Chuck pin intelligent detection early warning method and device and medium

Abstract

The invention discloses an intelligent detection and early warning method and equipment for a chuck pin and a medium, and belongs to the technical field of chuck pins. The method comprises the steps of integrating three-dimensional gap distribution data, surface wear feature data and a dynamic clamping force curve, constructing a comprehensive state matrix through a data fusion algorithm, and performing dimension reduction processing on the matrix to extract key features; inputting the comprehensive state matrix into a preset deep learning model, and outputting the wear grade, the residual life and the failure risk tag of the chuck pin based on a multi-modal fusion training strategy; according to the abrasion grade, the remaining life and the failure risk label, the nanoscale displacement compensation amount is calculated through a piezoelectric actuator to correct the position deviation, and the pressure of a hydraulic system is controlled and adjusted through a closed-loop PID to stabilize clamping force output; and establishing a digital twin model of the chuck pin, and automatically pushing a maintenance plan when the residual life is lower than a threshold value. By means of the method, automatic detection and early warning of the chuck pin are achieved.

Description

Technical Field

[0001] The present application relates to the technical field of chuck pins, and in particular to a method, device and medium for intelligent detection and early warning of chuck pins. Background Art

[0002] In mechanical manufacturing and semiconductors, chuck pins are critical components for connecting and clamping workpieces. Their condition is directly related to machining quality and production efficiency. However, chuck pins are affected by various factors during use, such as wear, corrosion, and changes in clamping force. These factors can lead to performance degradation and even equipment failure. Existing techniques rely primarily on manual experience and simple measuring tools. This method is not only inefficient but also difficult to fully and accurately assess the actual condition of the chuck pins.

[0003] Therefore, how to automatically detect and warn the chuck pin has become a technical problem that needs to be solved urgently. Summary of the Invention

[0004] The embodiments of the present application provide a method, device and medium for intelligent detection and early warning of a chuck pin, which are used to solve the following technical problem: how to automatically detect and early warning of a chuck pin.

[0005] In the first aspect, the embodiment of the present application provides a chuck pin intelligent detection and early warning method, which is applied to the chuck pin intelligent detection system, and the chuck pin intelligent detection system includes a laser interferometer, a microscopic imaging device and a pressure sensing network. The method includes: based on the laser interferometer, the inner wall of the hole of the chuck pin is scanned in all directions to obtain the three-dimensional gap distribution data of the inner wall of the hole; based on the microscopic imaging device, the pin surface of the chuck pin is subjected to multi-spectral illumination and confocal microscopic imaging to extract the surface wear characteristic data; based on the pressure sensing network, the dynamic clamping force curve of the chuck pin during the clamping process is collected in real time, and the temperature drift and vibration interference are separated by the decoupling algorithm to generate a dynamic clamping force curve; the three-dimensional gap distribution data, surface wear characteristic data and dynamic clamping force curve are integrated to obtain the three-dimensional gap distribution data, surface wear characteristic data and dynamic clamping force curve through data fusion. The combined algorithm constructs a comprehensive state matrix and performs dimensionality reduction processing on the matrix to extract key features; the comprehensive state matrix is ​​input into a preset deep learning model, and the wear level, remaining life and failure risk label of the chuck pin are output based on the multimodal fusion training strategy; wherein, the deep learning model optimizes the generalization ability of unknown wear patterns through transfer learning; according to the wear level, remaining life and failure risk label, the nanometer-level displacement compensation amount is calculated by the piezoelectric driver to correct the position deviation, and the hydraulic system pressure is adjusted through closed-loop PID control to stabilize the clamping force output; a digital twin model of the chuck pin is established, the stress distribution and wear evolution are simulated based on real-time detection data, and the compensation strategy is verified through historical fault data, and the maintenance plan is automatically pushed when the remaining life is lower than the threshold.

[0006] In one implementation of the present application, a laser interferometer is used to perform a full-circumferential scan of the inner wall of the hole of the chuck pin to obtain three-dimensional gap distribution data of the inner wall of the hole, specifically including: controlling the rotating scanning probe to perform a full-circumferential, multi-level coverage scan along the axis of the pin hole, and driving the probe to rotate and move axially by a stepper motor or a servo motor; analyzing the height difference of each point on the inner wall of the hole based on the phase change of the interference fringes to generate initial three-dimensional gap distribution data; compensating the data of the scanning blind area by a pre-trained neural network model; wherein, the deep learning model is trained based on historical complete scanning data, and outputs a predicted gap value after inputting the blind area position information; the compensated data is normalized to eliminate noise interference, and a complete three-dimensional gap distribution data set for the inner wall of the hole is generated.

[0007] In one implementation of the present application, multispectral illumination and confocal microscopy are performed on the pin surface of the chuck pin based on a microscopic imaging device to extract surface wear feature data, specifically including: switching the multispectral light source to scan the pin surface point by point to obtain surface morphology images at different wavelengths; focusing layer by layer and recording the reflected light intensity through confocal microscopy technology to reconstruct the three-dimensional morphology information of the pin surface; denoising and contrast enhancement processing are performed on the three-dimensional morphology data to extract quantitative indicators of surface roughness, wear depth and wear area to generate surface wear feature data.

