Numerical control machine tool cutting parameter optimization method and system based on multi-source data fusion

Abstract

The invention discloses a numerical control machine tool cutting parameter optimization method and system based on multi-source data fusion. The method comprises the following steps: synchronously collecting machine tool state, dynamic response and machining result multi-source data, and carrying out preprocessing and feature association; based on historical data, deriving a material removal rate, a flutter risk, an acoustic emission energy ratio and a surface quality implementation coefficient feature, and constructing an enhanced feature data set; by taking the cutting parameters as input and the features as output, constructing and incrementally updating a machining process response surface agent model by adopting Gaussian process regression; and defining a search space by combining a mechanism hard constraint based on a tool capability, machine tool performance and a stability lobe graph, calling an agent model, and performing optimization by adopting a multi-objective evolutionary algorithm to obtain a Pareto optimal parameter solution set. According to the method, deep coupling of data and a mechanism is achieved, the efficiency can be improved through dynamic optimization on the premise that the machining stability and quality are guaranteed, and high adaptability and reliability are achieved.

Description

Technical Field

[0001] This invention relates to the field of CNC machining technology, and in particular to a method and system for optimizing CNC machine tool cutting parameters based on multi-source data fusion. Background Technology

[0002] CNC machine tools are core equipment in modern manufacturing. Their processing efficiency, accuracy, and cost directly depend on the rationality of the selection of cutting parameters (such as spindle speed, feed rate, and depth of cut). Traditional cutting parameter optimization mainly relies on operator experience, conservative parameters recommended by tool manufacturers, and simulations based on a single physical model. These methods have significant limitations: First, the "trial and error" method, which relies on human experience, is inefficient. Parameter selection varies from person to person, making it difficult to guarantee global optimality and easily leading to abnormal tool wear or even workpiece scrap. Second, the parameters recommended by tool manufacturers are usually conservative to ensure the broadest applicability and safety, failing to fully explore the processing potential of a specific machine tool-tool-workpiece combination, resulting in efficiency losses. Third, while simulation optimization based on a single theoretical model (such as a cutting force model or vibration model) can guide parameter selection to some extent, the model is often based on numerous simplifications and assumptions, making it difficult to accurately reflect the complex and ever-changing actual machining conditions.

[0003] In recent years, with the development of sensor technology, parameter optimization using machining process data has become a research hotspot. For example, monitoring machining status and adjusting parameters by collecting spindle power or vibration signals. However, such methods usually have the following problems: First, the data source is singular, relying on only one or a few types of sensor data (such as monitoring only vibration or only force), resulting in insufficient information dimensions and difficulty in comprehensively and accurately characterizing the multidimensional and coupled physical phenomena in the machining process (such as chatter, tool wear, and surface formation, which are comprehensive results of force, heat, and vibration). Second, data is disconnected from knowledge. Simple data monitoring and feedback control fail to be effectively combined with deeper machining mechanisms (such as cutting dynamics and material removal mechanisms), leading to a lack of theoretical support for optimization decisions, poor adaptability and robustness, and a sharp decline in performance when the working conditions change beyond the range of historical data. Third, the optimization objectives are one-sided, often pursuing only a single indicator (such as the highest material removal rate or the longest tool life), failing to achieve a dynamic balance among multiple constraints such as efficiency, quality, and cost.

[0004] Therefore, existing technologies lack a method for intelligently optimizing CNC machine tool cutting parameters that can deeply integrate multi-source heterogeneous process data with machining mechanism knowledge, thereby achieving high adaptability, high reliability, and multi-objective dynamic balance. This invention aims to solve the above problems. Summary of the Invention

[0005] To achieve the above objectives, this invention provides a method for optimizing CNC machine tool cutting parameters based on multi-source data fusion, the method comprising the following steps:

[0006] Step 1: During CNC machine tool cutting, synchronously collect machine tool status data, dynamic response data, and machining result data, and preprocess the collected raw data. Then, associate the preprocessed data with the corresponding cutting parameters to form machining process data records and store them.

