Cutting force signal synchronous calibration method based on multi-sensor data fusion

Abstract

The invention relates to the technical field of precision machining, in particular to a cutting force signal synchronous calibration method based on multi-sensor data fusion, which comprises the following steps of: 1, dynamically preprocessing acquired signals; step 2, constructing a reference time axis based on the spindle current signal, and realizing time domain alignment of multichannel signals by adopting a dynamic time warping algorithm; 3, establishing a feature incidence matrix containing a force-vibration-temperature coupling relation, and generating a fusion cutting force feature vector through frequency domain energy weight distribution; 4, a dynamic transfer function model is constructed, model parameters are updated in real time, and the fusion feature vector is mapped into a calibration cutting force signal; and 5, carrying out dynamic compensation on the calibration signal by adopting a nonlinear inverse compensation and residual modal screening strategy. According to the method, the accuracy of cutting force signals can be remarkably improved, errors and uncertainty in the machining process are reduced, and the production quality and efficiency are optimized.

Description

Technical Field

[0001] The present invention relates to the field of precision machining technology, and in particular to a cutting force signal synchronization calibration method based on multi-sensor data fusion. Background Art

[0002] In the field of precision machining, especially in high-speed cutting of difficult-to-machine materials (such as titanium alloys and high-temperature alloys) in aerospace and medical device manufacturing, accurate measurement and synchronous calibration of cutting force signals are the core technologies for intelligent monitoring of the machining process. Currently, multi-sensor fusion technology is mainly used for cutting force monitoring in this field, but existing technical solutions face the following technical bottlenecks under complex working conditions:

[0003] Existing synchronous calibration methods often use fixed delay compensation strategies. In high-speed milling of titanium alloys, when the spindle speed switches from 10,000 rpm to 15,000 rpm, the hardware response delay difference between the piezoelectric force sensor and the vibration sensor increases from microseconds to milliseconds, causing a cumulative phase shift. Traditional clock synchronization-based solutions cannot eliminate the random delay jitter caused by differences in sensor physical properties and environmental electromagnetic interference. This leads to waveform distortion when multi-channel signal fusion, seriously affecting the accuracy of cutting force harmonic component analysis.

[0004] Existing data fusion methods have difficulty in effectively separating the coupled interference of cutting force signals and process system vibration and temperature drift. In the processing of thin-walled parts, the instantaneous temperature change in the tool-workpiece contact area can reach 200°C / s, which triggers the thermal expansion effect of the sensor substrate material and causes the piezoelectric force sensor to generate low-frequency drift noise. At the same time, the periodic vibration component caused by the eccentric rotation of the tool (accounting for approximately 15%-30% of the total signal energy) will be reversely coupled to the force sensor through the mechanical transmission path, forming amplitude modulation interference. Due to the lack of multi-physical field correlation analysis, the traditional frequency domain filtering method will lose the high-frequency component of the effective cutting force (>2kHz) while filtering out the interference signal, resulting in the failure of the cutting vibration warning.

[0005] In aerospace engine impeller machining, deviations in cutting force phase detection caused by synchronization errors can cause the chatter suppression system to produce 180° anti-phase control, amplifying the vibration amplitude by up to 300%. In the micromachining of medical implants, time-domain misalignment can mask the transient force characteristics of early tool chipping, resulting in a detection rate of abnormal tool breakage of less than 60%. Therefore, developing a multi-sensor signal synchronization calibration method with dynamic time delay tracking and nonlinear compensation capabilities has become a primary technical challenge for improving the reliability of intelligent machining systems. Summary of the Invention

[0006] Based on the above objectives, the present invention provides a cutting force signal synchronization calibration method based on multi-sensor data fusion, comprising the following steps:

[0007] Step 1: Synchronously collect the triaxial piezoelectric force sensor signal, spindle current signal, tool vibration acceleration signal, and cutting zone infrared temperature signal, and dynamically pre-process the collected signals;

[0008] The dynamic preprocessing includes dynamically adjusting the sampling frequency of the force sensor according to the tool tooth frequency, eliminating the eccentric vibration component by arranging dual acceleration sensors, and calibrating the temperature signal using blackbody radiation;

[0009] Step 2: Construct a reference time axis based on the spindle current signal and use the dynamic time warping algorithm to achieve time domain alignment of multi-channel signals;

[0010] The dynamic time warping algorithm includes path constraints based on tool wear status;

[0011] Step 3: Establish a characteristic correlation matrix containing the force-vibration-temperature coupling relationship, and generate the fusion cutting force characteristic vector through frequency domain energy weight distribution;

[0012] The frequency domain energy weight distribution strategy activates the thermal drift compensation mode when the temperature change rate exceeds the limit and adjusts the low frequency band weight distribution ratio;

[0013] Step 4: Construct a dynamic transfer function model and update the model parameters in real time to map the fused feature vector into a calibrated cutting force signal;

[0014] Step 5: Use nonlinear inverse compensation and residual mode screening strategies to dynamically compensate the calibration signal.

