Machine tool intelligent control method and system based on big data analysis

Abstract

The invention discloses an intelligent machine tool control method and system based on big data analysis, and the method comprises the steps: collecting a key parameter set of a machining technology, decomposing a vibration signal in a machining process through combining successive variational mode decomposition (SVMD), and extracting the modal frequency, amplitude and fluctuation characteristics of the signal; the method comprises the following steps: on the basis of a BiLSTM-Enhanced Transformer (BET) model, capturing a nonlinear relationship between a key parameter set of a processing technology and carbon emission, economic cost and vibration loss, training an optimal parameter model by adopting an improved energy valley optimization algorithm IEVO, and constructing an IEVO-BET machine tool intelligent control system; a key parameter set of a machining process is used as input, adjustable variables of the machining process are adjusted in real time through the synergistic effect of an IEVO model and a BET model, and multivariable control over carbon emission, energy consumption and economic cost in the machining process of a machine tool is achieved. The intelligent and real-time machine tool control requirements under different process requirements are met, carbon emission is effectively reduced, the production efficiency is improved, and the product quality and the sustainability of the machine tool performance are enhanced.

Description

Technical Field

[0001] The present invention relates to a method and system for intelligent control of machine tools, and particularly to a method and system for intelligent control of machine tools based on big data analysis. Background Art

[0002] With the continuous development of the manufacturing industry and the continuous advancement of intelligence and automation, as a core production equipment, the operating efficiency and processing quality of machine tools directly affect the smooth progress of the entire production process and the quality of the final products. The intelligent control technology of machine tools has gradually become the key to improving production efficiency, reducing production costs, and enhancing product quality. However, traditional machine tool control methods mostly rely on empirical formulas and manual adjustment, with low efficiency and difficulty in coping with complex processing environments and diverse production requirements. Therefore, how to achieve precise control of machine tools through intelligent means has become an important issue faced by the manufacturing industry.

[0003] Traditional machine tool control methods usually set parameters based on experience to adjust the working state of the machine tool. Although this can ensure the stability of processing to a certain extent, there are also many drawbacks. For example, with changes in the production environment, such as temperature fluctuations and load changes, traditional methods often cannot adapt quickly, resulting in unstable processing quality, or even failures or waste. In addition, the simplicity of empirical formulas makes it difficult to fully consider multiple complex factors in the processing process, such as tool wear and material property changes, resulting in greater uncertainty in processing accuracy and effects. The manufacturing industry has gradually introduced data-driven intelligent control methods to improve the automation and intelligence level of the machine tool processing process. By comprehensively collecting machine tool operation data and conducting in-depth analysis, various factors involved in the processing process can be monitored in real time, and then the key parameters of the processing technology can be optimized to improve the processing quality. For example, by using sensors to collect data such as temperature, vibration, and force in real time, the machine tool control system can optimize the processing process according to real-time feedback to ensure its accuracy and stability.

[0004] However, the current intelligent machine tool control methods still have certain limitations in multi-objective optimization, real-time adjustment, and coping with complex processing environments. Most methods only rely on a single data source and fail to comprehensively consider multi-dimensional data in the machine tool operation process, making it difficult for them to adapt to changing production requirements. Summary of the Invention

[0005] Object of the Invention: The object of the present invention is to provide a method and system for intelligent control of machine tools based on big data analysis, to achieve efficient optimization control in the machine tool processing process, with the goals of optimizing carbon emissions, vibration losses, and processing costs during the processing process, and providing technical support for intelligent manufacturing.

[0006] Technical Solution: A method for intelligent control of machine tools based on big data analysis according to the present invention includes the following steps:

[0007] (1) Collect the key parameter set of the machine tool processing technology, including: cutting speed, feed speed, grinding depth, cutting fluid use, tool material, machine tool performance, and environmental temperature data and vibration signals. Use the SVMD technology to decompose the machine tool vibration signal, extract the modal frequency and the fluctuation characteristics of the signal, and construct a feature matrix;

[0008] (2) Design the IEVO algorithm based on the double-tokenization strategy for optimizing the key parameters of the processing technology and model training;

[0009] (3) Take the key parameter set of the processing technology as the input, and vibration loss, carbon emissions, and processing cost as the output, and establish a comprehensive objective function;

[0010] (4) Optimize the BET neural network model parameters through the IEVO algorithm, construct a machine tool intelligent control model based on IEVO-BET, establish a non-linear relationship between the key parameters of the processing technology and carbon emissions, vibration loss, and processing cost, and conduct model training and prediction;

[0011] (5) According to the trained IEVO-BET machine tool intelligent control model, as well as the new process requirements and the collected dynamic data, dynamically adjust the adjustable variables of the processing technology in real time through the IEVO algorithm to control multiple variables such as carbon emissions, vibration loss, and economic cost during the machine tool processing process.

[0012] Preferably, the carbon emissions are calculated through the electricity carbon emission factor, and the formula is as follows:

[0013] E c = P·t·F

[0014] Among them, Ec is the carbon emission during the processing process, P is the electric power during the processing process, t is the time required for processing, and F is the electricity carbon emission factor;

[0015] The total power consumption is calculated by integrating the time of the power signal, and the carbon emissions are quantified in combination with the electricity carbon emission factor; for dynamic signals, the power signal is integrated in segments, and the carbon emissions of each segment are accumulated to obtain the total value.