[0007] Step 2: Extract data from the processing data records, perform feature derivation and correlation analysis, calculate the material removal rate, analyze flutter risk, calculate the acoustic emission energy ratio per unit material removal rate, and calculate the surface quality achievement coefficient to form an enhanced feature dataset;

[0008] Step 3: Using cutting parameters as input variables and material removal rate, vibration intensity index, acoustic emission energy ratio per unit material removal rate, and surface quality achievement coefficient as output variables, construct and incrementally update the processing response surface proxy model based on the enhanced feature dataset;

[0009] Step 4: Determine the optimization objectives as maximizing material removal rate, minimizing vibration intensity index, and minimizing surface quality achievement coefficient, and use the acoustic emission energy ratio per unit material removal rate as a constraint. Combine the hard constraints based on the processing mechanism to define the cutting parameter search space. Within the search space, call the processing response surface proxy model to perform multi-objective optimization and obtain the Pareto optimal solution set.

[0010] Step 5: Select a cutting parameter scheme from the Pareto optimal solution set for trial machining, collect trial machining data and calculate the actual machining index, compare and verify the actual machining index with the predicted index of the machining process response surface surrogate model, use the verified cutting parameter scheme for mass production and store the trial machining data in the machining process data record.

[0011] Preferably, in step one:

[0012] The machine tool status data includes the actual spindle speed, the actual feed rate of each feed axis, and the spindle load current, which are read in real time through the internal data interface of the CNC system.

[0013] The dynamic response data includes the spindle radial vibration signal acquired by a vibration acceleration sensor mounted on the spindle box, and the acoustic emission signal acquired by an acoustic emission sensor mounted on the worktable.

[0014] The processing result data includes surface roughness profile data obtained by scanning the surface of the processed workpiece using a non-contact laser displacement sensor;

[0015] The preprocessing includes:

[0016] Calculate the root mean square value of the spindle load current over one machining cycle;

[0017] The radial vibration signal of the spindle is subjected to bandpass filtering related to the spindle rotation frequency and the characteristic frequency of the cutting process, and the root mean square value of the filtered signal in one machining cycle is calculated as the vibration intensity index.

[0018] Calculate the root mean square energy value of the acoustic emission signal over one processing cycle;

[0019] The arithmetic mean deviation value of the surface roughness profile data is calculated according to the standard algorithm and used as the surface roughness index.

[0020] The cutting parameters include the spindle set speed, set feed rate, and depth of cut.

[0021] Preferably, in step two:

[0022] The calculation of the material removal rate is specifically as follows: multiply the spindle set speed, set feed rate, and depth of cut by a constant coefficient determined by the tool geometry parameters.

[0023] The analysis of chatter risk specifically involves: calculating the product of the spindle set speed and the number of tool teeth to obtain the spindle tooth frequency per revolution; analyzing whether abnormal peak values ​​appear in the vibration intensity index near the spindle tooth frequency per revolution or its multiples; if so, marking the machining process as having a suspected chatter risk.

[0024] The calculation of the acoustic emission energy ratio per unit material removal rate is specifically performed by dividing the root mean square energy value of acoustic emission by the material removal rate.

[0025] The calculation of the surface quality achievement coefficient specifically involves: calculating the theoretical roughness value using a theoretical formula based on the cutting parameters and tool geometric radius, and then dividing the surface roughness index by the theoretical roughness value to obtain the surface quality achievement coefficient.

[0026] Preferably, in step three:

[0027] The process response surface proxy model is constructed using the Gaussian process regression method.

[0028] The Gaussian process regression method can provide a mean estimate and a variance estimate for each predicted output, where the variance estimate represents the uncertainty of the model's prediction at that input point.

[0029] The incremental update process response surface surrogate model refers to the method of updating the hyperparameters of a Gaussian process regression model by incremental learning when new enhanced feature data is added.

[0030] Preferably, in step four:

[0031] The hard constraints based on processing mechanisms include:

[0032] Calculate the maximum permissible spindle speed constraint based on the maximum permissible cutting line speed of the tool;

[0033] Based on the maximum torque of the machine tool spindle and the maximum thrust of the feed axis, and combined with the cutting force coefficient estimation model, the maximum allowable feed speed constraint is calculated under the current cutting depth.

[0034] Based on the modal parameters of the machine tool spindle system and workpiece system obtained through pre-modal tests, the maximum depth of cut constraint allowed to maintain stable cutting at different spindle speeds is calculated based on the stability lobe diagram theory.