[0015] Preferably, the process of dynamically adjusting the sampling frequency of the force sensor in step 1 is:

[0016] Obtain the spindle speed in real time and calculate the fundamental frequency of the tool tooth frequency and its third harmonic frequency;

[0017] Determining a minimum sampling frequency according to the maximum value of the frequency, wherein the determination method is: multiplying the maximum value by a frequency multiplication coefficient calibrated through a cutting test, wherein the frequency multiplication coefficient is calibrated according to the tool type and the material hardness classification;

[0018] Select the highest available sampling frequency that meets the minimum requirements based on the maximum sampling capability of the data acquisition system to simultaneously avoid high-frequency signal aliasing and reduce the amount of redundant data.

[0019] Preferably, the signal processing process of the dual acceleration sensor is:

[0020] Two acceleration sensors are arranged at radially symmetrical positions on the tool clamping handle, with their symmetry axes coinciding with the tool rotation axis;

[0021] Perform arithmetic averaging on the two vibration signals to eliminate the periodic interference component caused by the eccentric rotation of the tool;

[0022] The correlation coefficient of the two signals is calculated in real time. When the correlation coefficient is lower than the threshold calibrated by the idling test, the signal with smaller amplitude is switched to the valid signal to cope with the signal distortion caused by sudden sensor failure.

[0023] Preferably, the path constraint adjustment method of the dynamic time warping algorithm is:

[0024] Calculating the tool wear coefficient by the center-of-gravity offset of the cutting power spectrum, wherein the center-of-gravity offset of the power spectrum is calculated by the difference between the current cutting power spectrum and the initial reference power spectrum;

[0025] Dynamically adjusting the path slope constraint range according to the graded interval of the wear coefficient, wherein the graded interval is divided by tool life testing and includes low wear, medium wear and high wear states;

[0026] A wide range slope constraint is used in low wear state to preserve the signal morphology characteristics, while a narrow range constraint is used in high wear state to suppress abnormal fluctuation interference.

[0027] Preferably, the process of constructing the feature correlation matrix includes:

[0028] The time-aligned force, vibration, and temperature signals are intercepted at fixed durations to form a three-dimensional data cube. The duration is dynamically adjusted to an integer multiple of the single-tooth cutting cycle based on the current spindle speed.

[0029] The time-varying correlation coefficients between signals are calculated by sliding the window along the time axis, and the de-averaging cross-correlation algorithm is used to eliminate the influence of the DC component of the signal;

[0030] The principal component analysis is performed on the correlation coefficient matrix, and the first N principal components are extracted as the dominant coupling modes. The value of N is dynamically determined according to the preset cumulative variance contribution rate threshold.

[0031] Preferably, the parameter updating mechanism of the dynamic transfer function model includes:

[0032] In the tool presetting stage, the initial transfer function parameters are obtained by a multi-objective optimization algorithm, wherein the optimization objectives include the flatness of the amplitude-frequency characteristic of the force signal and the minimization of the energy of the vibration signal;

[0033] During the online processing stage, the fundamental frequency offset of the vibration spectrum is monitored in real time. The offset is calculated by comparing the relative distance between the current spectrum peak position and the reference position.

[0034] When the offset exceeds a threshold determined by the tool geometry parameters, a parameter update based on the recursive least squares method is triggered, and its forgetting factor decays exponentially according to the continuous cutting time.

[0035] Preferably, the implementation process of the nonlinear inverse compensation is:

[0036] An inverse model of the dynamic characteristics of the force sensor is established, and the truncated Volterra series is used to express the nonlinear relationship.

[0037] Dynamically adjust the order of the series based on the signal-to-noise ratio of the current signal, where the signal-to-noise ratio is calculated as the ratio of the power in the signal's effective frequency band to the power in the noise frequency band;

[0038] When the signal-to-noise ratio decreases, the series order is reduced to prevent noise amplification. The specific order switching threshold is determined by the compensation effect under different working conditions in the calibration test.

[0039] Preferably, the residual mode screening strategy includes:

[0040] Performing empirical mode decomposition on the compensated signal to obtain multiple intrinsic mode function components;

[0041] Calculating a reasonable threshold range for each component based on a cutting force theoretical model, wherein the theoretical model input parameters include tool rake angle, clearance angle, and material removal rate;

[0042] When the energy value of a modal component exceeds a corresponding threshold, it is determined to be an abnormal mode and is eliminated. The energy value is obtained by integrating the power spectrum density of the component within the characteristic frequency band.

