Intelligent metal cutting regulation and control method based on multi-sensor data fusion

Abstract

The invention discloses an intelligent metal cutting regulation and control method based on multi-sensor data fusion, and particularly relates to the technical field of intelligent manufacturing and precision machining. Extracting a metal physical feature set, performing physical failure mechanism matching, and outputting a metal cutting health state vector; constructing a process influence quantification model, and outputting a multi-target influence vector; constructing a multi-objective loss function, and outputting an optimized metal cutting parameter set value; and the optimized metal cutting parameter set value is issued to a machine tool to execute regulation and control, and a self-adaptive closed loop is formed. The intelligent metal cutting regulation and control method based on multi-sensor data fusion is constructed through multi-source heterogeneous sensing data acquisition and preprocessing, feature fusion and state diagnosis, process influence quantitative evaluation, multi-target collaborative optimization decision making and instruction issuing regulation and control. The problems of single monitoring dimension, weak feature association, lack of quantitative prediction and multi-target optimization and the like are solved.

Description

Technical Field

[0001] This invention relates to the field of intelligent manufacturing and precision machining technology, and more specifically, to an intelligent metal cutting control method based on multi-sensor data fusion. Background Technology

[0002] Metal cutting is one of the core processes in modern manufacturing, and its processing quality, efficiency, and stability directly affect the performance of the final product and production costs. Traditional metal cutting processes rely heavily on operator experience to set process parameters and on manual observation, offline detection, or single-sensor monitoring for status assessment and intervention during processing. This approach is not only highly dependent on operator experience but also struggles to achieve real-time, precise process adjustments, easily leading to fluctuations in processing quality, unexpected tool failures, and machine chatter, thus impacting processing efficiency and part consistency. While some existing multi-sensor monitoring solutions attempt to integrate multi-source signals, they largely remain at the level of feature extraction and simple threshold alarms, lacking systematic modeling and reasoning of the physical failure mechanisms of cutting, resulting in insufficient accuracy and interpretability of status diagnosis.

[0003] Therefore, existing metal cutting process monitoring and control technologies still have the following technical shortcomings:

[0004] First, the monitoring dimensions are limited: relying on signals from a single type of sensor makes it difficult to comprehensively and accurately identify various typical failure modes such as chatter, tool wear, and chipping, which can easily lead to missed or false alarms.

[0005] Second, the correlation between features and mechanisms is weak: the extracted signal features often lack clear physical failure mechanisms, state diagnosis relies on empirical thresholds, has poor interpretability and weak adaptability.

[0006] Third, there is a lack of quantitative prediction and multi-objective optimization: existing methods focus more on state identification and alarm, and fail to establish a quantitative prediction model between the processing state and the final processing target, and even more so lack the ability to optimize dynamic parameters under multi-objective constraints.

[0007] Fourth, the closed-loop adaptive capability is insufficient: most schemes have not formed a full closed-loop adaptive control that includes monitoring, diagnosis, prediction, optimization, regulation, and verification. They cannot iterate and optimize in real time based on the regulation effect, making it difficult to ensure the continuous stability and optimality of the processing process.

[0008] To address the aforementioned issues, there is an urgent need for an intelligent metal cutting control method that can integrate multi-source sensor information, perform accurate state diagnosis based on physical mechanisms, and achieve multi-objective dynamic optimization and closed-loop adaptive control. Summary of the Invention

[0009] To overcome the aforementioned deficiencies of the prior art, this invention provides an intelligent metal cutting control method based on multi-sensor data fusion, which solves the problems mentioned in the background art through the following scheme.

[0010] To achieve the above objectives, the present invention provides the following technical solution: an intelligent metal cutting control method based on multi-sensor data fusion, the method comprising:

[0011] S1: Synchronous acquisition and preprocessing of multi-source heterogeneous sensor data. Multi-source sensors are installed on the machine tool to synchronously acquire triaxial force signals, vibration signals and acoustic emission signals during the metal cutting process, and preprocess all signals by noise reduction, normalization and time-frequency domain transformation.

[0012] S2: Extract the metal physical feature set from the preprocessed signal. The metal physical feature set includes: vibration energy in a specific frequency band, cutting force ratio and fluctuation rate, and burst count rate of acoustic emission signal. Input the metal physical feature set into the knowledge base inference engine to perform physical failure mechanism matching and diagnosis, and output the metal processing health status vector.