[0035] Preferably, in step four:

[0036] The process of calling the process response surface proxy model within the search space for multi-objective optimization specifically employs a multi-objective evolutionary algorithm.

[0037] The multi-objective evolutionary algorithm uses the material removal rate, vibration intensity index, and surface quality realization coefficient predicted by the process response surface surrogate model as optimization objectives, and takes the acoustic emission energy ratio per unit material removal rate predicted by the process response surface surrogate model as a constraint that does not exceed an empirical threshold, and performs iterative optimization within the search space defined by hard constraints.

[0038] The Pareto optimal solution set is output by a multi-objective evolutionary algorithm and is a set of cutting parameter schemes that achieve the best trade-off among multiple objectives such as material removal rate, vibration intensity index, and surface quality achievement coefficient.

[0039] Preferably, in step five:

[0040] The specific method for selecting cutting parameters from the Pareto optimal solution set is as follows: selection is made according to the priority strategy of the current production task, including efficiency priority strategy, quality priority strategy or equilibrium strategy.

[0041] The efficiency-first strategy selects the solution with the highest material removal rate in the Pareto optimal solution set.

[0042] The quality-first strategy selects the solution in the Pareto optimal solution set whose surface quality realization coefficient is closest to 1;

[0043] The equilibrium strategy selects the solution with the optimal weighted sum of the normalized objective function values ​​in the Pareto optimal solution set.

[0044] Preferably, in step five:

[0045] The comparative verification specifically involves comparing the vibration intensity index and surface quality achievement coefficient actually obtained from the trial machining with the predicted values ​​of the same cutting parameters by the response surface surrogate model of the machining process.

[0046] If the deviations between the actual and predicted values ​​are both within the preset acceptable error range, the verification is successful.

[0047] If any deviation between the actual value and the predicted value exceeds the preset acceptable error range, the verification fails and an abnormal operating condition warning is issued.

[0048] Preferably, before the method is applied for the first time, modal tests need to be conducted in advance to obtain the modal parameters of the machine tool spindle system and the workpiece system, and a certain number of machining process data records covering different combinations of cutting parameters need to be accumulated in advance to complete the initial training of the machining process response surface proxy model.

[0049] A CNC machine tool cutting parameter optimization system based on multi-source data fusion for implementing the method, the system comprising:

[0050] A CNC machine tool, which is equipped with a CNC system, a spindle, a feed axis, a cutting tool, and a workpiece clamping device;

[0051] The data acquisition component includes a data interface that communicates with the CNC system, a vibration acceleration sensor mounted on the spindle box, an acoustic emission sensor mounted on the worktable, and a non-contact laser displacement sensor mounted on the machine tool.

[0052] A computing and storage device, which is communicatively connected to a data acquisition component and a numerical control system;

[0053] The computing and storage device is configured to perform the following operations:

[0054] The data acquisition component synchronously acquires and preprocesses the processing data to form a processing data record.

[0055] Data is extracted from processing data records for feature engineering and correlation analysis to construct an enhanced feature dataset.

[0056] A proxy model for the processing response surface is constructed and incrementally updated based on the enhanced feature dataset;

[0057] By combining multi-objective optimization algorithms with hard constraints of machining mechanisms, a proxy model of the machining process response surface is invoked to optimize cutting parameters and generate a Pareto optimal solution set.

[0058] Implement a closed-loop process for selecting, verifying, applying, and feeding back the parameter scheme, and send the verified cutting parameter scheme to the CNC system for execution.

[0059] The beneficial effects of this invention are:

[0060] 1. This invention overcomes the information limitations caused by traditional methods relying on a single data source by simultaneously acquiring multi-source heterogeneous data such as machine tool electrical status, dynamic response, and machining results, and performing targeted feature engineering on these data. This fusion enables the system to simultaneously characterize multi-dimensional physical effects of the cutting process, including force, heat, vibration, and surface formation, providing a richer and more accurate information foundation for subsequent optimization. This allows optimization decisions to more closely reflect the complex and ever-changing actual working conditions.