[0043] When reconstructing the residual modes, the weight coefficient of each component is positively correlated with the energy proportion of the vibration signal in the corresponding frequency band.

[0044] Preferably, the starting condition of the thermal drift compensation mode is:

[0045] Calculate the first-order derivative of the temperature signal in real time and use the sliding window difference method to eliminate the influence of measurement noise;

[0046] When the absolute value of the derivative continuously exceeds a threshold value calibrated by thermodynamic tests, it is determined that the system has entered a thermal drift state;

[0047] The compensation intensity coefficient is determined by a two-dimensional lookup table, the input dimensions of which are the current temperature value and the temperature change rate, and the data of which are filled through calibration tests in a temperature-controlled environment.

[0048] Preferably, the decay law of the forgetting factor is:

[0049] The initial forgetting factor value is set to the initial value determined by parameter sensitivity analysis;

[0050] The speed decays exponentially after a fixed time interval, and the time interval is inversely proportional to the current cutting feed rate;

[0051] When the forgetting factor value is lower than the lower limit value determined by steady-state error analysis, it remains at the lower limit value until the next parameter update is triggered.

[0052] Beneficial effects of the present invention:

[0053] The use of multi-sensor data fusion technology can obtain comprehensive cutting process information through different types of signals, improving the accuracy and robustness of signal calibration. Secondly, the combination of dynamic time warping and feature fusion strategies can effectively cope with the challenges brought by different wear states and temperature changes, further enhancing the ability of signal synchronization. In addition, the real-time update of the dynamic transfer function and nonlinear inverse compensation can adaptively adjust parameters according to the actual processing state, effectively improving the response speed and accuracy of the system. Overall, this method can significantly improve the accuracy of the cutting force signal, reduce errors and uncertainties in the processing process, and optimize production quality and efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only for the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0055] Figure 1 is a flow chart of the steps of the method of the present invention;

[0056] Figure 2 A flow chart showing the steps of starting conditions for the thermal drift compensation mode of the method of the present invention;

[0057] Figure 3 The figure is a flow chart of the steps of the attenuation law of the forgetting factor of the method of the present invention. DETAILED DESCRIPTION

[0058] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments. It is also noted that, to provide a more detailed description, the following embodiments are best and preferred embodiments, and those skilled in the art may employ alternative methods for implementing certain known technologies. Furthermore, the accompanying drawings are intended only to provide a more detailed description of the embodiments and are not intended to limit the present invention.

[0059] See Figure 1-Figure 3An embodiment of the present invention provides a method for synchronous calibration of cutting force signals based on multi-sensor data fusion. In the first step, a three-axis piezoelectric force sensor, a spindle current sensor, a tool vibration acceleration sensor, and a cutting zone infrared temperature sensor are used to synchronously collect signals. The preprocessing process is very critical. Its purpose is to adapt to changes in tool tooth frequency by dynamically adjusting the sampling frequency, and to eliminate interference caused by eccentric tool vibration through a dual acceleration sensor layout. The temperature signal is calibrated through blackbody radiation to improve measurement accuracy. This process ensures the consistency and high quality of data collected by all sensors, providing a stable foundation for subsequent signal alignment and fusion.

[0060] Step 2 constructs a reference time axis based on the spindle current signal and uses the dynamic time warping (DTW) algorithm to align the signals from different channels in the time domain. During this process, path constraints based on tool wear are introduced to dynamically adjust the alignment strategy as tool wear changes. This ensures precise signal matching on the time axis and reduces errors caused by time misalignment. The application of this algorithm significantly improves the synchronization accuracy of the different sensor signals.

[0061] Step 3 combines the multiple time-domain aligned signals (force, vibration, and temperature) to establish a characteristic correlation matrix for the coupling relationship. A fused cutting force feature vector is generated using a frequency-domain energy weighting strategy. In this step, when the temperature change rate exceeds the limit, the system activates thermal drift compensation mode and adjusts the weight ratio of the low-frequency band to address signal drift caused by temperature fluctuations. This process ensures the rationality of signal fusion and further improves the accuracy of signal calibration.

[0062] Step 4 constructs a dynamic transfer function model based on the fused eigenvectors and updates the model parameters in real time. By monitoring the offset of the fundamental frequency of the vibration spectrum in real time and triggering a recursive least squares method to update the parameters when the offset exceeds a certain threshold, this step ensures that the transfer function model can adaptively respond to changes in the cutting process, further improving the calibration accuracy of the cutting force signal.

