Self-adaptive compensation system based on cutter wear of numerical control machine tool

Abstract

The invention discloses a self-adaptive compensation system based on cutter wear of a numerical control machine tool, and particularly relates to the field of machining precision control of the numerical control machine tool, and the self-adaptive compensation system comprises a multi-dimensional data acquisition module, a progressive data processing module, a hierarchical analysis module, a multi-stage compensation output module, a data storage module and a self-diagnosis module. The multi-dimensional data acquisition module synchronously acquires cutting force, spindle dynamic state and vibration parameters; the progressive data processing module performs preprocessing, feature extraction and data fusion on the acquired parameters; the hierarchical analysis module is used for realizing wear grade evaluation through single-mode identification, multi-mode association verification and fuzzy comprehensive evaluation; the multi-stage compensation output module outputs a state snapshot and a graded compensation instruction and predicts the service life of the cutter; the data storage module adopts local and cloud dual backup to store full-process data; and the self-diagnosis module detects the state of each module and pushes a diagnosis report when the state is abnormal. According to the system, the wear evaluation precision and real-time performance are improved, the misjudgment rate is reduced, and the machining stability is guaranteed.

Description

Technical Field

[0001] This invention relates to the field of CNC machine tool machining accuracy control technology, and more specifically, to an adaptive compensation system based on CNC machine tool tool wear. Background Technology

[0002] As high-end manufacturing develops towards higher precision and stability, CNC machine tools, as core processing equipment, directly determine the dimensional accuracy and surface quality of workpieces due to the wear condition of their cutting tools. During long-term cutting, cutting tools are subjected to the coupled effects of mechanical friction, thermal shock, and chemical corrosion, resulting in aging phenomena such as dulling and chipping of the cutting edge. If not compensated for in a timely and accurate manner, this can easily lead to batches of workpieces exceeding tolerances, or even cause serious failures such as tool breakage and spindle damage.

[0003] Tool wear control technology in the industry has gone through three stages of development: In the early stage, it relied on manual timed detection, which had a response lag of 2 to 4 hours and was prone to causing scrap accumulation; in the middle stage, it used single parameter threshold monitoring, which could only qualitatively judge wear and could not quantify the degree, with compensation accuracy error exceeding 0.01mm; in recent years, although attempts have been made to introduce intelligent algorithms, hundreds of sets of wear samples need to be collected to train the model, and different tool-workpiece combinations need to be re-adapted, resulting in poor universality and high deployment costs.

[0004] The existing technology has significant drawbacks: First, parameter acquisition is incomplete, ignoring the coupling relationship between cutting force, torque, and vibration, which can easily lead to electromagnetic interference being misjudged as wear signals. Second, spatiotemporal synchronization is poor, with no unified clock reference, resulting in a time difference of several hours between real-time cutting data and offline detection data, making it impossible to establish a causal relationship. Third, data processing is fragmented, with a single denoising method, incomplete feature extraction, high data dimensionality after fusion, and an evaluation delay exceeding 30 seconds, making it difficult to meet real-time compensation requirements. Fourth, the output results are limited, providing only a vague "whether to compensate" instruction without a tiered compensation strategy, making it difficult for maintenance personnel to quickly locate faults and prioritize repairs, resulting in severely insufficient engineering practicality. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide an adaptive compensation system based on CNC machine tool tool wear, which solves the problems mentioned in the background art through the following solutions.

[0006] To achieve the above objectives, the present invention provides the following technical solution: an adaptive compensation system based on CNC machine tool tool wear, comprising: Multi-dimensional data acquisition module, progressive data processing module, hierarchical analysis module, multi-level compensation output module, data storage module and self-diagnosis module; The multi-dimensional data acquisition module is used to synchronously acquire cutting force parameters, spindle dynamic parameters, and vibration parameters during the CNC machine tool machining process; The progressive data processing module is connected to the multi-dimensional data acquisition module and performs three levels of processing on the acquired parameters: first-level preprocessing, second-level feature extraction, and third-level data fusion. The hierarchical analysis module is connected to the progressive data processing module and performs three levels of analysis on the processed data, namely, first-level single-modal state identification, second-level multimodal correlation verification, and third-level wear level assessment. The multi-level compensation output module is connected to the hierarchical analysis module and outputs basic status information, hierarchical compensation instructions and tool life prediction reports based on the analysis results. The data storage module is connected to the multi-dimensional data acquisition module, the progressive data processing module, and the hierarchical analysis module, and is used to store the entire process data. The self-diagnosis module is used to periodically detect the hardware status, data transmission links, and algorithm operation of each module. The self-diagnosis cycle is 24 hours. When a hardware failure, data interruption, or algorithm abnormality is detected, a diagnostic report is automatically generated and pushed to the operation and maintenance terminal.