[0013] S3: Based on the output metal processing health status vector and combined with the current process parameters, a process influence quantification model consisting of a semi-empirical analytical sub-model is used to predict its quantitative influence value on the final processing target, and the output is a multi-objective influence vector.

[0014] S4: Receive the multi-objective influence vector and construct a multi-objective loss function based on the preset and dynamically adjustable optimization weight matrix, and output the optimized metal cutting parameter settings;

[0015] S5: Send the optimized metal cutting parameter settings to the machine tool for control and continuously collect the controlled sensor data, return to S2 to verify the control effect and update the metal processing health status diagnosis, forming an adaptive closed loop.

[0016] The technical effects and advantages of this invention are as follows:

[0017] 1. This invention, by installing multi-source sensors and achieving synchronous acquisition, can simultaneously acquire triaxial force, vibration and acoustic emission signals during the metal cutting process, breaking through the limitations of traditional methods that rely on a single signal source and improving failure identification capabilities.

[0018] 2. This invention constructs a knowledge base reasoning engine based on physical failure mechanisms. It adopts production rule expression (IF-THEN) and feature threshold matching to realize the mechanism mapping from signal features to specific failure modes. Through the rule priority conflict resolution mechanism, it supports multi-feature fusion reasoning, so that the state diagnosis process has clear physical interpretability and enhances the mechanism-driven diagnosis and state interpretability and adaptability.

[0019] 3. This invention establishes a process influence quantification model composed of semi-empirical analytical sub-models, which can combine the processing health status with process parameters to quantify and predict its impact on key objectives such as surface roughness, material removal rate, and tool life. On this basis, a multi-objective loss function is constructed through dynamically adjusted optimization weight matrix, and combined with gradient descent algorithm and model predictive control rolling optimization, intelligent optimization of cutting parameters is achieved.

[0020] 4. This invention realizes closed-loop control of the entire process from signal acquisition, status diagnosis, impact prediction, parameter optimization to execution verification; the optimized parameters are sent to the machine tool for execution, and the adjusted sensor data is collected in real time for effect verification and status iterative update, promoting the development of metal cutting towards intelligence and autonomy. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the overall method structure of the present invention.

[0022] Figure 2 This is a schematic diagram of the physical feature extraction and health status diagnosis structure of S2 in this invention.

[0023] Figure 3 This is a schematic diagram illustrating the prediction of multi-objective impact results using the quantification model of S3 in this invention.

[0024] Figure 4 This is a schematic diagram of the multi-objective loss function construction and parameter optimization structure of S4 in this invention.

[0025] Figure 5 This is a graph showing the relationship between experimental data and the multi-objective loss function of this invention. Detailed Implementation

[0026] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] Please see Figure 1 - Figure 5 As shown, this embodiment of the invention provides an intelligent metal cutting control method based on multi-sensor data fusion, the method comprising:

[0028] S1: Synchronous acquisition and preprocessing of multi-source heterogeneous sensor data. Multi-source sensors are installed on the machine tool to synchronously acquire triaxial force signals, vibration signals and acoustic emission signals during the metal cutting process, and preprocess all signals by noise reduction, normalization and time-frequency domain transformation.

[0029] S2: Extract the metal physical feature set from the preprocessed signal. The metal physical feature set includes: vibration energy in a specific frequency band, cutting force ratio and fluctuation rate, and burst count rate of acoustic emission signal. Input the metal physical feature set into the knowledge base inference engine to perform physical failure mechanism matching and diagnosis, and output the metal processing health status vector.

[0030] S3: Based on the output metal processing health status vector and combined with the current process parameters, a process influence quantification model consisting of a semi-empirical analytical sub-model is used to predict its quantitative influence value on the final processing target, and the output is a multi-objective influence vector.

[0031] S4: Receive the multi-objective influence vector and construct a multi-objective loss function based on the preset and dynamically adjustable optimization weight matrix, and output the optimized metal cutting parameter settings;

[0032] S5: Send the optimized metal cutting parameter settings to the machine tool for control and continuously collect the controlled sensor data, return to S2 to verify the control effect and update the metal processing health status diagnosis, forming an adaptive closed loop.

[0033] In S1, multi-source heterogeneous sensor data is synchronously acquired and preprocessed. Multi-source sensors are installed on the machine tool to synchronously acquire triaxial force signals, vibration signals and acoustic emission signals during the metal cutting process, and preprocess all signals by noise reduction, normalization and time-frequency domain transformation.