[0063] Finally, the calibrated signal is optimized using nonlinear inverse compensation and residual mode screening strategies. During the nonlinear inverse compensation phase, an inverse model of the force sensor's dynamic characteristics is constructed, and the nonlinear relationship is compensated using the Volterra series. During the residual mode screening phase, empirical mode decomposition (EMD) is used to eliminate abnormal modes, ensuring that the signal is free of unnecessary interference. These two steps further eliminate noise and irregular modal components, thereby improving the accuracy and stability of the final cutting force signal.

[0064] In one possible implementation, when dynamically adjusting the force sensor's sampling frequency, it is first necessary to obtain the spindle speed in real time. This information is obtained using a rotary encoder or other speed measurement device. After obtaining the spindle speed, the fundamental frequency of the tool's tooth frequency can be calculated. The fundamental frequency of the tool's tooth frequency is calculated based on the spindle speed and the number of tool teeth. This frequency is the primary frequency of contact between the tool and the workpiece during the cutting process. Furthermore, the third harmonic frequencies of this fundamental frequency must be calculated; these frequencies contain important information related to tool vibration in the cutting force signal.

[0065] Based on the maximum frequency calculated in the previous step (i.e., the maximum value of the tool tooth fundamental frequency and its third harmonic frequency), the system's minimum sampling frequency is determined. This minimum sampling frequency is determined by multiplying the maximum frequency by a calibrated multiplication factor. This multiplication factor is calibrated through cutting tests and is categorized based on the tool type and material hardness. The multiplication factor is chosen to ensure that the sampling frequency is at least twice the frequency of interest in the signal to avoid information loss.

[0066] Finally, the system needs to select a sampling frequency based on the maximum sampling capacity of the data acquisition system. The selected sampling frequency must meet the minimum sampling frequency requirements while also avoiding signal aliasing caused by excessively high sampling frequencies and reducing unnecessary redundant data. Redundant data not only increases the computational burden but can also reduce processing efficiency. Therefore, selecting an appropriate sampling frequency can optimize system efficiency while ensuring data accuracy.

[0067] By dynamically adjusting the force sensor's sampling frequency, we can ensure the quality of signal acquisition and the timeliness of data. In practical applications, the frequency characteristics of the cutting process vary significantly. Factors such as tool wear, cutting conditions, and material properties can cause frequency fluctuations. Dynamically adjusting the sampling frequency allows real-time adaptation to varying cutting conditions, avoiding the errors introduced by a fixed sampling frequency. Furthermore, selecting the highest available sampling frequency avoids signal aliasing and reduces redundant data, significantly improving system efficiency and response speed.

[0068] In one possible implementation, two acceleration sensors are placed on the tool holder shank, symmetrically distributed in the radial direction. Specifically, the two sensors are located on opposite sides of the tool holder, with their axes of symmetry strictly aligned with the tool's rotational axis. This arrangement ensures that the two sensors can synchronously detect interference signals caused by eccentricity, vibration, and other factors as the tool rotates.

[0069] The two acquired acceleration signals are first arithmetic averaged. This process effectively eliminates periodic vibration interference caused by eccentric tool rotation. Ideally, the interference signals caused by eccentricity have opposite phases at two symmetrical points. After averaging, they cancel each other out, thus improving signal purity and effectiveness.

[0070] The system monitors the correlation coefficient between the two acceleration signals in real time. This correlation coefficient reflects the consistency of the vibrations sensed by the two sensors. Under normal operating conditions, the correlation coefficient should remain high because the two sensors are subjected to the same force environment. An empirical threshold can be calibrated through idling tests (i.e., when the tool is running without load and not cutting). When the correlation coefficient is detected to be lower than this threshold, it indicates an abnormality such as sensor detachment, failure, or poor contact. At this point, the system automatically switches to a signal with a smaller amplitude as the valid input to prevent erroneous information caused by a single sensor failure from entering the subsequent processing module.

[0071] In one possible implementation, the currently collected cutting force signal is subjected to spectral analysis to obtain its power spectrum. This power spectrum is then compared with a baseline power spectrum of the tool's initial state, and the offset of the spectrum's center of gravity is calculated. The center of gravity reflects the center of the signal's energy distribution, and its offset can be used to reflect the changing trend of tool wear.

[0072] According to the amplitude of the spectrum center of gravity deviation, the wear coefficient is calculated and matched to the classification intervals previously divided by life test. These intervals are divided into:

[0073] Low wear range: tool performance is close to new state;

[0074] Medium wear range: the tool is in the late stage of normal use;

[0075] High wear range: The tool is about to fail and the signal fluctuation increases significantly.

