Cutting tool residual life online detection method based on multi-source sensing data fusion

Abstract

The invention provides a cutting tool residual life online detection method based on multi-source sensing data fusion, and relates to the technical field of intelligent manufacturing, and the method comprises the steps: collecting a sensor signal in a cutting machining process, carrying out the intrinsic mode decomposition based on Hilbert-Huang transformation, and obtaining the natural vibration mode of the sensor signal; analyzing the instantaneous frequency spectrum, extracting a drift component, eliminating the influence of the drift component by adopting a self-compensation mechanism, and obtaining a temperature-compensated sensor signal; second-order cyclic spectrum analysis is carried out, coupling frequency components among the sensors are identified, a decoupling mapping matrix is constructed, and independent wear characteristics of the sensors are separated; and constructing a time sequence phase spectrum matrix, carrying out singular value decomposition on the time sequence phase spectrum matrix to obtain a feature importance degree sequence, carrying out probability density modeling on the sorted features by adopting a kernel density estimation method, and calculating the residual life of the cutting tool. The problem of non-linear coupling among signals of multiple sensors is solved, and accurate prediction of the residual life of the tool under the complex cutting working condition is achieved.

Description

Technical Field

[0001] This invention relates to the field of intelligent manufacturing technology, and more specifically, to an online detection method for the remaining life of cutting tools based on multi-source sensor data fusion. Background Technology

[0002] As machining has become one of the most important processing technologies in modern manufacturing, its accuracy and efficiency directly affect product quality and production costs. In recent years, with the development of intelligent manufacturing and Industry 4.0 technologies, intelligent monitoring and diagnostic technologies for machining processes have received widespread attention. Especially in high-end manufacturing fields such as aerospace, automotive, and precision instrument manufacturing, tool condition monitoring and remaining life prediction have become key aspects of improving machining quality and reducing costs. Traditional tool condition monitoring methods mostly rely on experience-based periodic replacement strategies or single-sensor signal evaluation methods. These methods often fail to accurately reflect the actual wear state of tools under complex machining conditions, leading to premature tool replacement or overuse. In recent years, with the development of multi-sensor fusion technology, using multiple sensor signals such as acceleration, acoustic emission, and cutting force for tool condition monitoring has become an effective research direction. However, sensor signals are often affected by complex working conditions, environmental temperature changes, and nonlinear coupling between multiple physical quantities. How to extract independent features related to tool wear from multi-source data and accurately predict the remaining tool life remains a key research focus and challenge.

[0003] Existing tool condition monitoring and remaining life prediction technologies mainly rely on signal processing and feature extraction methods, such as Fast Fourier Transform (FFT), Wavelet Transform (WT), and Empirical Mode Decomposition (EMD). However, while FFT is suitable for frequency domain analysis of steady-state signals, its ability to handle non-stationary and nonlinear signals during machining is limited. Wavelet Transform performs well in time-frequency analysis, but its decomposition results are highly dependent on preset wavelet basis functions, lacking adaptability. Empirical Mode Decomposition is an adaptive method for non-stationary signal analysis, but its intrinsic mode component (IMF) decomposition results are susceptible to endpoint effects and mode aliasing. Furthermore, existing technologies often fail to adequately consider temperature drift, nonlinear coupling between signals, and correlations between multiple physical quantity features in the fusion and analysis of multi-sensor signals, resulting in inaccurate extracted tool wear characteristics and thus limiting the accuracy of remaining life prediction. Especially under high-temperature, high-speed, and complex cutting conditions, temperature drift and frequency coupling problems are significant, and existing static data compensation methods and single-frequency domain decoupling methods are insufficient to meet practical application requirements. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention is proposed. This invention provides an online detection method for the remaining life of cutting tools based on multi-source sensor data fusion. This method can, to some extent, solve the problems caused by drift components due to temperature changes in multi-sensor signals, which fail to accurately reflect the actual tool condition, resulting in low feature extraction accuracy. It also addresses the difficulty in separating independent wear features due to nonlinear coupling between multi-sensor signals, thus affecting the accuracy of remaining life prediction.