[0008] In one implementation of the present application, the dynamic clamping force curve during the clamping process of the chuck pin is collected in real time based on a pressure sensing network, and the temperature drift and vibration interference are separated by a decoupling algorithm to generate a dynamic clamping force curve. Specifically, it includes: arranging a distributed pressure sensor array inside the pin and key parts of the chuck to collect force signals from multiple nodes in real time; constructing a mathematical model of temperature drift and vibration interference, and separating the environmental interference component from the original force signal through a decoupling algorithm; performing time series analysis on the denoised force signal to generate a dynamic curve of the clamping force changing with time, and marking the location and intensity of abnormal fluctuation events.

[0009] In one implementation of the present application, the three-dimensional gap distribution data, surface wear characteristic data and dynamic clamping force curve are integrated, a comprehensive state matrix is ​​constructed through a data fusion algorithm, and the matrix is ​​subjected to dimensionality reduction processing to extract key features, specifically including: standardizing the three-dimensional gap distribution data, surface wear characteristic data and clamping force timing signals and unifying the data format and dimension; fusing multi-source data using a principal component analysis algorithm to construct a comprehensive state matrix including gap distribution, wear characteristics and dynamic changes in clamping force; and reducing the dimensionality of the comprehensive state matrix based on linear discriminant analysis to extract key feature vectors that are strongly correlated with wear level and failure risk.

[0010] In one implementation of the present application, the comprehensive state matrix is ​​input into a preset deep learning model, and the wear level, remaining life and failure risk label of the chuck pin are output based on a multimodal fusion training strategy, specifically including: constructing a multimodal fusion convolutional neural network model, the input layer receives gap distribution data, three-dimensional wear feature map and clamping force timing signal respectively; loading pre-trained mechanical parts wear prediction model parameters through transfer learning to optimize the network's recognition ability of unknown wear patterns; the output layer generates a visual warning map containing high-risk area coordinates and risk levels, and associates the remaining life prediction value and failure risk probability label.

[0011] In one implementation of the present application, based on the wear level, remaining life and failure risk label, the piezoelectric driver is used to calculate the nanometer-scale displacement compensation amount to correct the position deviation, and the hydraulic system pressure is adjusted through closed-loop PID control to stabilize the clamping force output, specifically including: calculating the nanometer-scale displacement compensation amount of the piezoelectric driver based on the position error, generating a real-time control signal to drive the piezoelectric element to adjust the pin position; dynamically adjusting the hydraulic system pressure through the PID controller, adjusting the proportional, integral and differential parameters according to the clamping force deviation value to stabilize the output; recording the displacement adjustment amount and pressure change data during the compensation process, and feeding back to the deep learning model to optimize the prediction accuracy and compensation strategy.

[0012] In one implementation of the present application, a digital twin model of the chuck pin is established, stress distribution and wear evolution are simulated based on real-time detection data, and the compensation strategy is verified through historical fault data. When the remaining life is lower than a threshold, a maintenance plan is automatically pushed. Specifically, it includes: constructing a finite element digital twin model of the pin, and synchronizing three-dimensional gap distribution, surface wear characteristics and clamping force data to the model in real time; injecting historical fault data into the finite element digital twin model, simulating stress distribution and wear evolution trends under different working conditions, and evaluating the effectiveness of the compensation strategy; when the finite element digital twin model predicts that the remaining life is lower than a preset threshold, the maintenance system is triggered to automatically generate a spare parts replacement work order and push it to the maintenance terminal.

[0013] In the second aspect, the embodiment of the present application also provides an intelligent detection and early warning device for a chuck pin, characterized in that the device includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can: perform a full-circumferential scan of the inner wall of the hole of the chuck pin based on a laser interferometer to obtain three-dimensional gap distribution data of the inner wall of the hole; perform multi-spectral illumination and confocal microscopy imaging of the pin surface of the chuck pin based on a microscopic imaging device to extract surface wear feature data; collect the dynamic clamping force curve of the chuck pin in real time during the clamping process based on a pressure sensing network, and separate temperature drift and vibration interference through a decoupling algorithm to generate a dynamic clamping force curve; integrate the three-dimensional gap distribution The data, surface wear characteristic data and dynamic clamping force curve are used to construct a comprehensive state matrix through a data fusion algorithm, and the matrix is ​​subjected to dimensionality reduction processing to extract key features; the comprehensive state matrix is ​​input into a preset deep learning model, and the wear level, remaining life and failure risk label of the chuck pin are output based on the multimodal fusion training strategy; wherein, the deep learning model optimizes the generalization ability of unknown wear patterns through transfer learning; according to the wear level, remaining life and failure risk label, the nano-level displacement compensation amount is calculated by the piezoelectric driver to correct the position deviation, and the hydraulic system pressure is adjusted through closed-loop PID control to stabilize the clamping force output; a digital twin model of the chuck pin is established, the stress distribution and wear evolution are simulated based on real-time detection data, and the compensation strategy is verified through historical fault data, and a maintenance plan is automatically pushed when the remaining life is lower than the threshold.