[0076] Dynamic path slope constraint adjustment:

[0077] In the DTW algorithm, the path slope constraint is used to limit the tilt of the alignment path to avoid unreasonable time matching. Based on the current wear level, the system dynamically selects the constraint strategy:

[0078] Low wear state: Set loose slope constraints (such as increasing the Sakoe-Chiba bandwidth) to allow for larger deformations between signals, helping to preserve the complete signal shape and key features.

[0079] High wear state: Set strict slope constraints to suppress sharp fluctuations or abnormal noise caused by wear, so as to avoid affecting the stability and credibility of the matching results.

[0080] Among them, the path constraint adjustment mechanism is an important part of the entire multi-sensor data fusion and synchronous calibration system. Specifically:

[0081] The power spectrum is obtained by processing multi-source sensor data such as vibration and cutting force, and is the output of the signal feature extraction step;

[0082] The DTW algorithm is used for time alignment of cutting force signals and is the core method of synchronous calibration.

[0083] The proposed path constraint adjustment mechanism serves as a bridge to organically combine the physical state of tool wear with the signal alignment strategy, realizing a dynamic adaptive matching strategy based on working condition changes.

[0084] In a possible implementation, first, signal data collected from multiple sensors (such as force sensors, vibration sensors, temperature sensors, etc.) during the cutting process are collected.

[0085] These signal data need to be aligned in the time domain, that is, the signals of all sensors need to be synchronized in time to facilitate subsequent joint analysis.

[0086] These time-aligned signals are then sliced at fixed intervals to form a three-dimensional data cube, where each dimension corresponds to a signal (force, vibration, temperature). The time dimension is dynamically adjusted to an integer multiple of the single-tooth cutting cycle based on the spindle speed. This ensures synchronization between the signals and the actual cutting process, avoiding analysis errors caused by inconsistent sampling.

[0087] After constructing the data cube, a sliding window is used along the time axis to calculate the time-varying correlation coefficients between the various signals. This process helps quantify the correlation between different signals and reveals their coupling behavior during the cutting process.

[0088] A de-averaged cross-correlation algorithm (removing the DC component of the signal) is used to eliminate the influence of non-cutting behaviors such as low-frequency drift, thereby focusing on the real dynamic coupling mode.

[0089] The calculated correlation coefficient matrix is used as input to perform principal component analysis (PCA). PCA is a commonly used data dimensionality reduction technique that can reveal the main coupling patterns between signals by extracting the principal components in the data.

[0090] Specifically, the first N principal components are selected, and the value of N is dynamically determined according to a preset cumulative variance contribution rate threshold to ensure that the extracted principal components can effectively describe the main features of the signal without containing excessive noise.

[0091] In a possible implementation, during the tool presetting stage, initial parameters of the dynamic transfer function are obtained through a multi-objective optimization algorithm.

[0092] The goals of this optimization algorithm include two aspects: one is the flatness of the amplitude-frequency characteristics of the force signal, and the other is the minimization of the energy of the vibration signal.

[0093] The flatness of the force signal's amplitude-frequency characteristics requires that the transfer function accurately describe the force signal's frequency response to avoid unwanted fluctuations in the spectrum. Minimizing vibration signal energy, on the other hand, aims to minimize the noise component in the vibration signal by adjusting the transfer function, thereby enhancing signal stability and accuracy.

[0094] Through these optimization objectives, preliminary transfer function parameters can be obtained, providing a basis for the subsequent online processing stage.

[0095] During the online processing stage, the vibration spectrum is monitored in real time and the offset of the spectrum base frequency is calculated.

[0096] The offset is determined by comparing the relative distance between the current spectrum peak position and the reference position. This offset reflects the tool vibration during the machining process. If the fundamental frequency of the spectrum shifts too much, it may indicate tool wear, abnormal vibration, or other factors affecting machining accuracy.

[0097] When the detected spectrum deviation exceeds a preset threshold determined by the tool geometry parameters, a parameter update mechanism based on recursive least squares (RLS) is triggered.

[0098] Recursive least squares method is an adaptive filtering algorithm that can continuously update the transfer function parameters according to real-time data to ensure that the transfer function matches the dynamic characteristics of the actual cutting process.

[0099] During the update, the forgetting factor decays exponentially according to the continuous cutting time, which means that the parameters that have not changed for a long time will gradually reduce their impact on the update results, thereby enhancing the adaptability and real-time performance of the algorithm.

[0100] In one possible implementation, first, an inverse model of the dynamic characteristics of the force sensor is established through experiments and data collection. This model is used to reversely calculate the nonlinear characteristics of the force sensor, thereby achieving nonlinear inverse compensation of the signal.