[0005] According to one aspect of the present invention, an online detection method for the remaining life of cutting tools based on multi-source sensor data fusion is provided, comprising: Sensor signals are acquired during the cutting process, and intrinsic mode decomposition is performed on the sensor signals based on Hilbert-Huang transform to obtain the inherent vibration modes of the sensor signals. The instantaneous frequency spectrum of the inherent vibration mode is analyzed, the drift component related to temperature change is extracted, and a self-compensation mechanism is used to eliminate the influence of the drift component on the sensor signals to obtain the temperature-compensated sensor signal. Second-order cyclic spectrum analysis is performed on the temperature-compensated sensor signal to identify the coupling frequency components between sensors, a decoupling mapping matrix is ​​constructed, and the independent wear characteristics of each sensor are separated through the decoupling mapping matrix. Based on the independent wear characteristics, a time-series phase spectrum matrix is ​​constructed. Singular value decomposition is performed on the time-series phase spectrum matrix to obtain the feature importance ranking. The kernel density estimation method is used to model the probability density of the ranked features and calculate the remaining life of the cutting tool.

[0006] Furthermore, the drift components include acceleration signal drift, acoustic emission signal drift, and cutting force signal drift.

[0007] Furthermore, the acceleration signal drift is expressed by the formula: , in, The actual measured acceleration signal. For reference temperature, For an ideal acceleration signal, For the random error term of the acceleration signal, and These are the first-order and second-order temperature coefficients, respectively; The acoustic emission signal drift is expressed by the formula: , in, The original acoustic emission signal frequency, The temperature-velocity of sound coupling coefficient. For frequency disturbance terms; The drift of the cutting force signal is expressed by the following formula: , in, This is the actual measured cutting force signal. The static temperature coefficient, For dynamic temperature coefficient, For the actual cutting force, For measuring noise.

[0008] Furthermore, a unified temperature drift compensation framework is constructed based on the drift components, and the temperature drift factor is defined as: , in, For the first The generalized temperature drift factor of this type of sensor This refers to the sensor signal under the current operating conditions. The sensor signal is under reference conditions.

[0009] Furthermore, a nonlinear compensation algorithm based on the Volterra series is constructed based on the temperature drift factor, expressed by the following formula: , in, For the compensated signal, The original signal contains temperature drift. As a first-order compensation factor, for The Volterra kernel function describes the nonlinear characteristics of the system. For the time delay variable group, For the first Temperature change at each time delay point It is the highest order of the Volterra series.

[0010] Furthermore, the kernel function The formula for updating adaptively is expressed as follows: , in, for First-order compensation gain coefficient, for Damping coefficient.

[0011] Furthermore, the temperature-compensated sensor signal is output after adaptive filtering, as expressed by the formula: , in, For the final output signal, This is a frequency domain correction function. It is a time-domain smoothing function. The angular frequency of the signal. The temperature-dependent cutoff frequency, The time constant is related to the magnitude of the temperature gradient. This represents the absolute value of the temperature gradient.

[0012] Furthermore, the second-order cyclic spectrum analysis transforms the signal from the time domain to a dual domain of cyclic frequency and spectral frequency. For a single sensor signal, its cyclic spectrum is defined as: , in, This is the sensor signal after temperature compensation; For time delay variables; For time variables, The length of the observation time window; The cycle frequency, Spectral frequency;

[0013] Furthermore, a wavelet-enhanced cyclic spectrum estimation method is introduced, using an adaptive weighting function. Enhance the effective features.

[0014] Furthermore, the decoupling mapping matrix is ​​expressed by the following formula: , in, Let Kronecker function be used when = The value is 1 if it is true, and 0 otherwise. For regularization parameters, It is a nonlinear coherence function.

[0015] Furthermore, the singular value decomposition is expressed by the following formula: , in, It is a left singular matrix. This is the transpose of the right singular matrix. For regularization strength, For time-phase balance coefficient, The first element in the time-series phase spectrum matrix OK The element values ​​of the column, The first element in the time-series phase spectrum matrix +1 line The element values ​​of the column, The first element in the time-series phase spectrum matrix OK +1 column element values.