[0061] 2. This invention does not rely solely on data modeling. Instead, it utilizes mechanistic knowledge to guide feature construction at the feature layer and directly applies physics-based stability leaflet diagrams and machine tool / tool ​​capability limits as hard boundaries for the optimization search space at the constraint layer. This dual-layer fusion mechanism of "data features + mechanistic constraints" ensures that the optimization scheme not only conforms to data patterns but also strictly satisfies physical feasibility and safety requirements. This significantly improves the system's adaptability and decision-making robustness when facing unfamiliar working conditions, avoiding absurd or dangerous outputs that might arise from pure data models.

[0062] 3. This invention forms a complete closed loop, from multi-source data acquisition, surrogate model construction and incremental updates, to multi-objective Pareto optimization, solution verification and feedback. The system can automatically verify and update the model based on actual processing results, adapting to slow time-varying factors such as tool wear. Simultaneously, through multi-objective optimization algorithms, it automatically finds the optimal balance point among multiple competing objectives such as efficiency, stability, and quality, and provides a solution set that can be flexibly selected according to production strategies. Thus, while ensuring processing safety and quality, it continuously and dynamically explores the processing potential of specific machine tool-tool-workpiece combinations, achieving a fundamental shift from static experience-based settings to dynamic intelligent optimization. Attached Figure Description

[0063] To more clearly illustrate the technical solutions in this invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, those skilled in the art can obtain other drawings based on these drawings without creative effort.

[0064] Figure 1 This is a flowchart of the steps of the method of the present invention. Detailed Implementation

[0065] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. It should also be noted that, to make the embodiments more comprehensive, the following embodiments are the best and preferred embodiments, and those skilled in the art can use other alternative methods to implement some well-known technologies; moreover, the accompanying drawings are only for more specific description of the embodiments and are not intended to specifically limit the present invention.

[0066] Please see Figure 1 This invention provides a method for optimizing CNC machine tool cutting parameters based on multi-source data fusion. The method first involves data acquisition and preprocessing. During milling, the CNC system's internal data interface reads the actual spindle speed, the actual feed rates of the X / Y / Z axes, and the load current of the spindle motor in real time at a sampling frequency of 100 Hz.

[0067] Simultaneously, a triaxial vibration accelerometer mounted on the side of the spindle box collects vibration signals at a sampling frequency of 10 kHz, while an acoustic emission sensor mounted on the worktable collects signals at a sampling frequency of 1 MHz. After one machining operation is completed, the machine tool is controlled to move so that a non-contact laser displacement sensor mounted next to the tool magazine scans the machined surface of the workpiece along a direction perpendicular to the tool marks.

[0068] During preprocessing, the spindle load current data is divided into machining cycles based on a single tool path, and the root mean square (RMS) value of the current within each cycle is calculated. The vibration signal, after being bandpass filtered from 100 Hz to 2000 Hz, also has its RMS value calculated for each machining cycle as a vibration intensity index. The acoustic emission signal has its RMS energy calculated for each machining cycle. The contour data obtained from laser scanning is used to calculate the arithmetic mean deviation Ra value of the contour according to the national standard GB / T1031.

[0069] Finally, the root mean square value of current, vibration intensity, acoustic emission energy, and surface roughness Ra value obtained from the above processing are bound to the cutting parameter combination specifically used in this processing and a unique test number to form a complete record and stored in the database table.

[0070] In one possible implementation, all historical records are extracted from a machining process database table. For each record, the material removal rate is calculated based on the spindle speed setting, feed per tooth, axial depth of cut, and tool diameter in the record. Specifically, this is achieved by multiplying the spindle speed, feed rate, depth of cut, and a coefficient related to the tool diameter and number of cutting edges.

[0071] To analyze chatter, the tooth passage frequency is calculated based on the spindle speed setting and the number of cutting edges. The vibration signal spectrum corresponding to the record is then checked for peaks significantly higher than the background noise at the tooth passage frequency and its harmonics. If any are found, a "chatter risk" flag is marked in the record. Furthermore, the acoustic emission energy value of this record is divided by its material removal rate to obtain the "specific acoustic emission energy."

[0072] Simultaneously, based on the feed per tooth and tool radius used, the theoretical residual height is calculated through geometric relationships and used as a theoretical surface roughness reference value. Then, the actual measured Ra value is divided by this theoretical value to obtain a "surface forming coefficient". Finally, for each record, four derived features are added to the original data: material removal rate, chatter risk indicator, specific acoustic emission energy, and surface forming coefficient, which together constitute an enhanced dataset for modeling.