[0014] On the third aspect, the embodiment of the present application also provides a non-volatile computer storage medium for intelligent detection and early warning of chuck pins, which stores computer executable instructions, and is characterized in that the computer executable instructions are set to: perform full-circle scanning of the inner wall of the hole of the chuck pin based on a laser interferometer to obtain three-dimensional gap distribution data of the inner wall of the hole; perform multi-spectral illumination and confocal microscopy imaging of the pin surface of the chuck pin based on a microscopic imaging device to extract surface wear characteristic data; collect the dynamic clamping force curve of the chuck pin in the clamping process in real time based on a pressure sensing network, and separate temperature drift and vibration interference through a decoupling algorithm to generate a dynamic clamping force curve; integrate the three-dimensional gap distribution data, surface wear characteristic data and dynamic clamping force curve, and construct a dynamic clamping force curve through a data fusion algorithm. A comprehensive state matrix is ​​generated, and the matrix is ​​subjected to dimensionality reduction processing to extract key features; the comprehensive state matrix is ​​input into a preset deep learning model, and the wear level, remaining life and failure risk label of the chuck pin are output based on a multimodal fusion training strategy; wherein, the deep learning model optimizes the generalization ability of unknown wear patterns through transfer learning; according to the wear level, remaining life and failure risk label, the nanometer-level displacement compensation amount is calculated by the piezoelectric driver to correct the position deviation, and the hydraulic system pressure is adjusted through closed-loop PID control to stabilize the clamping force output; a digital twin model of the chuck pin is established, the stress distribution and wear evolution are simulated based on real-time detection data, and the compensation strategy is verified through historical fault data, and a maintenance plan is automatically pushed when the remaining life is lower than the threshold.

[0015] The embodiments of the present application provide a method, device and medium for intelligent detection and early warning of chuck pins, which realize comprehensive monitoring of the three-dimensional gap distribution of the inner wall of the chuck pin hole, the wear characteristics of the pin surface and the changes in the clamping force during the clamping process through a laser interferometer, a microscopic imaging device and a pressure sensing network. This improves the accuracy and comprehensiveness of the chuck pin status assessment. By integrating these multi-dimensional data, a comprehensive state matrix is ​​constructed, which provides a rich and accurate information basis for subsequent deep learning model processing. Based on the deep learning model to determine the wear level, remaining life and failure risk label of the chuck pin, the position deviation and clamping force parameters of the chuck pin are automatically corrected according to the evaluation results to extend the service life of the chuck pin, improve the stability and reliability of the equipment operation, and reduce the production interruption and maintenance costs caused by the failure of the chuck pin. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0017] Figure 1 A flow chart of a chuck pin intelligent detection and early warning method provided in an embodiment of the present application;

[0018] Figure 2 A schematic diagram of the internal structure of a chuck pin intelligent detection and early warning device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0019] To make the purpose, technical solutions, and advantages of this application more clear, the technical solutions of this application will be clearly and completely described below in conjunction with the specific embodiments of this application and the corresponding drawings. Obviously, the embodiments described are only part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0020] The application embodiment provides a chuck pin intelligent detection and early warning method, device and medium to solve the following technical problem: how to automatically detect and warn the chuck pin.

[0021] The technical solutions proposed in the embodiments of the present application are described in detail below with reference to the accompanying drawings.

[0022] Figure 1 This is a flowchart of an intelligent detection and warning of a chuck pin provided in an embodiment of the present application. Figure 1 As shown, an embodiment of the present application provides an intelligent detection and early warning method for a chuck pin, which specifically includes the following steps:

[0023] Step 1: Scan the inner wall of the hole of the chuck pin in all directions using a laser interferometer to obtain three-dimensional gap distribution data of the inner wall of the hole.

[0024] Step 1.1, control the rotary scanning probe to perform full-circumferential, multi-level coverage scanning along the axis of the pin hole, and drive the probe to rotate and move axially through a stepper motor or a servo motor.

[0025] Rotating scanning probe: A precision probe with integrated laser transmitting and receiving devices that can rotate around the axis of the pin hole and move axially to achieve full coverage scanning of the inner wall of the hole.

[0026] Full circumferential scanning: The probe rotates 360 degrees around the axis of the pin hole to ensure that the scan covers the entire circumference of the hole wall.

[0027] Multi-level coverage scanning: The probe moves in layers along the axis of the pin hole while rotating, scanning the inner wall areas of different depths layer by layer.

[0028] Operation process:

[0029] Fix the chuck pin on the detection platform and adjust the probe of the laser interferometer to the entrance of the pin hole.