[0016] Compared with existing technologies, this invention effectively solves the end-point effect and mode aliasing problems in sensor signal processing by using an adaptive decomposition method based on Hilbert-Huang transform, thereby improving the accuracy and stability of signal decomposition. It also introduces a temperature compensation mechanism to eliminate the interference of temperature drift on sensor signals, ensuring the accuracy of feature extraction. Furthermore, it employs second-order cyclic spectrum analysis and decoupling mapping matrix to solve the nonlinear coupling problem between multi-sensor signals, extracting independent tool wear features. Finally, by constructing a time-series phase spectrum matrix and combining singular value decomposition and kernel density estimation methods, a dynamic feature importance ranking and life prediction model is established, enabling accurate online prediction of tool remaining life under complex cutting conditions, significantly improving machining efficiency and quality control. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In the drawings: Figure 1 This is a flowchart of an online detection method for the remaining life of cutting tools based on multi-source sensor data fusion according to an embodiment of the present invention; Figure 2 This is a comparison chart of the online detection method for remaining tool life based on multi-source sensor data fusion according to an embodiment of the present invention and the prediction method of traditional methods. Detailed Implementation

[0018] Hereinafter, exemplary embodiments according to the present invention will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of the present invention, and not all embodiments of the present invention. It should be understood that the present invention is not limited to the exemplary embodiments described herein.

[0019] As mentioned in the background section, the existing technology has two main problems: first, it fails to effectively eliminate the drift effect of temperature changes on sensor signals, resulting in poor accuracy of feature extraction; second, it fails to solve the nonlinear coupling problem between multiple sensor signals, resulting in insufficient feature independence; and third, the construction of feature importance ranking and lifetime prediction models fails to fully incorporate the dynamic evolution law of the signals, and the prediction accuracy needs to be improved.

[0020] Figure 1 This is a system block diagram of an online detection method for the remaining life of cutting tools based on multi-source sensor data fusion according to an embodiment of the present invention. Figure 1As shown, the online detection method for remaining tool life based on multi-source sensor data fusion includes: S1: Acquire sensor signals during the cutting process, and perform intrinsic mode decomposition on the sensor signals based on Hilbert-Huang transform to obtain the inherent vibration modes of the sensor signals.

[0021] A sensor array system, including a triaxial accelerometer, an acoustic emission sensor, and a cutting force sensor, is deployed at key locations on the machine tool. The triaxial accelerometer uses a PCB 356A32 model and is installed on the tool holder 15mm from the tool tip. It is fastened with an M6 thread to ensure a rigid connection between the sensor and the tool holder. The sensitivity is 100mV / g, the measurement range is ±50g, and the sampling frequency is set to 20kHz. The acoustic emission sensor is a PICO model sensor, which is installed on the workpiece surface 50mm away from the cutting area by a special magnetic base. Vacuum oil is used as a coupling agent to improve the acoustic wave transmission efficiency. The sensor's operating frequency range is 100kHz-1MHz, the sampling frequency is set to 2MHz, and the preamplifier gain is set to 40dB. The cutting force sensor uses a Kistler 9257B triaxial force gauge, which is installed at the bottom of the workpiece fixture and fixed with a bolt with a preload of 15kN. The range is 0-5000N, the sampling frequency is set to 10kHz, and the signal is collected after being conditioned by a 5070A charge amplifier.

[0022] After acquiring the raw sensor signal, a modified Hilbert-Huang transform is used for signal processing. The modified Hilbert-Huang transform includes: After acquiring the original sensor signal, a data preprocessing method based on variational mode decomposition is adopted. By setting the decomposition scale K, the signal is decomposed into sub-modes of different frequency bands. When the signal exhibits end-point effects, the mirror extension method is used for processing, and the extension length is taken as 15% of the original signal length.