[0073] In one possible implementation, the spindle speed setting, feed per tooth, and axial depth of cut from the augmented dataset are selected as the three input dimensions of the model. Material removal rate, vibration intensity index, specific acoustic emission energy, and surface forming coefficient are selected as the four output dimensions of the model. A Gaussian process regression algorithm is used to establish a nonlinear mapping relationship between the inputs and outputs. The model is trained using a historical augmented dataset. During training, the Gaussian process regression learns a covariance function and its hyperparameters to characterize the inherent patterns in the data.

[0074] After training, for any new set of cutting parameters, the model can not only predict the mean of the four output values, but also provide the variance of each predicted value, which represents the confidence level of the model's prediction at that point. When the system conducts new machining experiments and obtains new augmented data, an incremental update algorithm is used to fine-tune the hyperparameters of the original Gaussian process model based on the new data, enabling the model to track the gradual changes in machining system performance caused by factors such as tool wear.

[0075] In one possible implementation, the optimization objectives are first set: maximize material removal rate, minimize vibration intensity index, and make the surface finish coefficient as close to 1 as possible. Simultaneously, constraints are set: the specific acoustic emission energy must not exceed a threshold set according to the tool coating material. Then, mechanistic hard constraints are applied to define the search boundary: the maximum permissible spindle speed is calculated by dividing the maximum linear velocity indicated on the tool holder by the current tool diameter. Based on the maximum spindle torque and maximum feed axis thrust in the machine tool manual, combined with the cutting force coefficient calibrated for the current workpiece material, the maximum permissible feed rate curve without overload at different cutting depths is derived.

[0076] Using the frequency response function of the spindle-tool holder-tool combination obtained in advance through hammer impact experiments, a stability lobe diagram is plotted to determine the maximum allowable depth of cut without chatter at different rotational speeds. Finally, within the parameter space formed by the aforementioned multidimensional boundaries, a trained Gaussian process surrogate model is invoked, and the NSGA-II multi-objective genetic algorithm is used for searching. The algorithm iterates with the output predicted by the surrogate model as the objective, ultimately outputting a set of non-dominated solutions, i.e., the Pareto optimal solution set.

[0077] In one possible implementation, from the Pareto optimal solution set, the cutting parameter scheme with the highest predicted material removal rate is selected according to an "efficiency-first" strategy. The spindle speed, feed per tooth, and axial depth of cut settings of this scheme are sent to the CNC system, and trial cuts are performed on the same workpiece material. During the trial cut, current, vibration, and acoustic emission data are simultaneously collected strictly according to step one, and the surface roughness is measured after machining. After data processing, the actual material removal rate, vibration intensity, specific acoustic emission energy, and surface forming coefficient of the trial cut are calculated.

[0078] The actual vibration intensity was compared with the predicted vibration intensity of the surrogate model for this set of parameters. The absolute value of the deviation was less than 15% of the predicted value; at the same time, the deviation between the actual surface forming coefficient and the predicted value was less than 0.1. Both key indicators were within the allowable range, and the verification was deemed successful. This set of parameters was confirmed to be usable for subsequent mass production. All data from this trial cut was assigned a new test number and stored as a new record in the database to trigger incremental updates to the model.

[0079] In one possible implementation, to achieve high-precision time synchronization, all external sensors are connected to the same synchronous data acquisition unit, which receives the pulse signal per revolution from the CNC system as an external clock reference. The CNC system transmits internal data outward via the OPCUA protocol at 10-millisecond intervals. Upon receiving the rising edge of each revolution pulse, the data acquisition unit timestamps the data from all channels.

[0080] For vibration signals, an additional trend term removal step is performed before calculating the root mean square value to eliminate the influence of temperature drift. For acoustic emission signals, a 100 kHz high-pass filter is applied before calculating the root mean square energy to remove low-frequency mechanical noise interference. Before scanning, the laser displacement sensor requires height calibration using standard gauge blocks to ensure accurate measurement reference. All preprocessing algorithms are encapsulated in independent software modules to ensure consistency and repeatability of the data processing flow.