[0007] The technical effects and advantages of this invention are as follows: 1. This invention constructs a multi-dimensional parameter acquisition system for "cutting force - spindle dynamics - vibration". It synchronously acquires three core parameters using a triaxial piezoelectric force sensor (9257B), a torque sensor (JN338A), and a triaxial accelerometer (ADXL355), and achieves spatiotemporal synchronization using a PLC-built-in GPS clock (accuracy ±1μs), establishing a coupled correlation model of "force - torque - vibration". For example, through the coordinated judgment of "increased cutting force coefficient of variation + increased torque coefficient + increased vibration kurtosis value", it can effectively distinguish between real wear and interference signals such as electromagnetic interference and uneven blank allowance, reducing the false judgment rate. 2. This invention designs a three-level progressive data processing flow of "preprocessing - feature extraction - data fusion" to specifically address the fragmentation problem of existing data processing technologies: The first-level preprocessing adopts differentiated denoising methods such as db6 wavelet denoising and 50Hz notch filtering to reduce the data noise rate; the second-level feature extraction fully covers the time domain, frequency domain (such as the main frequency f0 and spectral entropy), and time and frequency domain (such as the energy spectrum of 6-band wavelet packets) features to ensure that wear information is not lost; the third-level data fusion uses PCA dimensionality reduction to compress the 28-dimensional feature vector into 7-dimensional principal components, reducing computational complexity and shortening evaluation delay, fully meeting the response requirements of real-time compensation for CNC machine tools. 3. This invention constructs a hierarchical analysis logic of "single-modal recognition - multi-modal correlation verification - wear level assessment". First, single-modal recognition is used to preliminarily determine the wear state corresponding to each parameter. Then, a correlation matrix is ​​established using the Pearson correlation coefficient to verify the authenticity of the signal. Finally, a fuzzy comprehensive evaluation model is used to divide the wear into four levels to achieve accurate quantification. Correspondingly, the multi-level compensation output module triggers differentiated actions for different levels: Level V2 (slight wear) requires only a small displacement of the turret without interrupting processing; Level V3 (moderate wear) simultaneously executes "displacement compensation + feed rate adjustment"; Level V4 (severe wear) triggers audible and visual alarms and shutdown. This strategy ensures processing accuracy while avoiding excessive shutdown. 4. This invention breaks through the limitations of the single output of existing systems and provides three types of core outputs: First, basic output, which includes 7-dimensional principal component values ​​and single-modal status identifiers, making it easy for technicians to trace and verify; second, hierarchical compensation instructions, which directly connect to the CNC machine tool actuator to achieve "immediate judgment and compensation"; and third, trend output, which uses the gray model GM(1,1) to predict the remaining life with an error of ≤5%, allowing maintenance personnel to plan tool procurement and replacement plans 3-6 months in advance to avoid resource waste. 5. This solution does not rely on a large number of samples for training. It determines the weights of the fuzzy evaluation model through the analytic hierarchy process. For different tool-workpiece combinations, only the model coefficients need to be adjusted (e.g., when machining aluminum alloys with a high-speed steel tool, the weight vector can be adjusted to A=[0.35,0.4,0.25]). The adaptation period is ≤2 hours, which is 24 times better than deep learning models (which require more than 48 hours of sample training). At the same time, the hardware uses mature industrial-grade products, with no special high-performance computing requirements. The deployment cost is lower than that of intelligent algorithm solutions. It can be widely adapted to various equipment such as CNC lathes, milling machines, and machining centers, as well as various machining materials such as steel, aluminum, and titanium alloys. Attached Figure Description

[0008] Figure 1 This is a schematic diagram of the overall structure of the present invention. Detailed Implementation

[0009] 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.