[0023] Subsequently, integrated empirical mode decomposition is performed on each sub-mode signal. During the decomposition process, an adaptive noise-assisted technique is introduced, which dynamically adjusts the amplitude of the added white noise according to the local signal-to-noise ratio of the signal. When the local signal-to-noise ratio is lower than 10dB, the white noise amplitude is set to 0.2 times the signal standard deviation. When the signal-to-noise ratio is higher than 20dB, the white noise amplitude is reduced to 0.1 times the signal standard deviation.

[0024] Perform standard EMD decomposition on the noisy signal, construct the signal envelope using cubic spline interpolation, and introduce an adaptive step size control strategy. Specifically, when starting the signal decomposition, first set the initial step size coefficient α. At this time, calculate the upper and lower envelope lines through cubic spline interpolation and obtain the first screening result h1. Then, calculate the standard deviation ratio SD of the adjacent two screening results. If SD is greater than 0.3, it indicates that there are significant differences between the signal components. At this time, reduce the step size coefficient α to 0.6 times the original value and perform fine screening with a smaller step size. If 0.1 < SD ≤ 0.3, keep the current step size coefficient unchanged. If SD ≤ 0.1, it means that the decomposition tends to be stable, then increase the step size coefficient α to 1.2 times the original value to accelerate convergence. If it is found during the iteration process that the change in the number of local extreme points compared to the previous iteration exceeds 20%, trigger the step size emergency adjustment mechanism and immediately reduce the step size coefficient α to 0.5 times the current value to prevent signal distortion caused by over-screening. When the standard deviation ratio SD of five consecutive iterations is less than the preset threshold, it is determined that the current IMF component extraction is completed.

[0025] During the entire decomposition process, the step size coefficient α is restricted within the range of [0.2, 0.9] to ensure the stability of the algorithm. When dealing with high-frequency components, in order to improve the ability to characterize local features, the upper limit of the step size coefficient is reduced to 0.7. When dealing with low-frequency components, in order to reduce the computational amount, the lower limit of the step size coefficient is increased to 0.3. If the mode mixing phenomenon is found, start the local EMD algorithm based on windowing processing, add white noise with an amplitude of 0.2 times the standard deviation of the signal, set the integration number to 100 times, and set the residual threshold to 0.1% of the standard deviation of the original signal.

[0026] To ensure the statistical stability of the results, the above EEMD decomposition process is repeated 100 times, and the IMFs obtained from each decomposition are integrated and averaged to eliminate the influence of random noise. During the entire decomposition process, the residual threshold is set to 0.1% of the standard deviation of the original signal, and the iteration stops when the difference between the IMF components obtained from two adjacent screenings is less than this threshold.

[0027] For acceleration signals, EEMD decomposition usually obtains 8 - 10 IMF components. Mainly focus on the IMF components in the frequency band of 1 - 5 kHz. This frequency band is divided into three sub-bands: IMF3 in the frequency band of 1 - 2 kHz reflects the fundamental vibration of the machine tool spindle - tool system. IMF4 in the frequency band of 2 - 3 kHz corresponds to the torsional vibration mode of the tool - workpiece system. IMF5 in the frequency band of 3 - 4 kHz represents the second-order bending vibration mode of the tool, which has a significant correlation with the tool wear state. IMF6 in the frequency band of 4 - 5 kHz reflects the coupled vibration mode of the tool - chip system. The acoustic emission signal can be decomposed by EEMD to obtain 12-15 IMF components. The focus is on extracting the IMF components in the 200kHz-800kHz frequency band. Among them, IMF2 in the 200-400kHz frequency band reflects the plastic deformation process of the material, IMF4 in the 500-700kHz frequency band characterizes the dynamic characteristics of the crack propagation process, and IMF5 in the 600-800kHz frequency band reflects the transient acoustic emission response during the chip separation process. For cutting force signals, EEMD decomposition typically yields 6-8 IMF components. Therefore, the focus is on analyzing the IMF components in the 0-500Hz frequency band. The IMF4 in the 0-100Hz frequency band reflects the quasi-static force characteristics of the cutting process and is directly related to cutting parameters and workpiece material properties. The IMF5 in the 100-300Hz frequency band characterizes the dynamic response characteristics of the cutting system and includes coupled vibration information of the tool-workpiece-machine tool system. The IMF6 in the 300-500Hz frequency band corresponds to the dynamic characteristics of the periodic changes in chips and is closely related to the chip formation mechanism and tool wear state.