[0016] Preferably, the calculation method of the vibration loss and the construction method of the feature matrix are as follows:

[0017] Decomposition of the vibration signal: Use the SVMD technology to decompose the collected vibration signal x(t) into multiple intrinsic mode functions, and the formula is as follows:

[0018]

[0019] Among them, IMF i(t) is the i-th mode function, and r(t) is the residual term;

[0020] Extract frequency components: By performing Hilbert transform on each intrinsic mode function signal, the instantaneous frequency is obtained:

[0021]

[0022] Among them, φ i (t) is the instantaneous phase of the i-th mode function;

[0023] Vibration frequency calculation: Conduct statistical analysis on the instantaneous frequencies of the intrinsic mode functions, and take their average frequency values as the vibration frequencies during this time period to obtain the main frequency components of the machine tool vibration during the processing;

[0024] Assume that the average frequency of the i-th mode function is f i , then the total vibration frequency f total is:

[0025]

[0026] Among them, m is the number of intrinsic mode functions, and f i is the average frequency of the i-th mode; (t) is the instantaneous phase of the i-th mode function;

[0027] Estimation of vibration loss: Use the vibration frequency and amplitude to calculate the vibration loss L vibration , and the vibration energy E i of each mode and the vibration amplitude A i and frequency f i are related as:

[0028]

[0029] Among them, A i is the amplitude of the i-th mode, and f i is the frequency of the i-th mode;

[0030] Effect of damping on energy loss: Let the damping ratio of each mode be ζ i , then the corresponding energy loss rate is:

[0031]

[0032] Among them, is the energy loss rate of the i-th mode, and ζ i is the damping ratio of the i-th mode;

[0033] Comprehensive vibration loss: Perform weighted summation on the vibration losses of all modes to obtain the total vibration loss of the system:

[0034]

[0035] Among them, E i is the vibration energy of the i-th mode, and ζ i is the damping ratio of the corresponding mode, and L vibration is the total vibration loss;

[0036] The feature matrix X structurally integrates the key parameters of the machining process and the vibration signals collected, forming input data that can be used for model training and optimization. The matrix includes the following feature items:

[0037] X = [E c , P, t, F, f total , A1, A2, … A m , ζ1, ζ2, … ζ m

[0038] Among them, Ec is the carbon emission during the machining process, P is the electric power during the machining process, t is the time required for machining, F is the electric energy carbon emission factor, and f total is the total vibration frequency, A1, A2, … A m is the amplitude of the m-th mode, and ζ1, ζ2, … ζ m is the damping ratio of the m-th mode.

[0039] Preferably, the IEVO algorithm described in step (2) is designed as follows:

[0040] Initialize the population P as N candidate solutions. Each candidate solution represents a set of key parameter combinations of the machining process, including the cutting speed v s , the feed rate f s , the grinding depth d s , and the tool material. The fitness of each candidate solution is evaluated through the objective function, and the objective function contains multiple optimization objectives, including vibration loss, carbon emission, and machining cost;

[0041] Chromosome coding and solution space representation: Each candidate solution represents the key parameters of the machining process through real number coding, as shown below:

[0042] X = [E c , P, t, F, f total , A1, A2, … A m , ζ1, ζ2, … ζ m

[0043] The implementation of the double tokenization strategy is as follows:

[0044] Global exploration token: Introduce non-local solutions to increase the coverage of the solution space. The formula is as follows:

[0045] x new = x current ​​+α·Δx

[0046] where α is the step factor and Δx is the random perturbation value;

[0047] Local refinement token: Based on gradient information or adaptive mutation operation, local optimization is performed, and the formula is as follows:

[0048]

[0049] where β is the local refinement step size, is the cotton degree of the fitness function;

[0050] The non-linear convergence factor controls the balance between global search and local optimization, and the adjustment formula is:

[0051]

[0052] where γ and δ are adjustment parameters to control the convergence rate.

[0053] During the optimization process, cyclic iteration is performed. The implementation method is as follows: selection is made according to the fitness value, and the selection methods include roulette wheel selection and tournament selection. Individuals with high fitness values are selected for crossover or replication; the selected individuals are subjected to gene exchange to generate new candidate solutions to achieve the crossover operation; the newly generated solutions are randomly perturbed; the population is updated according to the fitness value to find the optimal solution; the optimization process is aborted when the maximum number of iterations is reached or the fitness value exceeds the preset threshold.

[0054] Preferably, the implementation process of the objective function in step (3) is as follows:

[0055] (31) Input the set of key parameters of the processing technology into the neural network, and the input data is:

[0056] X = [E c , P, t, F, f total , A1, A2, … A m , ζ1, ζ2, … ζ m

[0057] Each sample is standardized to be within a reasonable range:

[0058]

[0059] where μ is the mean of the sample, σ is the standard deviation, and x i is each parameter sample of the input data;

[0060] (32) Establish a comprehensive objective function, defined as follows:

[0061] L = w1·L1 + w2·L2 + w3·L3 ​

[0062] Among them, L1 is the mean square error of vibration loss, L2 is the mean square error of carbon emissions, L3 is the mean square error of processing cost, and w1, w2, w3 are weight coefficients;

[0063] (33) Calculate the mean square error of vibration loss L1:

[0064]

[0065] Among them, the vibration loss predicted by the model is The true vibration loss is

[0066] Calculate the mean square error of carbon emissions L2:

[0067]

[0068] Among them, the carbon emissions predicted by the model are The true carbon emissions are

[0069] Calculate the mean square error of processing cost L3:

[0070]

[0071] Among them, the processing cost predicted by the model is The true processing cost is

[0072] Preferably, the implementation steps of constructing the machine tool intelligent control model based on IEVO - BET in step (4) are as follows:

[0073] (41) Use the population initialization method of the IEVO algorithm to generate an initial parameter set, and the initial population is expressed as:

[0074]

[0075] Among them, each θ i =(d e , d r , L, T pred , η) is a randomly generated parameter set;

[0076] (42) Define the prediction model:

[0077]

[0078] Among them, is the prediction result of the BET model based on the parameter set θ;

[0079] (43) Update the parameter set using the IEVO algorithm. The candidate solution θ is continuously adjusted according to the strategies in the exploration and exploitation stages to optimize the performance of the BET model;

[0080]

[0081] Where: and represent the adjustments of global search and local search respectively, and α1, α2 are the weight coefficients of global and local search;

[0082] (44) Input the updated parameter set into the BET model, calculate the mean square error, and evaluate the fitness. Among them, the smaller the L, the lower the fitness value, indicating that the parameter set can better optimize the prediction performance of the BET model. Update the optimal parameter set. If the fitness of the current solution is better, update the optimal solution:

[0083]

[0084] After reaching the predetermined termination condition, output the optimal parameter set This parameter set is used as the final configuration of the BET model and applied to the machine tool intelligent control system.