[0010] As attached Figure 1 The adaptive compensation system based on CNC machine tool tool wear shown includes a multi-dimensional data acquisition module, a progressive data processing module, a hierarchical analysis module, a multi-level compensation output module, a data storage module, and a self-diagnosis module; The multi-dimensional data acquisition module is used to synchronously acquire cutting force parameters, spindle dynamic parameters, and vibration parameters during the CNC machine tool machining process; The multi-dimensional data acquisition module includes: Cutting force parameter acquisition unit: adopts a triaxial piezoelectric force sensor (model 9257B), installed on the transition flange between the turret and the tool holder, with a measurement range of 0-2000N in the X / Y direction and 0-3000N in the Z direction, a resolution of 0.1N, a sampling rate of 500Hz, a data length of 4096 points per frame, and a timestamp accuracy of ±1μs; The installation method of the cutting force parameter acquisition unit is as follows: The 9257B triaxial force sensor is fixed to the turret transition flange with four M8 hex bolts (torque 25 N·m). A 0.1 mm copper foil pad is placed between the top surface of the sensor and the bottom surface of the tool holder to ensure tight contact. A stainless steel protective cover (thickness 1.5 mm) is installed around the sensor to prevent cutting fluid from entering. The ±5V signal output by the sensor is amplified by the AD620 gain module and then connected to the PCI-1710 data acquisition card (sampling rate 500 Hz). Spindle dynamic parameter acquisition unit: includes a torque sensor (model JN338A) and a photoelectric encoder (model E6B2-CWZ6C). The torque sensor is installed at the coupling between the spindle motor and the spindle box, with a measurement range of 0-50 N·m, an accuracy of 0.5%FS, and a sampling interval of 10 ms. The encoder is installed at the end of the spindle, with a resolution of 1000 P / R, and acquires the spindle speed (unit r / min) in real time. The installation method of the spindle dynamic parameter acquisition unit is as follows: The JN338A torque sensor is installed at the coupling between the spindle motor and the spindle box through a Z2 type expansion sleeve (inner diameter 30mm, outer diameter 50mm), with a locking washer and Loctite 243 thread locking glue to prevent loosening; the E6B2-CWZ6C encoder is connected to the spindle extension shaft through a flexible coupling (LK2-14C), and the L-shaped bracket is fixed to the spindle box with a coaxiality error ≤0.02mm; Vibration parameter acquisition unit: A triaxial accelerometer (model ADXL355) is used, which is attached to the side wall of the spindle box. The measurement range is ±20g, the sensitivity is 0.1mg / LSB, the sampling rate is 1000Hz, and a vibration spectrum diagram is generated every 30s. The tool number and the model of the workpiece being processed are recorded simultaneously.

[0011] The progressive data processing module is connected to the multi-dimensional data acquisition module and performs three levels of processing on the acquired parameters: first-level preprocessing, second-level feature extraction, and third-level data fusion. The primary preprocessing of the progressive data processing module includes: The cutting force signal is denoised using a 3-level decomposition of db6 wavelet with a threshold of λ=0.8σ (σ is the noise standard deviation) to remove high-frequency interference of 100Hz±5Hz. The spindle torque signal is filtered by a 50Hz notch filter, and a 10-minute effective value sequence is calculated to remove outliers that are greater than 1.1 times the rated torque. The vibration signal was filtered by a moving average with a window size of 20 sampling points to remove abrupt changes that deviated from the adjacent points by ±0.5g. The first-order difference method is used for the rotational speed data. When |Δn / Δt|>50r / min・s, a second sampling calibration is initiated to ensure the authenticity of the data.