[0101] The main purpose of this inverse model is to obtain a more accurate cutting force signal by compensating for the nonlinear effects in the force sensor. This is usually established by analyzing the relationship between the sensor output and input and constructing a mathematical model to reflect this nonlinear behavior.

[0102] To accurately describe the nonlinear relationship of the force sensor, the truncated Volterra series is used as a mathematical tool. The Volterra series is a high-order polynomial function widely used to describe nonlinear systems. By convolving the input signal with functions of multiple orders, it can well express the dynamic characteristics of nonlinear systems.

[0103] Truncating the Volterra series means making approximations in the high-order terms and retaining the most influential low-order terms, thereby simplifying the calculation and ensuring the accuracy of the compensation effect.

[0104] The signal-to-noise ratio (SNR) is an important indicator of signal quality and plays a crucial role in nonlinear compensation. The SNR is determined by calculating the ratio of the power in the signal's effective frequency band to the power in the noise frequency band.

[0105] When the signal-to-noise ratio is high, the nonlinear relationship can be expressed more accurately, so higher-order Volterra series terms can be retained; when the signal-to-noise ratio is low, in order to avoid noise amplification, the series order will be reduced, making the compensation process more stable and reducing the interference of noise on the compensation results.

[0106] When the signal-to-noise ratio is detected to be lower than a preset threshold, the system will automatically reduce the order of the Volterra series to prevent inaccurate compensation results due to excessively high series when the noise influence is large.

[0107] The order switching threshold is determined through calibration tests under different working conditions. Usually, the order switching rules are optimized through experimental data and feedback from compensation effects to ensure the stability and effectiveness of the compensation algorithm under various working conditions.

[0108] In one possible implementation, empirical mode decomposition (EMD) is an adaptive method for signal analysis that decomposes a signal into multiple intrinsic mode functions (IMFs), each representing an independent mode in the signal. The purpose of this step is to decompose the compensated cutting force signal to extract the modal characteristics of each frequency band.

[0109] The advantage of this process is that it does not require any prior assumptions and is particularly effective for nonlinear and non-stationary signals. Therefore, it can accurately extract the various modal components in the signal and provide clear component characteristics for subsequent analysis.

[0110] Using a cutting force theoretical model, we calculate the appropriate threshold range for each modal component based on operating parameters such as the tool rake angle, clearance angle, and material removal rate. This cutting force theoretical model reflects the dynamics of force during machining. By inputting tool parameters and machining conditions, it predicts the normal energy distribution of the signal across different frequency bands.

[0111] The purpose of this step is to provide a basis for subsequent modal screening, define the energy range of each modal component, and ensure that no valid signal components are missed during the screening process.

[0112] After obtaining the threshold range for each modal component, the energy value of each modal component is calculated and compared with its corresponding reasonable threshold. When the energy value of a modal component exceeds its preset threshold, the mode is considered to be an abnormal mode.

[0113] This abnormal mode is usually caused by noise, sensor interference or other abnormal signal sources. By eliminating these abnormal modes, you can effectively remove unnecessary noise components and ensure the quality of subsequent signals.

[0114] After removing the anomalous modes, the remaining modes are reconstructed. During the reconstruction process, the weight coefficients of each component are determined based on the energy contribution of each component within the characteristic frequency band. This gives higher energy components greater weights, ensuring they dominate the signal reconstruction.

[0115] This weight distribution strategy is based on the principle of energy proportion, which helps to ensure that the mode with the greatest impact on the cutting force signal is retained during reconstruction, thereby improving the accuracy and reliability of the signal.

[0116] In one possible implementation, the temperature signal collected by the sensor during the cutting process is monitored in real time and its first-order derivative is calculated. This derivative reflects the rate of temperature change and can indicate the thermal dynamics of the system.

[0117] To eliminate the effects of measurement noise, a sliding window differencing method is used. By differentiating and smoothing consecutive data points, this method effectively removes high-frequency noise interference, accurately reflecting the true trend of temperature changes. This operation ensures that the calculated temperature change rate is more stable and reliable.

[0118] When the absolute value of the temperature change rate continuously exceeds the threshold value calibrated by thermodynamic testing, the system determines that it has entered a thermal drift state. The thermodynamic testing is conducted in a temperature-controlled environment to ensure the rationality and accuracy of the threshold value.

[0119] The key to this step is to monitor the rate of temperature change to determine whether there is any cutting force signal drift caused by temperature changes. This method can respond to temperature changes during the cutting process in real time, ensuring that compensation mode is activated in a timely manner in the event of thermal drift.

[0120] Once the system enters a thermal drift state, temperature compensation is required to eliminate the temperature effect on the cutting force signal. The compensation intensity coefficient is determined by searching a two-dimensional lookup table. The input dimensions of the lookup table include the current temperature value and the temperature change rate.

