Based on the fused feature data, the LightGBM neural network model is used for real-time state assessment and future discharge trend prediction, outputting processing prediction results, including: Nearly 10 batches of fused feature data were extracted from the fused feature data and used as the input vector for the LightGBM model; Load the pre-trained LightGBM gradient boosting tree model, set the model's learning rate to 0.05, the number of trees to 100, and the number of leaf nodes to 31; Input the input vector into the LightGBM model, and output the real-time processing status evaluation results, including processing stability level, surface quality prediction value, and anomaly risk probability.

5. The data processing method for a precision automatic multi-axis CNC EDM forming machine as described in claim 1, characterized in that, Based on the fused feature data, the LightGBM neural network model is used for real-time state assessment and future discharge trend prediction, outputting processing prediction results, including: Fifty batches of fused feature data were extracted to construct a time-series input sequence. The LightGBM time-series prediction model was used to predict the discharge trend of the next three batches, including at least the discharge frequency change trend, the discharge energy fluctuation range, and the time window of anomaly occurrence.

6. The data processing method for a precision automatic multi-axis CNC EDM forming machine as described in claim 1, characterized in that, The process involves establishing a multi-objective optimization function and using an improved NSGA-II algorithm with the current machining parameters as the initial population for rapid optimization search to generate multi-axis cooperative motion commands, including: The initial population was generated using the current processing parameters as a baseline and the Latin hypercube sampling method to generate 100 initial population individuals. Set the basic parameters of the improved NSGA-II, including 50 iterations, 100 population size, simulated binary crossover, polynomial mutation, and initialize the non-dominated sorting counter and crowding calculation matrix. After the iteration, the non-dominated individuals of level 1 are extracted to generate the Pareto optimal solution set. The optimal parameters are converted into motion control commands. The multi-axis motion parameters include the target position, motion velocity and acceleration of each axis, and the pulse power supply parameters include the discharge voltage, peak current and pulse width.

7. The data processing method for a precision automatic multi-axis CNC EDM forming machine as described in claim 1, characterized in that, The step of transmitting the multi-axis coordinated motion command to the multi-axis motion control module and the pulse power supply module for execution includes: After receiving the command, the axis motion parameters are analyzed, and the servo motors of each axis are driven to move along the target trajectory. The pulse parameters are also analyzed, and the power output is adjusted.

8. A data processing system for a precision automatic multi-axis CNC electrical discharge machining (EDM) machine, characterized in that, The data processing system of the precision automatic multi-axis CNC EDM forming machine includes the following modules: The data acquisition and processing module is used to collect real-time data on discharge status, motion axes, machine tool status, and environment, and to perform standardized processing to obtain pre-processed data. The feature analysis and fusion module is used to map preprocessed data to a digital twin model for fusion analysis, extracting fusion features including instantaneous energy density, effective machining angle, thermal error compensation amount, and chip removal efficiency evaluation value to obtain fusion feature data; The processing result prediction module is used to perform real-time state assessment and future discharge trend prediction based on the fused feature data using the LightGBM neural network model, and output the processing prediction result. The motion command generation module is used to establish a multi-objective optimization function and use an improved NSGA-II algorithm to perform a fast optimization search with the current processing parameters as the initial population to generate multi-axis cooperative motion commands. The crossover rate and mutation rate of the improved NSGA-II algorithm are dynamically adjusted according to the processing prediction results. The multi-axis collaborative control module is used to transmit the multi-axis collaborative motion commands to the multi-axis motion control module and the pulse power supply module for execution.

9. The data processing system for a precision automatic multi-axis CNC EDM forming machine as described in claim 8, characterized in that, The processing result prediction module includes the following sub-modules: The extraction submodule is used to extract fusion feature data from 50 batches, construct a time-series input sequence, and use LightGBM's time-series prediction mode to predict the discharge trend of the next 3 batches, including at least the discharge frequency change trend, discharge energy fluctuation range, and the time window of anomaly occurrence.

10. The data processing system for a precision automatic multi-axis CNC EDM forming machine as described in claim 8, characterized in that, The multi-axis collaborative control module includes the following sub-modules: The execution submodule is used to receive instructions, parse axis motion parameters, drive each axis servo motor to move along the target trajectory, parse pulse parameters, and adjust power output.

Images