[0085] Preferably, the BET model described in step (4) consists of a shallow BiLSTM module and a deep Transformer module;

[0086] BiLSTM layer: The bidirectional LSTM extracts features from time series data through forward and backward learning. Let the input be x t , and the output of the BiLSTM layer be h t . Then the output formula at each moment t is expressed as follows:

[0087] h t = LSTM f (x t ) + LSTM b (x t )

[0088] Among them, LSTM f (x t ) is the output of the forward LSTM, and LSTM b (x t ) is the output of the backward LSTM;

[0089] Transformer layer: Used to capture long-range dependencies in the sequence. Among them, the output of the self-attention mechanism is:

[0090] Q, K, V = W q h t , W kh t ,W v h t

[0091]

[0092] Among them, W q ,W k ,W v is the weight matrix, and Q, K, and V are the query, key, and value matrices respectively. d k is the dimension of the key.

[0093] Preferably, the new process requirements described in step (5) include the product size d product , material hardness H, and surface finish Ra. The input data format is as follows:

[0094] R = [d product ,H,Ra]

[0095] The dynamic data is collected in real time through sensors and a monitoring system, including the cutting speed v s , feed rate f s , and grinding depth d s , which is expressed as follows:

[0096] D real = [v sreal ,f sreal ,d real **]

[0097] Among them, D real is the dynamic data collected in real time.

[0098] Preferably, the dynamic prediction implementation method described in step (5) is as follows:

[0099] Input the process requirements and the dynamically collected real-time data into the pre-trained BET model. The BET model predicts the change trends of key characteristics during the machining process based on the process requirements and real-time data, including carbon emissions economic cost , and vibration loss

[0100]

[0101] Among them, represents the predicted key machining characteristic value;

[0102] After obtaining the dynamic prediction results, use the IEVO algorithm to optimize the current machining process. Calculate the objective function using the direct summation method. The comprehensive objective function F real is:

[0103]

[0104] Among them, F real represents the comprehensive evaluation value of the current machining process.

[0105] An intelligent control system for machine tools based on big data analysis according to the present invention includes:

[0106] A collection module: used to collect a set of key parameters of the machine tool processing technology, including: cutting speed, feed speed, grinding depth, cutting fluid use, tool material, machine tool performance, and environmental temperature data and vibration signals. The SVMD technology is used to decompose the machine tool vibration signal, extract the modal frequency and the fluctuation characteristics of the signal, and construct a feature matrix;

[0107] A model training module: used to optimize the BET model parameters by the IEVO algorithm, construct an intelligent control model for the machine tool based on IEVO-BET, establish a non-linear relationship between the key parameters of the processing technology and carbon emissions, vibration loss, and processing cost, and perform model training and prediction;

[0108] A control module: used to adjust the adjustable variables of the processing technology in real time and dynamically according to the trained IEVO-BET intelligent control model for the machine tool, new process requirements, and working condition data through the IEVO algorithm, and perform multi-variable control on carbon emissions, vibration loss, and economic cost during the machine tool processing process.

[0109] Beneficial effects: Compared with the prior art, the present invention has the following remarkable advantages: Combining multi-dimensional data collection, SVMD signal decomposition, BET neural network, and IEVO algorithm to achieve comprehensive monitoring and multi-objective real-time optimization of the machine tool processing process; Through multi-source data fusion, comprehensively capture the key features in the processing process, provide accurate decision-making support for the dynamic adjustment of the adjustable parameters of the processing technology, and ensure processing efficiency, quality, and stability; The combination of BET neural network and IEVO algorithm adaptively adjusts the model weights and optimization strategies, enhances the model robustness and abnormal adaptation ability, and the IEVO algorithm combines global search and local optimization to improve the stability and accuracy of parameter optimization; Real-time collect processing data, combine the BET model and IEVO algorithm, dynamically predict and adjust the adjustable parameters of the processing technology, ensure efficient operation, quickly respond to environmental changes, improve production stability and flexibility; Based on multi-objective optimization, balance carbon emissions, economic cost, and vibration loss, maximize the overall benefit of the processing process, reduce resource waste, and promote sustainable manufacturing; Have good scalability and versatility, applicable to machine tool intelligent control and other real-time monitoring and optimization scenarios, such as industrial equipment health monitoring, intelligent manufacturing, robot control, etc.; The optimization algorithm and model structure reduce the computational complexity, save computational resources, improve the system response speed and computational efficiency, applicable to energy-efficient intelligent manufacturing and real-time monitoring systems, and promote green manufacturing. Brief Description of the Drawings

[0110] Figure 1 This is the flowchart of the method described in the present invention.

[0111] Figure 2 This is the flowchart of the IEVO algorithm described in the present invention. Detailed Embodiments

[0112] The technical solution of the present invention will be further described below in conjunction with the accompanying drawings.

[0113] As Figure 1 shown, a machine tool intelligent control method based on big data analysis includes the following steps:

[0114] (1) Collect the key parameter set of the machine tool processing technology, including: cutting speed, feed speed, grinding depth, cutting fluid use, tool material, machine tool performance, and environmental temperature data and vibration signals. Use the SVMD technology to decompose the machine tool vibration signal, extract the modal frequency and the fluctuation characteristics of the signal, and construct a feature matrix;

[0115] Among them, the vibration frequency and carbon emissions are key indicators to measure the machine tool processing process, reflecting the dynamic stability and energy consumption level respectively. To accurately capture the vibration characteristics and quantify carbon emissions, the present invention uses a method combining ensemble empirical mode decomposition (SVMD) and electric energy carbon emission factor calculation for comprehensive analysis.