[0030] The stepper motor drives the probe to rotate at a constant speed (e.g., 10 revolutions per second) around the axis of the pin hole, while the servo motor controls the probe to advance layer by layer along the axial direction at a preset step length (e.g., 0.1 mm).

[0031] After each layer of scanning is completed, the probe moves to the next axial position and repeats the rotation scan until the entire length of the pin hole is covered.

[0032] Step 1.2: Analyze the height difference of each point on the inner wall of the hole based on the phase change of the interference fringes to generate initial three-dimensional gap distribution data.

[0033] Interference fringe phase change: The laser beam is reflected after hitting the inner wall of the hole, interfering with the reference beam to form fringes. The phase change reflects the height difference of the measured surface.

[0034] Initial 3D gap distribution data: The original data set of relative heights of each point on the hole wall generated by phase analysis.

[0035] In a specific example, a laser interferometer transmits a laser beam to the inner wall of a hole, receives the reflected light, and interferes with the reference light to generate an interference fringe image. A phase-decomposition algorithm (such as the phase-shift method) analyzes the phase changes in the interference fringes and calculates the height difference of each point on the inner wall of the hole. This height difference data is converted into point cloud data in a three-dimensional coordinate system to form an initial three-dimensional gap distribution map.

[0036] Step 1.3: Compensate the data of the scanning blind spot through a pre-trained neural network model; wherein the deep learning model is trained based on historical complete scanning data, and outputs a predicted gap value after inputting the blind spot position information.

[0037] Scanning blind area: An area that cannot be directly scanned due to physical obstruction or structural limitations of the probe.

[0038] Pre-trained neural network model: A deep learning model trained on historical complete scan data to predict gap distribution in blind spots.

[0039] In a specific example, complete, blind-spot-free chuck pin scan data was selected from a historical database, and the blind spot locations and their corresponding true gap values ​​were annotated. A convolutional neural network (CNN) was constructed, which input the known gap data and location coordinates around the blind spot and output a predicted blind spot value. The currently scanned blind spot location information was fed into the trained model to generate a predicted gap value, which was then populated into the initial three-dimensional gap distribution data.

[0040] Step 1.4: Normalize the compensated data to eliminate noise interference and generate a complete three-dimensional gap distribution data set of the hole inner wall.

[0041] Normalization: Scale the data to a uniform dimension to eliminate sensor noise and environmental interference.

[0042] In a specific example, the compensated 3D gap data was subjected to denoising (e.g., Gaussian filtering) to remove abnormal fluctuations. A minimum-maximum normalization algorithm was used to adjust the data range to the 0-1 range, eliminating dimensional differences between different scan layers. The data from all layers was then integrated to generate a complete 3D gap distribution dataset for the inner wall of the hole.

[0043] In a specific example, precision manufacturing company A required regular inspection of chuck pins on its production line. The process was as follows: The chuck pin to be tested was fixed to a testing platform. The laser interferometer's rotary scanning probe, driven by a servo motor, moved axially along the pin hole in 0.1 mm increments while simultaneously rotating around its axis at 10 revolutions per second. The probe scanned the entire 20 mm length of the pin hole, dividing it into 200 layers, generating 360 sampling points per layer. The interference fringe image captured by the laser interferometer was phase-resolved to generate initial 3D gap distribution data, revealing a localized depression (maximum depth 0.05 mm) on the inner wall of the hole. A 2 mm scanning blind spot was detected at the bottom of the pin hole. The coordinates of this blind spot were input into a pre-trained CNN model, which output a predicted gap value of 0.03 mm, consistent with the data trend of the adjacent area. The compensated data was then Gaussian filtered for denoising and normalized to a uniform dimension, ultimately generating a complete 3D gap distribution dataset for the inner wall of the hole for subsequent wear level assessment.

[0044] Step 2: Perform multispectral illumination and confocal microscopy on the pin surface of the chuck pin using a microscopic imaging device to extract surface wear feature data.

[0045] Step 2.1: Switch the multispectral light source to scan the pin surface point by point to obtain surface topography images at different wavelengths.

[0046] "Switching a multispectral light source" involves sequentially loading narrowband filters through a programmable filter wheel to generate monochromatic illumination of a specific wavelength. "Point-by-point scanning" involves using a motorized translation stage to move pins at preset intervals in the XY plane, triggering a camera to capture an image at each position. The light source wavelength ranges from 400nm to 800nm, encompassing the ultraviolet, visible, and near-infrared bands. Different wavelengths have different reflective responses to features such as metal oxide layers and microcracks. For example, short-wavelength blue light is sensitive to fine scratches, while near-infrared light can penetrate surface oil films to detect substrate wear.

[0047] In a specific example, testing organization A used an LED light source equipped with a six-channel filter wheel to inspect a tungsten carbide pin produced by Company B. The filter wheel sequentially switched to three characteristic wavelengths: 450nm, 550nm, and 650nm. The pin was fixed on a two-dimensional translation stage with a travel accuracy of 1μm, scanning its entire surface at 5μm intervals. The image acquisition module simultaneously recorded brightfield reflectance images of a localized area at each wavelength, generating three sets of two-dimensional surface topography datasets covering 360° along the pin's axis.