[0034] This embodiment is used to achieve high-precision synchronous acquisition and standardized preprocessing of multi-source sensor signals, providing a high-quality, non-redundant data source for subsequent feature extraction and state diagnosis, specifically including:

[0035] S101: Sensor Hardware Deployment and Parameter Matching

[0036] S1011: Sensor selection and installation location determined. Dynamic cutting force sensor: A piezoelectric triaxial force sensor is selected and installed at the connection between the machine tool tool holder and the spindle to ensure that the sensor can directly collect the feed direction during metal cutting. , depth of cut Main cutting direction Dynamic force signal , , The sensor range must match the actual cutting force range, such as 0 to 50 kN, and the linearity error must be controlled within 1% of the actual cutting force range.

[0037] Vibration accelerometer: A triaxial piezoelectric accelerometer is selected and installed in a fixture near the machine tool spindle box or the workpiece machining surface to collect the triaxial acceleration signals of cutting vibration. , , The frequency response range covers 0.1Hz to 10kHz, meeting the detection requirements of high-frequency vibration during metal cutting.

[0038] Acoustic emission sensor: A wideband acoustic emission sensor with a frequency range of 20kHz to 1MHz is selected. It is mounted on the tool holder via a magnetic base to collect acoustic emission RMS signals generated by material plastic deformation and tool wear during the cutting process. The sensor needs to be grounded and shielded to eliminate electromagnetic interference.

[0039] S1012: Synchronous matching of multi-source sensor parameters and unified sampling frequency: Based on the characteristic frequencies of the cutting process, such as the fundamental frequency corresponding to the spindle speed and the chatter characteristic frequency band, the sampling frequency of all sensors is set to [value missing]. Ensure that the sampling frequency satisfies the Nyquist sampling theorem. ,in Represented as the highest characteristic frequency of the signal;

[0040] Synchronous triggering mechanism: The CNC pulse triggering method is adopted, and the machine tool spindle start signal is used as the synchronous triggering source. All sensors start to collect data simultaneously after receiving the same triggering signal, eliminating the time difference of multiple source signals, and the time synchronization accuracy is controlled within 1ms.

[0041] Data interface adaptation: Connect the output signals of all sensors to the multi-channel data acquisition card, convert the analog signals to digital signals through A / D conversion, and unify the output data format to 16-bit binary data. The data is cached to the local industrial control computer.

[0042] S102: Synchronous Acquisition of Multi-Source Heterogeneous Signals

[0043] When the machine tool receives the cutting command, it simultaneously triggers the dynamic cutting force sensor, vibration accelerometer, and acoustic emission sensor to start collecting data. The industrial control computer receives and stores the triaxial force signals in real time. , , Triaxial vibration acceleration signal , , Acoustic emission RMS signal The data acquisition time is consistent with the metal cutting processing time, and it supports segmented data acquisition for intermittent cutting processes.

[0044] Data integrity verification: Data integrity is verified in real time during the acquisition process. If signal loss or abnormal amplitude occurs, a re-acquisition command is immediately triggered. After acquisition is completed, a data acquisition log is generated, recording key parameters such as sampling time, sampling frequency, and signal amplitude range to ensure data traceability.

[0045] S103: Signal Preprocessing

[0046] The acquired raw multi-source signals are preprocessed in three steps: noise reduction, normalization, and time-frequency domain transformation, to eliminate noise interference, unify data dimensions, and extract time-frequency domain features.

[0047] S1031: Differentiated noise reduction algorithms are adopted for different noise types of signals. Cutting force signal: mainly contains low-frequency noise from machine tool transmission system. Wavelet thresholding is used for noise reduction. The db4 wavelet basis is selected and the decomposition level is 5. Soft thresholding is applied to the high-frequency wavelet coefficients to eliminate low-frequency noise interference.

[0048] Vibration acceleration signal: mainly contains random noise from environmental vibration. Adaptive Wiener filtering is used, and the filter coefficients are adjusted based on the local variance of the signal to retain the high-frequency characteristics related to flutter.

[0049] Acoustic emission RMS signal: mainly contains electromagnetic interference noise. Bandpass filtering is used, with the passband frequency set from 20kHz to 500kHz to filter out interference signals outside the passband.

[0050] The output after noise reduction is the denoised triaxial force signal. , , Vibration acceleration signal , , Acoustic emission RMS signal .