[0116] Use the SVMD technology to decompose the collected vibration signal x(t) into multiple intrinsic mode functions (IMFs):

[0117]

[0118] where, IMF i (t) is the i-th mode function, r(t) is the residual term, and each IMF represents a vibration mode in the signal, which can reflect different frequency components of the signal.

[0119] Extract the frequency components: By performing Hilbert transform on each IMF signal, the instantaneous frequency is obtained:

[0120]

[0121] Vibration frequency calculation: Conduct statistical analysis on the instantaneous frequencies of all IMF mode functions, and take their average frequency value as the vibration frequency during this time period. Finally, the main frequency components of the machine tool vibration during the processing are obtained. Assume that the average frequency of the i-th mode function is f i , then the total vibration frequency f total is:

[0122]

[0123] Among them, m is the number of IMFs, and f i is the average frequency of the i-th mode

[0124] Combined with the carbon emission factor of electric energy, the carbon emissions during the processing are calculated, and the formula is:

[0125] E c = P·t·F

[0126] Among them, Ec is the carbon emissions during the processing, P is the electric power during the processing, t is the time required for processing; F is the carbon emission factor of electric energy

[0127] Estimation of vibration loss: Vibration loss is caused by the dissipation of vibration energy in the system. Using the vibration frequency and amplitude, the energy loss in different modes can be estimated. It is assumed that the energy loss of the system in each mode is related to the vibration amplitude and frequency. The vibration loss L vibration can be calculated through the following steps:

[0128] Estimation of vibration energy: The vibration energy E i of each mode is related to the vibration amplitude A i and the frequency f i The relationship is:

[0129]

[0130] Among them, A i is the amplitude of the i-th mode, and f i is the frequency of the i-th mode

[0131] Effect of damping on energy loss: Vibration loss is usually related to the damping characteristics of the system. It is assumed that the damping ratio of each mode is ζ i , then the corresponding energy loss rate is:

[0132]

[0133] Among them, is the energy loss rate of the i-th mode, and ζ i is the damping ratio of the i-th mode

[0134] Comprehensive vibration loss: The vibration losses of all modes are weighted and summed to obtain the total vibration loss of the system:

[0135]

[0136] Among them, E i is the vibration energy of the i-th mode, ζ i is the damping ratio of the corresponding mode, and L vibration is the total vibration loss

[0137] The feature matrix X structurally integrates the key parameters of the machining process and the vibration signals collected to form input data for model training and optimization. The matrix includes the following feature items:

[0138] X = [E c , P, t, F, f total , A1, A2, … A m , ζ1, ζ2, … ζ m

[0139] Among them, Ec is the carbon emission during the machining process, P is the electric power during the machining process, t is the time required for machining, F is the carbon emission factor of electric energy, and f total is the total vibration frequency, A1, A2, … A m is the amplitude of the m-th mode, and ζ1, ζ2, … ζ m is the damping ratio of the m-th mode.

[0140] (2) Design the IEVO algorithm based on the double-tokenization strategy for the optimization of key parameters of the machining process and model training;

[0141] The improved energy valley optimization algorithm (IEVO) adopts the double-tokenization strategy, combines global exploration and local optimization, and solves the multi-objective optimization of the key parameters of the machining process.

[0142] As Figure 2 shown, the improvement process of the IEVO algorithm includes the following key steps:

[0143] Initialize the population P as N candidate solutions. Each candidate solution represents a set of combinations of key parameters of the machining process, such as the cutting speed v s , the feed rate f s , the grinding depth d s , the tool material, etc.

[0144] The fitness of each candidate solution is evaluated through the objective function, and the objective function includes multiple optimization objectives, such as the vibration frequency f total , the carbon emission, the machining economic cost, etc.

[0145] Chromosome encoding and solution space representation: Each candidate solution represents the key parameters of the machining process through real number encoding as follows:

[0146] X = [E c , P, t, F, f total , A1, A2, … A m , ζ1, ζ2, … ζ m

[0147] ​​With this encoding method, the key parameters of the processing technology for each candidate solution are directly represented as a real number vector, simplifying subsequent optimization operations.

[0148] Application of the double-tokenization strategy: During the evolutionary process, the double-tokenization strategy is adopted to balance global search and local optimization;

[0149] Global exploration token: By introducing non-local solutions, such as randomly generating new solutions or introducing external information, the coverage of the solution space is increased to prevent the solution from falling into a local optimal solution. The formula is as follows:

[0150] x new = x current + α·Δx

[0151] where α is the step factor and Δx is the random perturbation value.

[0152] Local refinement token: Based on gradient information or adaptive mutation operations, local optimization is performed to obtain a better local optimal solution. The formula is as follows:

[0153]

[0154] where β is the local refinement step size, is the gradient of the fitness function.

[0155] Adjustment of the non-linear convergence factor: During the optimization process, the non-linear convergence factor σ(t) gradually decreases as the number of iterations t increases to control the balance between global search and local optimization. The specific adjustment formula is:

[0156]

[0157] where γ and δ are adjustment parameters.

[0158] The non-linear convergence factor plays a key role in the optimization process, ensuring that the optimization algorithm gradually converges and avoiding premature entrapment in local optimal solutions at an early stage. By adjusting γ and δ, the convergence rate can be flexibly controlled.

[0159] Loop iteration of the evolutionary process: In each iterative optimization, the IEVO algorithm evaluates the fitness values of each candidate solution and selects better solutions for crossover, mutation, etc. according to the fitness values. The specific operations include:

[0160] Selection operation: Selection is performed according to the fitness value. Common selection methods include roulette wheel selection, tournament selection, etc. Through these methods, individuals with higher fitness are selected for crossover or replication.