[0048] Step 2.2: Use confocal microscopy to focus and record the reflected light intensity layer by layer to reconstruct the three-dimensional morphology of the pin surface.

[0049] "Layer-by-layer focusing" involves controlling the objective lens's axial movement along the Z axis in 0.1μm increments using a piezoelectric ceramic driver, capturing the reflected light intensity distribution at each height level. "Confocal microscopy" employs pinhole spatial filtering to capture only the reflected light signal from the focal plane, suppressing stray light interference from out-of-focus areas. By traversing the entire depth of field, the axial light intensity distribution curve for each pixel is obtained, and the Z coordinate corresponding to the peak value represents the height of that point.

[0050] In the example of step 2.1, test facility A used confocal mode on the same pin sample. The objective lens scanned 50 layers at 20× magnification with an axial step size of 0.5 μm. The exposure time per layer was set to 10 ms to minimize motion blur. A 3D reconstruction algorithm fused the intensity data from each layer to generate a 3D point cloud model containing 200,000 spatial points. The axial resolution reached 0.3 μm, clearly capturing both concave and convex features larger than 2 μm in diameter.

[0051] Step 2.3: De-noise and contrast enhance the three-dimensional morphology data, extract quantitative indicators of surface roughness, wear depth and wear area, and generate surface wear feature data.

[0052] Denoising uses an anisotropic diffusion filter algorithm to eliminate random noise while preserving edge features. Contrast enhancement improves visual recognition of microscopic features through adaptive histogram equalization. Wear depth calculation uses a reference plane fitting method, using the unworn area plane as the reference plane and measuring the maximum relative height difference at each point. Wear area is segmented using a threshold method to identify the projected area of ​​areas where height deviates from the normal range.

[0053] Continuing with the previous example, Test Institution A imported the 3D point cloud into analysis software. Wavelet noise reduction was first applied to eliminate high-frequency noise components, followed by local contrast stretching to enhance the visibility of micron-level wear marks. The software automatically calculated the surface roughness (Sa) as 0.82 μm, detected a maximum wear depth of 12.5 μm, and a cumulative wear area of ​​7.3%. Finally, a test report was generated, including a 3D topography image, a pseudo-color wear distribution map, and a quantitative parameter table for pin service life assessment.

[0054] Step 3: Based on the pressure sensing network, the dynamic clamping force curve during the clamping process of the chuck pin is collected in real time, and the temperature drift and vibration interference are separated by the decoupling algorithm to generate the dynamic clamping force curve.

[0055] Step 3.1: Deploy a distributed pressure sensor array inside the pin and key locations of the chuck to collect force signals from multiple nodes in real time.

[0056] The "distributed pressure sensor array" refers to micro-MEMS piezoresistive sensors embedded equidistantly along the pin's axis, as well as thin-film pressure sensors distributed in a circular pattern on the contact surface of the chuck jaws. "Real-time acquisition" refers to the simultaneous recording of pressure changes at each node at a sampling frequency of at least 1kHz. The sensor layout must cover stress concentration areas along the clamping force transmission path, such as the chamfer at the pin's root and the bottom of the V-groove in the jaws. Each sensor node is connected to the data acquisition module via an anti-interference shielded cable to ensure the fidelity of microvolt-level signal transmission.

[0057] In one specific example, testing organization A embedded three MEMS sensors with an axial spacing of 2 mm inside a titanium alloy pin manufactured by Company B. It also attached two thin-film sensors to each of the four jaws of a matching chuck. When the chuck clamped the pin with a torque of 20 N·m, all sensors were synchronized at a sampling rate of 1200 Hz, continuously recording pressure fluctuations at each node during the clamping process, generating a raw data set containing 14 channels of force signals.

[0058] Step 3.2: Construct a mathematical model of temperature drift and vibration interference, and separate the environmental interference component from the original force signal through a decoupling algorithm.

[0059] The "temperature drift model" establishes a second-order polynomial regression relationship based on the sensor's temperature compensation coefficient, describing the nonlinear relationship between sensor zero-point drift and temperature changes. The "vibration interference model" uses frequency-domain analysis to identify harmonic components that match the device's natural frequency. The decoupling algorithm first uses a wavelet transform to separate the low-frequency temperature drift component from the signal. An adaptive notch filter then eliminates mechanical vibration noise in specific frequency bands, retaining the intermediate-frequency signal components that reflect actual clamping force changes.

[0060] Continuing with the previous example, Test Unit A collected raw data containing interference from ambient temperature fluctuations (25±5°C) and spindle vibration (fundamental frequency 800Hz). The zero drift of each sensor was compensated using a temperature-sensitivity calibration curve, while a recursive least squares algorithm was used to update the amplitude-frequency characteristic parameters of the vibration interference online. The processed signal-to-noise ratio improved by 15dB, effectively eliminating the 120Hz frequency multiplication interference caused by spindle imbalance.