[0012] The secondary feature extraction of the progressive data processing module includes: Time-domain characteristics: average amplitude of cutting force, peak factor (peak / effective value), torque change rate (ΔM / Δt), root mean square value of vibration signal; Frequency domain characteristics: the dominant frequency f0 of the cutting force signal after FFT transformation, and the 8-hour spectral entropy of the torque signal (H=-Σp). i logp i p i (i) represents the probability of the i-th frequency component and the three main resonant frequencies of the vibration signal; Time-frequency domain characteristics: The wavelet packet energy spectrum of the vibration signal contains 6 frequency bands (center frequency 100Hz-5kHz), and the energy ratio of each frequency band (E1 / Etotal ~ E6 / Etotal) is calculated.

[0013] The three-level data fusion of the progressive data processing module includes: Time alignment: Based on the PLC's built-in GPS clock, all data are interpolated to a 1-second time scale, with a synchronization error of ≤5ms; Spatial matching: Establish a three-dimensional coordinate system between the tool and the workpiece (X-axis for feed direction, Y-axis for radial direction, and Z-axis for depth of cut), and map the cutting force and vibration position to the corresponding coordinate points; Feature fusion: Construct a 28-dimensional feature vector (cutting force 8-dimensional, spindle parameters 10-dimensional, vibration 10-dimensional), reduce the dimensionality to ≥95% using PCA, and finally retain 7 principal component features to reduce computational complexity.

[0014] The hierarchical analysis module is connected to the progressive data processing module and performs three levels of analysis on the processed data, namely, first-level single-modal state identification, second-level multimodal correlation verification, and third-level wear level assessment. The first-level single-modal state identification of the hierarchical analysis module includes: Cutting force condition: Calculate the coefficient of variation Cv (standard deviation / mean) of the cutting force amplitude. When Cv > 0.3, it is judged as a dull cutting edge. According to the ratio of Z-direction to radial cutting force (Fz / (Fx+Fy)), when the ratio > 1.5, it is marked as an abnormal cutting condition. Spindle status: The wear coefficient k (k = current torque / initial torque) is calculated using a torque decay model. When k > 1.2, a spindle load warning is triggered. Based on the speed fluctuation Δn (|real-time speed - set speed|), when Δn > 20 r / min, the load is determined to be unstable. Vibration state: Calculate the kurtosis value of the vibration signal (kurtosis value = 4th moment / root mean square value). 4 When the kurtosis value is greater than 3.5, it is determined to be a minor chipping of the tool; according to the main resonant frequency offset Δf (|real-time main frequency - initial main frequency|), when Δf > 50Hz, a vibration abnormality warning is triggered.

[0015] The second-level multimodal correlation verification of the hierarchical analysis module includes: A 3×3 correlation matrix Mpq is established (P / Q represent cutting force, spindle speed, and vibration mode, respectively), and the correlation coefficient is calculated using Pearson correlation coefficient: , where n sam X is the total number of samples. p,r Let r be the feature value of the r-th sample in the p-th mode. The mean of the p-th modality samples; When |Mpq|>0.7, a strong correlation is determined (such as the coordinated change of "increased cutting force - increased torque", which is confirmed as real wear); when |Mpq|<0.3, a historical database comparison is initiated (retrieving multimodal data of the same tool-workpiece combination in the past 3 years; if 80% of the historical records are "interference signals", the current data is marked as "to be verified").

[0016] The hierarchical analysis module employs a fuzzy comprehensive evaluation model for its three-level wear level assessment, including: Factor set U = {u1, u2, u3} (u1 is the cutting force characteristic, u2 is the main spindle characteristic, and u3 is the vibration characteristic); The weight vector A = [0.4, 0.35, 0.25] is determined based on the analytic hierarchy process: a judgment matrix is ​​constructed, the eigenvector corresponding to the largest eigenvalue is calculated and normalized, and the consistency test is passed (CR < 0.1). The evaluation set V = {V1, V2, V3, V4} (V1 = healthy, V2 = slight wear, V3 = moderate wear, V4 = severe wear) is calculated using the membership function: V1 (Healthy): Cutting force variation coefficient Cv≤0.1, torque coefficient k≤1.05, vibration kurtosis value≤2.5; V2 (slight wear): 0.1 < Cv ≤ 0.2, 1.05 < k ≤ 1.1, 2.5 < kurtosis value ≤ 3.0; V3 (moderate wear): 0.2 < Cv ≤ 0.3, 1.1 < k ≤ 1.2, 3.0 < kurtosis value ≤ 3.5; V4 (Severe Wear): Cv > 0.3, k > 1.2, kurtosis > 3.5.