[0051] S1032: Signal Normalization

[0052] To eliminate the difference in signal dimensions between different sensors, min-max normalization is performed on all denoised signals, mapping the signal amplitude to the [0,1] interval. The calculation formula is as follows: , where X is the original amplitude of a certain signal after noise reduction; These represent the minimum and maximum values ​​of the signal during the acquisition period, respectively. The normalized signal amplitude; the normalized output is a standardized signal set with uniform dimensions.

[0053] S1033: Time-frequency domain transformation processing

[0054] To address the non-stationary characteristics of signal features changing over time during the cutting process, such as the emergence of chatter, a time-frequency domain transformation is performed on the normalized signal, preserving both time and frequency domain features:

[0055] Short-Time Fourier Transform (STFT) is used: Hanning window function is set with a window length of 256 sampling points and a window shift of 128 sampling points to convert the one-dimensional time-domain signal into a two-dimensional time-frequency matrix. The horizontal axis of the matrix is ​​time, the vertical axis is frequency, and the amplitude is the energy value of the corresponding time-frequency point.

[0056] Feature frequency band extraction: Based on the failure mechanism of metal cutting, key feature frequency bands are preset, such as chatter feature frequency band from 200Hz to 800Hz and tool wear feature frequency band from 50Hz to 200Hz. The signal components of the corresponding frequency bands are extracted from the time-frequency matrix to provide input for the metal physical feature set extraction in step S2.

[0057] In S2, a metal physical feature set is extracted from the preprocessed signal. The metal physical feature set includes: vibration energy in a specific frequency band, cutting force ratio and fluctuation rate, and burst count rate of acoustic emission signal. The metal physical feature set is input into the knowledge base inference engine to perform physical failure mechanism matching and diagnosis, and outputs a metal processing health status vector.

[0058] This embodiment takes the standardized multi-source time-frequency domain signal dataset output by S1. The core objective is to generate a quantified processing health state vector H through physically meaningful feature extraction and mechanism-driven rule reasoning, providing accurate input for the quantitative assessment of the process impact in S3.

[0059] Further explanation is needed regarding the normalized multi-source time-frequency domain signal dataset output by S1:

[0060] A: Time-frequency matrix of the triaxial force signal after normalization and time-frequency domain transformation , , ;

[0061] B: Time-frequency matrix of the triaxial vibration acceleration signal after normalization and time-frequency domain transformation , , ;

[0062] C: Time-frequency matrix of acoustic emission RMS signal after normalization and time-frequency domain transformation ;

[0063] S201: First-level primary feature extraction and physical failure mechanism matching

[0064] This layer is divided into two stages: precise feature extraction and mechanism rule reasoning, to achieve accurate mapping between signal features and physical failures. Based on the failure mechanism of metal cutting (chatter, tool wear, micro fracture), three types of core features are extracted from the S1 input data. All features need to be quantized.

[0065] S2021: Vibrational Energy in a Specific Frequency Band This reflects the degree of chatter in metal cutting; the characteristic frequency band is selected based on the mechanism of chatter in cutting, from the vibration acceleration time-frequency matrix. , , In the process, the characteristic frequency band of flutter is selected, and the vibration energy of a specific frequency band is analyzed. The calculation, specifically the formula, is as follows:

[0066]

[0067] in Indicates the direction of vibration, Indicates the characteristic frequency band range of flutter, Indicates the duration of a single cutting data acquisition, Indicates frequency resolution, Indicates time resolution;

[0068] The results output includes the calculation of the mean energy of the characteristic frequency bands of the triaxial vibration. As a quantitative indicator of the degree of flutter.

[0069] S2022: Cutting Force Ratio and Fluctuation This reflects tool wear and chipping during the cutting process. The cutting force ratio is calculated by extracting the time-domain mean from the three-dimensional force time-frequency matrix and calculating the force ratio in the key directions. The specific calculation formula is as follows: ,in Represents the time-domain mean of the force in the depth of cut direction. This represents the time-domain mean of the force in the main cutting direction; changes in the ratio reflect changes in the cutting force distribution caused by tool flank wear.

[0070] Cutting force fluctuation The stability of the force signal is calculated using the coefficient of variation, as shown in the following formula:

[0071]

[0072] in This represents the time-domain standard deviation of the main cutting force; a sudden increase in volatility reflects sudden failures such as tool chipping. The output ratio is... and volatility This is the core characteristic of tool wear.