[0028] By combining EEMD decomposition and Hilbert transform, a set of inherent vibration modes that can characterize the dynamic properties of the cutting process are obtained.

[0029] S2: Analyze the instantaneous frequency spectrum of the inherent vibration mode, extract the drift component related to temperature change, and use a self-compensation mechanism to eliminate the influence of the drift component on the sensor signals to obtain temperature-compensated sensor signals.

[0030] After obtaining the inherent vibration modes, the influence of temperature changes on various sensor signals is further analyzed. By performing precise time-frequency analysis on each IMF component, temperature-induced drift characteristics are extracted, and a temperature influence model is established based on the physical mechanism. An adaptive compensation algorithm is then used to achieve signal correction.

[0031] Specifically, instantaneous amplitude and instantaneous phase information are obtained by calculating the analytical signals of each IMF. For the first... The analytical signal representation of each IMF component is as follows: , in, For the first The real part of each IMF component No. Hilbert transform results of each IMF component This is the instantaneous amplitude.

[0032] The instantaneous frequency is then calculated as follows: , in, It is the instantaneous phase.

[0033] It is important to note that temperature changes typically manifest as a slow drift in signal characteristics, and this drift exhibits different characteristics on different types of sensors.

[0034] For accelerometers, the temperature effect is mainly manifested as the temperature dependence of the piezoelectric constant, expressed by the formula: , in, For temperature The piezoelectric constant of the following, For reference temperature, and These are the first-order and second-order temperature coefficients, respectively.

[0035] This leads to the following drift in the acceleration signal, expressed by the formula: , in, The actual measured acceleration signal. For an ideal acceleration signal, This represents the random error term of the acceleration signal.

[0036] For acoustic emission signals, temperature primarily affects the propagation speed of sound waves within materials, as expressed by the formula: , in, For temperature The speed of sound wave propagation at this time The speed of sound at the reference temperature, The temperature-velocity of sound coupling coefficient is given.

[0037] Consequently, the frequency of the acoustic emission signal drifts, as expressed by the formula: , in, The original acoustic emission signal frequency, This is the frequency disturbance term.

[0038] For cutting force signals, the temperature effect mainly manifests as a change in the sensitivity of the force gauge, expressed by the formula: , in, To calibrate sensitivity, The static temperature coefficient, This is the dynamic temperature coefficient.

[0039] This leads to the following drift in the cutting force signal, expressed by the formula: , in, This is the actual measured cutting force signal. For the actual cutting force, For measuring noise.

[0040] Based on the above model, a unified temperature drift compensation framework is constructed. First, the temperature drift factor is defined as: , in, For the first The generalized temperature drift factor of this type of sensor This refers to the sensor signal under the current operating conditions. The sensor signal is under reference conditions.

[0041] Furthermore, a nonlinear compensation algorithm based on Volterra series is constructed, expressed by the following formula: , in, For the compensated signal, The original signal contains temperature drift. It is a first-order (linear) compensation factor. for The Volterra kernel function describes the nonlinear characteristics of the system. For the time delay variable group, For the first Temperature change at each time delay point It is the highest order of the Volterra series.

[0042] Compensation kernel function The formula for updating adaptively is expressed as follows: , in, for First-order compensation gain coefficient, for Damping coefficient.

[0043] Finally, the temperature-compensated signal is output after adaptive filtering, as expressed by the formula: , in, For the final output signal, This is a frequency domain correction function. It is a time-domain smoothing function. The angular frequency of the signal. The temperature-dependent cutoff frequency, The time constant is related to the magnitude of the temperature gradient. This represents the absolute value of the temperature gradient.