[0017] Taking the machining of 45# steel workpieces on a CNC milling machine (using a high-speed steel end mill with a diameter of 20mm) as an example: Level 1 single-modal recognition: Cutting force condition: Cv=0.25 (>0.2), judged as dull cutting edge; Fz / (Fx+Fy)=1.6 (>1.5), marked as abnormal cutting; Spindle status: Torque coefficient k = 1.15 (> 1.1), triggering load warning; Speed ​​fluctuation Δn = 15 r / min (< 20 r / min), spindle status is normal; Vibration status: Kurtosis value = 3.2 (>3.0), judged as micro-collapse; dominant frequency offset Δf = 30Hz (<50Hz), vibration is basically normal; Second-level association verification: Calculate the correlation matrix: M(cutting force - spindle) = 0.82 (strong correlation), M(cutting force - vibration) = 0.75 (strong correlation), M(spindle - vibration) = 0.68 (moderate correlation), confirming the wear signal is real; Level 3 wear rating assessment: Membership calculation: Membership of cutting force to V3 = 0.8, membership of spindle to V3 = 0.6, membership of vibration to V3 = 0.7; Fuzzy synthesis: V3 level score = 0.4×0.8+0.35×0.6+0.25×0.7=0.705, judged as V3 (moderate wear).

[0018] The multi-level compensation output module is connected to the hierarchical analysis module and outputs basic status information, hierarchical compensation instructions and tool life prediction reports based on the analysis results. The multi-level compensation output module includes: Basic output: A status snapshot is generated every 10 minutes, including 7-dimensional principal component values ​​and single-mode status indicators (such as "edge dull / no dullness" and "vibration abnormal / normal"). The snapshot is in both PDF and Excel formats and stored on the local server. Tiered compensation instructions: When the level is determined to be V2, level 1 compensation is triggered: the turret is controlled by a servo driver (model MR-J4-200A) to make a small displacement (compensation accuracy 0.001mm) without interrupting the machining process; When the level is determined to be V3, secondary compensation is triggered: "displacement compensation + parameter adjustment" is executed simultaneously. The displacement compensation amount Δr = initial tool radius - (target workpiece radius + wear amount), and the feed rate is adjusted to v_f = initial feed × (1 - 0.15 × wear coefficient k). When the level is determined to be V4, the third-level compensation is triggered: output audible and visual alarm (sound pressure level ≥85dB, red strobe light frequency 2Hz), pause machine tool feed, and push maintenance work order containing tool number and wear position (three-dimensional coordinates); Trend Output: Monthly tool wear rate reports are generated, and the remaining tool life is predicted using a grey model GM(1,1). The model construction steps are as follows: S1.1, Accumulate and generate the wear level data {x(1),x(2),…,x(12)} from the past 12 months: (k=1,2,…,12); S1.2 Establish the differential equation: , where a is the development coefficient (reflecting the wear rate) and b is the amount of ash action (reflecting the influence of external factors); S1.3 Solving for parameters [a, b] using the least squares method T =(B T B) -1 B T y, where: , ; S1.4 Calculate the cumulative predicted value at time k+1 based on parameters a and b: ; The original wear level prediction value is obtained by cumulative subtraction and restoration: .

[0019] The data storage module is connected to the multi-dimensional data acquisition module, the progressive data processing module, and the hierarchical analysis module, and is used to store the entire process data. It should be specifically noted that the data storage module includes: Local storage: Uses industrial-grade SSDs (1TB capacity, 500MB / s read / write speed) to store nearly 3 months of raw data, processed features and analysis results. The data retention period can be configured through the human-machine interface. Cloud storage: Monthly trend reports and fault records are synchronized to the cloud server via a 5G industrial gateway (upload speed 100Mbps), supporting historical data retrieval and multi-device data comparison and analysis; Data encryption: AES-256 encryption algorithm is used to encrypt transmitted data, and access permissions are set for local data (administrator / operator hierarchical authorization) to ensure data security.