[0073] S2022: Burst Count Rate of Acoustic Emission Signals This reflects microscopic fractures in metal cutting. A sudden fault threshold is set based on the baseline value of the acoustic emission signal from fault-free cutting. For example, 3 times the baseline value;

[0074] Burst frequency statistics, from acoustic emission time-frequency matrix In the statistical unit of time, the amplitude exceeds the threshold. Number of times The specific formula for calculating the count rate is as follows: The higher the count rate, the more severe the microscopic fracture of the material and the wear and spalling of the tool during the cutting process.

[0075] Output the burst count rate per unit time. Based on the time window in which the tool tip passes through the workpiece, and calibrated by the machine tool spindle position signal, the peak count rate within the marked window is... ;

[0076] S203: Knowledge-based physical failure mechanism reasoning

[0077] Construct an inference engine that includes a failure mechanism rule base and a feature threshold base to achieve accurate matching between features and failure modes, specifically including:

[0078] The knowledge base is initialized, the failure mechanism rule base is constructed, and discrimination rules based on cutting theory and engineering experience are entered. The rules are expressed using the production rule of IF feature condition combination THEN failure mode. Specific rules include:

[0079] Rule 1: Growth of more than 20% over three consecutive cutting cycles and If the synchronous growth rate is greater than 15%, the failure mode is determined to be intensified cutting regenerative chatter.

[0080] Rule Two: Trend change >10% (for 5 consecutive periods) and Fluctuation <5% and If the tool tip rises by more than 50% when passing the workpiece, the failure mode is determined to be tool flank wear.

[0081] Rule 3: Instantaneous increase >200% and A sudden increase of more than 50% and If the drop is greater than 30%, the failure mode is determined to be tool breakage or chipping.

[0082] Further explanation is needed regarding the knowledge base inference engine, specifically the feature input: the extracted quantified features ( , , , Input into the inference engine;

[0083] Threshold comparison: The feature values ​​are compared with the benchmark values ​​in the threshold library, and feature items that exceed the threshold are marked;

[0084] Rule matching: A forward reasoning strategy is adopted to traverse the rule base and match rules that satisfy the combination of feature conditions;

[0085] Conflict resolution: If multiple rules are triggered simultaneously (such as the coexistence of flutter and wear characteristics), the final judgment result is output based on the rule priority, such as chipping > flutter > normal wear.

[0086] Quantification assignment: Assign a quantification level value to the determined failure mode.

[0087] S204: Second-layer feature fusion and health status vector generation

[0088] The dispersed failure mode quantification results are integrated into a processing health state vector with clear physical meaning. Specifically, it includes:

[0089] Based on the core failure modes of metal cutting, the state vector dimension is defined. ,in The flutter intensity level is divided into 1 to 5. The tool wear stage is indicated by 0 = no wear, 1 = initial stage, 2 = intermediate stage, and 3 = late stage. This indicates that the tool is damaged; This indicates the cutting stability score;

[0090] Feature fusion assignment: The failure modes obtained from the first-level inference are quantized into their respective levels and filled into the corresponding vector dimensions to generate a structured state vector. ;

[0091] Vector validity verification: Compare the state vectors of historical cutting data. If the deviation between the current vector and the normal machining vector exceeds a preset threshold, it is marked as an abnormal state; otherwise, it is judged as a normal state.

[0092] Output: Outputs the final processed health state vector. It also generates a status diagnostic report, including failure mode, quantification level, and anomaly cause analysis.

[0093] Final output: After step S2 is completed, the following data will be output as input for the quantitative assessment of the process impact in step S3:

[0094] Structured processing health state vector ; Full process log of feature extraction and mechanism reasoning (including feature values, rule matching records, and threshold parameters); Processing status diagnostic report.

[0095] In S3, based on the output metal processing health status vector and combined with the current process parameters, a process influence quantification model consisting of a semi-empirical analytical sub-model is used to predict the quantified influence value on the final processing target, and the output is a multi-objective influence vector.

[0096] In this embodiment, taking the machining health status vector H output by S2 and the current cutting process parameters, the core objective is to establish a quantitative correlation between machining health status, process parameters, and machining objectives through a combination of semi-empirical analytical sub-models, and output a multi-objective influence vector. This provides quantifiable decision-making basis for S4 multi-objective collaborative optimization, specifically including:

[0097] S301: Input processing health state vector H

[0098] Receive structured processing health state vector Full log of feature extraction and mechanism reasoning (including feature values, rule matching records, and threshold parameters); processing status diagnostic report;

[0099] Real-time process parameters, cutting speed Feed rate Depth of cut Auxiliary parameters include: tool type, workpiece material properties (hardness, tensile strength), and cumulative machining time. ;