[0161] Crossover operation: The crossover operation exchanges genes of the selected individuals to generate new candidate solutions. Crossover methods include single-point crossover, multi-point crossover, etc. For example:

[0162] x offspring = Crossover(x1, x2)

[0163] where x1 and x2 are two parental solutions, and the crossover operation generates a new offspring solution xoffspring.

[0164] Mutation operation: Randomly perturb the newly generated solution to increase the diversity of the solution. The mutation operation formula is as follows:

[0165] x new = x offspring + ∈·Δx

[0166] where ∈ is the mutation factor and Δx is the random perturbation value.

[0167] Update operation: Update the population according to the fitness value and gradually find the optimal solution. The population update strategy usually includes retaining the optimal solution, eliminating the worst solution, etc.

[0168] Termination condition: The evolutionary process will terminate when any of the following conditions is met: reaching the maximum number of iterations T max or the fitness value exceeds the preset threshold.

[0169] (3) Take the set of key parameters of the processing technology as the input, and vibration loss, carbon emissions, and processing cost as the output to establish a comprehensive objective function;

[0170] The implementation process of the objective function is as follows:

[0171] The input data includes multi-dimensional time series signals such as cutting speed, feed speed, grinding depth, cutting fluid use, tool material, machine tool performance, and environmental temperature; the output is vibration loss, carbon emissions, and processing cost.

[0172] Input the set of key parameters of the processing technology into the neural network. The input data is:

[0173] X = [E c , P, t, F, f total , A1, A2, … A m , ζ1, ζ2, … ζ m

[0174] Each sample is normalized to ensure it is within a reasonable range:

[0175]

[0176] where μ is the mean of the sample, σ is the standard deviation, and x i is each parameter sample of the input data.

[0177] ​On this basis, a comprehensive objective function is established to simultaneously consider these three factors in subsequent optimization. The objective function is defined as follows:

[0178] L = w1·L1 + w2·L2 + w3·L3

[0179] where L1 is the mean square error of vibration loss, L2 is the mean square error of carbon emissions, L3 is the mean square error of processing cost, and w1, w2, w3 are weight coefficients to control the importance of each objective.

[0180] Objective function and its calculation:

[0181] The definition of the mean square error is as follows:

[0182]

[0183] where N is the number of samples; y i is the actual value of the i-th sample; is the predicted value of the i-th sample.

[0184] The calculation method for each objective is as follows:

[0185] Mean square error of vibration loss L1:

[0186]

[0187] where the vibration loss predicted by the model is the true vibration loss is

[0188] Mean square error of carbon emissions L2:

[0189]

[0190] Mean square error of processing cost L3:

[0191]

[0192] (4) Optimize the parameters of the BET neural network model through the IEVO algorithm, construct a machine tool intelligent control model based on IEVO-BET, establish a non-linear relationship between the key parameters of the processing technology and carbon emissions, vibration loss, and processing cost, and conduct model training and prediction;

[0193] The comprehensive objective function is used to evaluate the prediction performance of the BET model. The goal of the model is to minimize the objective function, that is, to minimize the error between the predicted value and the actual value by optimizing the hyperparameters. The parameters of the BET model include the embedding dimension d, the core representation dimension d′, the number of encoder layers N, the prediction window length H, and the learning rate L.

[0194] Initialize the population: Use the population initialization method of the IEVO algorithm to generate the initial parameter set, and the initial population is represented as:

[0195]

[0196] where each θ i =(d e , d r , L, T pred , η) is a randomly generated parameter set.

[0197] Define the prediction model:

[0198]

[0199] where is the prediction result of the BET model based on the parameter set θ.

[0200] In each iteration, apply the exploration and development stages of IEVO to update the parameter set.

[0201] During the update process, the candidate solution θ is continuously adjusted according to the strategies of the exploration and exploitation stages to optimize the performance of the BET model:

[0202]

[0203] where: and represent the adjustments of global search and local search respectively; α1, α2 are the weight coefficients of global and local search.

[0204] The parameter set θ after each update i is input into the BET model, calculate its mean square error L(θ), and evaluate the fitness according to this value. Among them, the smaller L is, the lower the fitness value, indicating that this parameter set can better optimize the prediction performance of the BET model. Update the optimal parameter set. If the fitness of the current solution is better, update the optimal solution:

[0205]

[0206] After reaching the predetermined termination condition, output the optimal parameter set This parameter set will be used as the final configuration of the BET model and applied to the intelligent control system of the machine tool.

[0207] The architecture of the BET model: It consists of a shallow BiLSTM module and a deep Transformer module. The shallow layer uses the method of replacing the multi-head self-attention mechanism (MHA) in the Transformer encoder with BiLSTM. This method captures the short-term dependencies of time-series data by using a bidirectional long short-term memory network (BiLSTM), replacing the global dependency modeling originally achieved by the MHA of the Transformer. The BiLSTM module is responsible for capturing short-term dynamic changes and local temporal dependencies, and can effectively handle short-term fluctuations and local features in time-series data; the deep Transformer module uses the global attention mechanism to model the long-term non-linear dependencies in the machine tool operation process, so as to achieve accurate prediction of carbon emissions, vibration loss, and processing costs, providing reliable support for optimization decisions.

[0208] The BET model consists of two parts:

[0209] BiLSTM layer: The bidirectional LSTM extracts features from time-series data through forward and backward learning. Assuming the input is x t , and the output of the BiLSTM layer is h t , then the output at each time t is represented by the following formula:

[0210] h t = LSTM f (x t ) + LSTM b (x t )

[0211] where LSTM f (x t ) and LSTM b (x t ) are the outputs of the forward and backward LSTMs respectively.