[0061] Step 3.3: Perform time series analysis on the denoised force signal to generate a dynamic curve of the clamping force changing over time, and mark the location and intensity of abnormal fluctuation events.

[0062] "Time Series Analysis" uses a sliding window statistical method to calculate the mean, variance, and peak values ​​of the force signal. The window width is set to 10ms to match the time-varying characteristics of typical clamping movements. "Abnormal Fluctuation Events" are defined as transient pulses exceeding a threshold of three standard deviations or amplitude anomalies lasting more than 5ms. The marker information includes the event's timestamp, phase position relative to the clamping cycle, and a normalized intensity level, which is categorized from 1 to 5 based on the ratio of the fluctuation amplitude to the rated clamping force.

[0063] In the example of step 3.1, the processed force signal showed three abnormal fluctuations exceeding 4.2 N during the clamping process. The analysis software automatically identified the second fluctuation as occurring at 85 ms of the clamping cycle (corresponding to the initial full contact between the pin and the jaw), lasting 8 ms and reaching intensity level 4. The system generates a report containing a time-force curve, a distribution diagram of abnormal events, and a statistical summary, providing a quantitative basis for evaluating the stability of the clamping system.

[0064] Step 4: Integrate the three-dimensional gap distribution data, surface wear characteristic data and dynamic clamping force curve, construct a comprehensive state matrix through a data fusion algorithm, and perform dimensionality reduction processing on the matrix to extract key features.

[0065] Step 4.1: Perform standardized preprocessing on the three-dimensional gap distribution data, surface wear characteristic data, and clamping force timing signal to unify the data format and dimension.

[0066] Among them, "Standardization Preprocessing" includes the following operations:

[0067] Normalization: linearly map the three-dimensional gap data (unit: micron) to the interval 0,1 to eliminate the influence of different dimensions;

[0068] Data alignment: resample the dynamic clamping force timing signal so that its timestamp is synchronized with the surface wear feature acquisition moment;

[0069] Format conversion: The 3D point cloud data is rasterized into a 100×100×100 voxel matrix, the wear quantitative index is converted into an 8-dimensional feature vector, and the peak and valley values, mean, and fluctuation frequency of the clamping force curve are extracted to form a 6-dimensional statistic.

[0070] In a specific example, testing organization A standardized the test data of chuck pin C:

[0071] The 3D gap data was converted into a regular grid with a resolution of 10 μm by voxelization of the original 25,000 point cloud;

[0072] Eight indicators including surface wear Sa roughness (0.82 μm) and wear area ratio (7.3%) were encoded as feature vectors;

[0073] The 120-second time series data of the dynamic gripping force curve was windowed, and the maximum force, variance, and spectral energy percentage within each 200ms window were extracted. All data were Z-score normalized to form a uniform input matrix.

[0074] Step 4.2: Use the principal component analysis algorithm to fuse the multi-source data and construct a comprehensive state matrix including the gap distribution, wear characteristics and dynamic changes of the clamping force.

[0075] The principal component analysis algorithm performs the following core operations:

[0076] Data splicing: Splice the standardized voxel matrix, eigenvectors, and time series statistics into a high-dimensional feature space according to the channel dimension;

[0077] Covariance calculation: Calculate the covariance matrix of cross-modal features based on 200 sets of historical sample data;

[0078] Principal component extraction: retain the first k principal components with cumulative contribution rate ≥ 85% to form a comprehensive state matrix after dimension compression.

[0079] Continuing with the previous example, Test Institution A concatenated the standardized 1×10^6 voxel data, 8-dimensional wear characteristics, and 6-dimensional clamping force statistics into a 1×1,000,014-dimensional input vector. Principal component analysis (PCA) selected the first 12 principal components (with a cumulative contribution of 87.2%) and compressed the data into a 12×1 comprehensive state matrix. The third principal component in this matrix was found to correlate simultaneously with gap asymmetry, mean wear depth, and clamping force fluctuation frequency, revealing an implicit coupling relationship between the three.

[0080] Step 4.3: Reduce the dimension of the comprehensive state matrix based on linear discriminant analysis and extract key feature vectors that are strongly correlated with wear level and failure risk.

[0081] Among them, the implementation points of "Linear Discriminant Analysis" include:

[0082] Category labeling: Based on historical failure records, samples are divided into three categories: normal (level I), slight wear (level II), and critical failure (level III);

[0083] Inter-class separation optimization: Calculate the distribution differences of samples with different wear levels in the comprehensive state matrix and maximize the ratio of inter-class divergence to intra-class divergence;

[0084] Feature projection: Project high-dimensional data into a low-dimensional discriminant space and select the top two discriminant vectors with the highest discrimination as key features.