[0100] S302: Construction of Semi-Empirical Analytical Sub-model

[0101] To address the three core machining objectives of metal surface roughness, material removal rate, and remaining tool life, a semi-empirical analytical sub-model is constructed. This semi-empirical analytical sub-model includes: a surface roughness prediction sub-model. Material removal rate calculation sub-model Tool Remaining Life Prediction Sub-model ;

[0102] The surface roughness prediction sub-model needs further explanation. , This is a semi-empirical model, and the specific calculation formula is as follows: ;

[0103] Material removal rate calculation sub-model , This is an analytical model, and the specific calculation formula is as follows: ;

[0104] Tool Remaining Life Prediction Sub-model , This is a semi-empirical model, and the specific calculation formula is as follows: ;

[0105] S303: Sub-model prediction calculation

[0106] Based on the calibrated semi-empirical analytical sub-model, the health state vector H of S2 and the real-time process parameters are substituted into the sub-model to complete the quantization calculation. The steps are as follows:

[0107] Extracting key parameters from the state vector: Extracting the tool wear stage from H Cutting stability score This serves as the basis for correcting the sub-model;

[0108] Surface roughness prediction: , , Substitution The sub-model calculates the predicted surface roughness value. ;

[0109] Material removal rate calculation: , , , Substitution The sub-model calculates the current effective material removal rate. ;

[0110] Tool remaining life prediction: , , , Substitution The sub-model calculates the predicted remaining tool life. ;

[0111] Outlier filtering: If the calculation result exceeds the allowable range of the process, it is marked as exceeding the process target and the reason for exceeding the target is recorded, such as severe tool wear or severe chatter.

[0112] S304: Multi-objective influence vector Generation and Verification

[0113] Based on the core machining objective, a multi-objective influence vector is constructed. ,in The process risk coefficient is represented by... and calculate, =2 or When ≥4, =1 indicates high risk; otherwise =0;

[0114] Vector validity verification: Compare the predicted values ​​in vector E with historical data under the same working conditions. If the deviation exceeds a preset threshold, such as... If the deviation is greater than 15%, return to the first layer and recalibrate the sub-model coefficients; if the deviation meets the requirements, confirm that vector E is valid.

[0115] Quantitative assessment report generation: For each indicator in E, analyze the degree of impact of machining health status on process objectives, such as tool wear in the middle stage. Increased by 80%, severe flutter caused Reduced by 30%;

[0116] In S4, a multi-objective influence vector is received, and a multi-objective loss function is constructed based on a preset and dynamically adjustable optimization weight matrix, outputting optimized metal cutting parameter settings.

[0117] In this embodiment, taking into account the multi-objective influence vector E and process target requirements output by S3, the core objective is to output the globally optimal cutting parameters through dynamic weight configuration, multi-objective loss function construction, and model prediction rolling optimization. This provides a basis for decision-making regarding the metal cutting control of the S5 machine tool;

[0118] S401: Receive multi-target influence vector

[0119] Input multi-objective influence vector Quantitative assessment report on process impact, and calculation log of sub-model;

[0120] Surface roughness target value Machine tool allowable cutting speed range Feed rate range ;

[0121] S402: Dynamic Weight Matrix Configuration

[0122] finishing ,in for Weight, for Weight, Tool life weighting; roughing The conditions for triggering the dynamic adjustment mechanism are as follows: When =1, the weight for increasing tool life is increased. Reduce the weight of other objectives to 0.3;

[0123] S403: Constructing a multi-objective loss function

[0124] Based on S402 dynamic weight matrix and multi-objective influence vector Construct a function with the objective of minimizing the loss, and the specific calculation formula is as follows:

[0125]

[0126] in =1, add risk penalty item ,in Risk penalty weights are used; the feasible solution of metal cutting parameters is spatially delineated, and the feasible region of the optimization parameters is delineated based on constraints of machine tool, cutting tool, and workpiece. , ;

[0127] The table below shows the experimental data for the multi-objective loss function, which is used for online rolling optimization of model prediction.

[0128]

[0129] The model predicts online rolling optimization solutions, selecting the gradient descent optimization algorithm and the loss function. As the optimization objective, model predictive control is introduced. Based on the semi-empirical sub-model of S3, the loss function value is extrapolated forward for the next 3 to 5 cutting cycles to avoid instantaneous optima. Iterative calculations are performed within the feasible solution space until the loss function value converges to a preset threshold, at which point the optimal cutting parameters are output. ;

[0130] Optimization result verification If the parameter is outside the feasible region, the weight matrix is ​​readjusted and the optimization is performed again. Once the verification is successful, a parameter optimization decision report is generated.