[0020] The self-diagnosis module is used to periodically detect the hardware status, data transmission links, and algorithm operation of each module. The self-diagnosis cycle is 24 hours. When a hardware failure, data interruption, or algorithm abnormality is detected, a diagnostic report is automatically generated and pushed to the operation and maintenance terminal.

[0021] Specifically, the self-diagnosis module includes a hardware detection unit, a link diagnosis unit, and an algorithm verification unit: The hardware detection unit periodically checks the sensor power supply voltage (normal range 24V±5%) and output signal amplitude (e.g., 4-20mA for torque sensor). When the voltage deviates or the signal is unresponsive, it is marked as "hardware fault". The link diagnostic unit: detects Ethernet connections between modules using the Ping command (a packet loss rate > 1% is considered a link anomaly), verifies data transmission integrity using CRC checksum, and triggers a retransmission mechanism when the checksum error rate > 0.1%. The algorithm verification unit inputs preset standard data (such as cutting force-torque samples with known wear levels) into the hierarchical analysis module. If the output result deviates from the standard value by more than 5%, it is determined to be "algorithm abnormal". The diagnostic report includes the name of the faulty module, the fault type (hardware / link / algorithm), the time of occurrence, and suggested repair measures, and is pushed to the person in charge via SMS and the operation and maintenance platform APP.

[0022] 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. 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. An adaptive compensation system for CNC machine tool tool wear, characterized in that, include: Multi-dimensional data acquisition module, progressive data processing module, hierarchical analysis module, multi-level compensation output module, data storage module and self-diagnosis module; The multi-dimensional data acquisition module is used to synchronously acquire cutting force parameters, spindle dynamic parameters, and vibration parameters during the CNC machine tool machining process; The progressive data processing module is connected to the multi-dimensional data acquisition module and performs three levels of processing on the acquired parameters: first-level preprocessing, second-level feature extraction, and third-level data fusion. The hierarchical analysis module is connected to the progressive data processing module and performs three levels of analysis on the processed data, namely, first-level single-modal state identification, second-level multimodal correlation verification, and third-level wear level assessment. The multi-level compensation output module is connected to the hierarchical analysis module and outputs basic status information, hierarchical compensation instructions and tool life prediction reports based on the analysis results. The data storage module is connected to the multi-dimensional data acquisition module, the progressive data processing module, and the hierarchical analysis module, and is used to store the entire process data. The self-diagnosis module is used to periodically detect the hardware status, data transmission links, and algorithm operation of each module. The self-diagnosis cycle is 24 hours. When a hardware failure, data interruption, or algorithm abnormality is detected, a diagnostic report is automatically generated and pushed to the operation and maintenance terminal.

2. The adaptive compensation system based on CNC machine tool tool wear according to claim 1, characterized in that: The multi-dimensional data acquisition module includes: Cutting force parameter acquisition unit: adopts a triaxial piezoelectric force sensor, installed on the transition flange between the turret and the tool holder, with a measurement range of 0-2000N in the X / Y direction and 0-3000N in the Z direction, a resolution of 0.1N, a sampling rate of 500Hz, a data length of 4096 points per frame, and a timestamp accuracy of ±1μs; The installation method of the cutting force parameter acquisition unit is as follows: the 9257B triaxial force sensor is fixed to the turret transition flange with four M8 hex bolts, the top surface of the sensor and the bottom surface of the tool holder are padded with 0.1mm copper foil, and a stainless steel protective cover is installed on the outside. The ±5V signal output by the sensor is amplified by the AD620 gain module and then connected to the PCI-1710 data acquisition card. Spindle dynamic parameter acquisition unit: includes a torque sensor and a photoelectric encoder. The torque sensor is installed at the coupling between the spindle motor and the spindle box, with a measurement range of 0-50 N·m, an accuracy of 0.5%FS, and a sampling interval of 10 ms. The encoder is installed at the end of the spindle, with a resolution of 1000 P / R, and acquires the spindle speed in real time. The installation method of the spindle dynamic parameter acquisition unit is as follows: The JN338A torque sensor is installed at the coupling between the spindle motor and the spindle box through a Z2 type expansion sleeve, and a locking washer + Loctite 243 thread locking glue is used to prevent loosening. The E6B2-CWZ6C encoder is connected to the spindle extension shaft through a flexible coupling. The L-shaped bracket is fixed to the spindle box, and the coaxiality error is ≤0.02mm. Vibration parameter acquisition unit: It adopts a triaxial accelerometer, which is attached to the side wall of the spindle box. The measurement range is ±20g, the sensitivity is 0.1mg / LSB, the sampling rate is 1000Hz, and a vibration spectrum diagram is generated every 30s. The tool number and the model of the workpiece are recorded simultaneously.