[0081] In one possible implementation, analyzing flutter risk involves not only examining the peak value at the tooth passage frequency but also performing the following steps: First, a short-time Fourier transform is performed on the vibration acceleration signal to generate a time-spectrum. Then, the presence of a "bird's wing" pattern with frequency increasing or decreasing over time is identified on the time-spectrum, a typical characteristic of flutter development. Second, the ratio of the energy of the vibration signal in the 200 Hz to 800 Hz frequency band to the energy in the 50 Hz to 200 Hz frequency band is calculated; an abnormally high increase in this ratio often indicates a decrease in stability.

[0082] Finally, by combining the results of tooth frequency peak detection, time-frequency pattern recognition, and frequency band energy ratio, a simple voting mechanism is adopted: if two of the three indicators are abnormal, the machining process is ultimately determined to have a chatter risk, and this is marked in the record. This multi-indicator comprehensive judgment improves the reliability of chatter identification.

[0083] In one possible implementation, the Gaussian process regression model uses the squared exponential covariance function as the kernel function. Before model training, the input and output data need to be standardized to have a mean of 0 and a standard deviation of 1 to improve numerical stability. The model's hyperparameters include length scale, signal variance, and noise variance, which are optimized by maximizing the marginal likelihood function. During the incremental update phase, a model update is triggered when the number of new data points accumulates to 10. The update is not retraining, but rather an approximation method based on Bayesian updates. A new posterior distribution is calculated based on the existing posterior distribution of the hyperparameters, combined with the new data, and the hyperparameters are adjusted accordingly. This method significantly reduces computational overhead while ensuring model adaptability, meeting the needs of online workshop applications.

[0084] In one possible implementation, the cutting force coefficient estimation model is obtained through a set of pre-designed orthogonal cutting experiments. In these experiments, milling is performed using different cutting parameters while the cutting force is measured using a force gauge. The radial and tangential cutting force coefficients for a specific tool-workpiece material pair are calibrated by linearly regressing the measured forces against the predicted forces based on the mechanical model.

[0085] The specific steps for plotting the stability lobe diagram are as follows: Substitute the frequencies, damping ratios, and modal stiffness parameters of the dominant modes of the spindle system, identified beforehand through modal testing, into the milling dynamics equations; calculate the stability limits under different combinations of spindle speeds and axial depths of cut by solving the eigenvalues ​​of the equations; finally, plot the lobe diagram curve with spindle speed as the abscissa and critical depth of cut as the ordinate. During the optimization process, the depth of cut for any candidate parameter must be less than the critical depth given by the lobe diagram at the corresponding spindle speed.

[0086] Accordingly, embodiments of the present invention also provide a CNC machine tool cutting parameter optimization system based on multi-source data fusion for implementing the method, the system comprising:

[0087] A CNC machine tool, which is equipped with a CNC system, a spindle, a feed axis, a cutting tool, and a workpiece clamping device;

[0088] The data acquisition component includes a data interface that communicates with the CNC system, a vibration acceleration sensor mounted on the spindle box, an acoustic emission sensor mounted on the worktable, and a non-contact laser displacement sensor mounted on the machine tool.

[0089] A computing and storage device, which is communicatively connected to a data acquisition component and a numerical control system;

[0090] The computing and storage device is configured to perform the following operations:

[0091] The data acquisition component synchronously acquires and preprocesses the processing data to form a processing data record.

[0092] Data is extracted from processing data records for feature engineering and correlation analysis to construct an enhanced feature dataset.

[0093] A proxy model for the processing response surface is constructed and incrementally updated based on the enhanced feature dataset;

[0094] By combining multi-objective optimization algorithms with hard constraints of machining mechanisms, a proxy model of the machining process response surface is invoked to optimize cutting parameters and generate a Pareto optimal solution set.

[0095] Implement a closed-loop process for selecting, verifying, applying, and feeding back the parameter scheme, and send the verified cutting parameter scheme to the CNC system for execution.

[0096] This invention encompasses any substitutions, modifications, equivalent methods, and solutions made within the spirit and scope of this invention. To provide the public with a thorough understanding of this invention, specific details are described in detail in the following preferred embodiments; however, those skilled in the art will fully understand the invention even without these details. Furthermore, to avoid unnecessary misunderstanding of the essence of this invention, well-known methods, processes, procedures, components, and circuits are not described in detail.