[0131] In S5, the optimized metal cutting parameter settings are sent to the machine tool for control, and the controlled sensor data is continuously collected and returned to S2 to verify the control effect and update the metal processing health status diagnosis, forming an adaptive closed loop.

[0132] In this embodiment, the optimized cutting parameters output by S4 are adopted. The core objective is to complete parameter adjustment, data collection and effect verification, and form an adaptive closed loop of collection, diagnosis, evaluation, optimization, adjustment and re-diagnosis to ensure the continuous stability of the processing process;

[0133] Input optimized cutting parameters Multi-objective optimization decision report; the machine tool CNC system is in a state that can receive commands, and the multi-source sensors deployed in S1 are kept powered on and their parameters are consistent with those of S1, including sampling frequency, synchronous triggering mechanism, etc.

[0134] Validation threshold preset: Processing health state vector deviation threshold Such as before and after regulation Vector differences controlled within 20% are considered valid.

[0135] S501: Issued to machine tool for execution control

[0136] Optimize cutting parameters Convert the commands to a format recognizable by the machine tool's CNC system; send the converted commands to the machine tool's CNC system, wait for the system to report the parameter's effectiveness status, and record the parameter switching time; after the machine tool runs for 1 to 2 metal cutting cycles according to the new parameters to ensure stable machining conditions, proceed to the next process data acquisition;

[0137] After adjustment, multi-source sensor data is synchronously acquired and preprocessed. The machine tool's stable operation signal is used as the trigger source to start synchronous acquisition of multi-source sensors. The acquired content is consistent with S1, including triaxial force signals. , , Triaxial vibration acceleration signal , , Acoustic emission RMS signal The data collection time covers one complete cutting cycle;

[0138] Standardized preprocessing follows the preprocessing flow of S1 (noise reduction, normalization, time-frequency domain transformation) to process the acquired data and output a standardized multi-source time-frequency domain signal dataset consistent with the S1 format, ensuring that it can be directly input into S2.

[0139] S502: Closed-loop verification and state iterative update

[0140] Data feedback diagnosis involves directly inputting the preprocessed dataset into the feature extraction and mechanism inference process of S2 to regenerate the regulated processing health state vector. ;

[0141] To verify the effectiveness of the regulation, the health status vectors before and after regulation were compared. To regulate the front vector, Calculate the vector deviation value after adjustment. ;

[0142] like If the regulation is deemed effective, maintain the current parameters and continue processing, continuously collecting data periodically for verification;

[0143] like : If the control is deemed ineffective, the re-optimization process from S3 to S4 is triggered, based on Quantitative assessment of the impact of process updates and decision-making on parameter optimization;

[0144] Closed-loop iteration marker: Record the verification result (valid / invalid), update the health status diagnosis log, and form an adaptive closed-loop iteration record.

[0145] The final output includes the regulation execution log (including parameter issuance time, effective status, and verification results) and the post-regulation health status vector. A closed-loop iteration trigger flag is used. If the flag is valid, the parameters are maintained; if it is invalid, S3 to S4 are triggered to re-optimize the metal cutting.

[0146] Secondly: The accompanying drawings of the embodiments disclosed in this invention only involve the structures involved in the embodiments disclosed in this invention. Other structures can refer to the general design. In the absence of conflict, the same embodiment and different embodiments of this invention can be combined with each other.

[0147] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A smart metal cutting control method based on multi-sensor data fusion, characterized in that, include: S1: Synchronous acquisition and preprocessing of multi-source heterogeneous sensor data. Multi-source sensors are installed on the machine tool to synchronously acquire triaxial force signals, vibration signals and acoustic emission signals during the metal cutting process, and preprocess all signals by noise reduction, normalization and time-frequency domain transformation. S2: Extract the metal physical feature set from the preprocessed signal. The metal physical feature set includes: vibration energy in a specific frequency band, cutting force ratio and fluctuation rate, and burst count rate of acoustic emission signal. Input the metal physical feature set into the knowledge base inference engine to perform physical failure mechanism matching and diagnosis, and output the metal processing health status vector. S3: Based on the output metal processing health status vector and combined with the current process parameters, a process influence quantification model consisting of a semi-empirical analytical sub-model is used to predict its quantitative influence value on the final processing target, and the output is a multi-objective influence vector. S4: Receive the multi-objective influence vector and construct a multi-objective loss function based on the preset and dynamically adjustable optimization weight matrix, and output the optimized metal cutting parameter settings; S5: Send the optimized metal cutting parameter settings to the machine tool for control and continuously collect the controlled sensor data, return to S2 to verify the control effect and update the metal processing health status diagnosis, forming an adaptive closed loop.