[0212] Transformer layer: The core of the Transformer is the self-attention mechanism, which is used to capture long-range dependencies in the sequence. The output of the self-attention mechanism is:

[0213] Q, K, V = W q h t , W k h t , W v h t

[0214]

[0215] where W q , W k , W vis the weight matrix, and Q, K, and V are the query, key, and value matrices respectively, where d k is the dimension of the key.

[0216] During the training process, the parameters of the BET model are optimized by the IEVO algorithm. The IEVO algorithm adjusts hyperparameters such as the learning rate and the number of encoder layers of the BET model during the training process to ensure that the goals of multi-objective optimization can be achieved simultaneously. When the optimization process of L reaches the predetermined fitness threshold, the training ends, and the machine tool intelligent control system is completed.

[0217] (5) According to the trained IEVO-BET machine tool intelligent control model, the new process requirements, and the collected dynamic data, the adjustable variables of the machining process are dynamically adjusted in real time through the IEVO algorithm to perform multi-variable control on carbon emissions, vibration losses, and economic costs during the machining process of the machine tool.

[0218] Input the new process requirements and predict the optimal parameter combination:

[0219] After training is completed, the machine tool intelligent control system can perform dynamic prediction based on the real-time collected working condition data and output the key parameters of the optimized machining process, such as cutting speed, feed speed, grinding depth, etc. These optimized parameters will directly affect the vibration losses, carbon emissions, and machining costs during the machining process, providing real-time feedback to ensure the continuous optimization of the process and the improvement of machining quality.

[0220] The new process requirements R include multiple dimensions, such as the product size d product , material hardness H, surface finish Ra, etc. The input data format is as follows:

[0221] R = [d product , H, Ra]

[0222] During the actual machining process, the machining dynamic data is collected in real time through sensors and monitoring systems, including, for example, the cutting speed v s , feed speed f s , grinding depth d s , etc. These data reflect the state of the current machining process and play a crucial role in the real-time prediction and parameter adjustment of the model:

[0223]

[0224] Input the input process requirements R and the real-time collected dynamic data D real into the trained BET model. The BET model predicts the change trends of various key characteristics during the machining process based on the process requirements and real-time data, including carbon emissions economic cost and vibration loss

[0225]

[0226] Among them, represents the predicted key machining characteristic value. The model provides a basis for dynamically adjusting the adjustable parameters of the machining process by capturing the non-linear relationship between the machining process parameters and the target characteristics.

[0227] After obtaining the dynamic prediction results, the IEVO algorithm is used to optimize the current machining process in real time. The goal of optimization is to adjust the adjustable parameters of the machining process, such as the cutting speed v s , feed rate f s , grinding depth d s , etc., to achieve the optimal machining effect. Among them, the calculation of the objective function adopts the direct summation method, and the comprehensive objective function F real is:

[0228]

[0229] Among them, F real represents the comprehensive evaluation value of the current machining process. By adjusting the adjustable parameters of the machining process through the optimization algorithm, the goal is to minimize F real or meet the process requirements to ensure the collaborative optimization of the efficiency, quality and stability of the machining process. The optimization algorithm automatically adjusts the adjustable parameters of the machining process according to the real-time feedback and the change of the objective function F real . By comparing the difference between the real-time machining characteristics and the process requirements, adjust the cutting speed v s , feed rate f s , grinding depth d s , etc., the adjustable variables of the machining process, to ensure that the key characteristics in the machining process, such as carbon emissions, economic costs and vibration characteristics, are optimized in real time. When the vibration loss exceeds the set range, the optimization algorithm will adjust the feed rate f s or the cutting speed v s to reduce vibration and improve machining stability.

[0230] Through the real-time feedback mechanism and dynamic adjustment strategy of the optimization algorithm, ensure the best balance of efficiency, quality and stability in the machining process. Whether it is reducing carbon emissions, lowering economic costs, or optimizing vibration characteristics, the algorithm will make precise adjustments among multiple objectives to achieve the optimal machining effect. Among them, the final result of real-time adjustment is the combination of adjustable variables of the optimal machining process This combination can meet the production requirements and ensure the high efficiency, low carbon and stability of the machining process.

[0231] An intelligent control system for machine tools based on big data analysis, including:

[0232] Acquisition module: It is used to collect the key parameter set of the machine tool processing technology, including: cutting speed, feed speed, grinding depth, cutting fluid use, tool material, machine tool performance, environmental temperature data and vibration signals. The SVMD technology is used to decompose the machine tool vibration signals, extract the modal frequency and the fluctuation characteristics of the signals, and construct a feature matrix;

[0233] Model training module: It is used to optimize the BET model parameters by the IEVO algorithm, construct a machine tool intelligent control model based on IEVO-BET, establish the non-linear relationship between the key parameters of the processing technology and carbon emissions, vibration loss and processing cost, and conduct model training and prediction;

[0234] Control module: It is used to adjust the adjustable variables of the processing technology in real time and dynamically according to the trained IEVO-BET machine tool intelligent control model, new process requirements and working condition data through the IEVO algorithm, and conduct multi-variable control over carbon emissions, vibration loss and economic cost during the machine tool processing process.

[0235] The present invention proposes a machine tool intelligent control method and system based on big data analysis, aiming to achieve multi-objective optimization among the processing process parameters, carbon emissions, economic cost and vibration characteristics. This method collects the key parameter set of the processing technology such as cutting speed, feed speed, grinding depth, cutting fluid use, tool material, machine tool performance and environmental temperature, combines the successive variational mode decomposition (SVMD) technology to extract the vibration frequency, amplitude and fluctuation characteristics, and quantifies the vibration frequency and carbon emission characteristics, and integrates these information into a multi-dimensional data set.

[0236] Based on the BET neural network model, a processing dynamic prediction model is constructed, which can capture the non-linear relationship between the key parameters of the processing technology and the optimization objectives. At the same time, the IEVO algorithm is introduced. This algorithm adopts a double-tokenization strategy, combines global exploration and local optimization, and solves the multi-objective optimization of the key parameters of the processing technology, which can effectively optimize high-dimensional multi-constraint problems and support the optimization of the key parameters of the processing technology and model training.