[0097] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A numerical control machine tool cutting parameter optimization method based on multi-source data fusion, characterized in that, The method comprises the following steps: Step one: during the cutting process of the numerical control machine tool, machine tool state data, dynamic response data and processing result data are synchronously collected, the collected original data is preprocessed, the preprocessed data is associated with corresponding cutting parameters to form processing process data records and is stored; Step two: data is extracted from the processing process data records, feature derivation and correlation analysis is performed, material removal rate is calculated, chatter risk is analyzed, unit material removal rate acoustic emission energy ratio is calculated, surface quality realization coefficient is calculated, and an enhanced feature data set is constituted; Step three: based on the enhanced feature data set, a processing process response surface proxy model is constructed and incrementally updated, taking the cutting parameters as input variables and taking the material removal rate, vibration intensity index, unit material removal rate acoustic emission energy ratio and surface quality realization coefficient as output variables; Step four: the search space of the cutting parameters is defined by taking the maximum material removal rate, the minimum vibration intensity index and the minimum surface quality realization coefficient as optimization objectives, taking the unit material removal rate acoustic emission energy ratio as a constraint, and combining the hard constraint based on the processing mechanism, the processing process response surface proxy model is called in the search space to perform multi-objective optimization, and a Pareto optimal solution set is obtained; Step five: a cutting parameter scheme is selected from the Pareto optimal solution set to perform trial processing, trial processing data is collected and actual processing indexes are calculated, the actual processing indexes are compared with the predicted indexes of the processing process response surface proxy model, and the cutting parameter scheme that passes the verification is used for batch production and the trial processing data is stored in the processing process data records.

2. The multi-source data fusion based cutting parameter optimization method for CNC machine tools according to claim 1, characterized in that, In step one: The machine tool state data includes the actual spindle speed, the actual feed speed of each feed shaft and the spindle load current which are read in real time through the internal data interface of the numerical control system; The dynamic response data includes the spindle radial vibration signal collected by the vibration acceleration sensor installed on the spindle box and the acoustic emission signal collected by the acoustic emission sensor installed on the workbench; The processing result data includes the surface roughness profile data obtained by scanning the machined workpiece surface using a non-contact laser displacement sensor; The preprocessing includes: The root mean square value of the spindle load current in a processing cycle is calculated; The spindle radial vibration signal is band-pass filtered in relation to the machine tool spindle rotation frequency and the cutting process characteristic frequency, and the root mean square value of the filtered signal in a processing cycle is calculated as the vibration intensity index; The root mean square energy value of the acoustic emission signal in a processing cycle is calculated; The arithmetic average deviation value of the surface roughness profile data is calculated as the surface roughness index according to the standard algorithm; The cutting parameters include the spindle set speed, the set feed speed and the cutting depth.

3. The multi-source data fusion based cutting parameter optimization method for CNC machine tools according to claim 2, characterized in that, In step two: The calculation of the material removal rate specifically is: multiplying the spindle set speed, the set feed speed and the cutting depth, and then multiplying by a constant coefficient determined by the tool geometric parameters; The analyzing chatter risk specifically comprises: calculating a product of the spindle set speed and the tool tooth number to obtain a tooth frequency per revolution of the spindle, analyzing whether an abnormal peak value of the vibration intensity index appears near the tooth frequency per revolution of the spindle or a multiple frequency thereof, and marking the current machining as having a suspected chatter risk if the abnormal peak value appears; The calculated unit material removal rate acoustic emission energy ratio specifically comprises: dividing the acoustic emission root mean square energy value by the material removal rate; The calculated surface quality realization coefficient specifically comprises: calculating a theoretical roughness value according to a theoretical formula based on the cutting parameters and the tool geometric radius, and dividing the surface roughness index by the theoretical roughness value to obtain the surface quality realization coefficient.

4. The multi-source data fusion based cutting parameter optimization method for CNC machine tools according to claim 1, characterized in that, In step three: The machining process response surface proxy model is constructed by using a Gaussian process regression method; The Gaussian process regression method can provide a mean value estimation and a variance estimation for each predicted output, and the variance estimation represents the uncertainty of the model in predicting at the input point; The incremental updating of the machining process response surface proxy model refers to that, when new enhanced feature data is added, the hyperparameters of the Gaussian process regression model are updated in an incremental learning manner.