2. The intelligent metal cutting control method based on multi-sensor data fusion according to claim 1, characterized in that, The multi-source sensor in S1 includes: A dynamic cutting force sensor is installed at the connection between the tool holder and the spindle to collect three-dimensional dynamic force signals; a vibration acceleration sensor is installed in the spindle housing or workpiece fixture to collect three-dimensional vibration acceleration signals; and an acoustic emission sensor is installed in the tool holder to collect acoustic emission signals.

3. The intelligent metal cutting control method based on multi-sensor data fusion according to claim 2, characterized in that, The signal preprocessing in S1 includes: Noise reduction is performed using wavelet thresholding, adaptive Wiener filtering, or bandpass filtering for different signal types; normalization is performed by mapping all signal amplitudes to the [0,1] interval; and time-frequency domain transformation is performed by using short-time Fourier transform to convert the time-domain signal into a time-frequency matrix.

4. The intelligent metal cutting control method based on multi-sensor data fusion according to claim 1, characterized in that, The extraction of the metal physical feature set in S2 includes: Vibration energy in a specific frequency band: Select the flutter characteristic frequency band from the vibration acceleration time-frequency matrix and calculate the average energy of the characteristic frequency band of the triaxial vibration; Cutting force ratio and volatility: The time-domain mean is extracted from the three-dimensional force time-frequency matrix, the ratio of the force in the depth of cut direction to the force in the main cutting direction is calculated, and the volatility of the main cutting force is calculated using the coefficient of variation; Acoustic emission signal burst count rate: Based on the baseline value of the acoustic emission signal of fault-free cutting, a burst judgment threshold is set, and the number of times the amplitude exceeds the threshold is counted per unit time. Combined with the time window calibrated by the machine tool spindle position signal, the peak count rate within the window is marked. The knowledge base reasoning engine includes a failure mechanism rule base and a feature threshold base. It adopts a forward reasoning strategy and a rule priority conflict resolution mechanism to match and diagnose the extracted quantitative features. The metal processing health status vector includes four dimensions: chatter intensity level, tool wear stage, tool breakage state, and cutting stability score.

5. The intelligent metal cutting control method based on multi-sensor data fusion according to claim 4, characterized in that, The failure mechanism rule base includes: The failure mechanism rule base adopts the production expression of the IF feature condition combination THEN failure mode, and the rule priority is chipping, chatter, and normal wear in metal cutting process.

6. The intelligent metal cutting control method based on multi-sensor data fusion according to claim 1, characterized in that, The semi-empirical analytical sub-model in S3 includes: Sub-model for surface roughness prediction, sub-model for material removal rate calculation, and sub-model for tool remaining life prediction; Current process parameters include cutting speed, feed rate, depth of cut, tool type, workpiece material properties, and cumulative machining time; The multi-objective influence vector includes the predicted surface roughness, the calculated material removal rate, the predicted tool life, and the process risk coefficient.

7. The intelligent metal cutting control method based on multi-sensor data fusion according to claim 1, characterized in that, The dynamically adjusted and optimized weight matrix in S4 includes: When the process risk coefficient is 1, the tool life weight is increased to 0.3, and the weights of other objectives are reduced. The multi-objective loss function adds a risk penalty term when the process risk coefficient is 1. During the optimization process, the feasible region of the cutting parameters is defined. The gradient descent optimization algorithm is combined with the idea of ​​model predictive control to calculate the loss function value for the next 3 to 5 cutting cycles until it converges to the preset threshold.

8. The intelligent metal cutting control method based on multi-sensor data fusion according to claim 1, characterized in that, S5 specifically includes: After the regulated sensor data is preprocessed by S1, it is input into S2 to regenerate the processing health status vector and calculate the deviation value of the vector before and after regulation. If the deviation value is ≤20%, the control is deemed effective, the current parameters are maintained, and periodic data collection and verification are continuously performed; if the deviation value is >20%, the control is deemed ineffective, and the optimization process from S3 to S4 is triggered.

Images