[0044] S3: Perform second-order cyclic spectrum analysis on the temperature-compensated sensor signal to identify the coupling frequency components between sensors, construct a decoupling mapping matrix, and separate the independent wear characteristics of each sensor through the decoupling mapping matrix.

[0045] Based on the acquired temperature-compensated sensor signal, a generalized second-order cyclic spectrum analysis framework is constructed to transform the signal from the time domain to a dual domain of cyclic frequency and spectral frequency. For a single sensor signal, its cyclic spectrum is defined as: ,

[0046] in, This is the sensor signal after temperature compensation; For time delay variables; For time variables, The length of the observation time window; The cycle frequency, is the spectral frequency.

[0047] Furthermore, considering the interaction of signals in a multi-sensor system, a generalized intercyclic spectrum is introduced to extend cyclic spectrum analysis to multi-sensor systems, expressed by the following formula: , in, and They represent the first The and the first Signals from each sensor, This represents the total number of sensors.

[0048] The intercyclic spectrum reflects the frequency coupling relationship between signals from different sensors through double integration. Furthermore, to improve analysis accuracy, wavelet-enhanced cyclic spectrum estimation is introduced, expressed by the following formula: , in, This is an adaptive weighting function used to enhance effective features and suppress noise.

[0049] Adaptive weight function The formula is expressed as: ,

[0050] in, and These are the center cyclic frequency and the spectral frequency, respectively, representing the frequency range of interest. and The corresponding bandwidth parameters control the precision of frequency selection.

[0051] Based on wavelet-enhanced cyclic spectrum, a sensor coupling metric matrix is ​​constructed: , Among them, matrix elements Indicates the first The and the first The coupling strength between the sensors.

[0052] Meanwhile, in order to capture the nonlinear coupling characteristics, a nonlinear coherence function is introduced, expressed by the formula: , in, For phase function, These are nonlinear modulation coefficients.

[0053] Based on the coupling analysis results, a decoupling mapping matrix is ​​constructed, expressed by the following formula: , in, Let Kronecker function be used when = The value is 1 if it is true, and 0 otherwise. This is the regularization parameter.

[0054] The coupled signal is mapped to an independent feature space using a decoupling mapping matrix, as follows:

[0055] in, These are the independent feature vectors after decoupling. This is the original sensor signal vector.

[0056] Feature extraction is performed on the decoupled signal to construct a wear feature vector: , in, The feature extraction operators include time-domain features such as mean, standard deviation, and peak factor; frequency-domain features such as power spectral density and spectral entropy; and time-frequency-domain features such as wavelet coefficients.

[0057] Furthermore, to improve the reliability of features, an adaptive feature selection mechanism is introduced to calculate feature weights: , in, The variance of the feature These are adaptive weighting coefficients.

[0058] Therefore, the final wear characteristics are expressed as follows: , in, The total number of characteristic operators, For the first Feature extraction operators.

[0059] S4: Construct a time-series phase spectrum matrix based on the independent wear characteristics, perform singular value decomposition on the time-series phase spectrum matrix to obtain the feature importance ranking, use kernel density estimation method to model the probability density of the ranked features, and calculate the remaining life of the cutting tool.

[0060] When predicting remaining lifetime based on acquired independent wear characteristics, it is necessary to construct a time-series phase spectrum matrix to analyze the dynamic evolution characteristics of the characteristics.

[0061] Specifically, after obtaining the wear characteristics of a continuous time series, the feature sequence is segmented using overlapping sliding windows. For the independent wear characteristics within each time window, an improved Hilbert transform is applied to obtain the analytical form of the signal, expressed as: , in, Let be the integral variable, representing the wear characteristic signal at time t within the integration interval. The value; For integration variables, For time window functions, For wear characteristics time series, The window function width parameter, For amplitude modulation coefficients, The gradient normalization factor is... This is the error function.

[0062] Based on the analytic signal, the enhanced instantaneous phase is calculated and expressed as: , in, For enhanced instantaneous phase, This represents the weighting coefficient for historical information.