[0085] In the embodiment of step 4.2, test organization A uses 50 groups of samples with known wear levels to train the discriminant model. Linear discriminant analysis projects the 12-dimensional comprehensive state matrix into a two-dimensional space. The first discriminant vector (contribution rate 78%) mainly reflects the coordinated change trend of the wear area and the clamping force fluctuation, and the second discriminant vector (contribution rate 19%) captures the interaction between the non-uniformity of the gap distribution and the surface roughness. In the projected two-dimensional feature space, the class center distance of the three types of samples increases by 3.7 times, providing highly discriminative feature input for subsequent failure risk assessment.

[0086] Step 5: Input the comprehensive state matrix into a preset deep learning model, and output the wear level, remaining life and failure risk label of the chuck pin based on the multimodal fusion training strategy; wherein, the deep learning model optimizes the generalization ability of unknown wear patterns through transfer learning.

[0087] Step 5.1: Construct a multimodal fusion convolutional neural network model, where the input layer receives gap distribution data, three-dimensional wear feature map, and clamping force time series signal respectively.

[0088] Among them, the "multimodal fusion convolutional neural network model" consists of three parallel feature extraction branches:

[0089] Gap distribution branch: 3D convolution kernel (5×5×5) is used to process voxelized gap data to capture spatial non-uniformity features;

[0090] Wear feature classification: The pseudo-color wear distribution map generated in step 4 is analyzed using a 2D convolutional network to extract local texture patterns.

[0091] Gripping force timing branch: Use time-frequency transformation to convert the original waveform into a Mel-spectrogram, which is then fed into a lightweight convolutional network to extract frequency domain features.

[0092] Multimodal features are weighted and added together in the fusion layer using attention weights, and the fully connected layer integrates cross-modal relationships. The network output includes three objectives: wear level classification, remaining life regression value, and failure risk probability.

[0093] In a specific example, the network structure designed by test organization A is:

[0094] The gap branch contains 4 layers of 3D convolution, and the number of filters increases from 16 to 64 layer by layer;

[0095] The wear branch uses the pre-trained ResNet18 network to extract a 1024-dimensional feature vector;

[0096] The clamping force branch generates a 128×128 spectrum through short-time Fourier transform, and then processes it with MobileNetV3 to obtain 256-dimensional features.

[0097] The attention weights of the three features calculated by the fusion layer are 0.35, 0.45, and 0.20 respectively. The final output dimension of the fully connected layer is (3 types of wear levels + 1 life value + 1 risk probability).

[0098] Step 5.2: Load the pre-trained mechanical parts wear prediction model parameters through transfer learning to optimize the network's ability to recognize unknown wear patterns.

[0099] Among them, the implementation methods of "transfer learning" include:

[0100] Parameter inheritance: Load the time series feature extraction module pre-trained on the TIMESERIES-2K public dataset;

[0101] Domain adaptation: Freeze the parameters of the first three convolutional layers and only fine-tune the fully connected layers and the classification head;

[0102] Data augmentation: Random occlusion, Gaussian noise, and channel shuffling are applied to the input data to improve model robustness.

[0103] Transfer learning enables the network to recognize unusual wear patterns that did not appear in the training set, such as butterfly-shaped wear caused by high-temperature oxidation.

[0104] Continuing with the previous example, test organization A initialized the network weights using a model pre-trained on a bearing wear dataset. Targeting the unique edge wear characteristics of the chuck pin:

[0105] Freeze the first two layers of MobileNetV3 parameters of the timing branch;

[0106] Add 20% synthetic wear samples to the training data to simulate the wear particle embedding effect of different materials;

[0107] The MixUp data enhancement strategy is applied to mix the spectral features of samples with different wear levels.

[0108] Step 5.3: The output layer generates a visual warning map containing the coordinates of high-risk areas and risk levels, and associates the remaining life prediction value and failure risk probability label.

[0109] The process of generating a "visualized early warning map" includes:

[0110] Feature back-projection: Mapping high-dimensional features back to the original gap distribution space through the deconvolution network;

[0111] Heat map generation: Use the Grad-CAM algorithm to locate spatial areas with high contribution to failure risk;

[0112] Multi-objective association: Jointly calibrate the predicted lifespan value (unit: times) and the risk probability (0-100%) to eliminate contradictory outputs.

[0113] The map uses red, yellow and green colors to mark risk areas, and the coordinates of the red areas are accurate to 0.1mm.

[0114] In the embodiment of step 5.1, when the detection data of the chuck pin C is input:

[0115] The model predicts a remaining life of 3500 ± 200 clamping cycles;

[0116] The failure risk probability reaches 92% (threshold >85% triggers a level 1 warning);

[0117] The heat map showed that there was an annular high-risk area at the axial position of 3.2-3.5 mm at the root of the pin.

[0118] The system automatically generates a comprehensive report including three-dimensional risk distribution, life curve and maintenance recommendations to guide users to replace high-risk components in a targeted manner.

[0119] Step 6: Based on the wear level, remaining life and failure risk label, the piezoelectric driver calculates the nanometer displacement compensation amount to correct the position deviation, and adjusts the hydraulic system pressure through closed-loop PID control to stabilize the clamping force output.