3. The adaptive compensation system based on CNC machine tool tool wear according to claim 1, characterized in that: The primary preprocessing of the progressive data processing module includes: The cutting force signal was denoised using a 3-level decomposition of db6 wavelet with a threshold of λ=0.8σ to remove high-frequency interference of 100Hz±5Hz. The spindle torque signal is filtered by a 50Hz notch filter, and a 10-minute effective value sequence is calculated to remove outliers that are greater than 1.1 times the rated torque. The vibration signal was filtered by a moving average with a window size of 20 sampling points to remove abrupt changes that deviated from the adjacent points by ±0.5g. The first-order difference method is used for the rotational speed data. When |Δn / Δt|>50r / min・s, the second sampling calibration is initiated. The secondary feature extraction of the progressive data processing module includes: Time-domain characteristics: average amplitude of cutting force, peak factor, rate of change of torque, and root mean square value of vibration signal; Frequency domain characteristics: the main frequency f0 of the cutting force signal after FFT transformation, the 8-hour spectral entropy of the torque signal, and the three resonant frequencies of the vibration signal; Time-frequency domain characteristics: The wavelet packet energy spectrum of the vibration signal contains 6 frequency bands; calculate the energy proportion of each frequency band. The three-level data fusion of the progressive data processing module includes: Time alignment: Based on the PLC's built-in GPS clock, all data are interpolated to a 1-second time scale, with a synchronization error of ≤5ms; Spatial matching: Establish a three-dimensional coordinate system between the tool and the workpiece, with the X-axis representing the feed direction, the Y-axis representing the radial direction, and the Z-axis representing the depth of cut; Feature fusion: Construct a 28-dimensional feature vector, reduce the dimensionality to ≥95% using PCA, and finally retain 7-dimensional principal component features.

4. The adaptive compensation system based on CNC machine tool tool wear according to claim 1, characterized in that: The first-level single-modal state identification of the hierarchical analysis module includes: Cutting force condition: Calculate the coefficient of variation Cv of the cutting force amplitude. When Cv > 0.3, it is judged as a dull cutting edge. According to the ratio of Z-direction to radial cutting force, when the ratio > 1.5, it is marked as an abnormal cutting condition. Spindle status: The wear coefficient k is calculated using a torque attenuation model. When k > 1.2, a spindle load warning is triggered. Based on the speed fluctuation Δn, when Δn > 20 r / min, the load is determined to be unstable. Vibration status: Calculate the kurtosis value of the vibration signal. When the kurtosis value is >3.5, it is determined to be a micro-breakage of the tool. Based on the main resonant frequency offset Δf, when Δf is >50Hz, a vibration abnormality warning is triggered. The second-level multimodal correlation verification of the hierarchical analysis module includes: Construct a 3×3 correlation matrix Mpq and calculate the correlation coefficient using Pearson's correlation coefficient: , where n sam X is the total number of samples. p,r Let r be the feature value of the r-th sample in the p-th mode. The mean of the p-th modality samples; When |Mpq|>0.7, a strong correlation is determined; when |Mpq|<0.3, a historical database comparison is initiated to retrieve multimodal data of similar tool-workpiece combinations from the past 3 years. If 80% of the historical records are "interference signals", the current data is marked as "to be verified". The hierarchical analysis module employs a fuzzy comprehensive evaluation model for its three-level wear level assessment, including: The factor set U = {u1, u2, u3}, where u1 is the cutting force characteristic, u2 is the main spindle characteristic, and u3 is the vibration characteristic; The weight vector A = [0.4, 0.35, 0.25] is determined based on the analytic hierarchy process: a judgment matrix is ​​constructed, the eigenvector corresponding to the largest eigenvalue is calculated and normalized, and consistency is verified (CR < 0.1). The evaluation set V = {V1, V2, V3, V4}, where V1 = healthy, V2 = slightly worn, V3 = moderately worn, and V4 = severely worn, is calculated using the membership function. V1 health: Cutting force variation coefficient Cv≤0.1, torque coefficient k≤1.05, vibration kurtosis value≤2.5; V2 Slight: 0.1 < Cv ≤ 0.2, 1.05 < k ≤ 1.1, 2.5 < kurtosis value ≤ 3.0; V3 Moderate wear: 0.2 < Cv ≤ 0.3, 1.1 < k ≤ 1.2, 3.0 < kurtosis value ≤ 3.5; V4 severe wear: Cv > 0.3, k > 1.2, kurtosis value > 3.