[0121] The data in the lookup table is populated through calibration tests in a temperature-controlled environment. The calibration test measures the variation of the cutting force signal under different temperatures and temperature change rates to obtain the compensation coefficient.

[0122] The function of the lookup table is to quickly find the corresponding compensation intensity coefficient according to the actually measured temperature and temperature change rate to ensure the accuracy of compensation.

[0123] In one possible implementation, an initial forgetting factor value is first determined based on parameter sensitivity analysis. Parameter sensitivity analysis is a method for evaluating the impact of system parameters on performance. The initial forgetting factor value can be determined based on different system settings and actual application requirements. This analysis allows the initial forgetting factor to more accurately reflect the actual changing characteristics of the cutting process.

[0124] The forgetting factor decays exponentially within each fixed time interval. The rate of decay is inversely proportional to the cutting feed rate: faster cutting feed rates result in slower decay, and vice versa. Changes in cutting feed rate directly affect the rate of change of cutting force. Therefore, attenuating the force inversely with feed rate allows for more precise signal calibration under varying cutting conditions.

[0125] The form of exponential decay is usually expressed as:

[0126] λ(t)=λ0·e -α·t ;

[0127] Where λ(t) is the current forgetting factor value, λ0 is the initial forgetting factor value, α is the decay rate constant, and t is time. By adjusting the decay rate constant α, the decay speed of the forgetting factor can be controlled.

[0128] The present invention encompasses any alternatives, modifications, equivalents, and solutions that fall within the spirit and scope of the present invention. To provide a thorough understanding of the present invention, specific details are described in detail below in connection with the preferred embodiments of the present invention, but those skilled in the art will be able to fully understand the present invention without these detailed descriptions. Furthermore, to avoid unnecessary confusion regarding the essence of the present invention, well-known methods, processes, procedures, components, and circuits have not been described in detail.

[0129] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. A cutting force signal synchronization calibration method based on multi-sensor data fusion, characterized in that: The following steps are involved: Step 1: Synchronously collect the triaxial piezoelectric force sensor signal, spindle current signal, tool vibration acceleration signal, and cutting zone infrared temperature signal, and dynamically pre-process the collected signals; The dynamic preprocessing includes dynamically adjusting the sampling frequency of the force sensor according to the tool tooth frequency, eliminating the eccentric vibration component by arranging dual acceleration sensors, and calibrating the temperature signal using blackbody radiation; Step 2: Construct a reference time axis based on the spindle current signal and use the dynamic time warping algorithm to achieve time domain alignment of multi-channel signals; The dynamic time warping algorithm includes path constraints based on tool wear status; Step 3: Establish a characteristic correlation matrix containing the force-vibration-temperature coupling relationship, and generate the fusion cutting force characteristic vector through frequency domain energy weight distribution; The frequency domain energy weight distribution strategy activates the thermal drift compensation mode when the temperature change rate exceeds the limit and adjusts the low frequency band weight distribution ratio; Step 4: Construct a dynamic transfer function model and update the model parameters in real time to map the fused feature vector into a calibrated cutting force signal; Step 5: Use nonlinear inverse compensation and residual mode screening strategies to dynamically compensate the calibration signal.

2. The method for synchronous calibration of cutting force signals based on multi-sensor data fusion according to claim 1, characterized in that: The process of dynamically adjusting the force sensor sampling frequency in step 1 is as follows: Obtain the spindle speed in real time and calculate the fundamental frequency of the tool tooth frequency and its third harmonic frequency; Determining a minimum sampling frequency according to the maximum value of the frequency, wherein the determination method is: multiplying the maximum value by a frequency multiplication coefficient calibrated through a cutting test, wherein the frequency multiplication coefficient is calibrated according to the tool type and the material hardness classification; Select the highest available sampling frequency that meets the minimum requirements based on the maximum sampling capability of the data acquisition system to simultaneously avoid high-frequency signal aliasing and reduce the amount of redundant data.

3. The method for synchronous calibration of cutting force signals based on multi-sensor data fusion according to claim 1, characterized in that: The signal processing process of the dual acceleration sensor is as follows: Two acceleration sensors are arranged at radially symmetrical positions on the tool clamping handle, with their symmetry axes coinciding with the tool rotation axis; Perform arithmetic averaging on the two vibration signals to eliminate the periodic interference component caused by the eccentric rotation of the tool; The correlation coefficient of the two signals is calculated in real time. When the correlation coefficient is lower than the threshold calibrated by the idling test, the signal with smaller amplitude is switched to the valid signal to cope with the signal distortion caused by sudden sensor failure.