[0237] In practical applications, with process requirements, cutting speed, etc. as inputs, through the collaborative action of the IEVO and BET models, the adjustable variables of the processing technology are adjusted in real time, and the key variables such as cutting speed, feed speed and grinding depth are dynamically optimized to ensure the efficiency, quality and stability of the processing process. The present invention meets the intelligent and real-time machine tool control requirements under different process requirements, effectively reduces carbon emissions, improves production efficiency, and enhances the sustainability of product quality and machine tool performance.

Claims

1. A machine tool intelligent control method based on big data analysis, characterized in that, It includes the following steps: (1) Collect the key parameter set of the machine tool processing technology, including: cutting speed, feed speed, grinding depth, cutting fluid use, tool material, machine tool performance, environmental temperature data and vibration signals. Use the SVMD technology to decompose the machine tool vibration signal, extract the modal frequency and the fluctuation characteristics of the signal, and construct a feature matrix; (2) Design an IEVO algorithm based on the double-tokenization strategy for optimizing the key parameters of the processing technology and model training; (3) Take the key parameter set of the processing technology as the input, and vibration loss, carbon emission, and processing cost as the output, and establish a comprehensive objective function; (4) Optimize the parameters of the BET neural network model through the IEVO algorithm, construct a machine tool intelligent control model based on IEVO-BET, establish a non-linear relationship between the key parameters of the processing technology and carbon emission, vibration loss, and processing cost, and perform model training and prediction; (5) According to the trained IEVO-BET machine tool intelligent control model, as well as the new process requirements and the collected dynamic data, dynamically adjust the adjustable variables of the processing technology in real time through the IEVO algorithm, and perform multi-variable control of carbon emission, vibration loss, and economic cost during the machine tool processing process.

2. The machine tool intelligent control method based on big data analysis according to claim 1, characterized in that The carbon emission is calculated through the electric energy carbon emission factor, and the formula is as follows: E c = P·t·F Among them, Ec is the carbon emission during the processing process, P is the electric power during the processing process, t is the time required for processing, and F is the electric energy carbon emission factor; The total electric energy consumption is calculated by integrating the power signal over time, and the carbon emission is quantified in combination with the electric energy carbon emission factor; for dynamic signals, the power signal is integrated in segments, and the carbon emissions of each segment are accumulated to obtain the total value.

3. The intelligent control method of a machine tool based on big data analysis according to claim 1, characterized in that The calculation method of the vibration loss and the construction method of the feature matrix are as follows: Decomposition of vibration signal: Use the SVMD technology to decompose the collected vibration signal x(t) into multiple intrinsic mode functions, and the formula is as follows: Among them, IMF i (t) is the i-th modal function, and r(t) is the residual term; Extraction of frequency components: By performing Hilbert transform on each intrinsic mode function signal, the instantaneous frequency is obtained: where, φ i (t) is the instantaneous phase of the i-th modal function; Vibration frequency calculation: Statistically analyze the instantaneous frequency of the intrinsic mode function, and take its average frequency value as the vibration frequency during this time period to obtain the main frequency components of the machine tool vibration during the processing process; Assume that the average frequency of the i-th modal function is f i , then the total vibration frequency f total is: where m is the number of intrinsic mode functions, and f i is the average frequency of the i-th mode; Estimation of vibration loss: Using the vibration frequency and amplitude, calculate the vibration loss L vibration , the vibration energy E of each mode i is related to the amplitude A of the vibration i and the frequency f i as follows: Among them, A i is the amplitude of the i-th mode, and f i is the frequency of the i-th mode; Effect of damping on energy loss: Let the damping ratio of each mode be ζ i , then the corresponding energy loss rate is: Among them, is the energy loss rate of the i-th mode, and ζ i is the damping ratio of the i-th mode; Comprehensive vibration loss: Perform weighted summation on the vibration losses of all modes to obtain the total vibration loss of the system: Among them, E i is the vibration energy of the i-th mode, ζ i is the damping ratio corresponding to the mode, and L vibration is the total vibration loss; The feature matrix X structurally integrates the collected key parameters of the processing technology and vibration signals to form input data that can be used for model training and optimization. The matrix includes the following feature items: X = [E c , P, t, F, f total , A1, A2, … A m , ζ1, ζ2, … ζ m ​ Among them, Ec is the carbon emission during the processing, P is the electric power during the processing, t is the time required for processing, F is the carbon emission factor of electric energy, f total is the total vibration frequency, A1, A2, … A m is the amplitude of the m-th mode, ζ1, ζ2, … ζ m is the damping ratio of the m-th mode.

4. The intelligent control method of a machine tool based on big data analysis according to claim 1, characterized in that, The IEVO algorithm described in step (2) is designed as: Initialize the population P as N candidate solutions, where each candidate solution represents a set of key parameter combinations of the machining process, including the cutting speed v s , the feed rate f s , the grinding depth d s , and the tool material. The fitness of each candidate solution is evaluated by the objective function, which contains multiple optimization objectives, including vibration loss, carbon emissions, and machining cost; Chromosome coding and solution space representation: Each candidate solution represents the key parameters of the processing technology through real number coding, and is expressed as follows: X = [E c , P, t, F, f total , A1, A2, … A m , ζ1, ζ2, … ζ m ​ The application implementation of the double-tokenization strategy is as follows: Global exploration token: Introduce non-local solutions to increase the coverage of the solution space, and the formula is as follows: x new = x current + α·Δx Among them, α is the step size factor, and Δx is the random perturbation value; Local refinement token: Based on gradient information or adaptive mutation operation, perform local optimization, and the formula is as follows: where β is the local refinement step size, is the cotton degree of the fitness function; The non - linear convergence factor controls the balance between global search and local optimization, and the adjustment formula is as follows: Among them, γ and δ are adjustment parameters that control the convergence rate; During the optimization process, it is iterated in a loop. The implementation method is as follows: selection is made according to the fitness value. The selection methods include roulette wheel selection and tournament selection, and individuals with high fitness values are selected for crossover or replication; the selected individuals are subjected to gene exchange to generate new candidate solutions to achieve the crossover operation; the newly generated solutions are randomly perturbed; the population is updated according to the fitness value to find the optimal solution; when the maximum number of iterations is reached or the fitness value exceeds the preset threshold, the optimization process stops.