5.

5. The adaptive compensation system based on CNC machine tool tool wear according to claim 1, characterized in that: The multi-level compensation output module includes: Basic output: A status snapshot is generated every 10 minutes, containing 7-dimensional principal component values ​​and single-modal status indicators. The snapshot is in both PDF and Excel formats and stored on the local server. Tiered compensation instructions: When the level is determined to be V2, level 1 compensation is triggered: the turret is controlled to make a small displacement through the servo driver so as not to interrupt the machining process; When the level is determined to be V3, secondary compensation is triggered: "displacement compensation + parameter adjustment" is executed simultaneously. The displacement compensation amount Δr = initial tool radius - (target workpiece radius + wear amount), and the feed rate is adjusted to v_f = initial feed × (1 - 0.15 × wear coefficient k). When the level is determined to be V4, the third-level compensation is triggered: output an audible and visual alarm, suspend the machine tool feed, and push a maintenance work order containing the tool number and wear location; Trend Output: Monthly tool wear rate reports are generated, and the remaining tool life is predicted using a grey model GM(1,1). The model construction steps are as follows: S1.1, Accumulate and generate the wear level data {x(1),x(2),…,x(12)} from the past 12 months: ，(k=1,2,…,12); S1.2 Establish the differential equation: , where a is the development coefficient and b is the gray action amount; S1.3 Solving for parameters [a, b] using the least squares method T =(B T B) -1 B T y, where: ， ； S1.4 Calculate the cumulative predicted value at time k+1 based on parameters a and b: ； The original wear level prediction value is obtained by cumulative subtraction and restoration: 。 6. The adaptive compensation system based on CNC machine tool tool wear according to claim 1, characterized in that: The data storage module includes: Local storage: Industrial-grade SSDs are used to store nearly 3 months of raw data, processed features and analysis results. The data retention period can be configured through a human-machine interface. Cloud storage: Monthly trend reports and fault records are synchronized to the cloud server via a 5G industrial gateway, supporting historical data retrieval and multi-device data comparison and analysis; Data encryption: AES-256 encryption algorithm is used to encrypt transmitted data, and access permissions are set for local data.

7. The adaptive compensation system based on CNC machine tool tool wear according to claim 1, characterized in that: The self-diagnosis module includes a hardware detection unit, a link diagnosis unit, and an algorithm verification unit. The hardware detection unit periodically detects the sensor's power supply voltage and output signal amplitude. When the voltage deviates or the signal is unresponsive, it is marked as a "hardware fault". The link diagnostic unit detects Ethernet connections between modules using the Ping command, verifies data transmission integrity using CRC checksum, and triggers a retransmission mechanism when the checksum error rate is greater than 0.1%. The algorithm verification unit inputs preset standard data into the hierarchical analysis module. If the output result deviates from the standard value by more than 5%, it is determined to be "algorithm abnormal". The diagnostic report includes the name of the faulty module, the fault type, the time of occurrence, and recommended repair measures, and is pushed to the person in charge via SMS and the operation and maintenance platform APP.

Images