[0063] Based on enhanced instantaneous phase, a generalized temporal phase spectrum matrix is ​​constructed by sliding the observation window on the time axis, as expressed by the formula: , in, For the first The center moment of a time window This is the time offset relative to the center of the window. For the feature-related time scale, The local entropy of the signal, This is the entropy weighting coefficient.

[0064] The singular value decomposition of the constructed time-series phase spectrum matrix is ​​expressed by the following formula: , in, It is a left singular matrix. This is the transpose of the right singular matrix. For regularization strength, For time-phase balance coefficient, The first element in the time-series phase spectrum matrix OK The element values ​​of the column, The first element in the time-series phase spectrum matrix +1 line The element values ​​of the column, The first element in the time-series phase spectrum matrix OK +1 column element values.

[0065] Furthermore, based on singular value decomposition, the dynamically weighted feature importance is calculated, and the formula for the dynamic weight value is expressed as: , in, For the first One characteristic in time The dynamic weight value at time. For the first singular value decomposition A singular value, This is the adjustment coefficient for time variability. For the first The time variability of each feature The total number of features involved in the calculation. The length of the observation time window. The first one obtained from singular value decomposition The left singular vector at time... The value, This is a time delay variable.

[0066] Enhanced kernel density estimation is performed on the features after importance ranking to construct a probability density model, expressed by the formula: , in, For at any time When, eigenvalues The probability density estimate at that location, For the sample size, For Gaussian kernel function, For the first Feature values ​​of each sample point It is a time-varying bandwidth function. Based on bandwidth, For bandwidth modulation coefficients, For reference time, For time scale parameters, It is a local modulation function. For local enhancement coefficients, It is a local mean. This refers to the spatial scale parameter.

[0067] Based on the constructed probability density model, the remaining life of the cutting tool is calculated, expressed by the formula: , in, For at any time The predicted remaining lifespan, For lifetime prediction weighting function, The failure threshold This represents the conditional probability density.

[0068] Lifetime prediction weighting function The formula is expressed as: , in, As a characteristic time scale, For reliability weighting coefficients, To predict confidence levels.

[0069] Furthermore, the failure threshold is calculated using an adaptive threshold update mechanism, expressed by the following formula: , in, As the initial threshold, This is the threshold adjustment coefficient. The threshold evolution timescale is denoted as .

[0070] Furthermore, the prediction confidence level is calculated using historical prediction errors, expressed by the following formula: , in, The length of the historical observation window. These are historical forecast values. This represents the historical true value.

[0071] In summary, the online detection method for remaining tool life based on multi-source sensor data fusion, as described in this invention, effectively solves the endpoint effect and mode aliasing problems in sensor signal processing through an adaptive decomposition method based on Hilbert-Huang transform, improving the accuracy and stability of signal decomposition. Furthermore, a temperature compensation mechanism is introduced to eliminate the interference of temperature drift on sensor signals, ensuring the accuracy of feature extraction. Second-order cyclic spectrum analysis and decoupling mapping matrix are used to solve the nonlinear coupling problem between multi-sensor signals, extracting independent tool wear features. Finally, by constructing a time-series phase spectrum matrix and combining singular value decomposition and kernel density estimation methods, a dynamic feature importance ranking and life prediction model is established, achieving accurate online prediction of remaining tool life under complex cutting conditions, significantly improving machining efficiency and quality control.

Claims

1. A method for online detection of remaining tool life based on multi-source sensor data fusion, characterized in that, include: Sensor signals are acquired during the cutting process, and intrinsic mode decomposition is performed on the sensor signals based on Hilbert-Huang transform to obtain the inherent vibration modes of the sensor signals. The instantaneous frequency spectrum of the inherent vibration mode is analyzed, the drift component related to temperature change is extracted, and a self-compensation mechanism is used to eliminate the influence of the drift component on the sensor signals to obtain the temperature-compensated sensor signal. Second-order cyclic spectrum analysis is performed on the temperature-compensated sensor signal to identify the coupling frequency components between sensors, a decoupling mapping matrix is ​​constructed, and the independent wear characteristics of each sensor are separated through the decoupling mapping matrix. Based on the independent wear characteristics, a time-series phase spectrum matrix is ​​constructed. Singular value decomposition is performed on the time-series phase spectrum matrix to obtain the feature importance ranking. The kernel density estimation method is used to model the probability density of the ranked features and calculate the remaining life of the cutting tool.