5. The multi-source data fusion based CNC machine tool cutting parameter optimization method according to claim 1, characterized in that, In step four: The hard constraints based on the machining mechanism include: calculating a highest allowable spindle speed constraint according to a maximum allowable cutting line speed of the tool; calculating a maximum allowable feed speed constraint at the current cutting depth according to a maximum torque of the machine tool spindle and a maximum thrust of the feed shaft, and in combination with a cutting force coefficient estimation model; calculating a maximum cutting depth constraint allowed for stable cutting at different spindle speeds based on a stability lobe diagram theory according to modal parameters of the machine tool spindle system and the workpiece system obtained through a pre-modal test.

6. The multi-source data fusion based cutting parameter optimization method for CNC machine tools according to claim 5, characterized in that, In step four: The multi-objective optimization in the search space by calling the machining process response surface proxy model specifically uses a multi-objective evolutionary algorithm; The multi-objective evolutionary algorithm takes the material removal rate, the vibration intensity index and the surface quality realization coefficient predicted by the machining process response surface proxy model as optimization objectives, takes the unit material removal rate acoustic emission energy ratio predicted by the machining process response surface proxy model being not more than an empirical threshold as a constraint, and performs iterative optimization in the search space defined by the hard constraints; The Pareto optimal solution set is output by the multi-objective evolutionary algorithm, and is a set of cutting parameter schemes that achieve the best trade-off among the material removal rate, the vibration intensity index and the surface quality realization coefficient.

7. The multi-source data fusion based CNC machine tool cutting parameter optimization method according to claim 1, characterized in that, In step five: The selection of the cutting parameter scheme from the Pareto optimal solution set specifically comprises: selecting according to a priority strategy of the current production task, and the priority strategy includes an efficiency priority strategy, a quality priority strategy or a balanced strategy; The efficiency priority strategy selects a scheme with the highest material removal rate in the Pareto optimal solution set; The quality priority strategy selects a scheme with the closest surface quality realization coefficient to 1 in the Pareto optimal solution set; The balanced strategy selects a scheme with the optimal weighted sum of the normalized objective function values in the Pareto optimal solution set.

8. The multi-source data fusion based CNC machine tool cutting parameter optimization method according to claim 1, characterized in that, In step five: The comparison and verification specifically comprise: comparing the vibration intensity index and the surface quality realization coefficient actually obtained in the trial machining with the predicted values of the machining process response surface proxy model for the same cutting parameters. If the deviation of the actual value from the predicted value is within the preset acceptable error range, the verification is passed; If the deviation of any actual value from the predicted value exceeds the preset acceptable error range, the verification is failed, and a working condition abnormality warning is issued.

9. The multi-source data fusion based CNC machine tool cutting parameter optimization method according to claim 1, characterized in that, The method needs to perform a modal test to obtain modal parameters of the machine tool spindle system and the workpiece system before being applied for the first time, and a certain number of machining process data records covering different cutting parameter combinations are accumulated in advance to complete the initial training of the machining process response surface proxy model.

10. A multi-source data fusion based CNC machine tool cutting parameter optimization system for implementing the method of any one of claims 1 to 9, characterized by, The system comprises: A numerical control machine tool equipped with a numerical control system, a spindle, a feed shaft, a tool, and a workpiece clamping device; A data acquisition assembly comprising a data interface in communication connection with the numerical control system, a vibration acceleration sensor installed on the spindle box, an acoustic emission sensor installed on the workbench, and a non-contact laser displacement sensor installed on the machine tool; A computing and storage device in communication connection with the data acquisition assembly and the numerical control system; The computing and storage device is configured to perform the following operations: Synchronously acquiring machining process data through the data acquisition assembly and preprocessing to form machining process data records; Extracting data from the machining process data records for feature engineering and correlation analysis to construct an enhanced feature data set; Constructing and incrementally updating the machining process response surface proxy model based on the enhanced feature data set; Combining a multi-objective optimization algorithm and machining mechanism hard constraints, calling the machining process response surface proxy model to optimize the cutting parameters, and generating a Pareto optimal solution set; Implementing a parameter scheme selection, verification, application, and data feedback closed-loop process, and sending the cutting parameter scheme that passes the verification to the numerical control system for execution.

Images