5. A machine tool intelligent control method based on big data analysis according to claim 1, characterized in that, The implementation process of the objective function described in step (3) is as follows: (31) Input the set of key parameters of the processing technology into the neural network. The input data is: X = [E c , P, t, F, f total , A1, A2, … A m , ζ1, ζ2, … ζ m ​ Each sample is standardized to be within a reasonable range: where μ is the mean of the sample, σ is the standard deviation, and x i is each parameter sample of the input data; (32) Establish a comprehensive objective function, defined as follows: L = w1·L1 + w2·L2 + w3·L3 Among them, L1 is the mean square error of vibration loss, L2 is the mean square error of carbon emissions, L3 is the mean square error of processing cost, and w1, w2, w3 are weight coefficients; (33) Calculate the mean square error of vibration loss L1: Among them, the vibration loss predicted by the model is The actual vibration loss is Calculate the mean square error of carbon emissions L2: Among them, the carbon emissions predicted by the model are The actual carbon emissions are Calculate the mean square error of processing cost L3: Among them, the processing cost predicted by the model is The actual processing cost is 6. The machine tool intelligent control method based on big data analysis according to claim 1, characterized in that, The implementation steps of constructing the machine tool intelligent control model based on IEVO - BET described in step (4) are as follows: (41) Use the population initialization method of the IEVO algorithm to generate an initial parameter set. The initial population is expressed as: where each θ i =(d e , d r , L, T pred , η) is a set of randomly generated parameters; (42) Define the prediction model: Among them, is the prediction result of the BET model based on the parameter set θ; (43) Use the IEVO algorithm to update the parameter set. The candidate solution θ is continuously adjusted according to the strategies in the exploration and exploitation stages to optimize the performance of the BET model; Wherein: and respectively represent the adjustments of global search and local search, and α1, α2 are the weight coefficients of global and local search; (44) Input the updated parameter set into the BET model, calculate the mean square error, and evaluate the fitness. Among them, the smaller L is, the lower the fitness value, indicating that the parameter set can better optimize the prediction performance of the BET model. Update the optimal parameter set. If the fitness of the current solution is better, update the optimal solution: After reaching the predetermined termination condition, output the optimal parameter set This parameter set is used as the final configuration of the BET model and applied to the intelligent control system of the machine tool.

7. A machine tool intelligent control method based on big data analysis according to claim 1, characterized in that The BET model described in step (4) consists of a shallow BiLSTM module and a deep Transformer module; BiLSTM layer: The bidirectional LSTM extracts features from time-series data through forward and backward learning. Let the input be x t , and the output of the BiLSTM layer is h t , then the output formula at each time t is expressed as follows: h t = LSTM f (x t ) + LSTM b (x t ) Among them, LSTM f (x t ) is the output of the forward LSTM, and LSTM b x t ) is the output of the backward LSTM; Transformer layer: used to capture long - range dependencies in the sequence. Among them, the output of the self - attention mechanism is: Q, K, V = W q h t , W k h t , W v h t Among them, W q , W k , W v is the weight matrix, and Q, K, and V are the query, key, and value matrices respectively. d k is the dimension of the key.

8. A machine tool intelligent control method based on big data analysis according to claim 1, characterized in that The new process requirements described in step (5) include the product size d product , the material hardness H, and the surface finish Ra. The input data format is as follows: R = [d product , H, Ra] The dynamic data is collected in real time by sensors and monitoring systems, including cutting speed v s , feed rate f s , grinding depth d s , which is expressed as follows: D real = [v sreal , f sreal , d real **] Among them, D real is the dynamic data collected in real time.

9. A machine tool intelligent control method based on big data analysis according to claim 1, characterized in that, The control implementation method described in step (5) is as follows: Input the process requirements and dynamically collected real-time data into the trained BET model. The BET model predicts the changing trends of key characteristics during the processing, including carbon emissions Economic cost and vibration loss Among them, represents the predicted key process characteristic value; After obtaining the results, the IEVO algorithm is used to optimize the current machining process, and the objective function is calculated by the direct summation method. The comprehensive objective function F real is as follows: Among them, F real represents the comprehensive evaluation value of the current processing process; Through this method, multi - variable control of carbon emissions, energy consumption, and economic costs during the machine tool processing process can be achieved, thereby more efficiently optimizing the processing process.

10. An intelligent control system for machine tools based on big data analysis, characterized in that, Including: Acquisition module: used to collect the set of key parameters of the machine tool processing technology, including: cutting speed, feed speed, grinding depth, cutting fluid use, tool material, machine tool performance, and environmental temperature data and vibration signals. Use the SVMD technology to decompose the machine tool vibration signal, extract the modal frequency and the fluctuation characteristics of the signal, and construct a feature matrix; Model training module: used to optimize the BET model parameters by the IEVO algorithm, construct a machine tool intelligent control model based on IEVO - BET, establish a non - linear relationship between the key parameters of the processing technology and carbon emissions, vibration loss, and processing cost, and perform model training and prediction; Control module: It is used to adjust the adjustable variables of the machining process in real time and dynamically through the IEVO algorithm according to the trained IEVO-BET machine tool intelligent control model, new process requirements and working condition data, and perform multivariable control on carbon emissions, vibration loss and economic cost during the machine tool machining process.

Images