2. The online detection method for remaining life of cutting tools based on multi-source sensor data fusion according to claim 1, characterized in that, The drift components include acceleration signal drift, acoustic emission signal drift, and cutting force signal drift.

3. The online detection method for remaining life of cutting tools based on multi-source sensor data fusion according to claim 2, characterized in that, The acceleration signal drift is expressed by the formula: ， in, The actual measured acceleration signal. For reference temperature, For an ideal acceleration signal, For the random error term of the acceleration signal, and These are the first-order and second-order temperature coefficients, respectively; The acoustic emission signal drift is expressed by the formula: ， in, The original acoustic emission signal frequency, The temperature-velocity of sound coupling coefficient. For frequency disturbance terms; The drift of the cutting force signal is expressed by the following formula: ， in, This is the actual measured cutting force signal. The static temperature coefficient, For dynamic temperature coefficient, For the actual cutting force, For measuring noise.

4. The online detection method for remaining tool life based on multi-source sensor data fusion according to claim 3, characterized in that, A unified temperature drift compensation framework is constructed based on the drift components, and the temperature drift factor is defined as: ， in, For the first The generalized temperature drift factor of this type of sensor This refers to the sensor signal under the current operating conditions. The sensor signal is under reference conditions.

5. The online detection method for remaining tool life based on multi-source sensor data fusion according to claim 4, characterized in that, Based on the temperature drift factor, a nonlinear compensation algorithm based on Volterra series is constructed, expressed by the following formula: ， in, For the compensated signal, The original signal contains temperature drift. As a first-order compensation factor, for The Volterra kernel function describes the nonlinear characteristics of the system. For the time delay variable group, For the first Temperature change at each time delay point It is the highest order of the Volterra series.

6. The online detection method for remaining tool life based on multi-source sensor data fusion according to claim 5, characterized in that, The kernel function The formula for updating adaptively is expressed as follows: ， in, for First-order compensation gain coefficient, for Damping coefficient.

7. The online detection method for remaining cutting tool life based on multi-source sensor data fusion according to claim 6, characterized in that, The temperature-compensated sensor signal is output after adaptive filtering, as expressed by the formula: ， in, For the final output signal, This is a frequency domain correction function. It is a time-domain smoothing function. The angular frequency of the signal. The temperature-dependent cutoff frequency, The time constant is related to the magnitude of the temperature gradient. This represents the absolute value of the temperature gradient.

8. The online detection method for remaining life of cutting tools based on multi-source sensor data fusion according to claim 7, characterized in that, The second-order cyclic spectrum analysis transforms the signal from the time domain to a dual domain of cyclic frequency and spectral frequency. For a single sensor signal, its cyclic spectrum is defined as: ， in, This is the sensor signal after temperature compensation; For time delay variables; For time variables, The length of the observation time window; The cycle frequency, Spectral frequency; Furthermore, a wavelet-enhanced cyclic spectrum estimation method is introduced, using an adaptive weighting function. Enhance the effective features.

9. The online detection method for remaining life of cutting tools based on multi-source sensor data fusion according to claim 8, characterized in that, The decoupling mapping matrix is ​​expressed by the following formula: ， in, Let Kronecker function be used when = The value is 1 if it is true, and 0 otherwise. For regularization parameters, It is a nonlinear coherence function.

10. The online detection method for remaining life of cutting tools based on multi-source sensor data fusion according to claim 1, characterized in that, The singular value decomposition is expressed by the following formula: ， in, It is a left singular matrix. This is the transpose of the right singular matrix. For regularization strength, For time-phase balance coefficient, The first element in the time-series phase spectrum matrix OK The element values ​​of the column, The first element in the time-series phase spectrum matrix +1 line The element values ​​of the column, The first element in the time-series phase spectrum matrix OK +1 column element values.

Images