4. The method for synchronous calibration of cutting force signals based on multi-sensor data fusion according to claim 1, characterized in that: The path constraint adjustment method of the dynamic time warping algorithm is: Calculating the tool wear coefficient by the center-of-gravity offset of the cutting power spectrum, wherein the center-of-gravity offset of the power spectrum is calculated by the difference between the current cutting power spectrum and the initial reference power spectrum; Dynamically adjusting the path slope constraint range according to the graded interval of the wear coefficient, wherein the graded interval is divided by tool life testing and includes low wear, medium wear and high wear states; A wide range slope constraint is used in low wear state to preserve the signal morphology characteristics, while a narrow range constraint is used in high wear state to suppress abnormal fluctuation interference.

5. The method for synchronous calibration of cutting force signals based on multi-sensor data fusion according to claim 1, characterized in that: The process of constructing the feature association matrix includes: The time-aligned force, vibration, and temperature signals are intercepted at fixed durations to form a three-dimensional data cube. The duration is dynamically adjusted to an integer multiple of the single-tooth cutting cycle based on the current spindle speed. The time-varying correlation coefficients between signals are calculated by sliding the window along the time axis, and the de-averaging cross-correlation algorithm is used to eliminate the influence of the DC component of the signal; The principal component analysis is performed on the correlation coefficient matrix, and the first N principal components are extracted as the dominant coupling modes. The value of N is dynamically determined according to the preset cumulative variance contribution rate threshold.

6. The method for synchronous calibration of cutting force signals based on multi-sensor data fusion according to claim 1, characterized in that: The parameter updating mechanism of the dynamic transfer function model includes: In the tool presetting stage, the initial transfer function parameters are obtained by a multi-objective optimization algorithm, wherein the optimization objectives include the flatness of the amplitude-frequency characteristic of the force signal and the minimization of the energy of the vibration signal; During the online processing stage, the fundamental frequency offset of the vibration spectrum is monitored in real time. The offset is calculated by comparing the relative distance between the current spectrum peak position and the reference position. When the offset exceeds a threshold determined by the tool geometry parameters, a parameter update based on the recursive least squares method is triggered, and its forgetting factor decays exponentially according to the continuous cutting time.

7. The method for synchronous calibration of cutting force signals based on multi-sensor data fusion according to claim 1, characterized in that: The implementation process of the nonlinear inverse compensation is: An inverse model of the dynamic characteristics of the force sensor is established, and the truncated Volterra series is used to express the nonlinear relationship. Dynamically adjust the order of the series based on the signal-to-noise ratio of the current signal, where the signal-to-noise ratio is calculated as the ratio of the power in the signal's effective frequency band to the power in the noise frequency band; When the signal-to-noise ratio decreases, the series order is reduced to prevent noise amplification. The specific order switching threshold is determined by the compensation effect under different working conditions in the calibration test.

8. The method for synchronous calibration of cutting force signals based on multi-sensor data fusion according to claim 1, characterized in that: The residual mode screening strategy includes: Performing empirical mode decomposition on the compensated signal to obtain multiple intrinsic mode function components; Calculating a reasonable threshold range for each component based on a cutting force theoretical model, wherein the theoretical model input parameters include tool rake angle, clearance angle, and material removal rate; When the energy value of a modal component exceeds a corresponding threshold, it is determined to be an abnormal mode and is eliminated. The energy value is obtained by integrating the power spectrum density of the component within the characteristic frequency band. When reconstructing the residual modes, the weight coefficient of each component is positively correlated with the energy proportion of the vibration signal in the corresponding frequency band.

9. The method for synchronous calibration of cutting force signals based on multi-sensor data fusion according to claim 1, characterized in that: The starting conditions of the thermal drift compensation mode are: Calculate the first-order derivative of the temperature signal in real time and use the sliding window difference method to eliminate the influence of measurement noise; When the absolute value of the derivative continuously exceeds a threshold value calibrated by thermodynamic tests, it is determined that the system has entered a thermal drift state; The compensation intensity coefficient is determined by a two-dimensional lookup table, the input dimensions of which are the current temperature value and the temperature change rate, and the data of which are filled through calibration tests in a temperature-controlled environment.

10. The method for synchronous calibration of cutting force signals based on multi-sensor data fusion according to claim 6, characterized in that: The decay law of the forgetting factor is: The initial forgetting factor value is set to the initial value determined by parameter sensitivity analysis; The speed decays exponentially after a fixed time interval, and the time interval is inversely proportional to the current cutting feed rate; When the forgetting factor value is lower than the lower limit value determined by steady-state error analysis, it remains at the lower limit value until the next parameter update is triggered.

Images