[0120] Step 6.1: Calculate the nanometer-level displacement compensation of the piezoelectric actuator based on the position error, and generate a real-time control signal to drive the piezoelectric element to adjust the pin position.

[0121] "Position error" refers to the difference between the pin axis offset measured by a laser displacement sensor and the theoretical centerline. "Nano-level displacement compensation" is calculated using a dual closed-loop control strategy: an outer loop determines the compensation baseline based on the wear level, while an inner loop makes fine adjustments based on real-time position deviations. The piezoelectric driver's built-in stacked piezoelectric ceramic actuator generates a linear displacement of 0-50μm at a drive voltage of 0-150V, with a response frequency of ≥1kHz, enabling real-time correction of micron-level position deviations caused by wear.

[0122] In a specific example, test facility A initiated a compensation procedure for pin C, which had a failure risk label of Level III. A laser sensor detected an 8.2μm radial offset at the pin's base. The compensation algorithm combined the baseline offset for Level III wear (5μm) with the real-time offset to generate a superimposed offset of 9.3μm. The piezoelectric actuator then output an 87V step voltage within 0.8ms, driving the actuator to produce a 9.3μm reverse displacement. Retesting after compensation revealed a reduction in residual offset to 0.7μm, meeting the accuracy requirement of ≤1μm.

[0123] Step 6.2: Dynamically adjust the hydraulic system pressure through the PID controller, and adjust the proportional, integral and differential parameters according to the clamping force deviation value to stabilize the output.

[0124] "Dynamic adjustment" refers to the adaptive adjustment of control parameters based on the real-time fluctuation amplitude of the clamping force:

[0125] Proportional coefficient (Kp): When the deviation is greater than 5N, it switches to high gain mode (Kp = 2.8) to quickly eliminate large deviations;

[0126] Integration time (Ti): Automatically extend the integration time to 200ms during continuous fluctuations to prevent integration saturation;

[0127] Derivative action (Td): For sudden impact loads, enable 10ms derivative control to suppress overshoot.

[0128] The piezoelectric nozzle flapper mechanism of the hydraulic servo valve receives a 0-10mA control current and controls the oil pressure regulation accuracy within ±0.05MPa.

[0129] Continuing with the above embodiment, during the clamping process after the pin C is compensated:

[0130] The pressure sensor detected that the clamping force dropped from the set value of 150N to 142N;

[0131] The adaptive PID switches to high-gain mode, increasing the oil pressure by 0.7 MPa within 12 ms;

[0132] The clamping force recovers to 149.8N within 35ms, and the steady-state error is less than 0.2%.

[0133] The control curve shows that the overshoot is controlled within 3%, eliminating the 5Hz oscillation phenomenon caused by traditional PID.

[0134] Step 6.3: Record the displacement adjustment and pressure change data during the compensation process and feed them back to the deep learning model to optimize the prediction accuracy and compensation strategy.

[0135] Among them, "data feedback" includes timestamps, compensation amount sequences, actual clamping force trajectories, and control parameter change records. The following features are extracted through a sliding window: compensation response delay (unit: ms); pressure regulation overshoot rate (%); steady-state error maintenance time (s);

[0136] The feedback data updates the weights of the fully connected layer of the deep learning model in step 5 in an incremental learning manner, optimizing the synergistic relationship between compensation calculation and pressure regulation.

[0137] After running 50 compensation cycles in the example of step 6.1:

[0138] The cumulative feedback data package contains 2,340 displacement compensation records and corresponding clamping force responses;

[0139] The online model update reduces the remaining life prediction error from ±15% to ±9%;

[0140] The overshoot suppression capability of the adaptive PID is improved by 22%, and the compensation trigger frequency is reduced by 40%.

[0141] The system automatically generates a compensation performance analysis report, recommending optimizing the upper limit of the piezoelectric drive voltage from 150V to 138V to extend the life of the actuator.

[0142] Step 7: Build a digital twin model of the chuck pin, simulate stress distribution and wear evolution based on real-time detection data, verify the compensation strategy through historical failure data, and automatically push a maintenance plan when the remaining life is below the threshold.

[0143] Step 7.1. Build a finite element digital twin model of the pin and synchronize the three-dimensional gap distribution, surface wear characteristics, and clamping force data to the model in real time.

[0144] Among them, the "finite element digital twin model" consists of the following core modules:

[0145] Geometry reconstruction module: converts the 3D topography data reconstructed in step 2.2 into a non-uniform rational B-spline (NURBS) surface, preserving the micron-level wear topography features;

[0146] Material property library: stores the elastic-plastic parameters, fatigue crack growth rate, and temperature dependence curve of the friction coefficient of the pin base material (such as tungsten carbide and titanium alloy);

[0147] Real-time data interface: Synchronize the integrated state matrix of step 4 and the compensation control parameters of step 6 through the OPC UA protocol to update the model boundary conditions.

[0148] The model uses tetrahedron-hexahedron hybrid meshing technology and performs 0.1mm-level local mesh encryption in the contact area to ensure the simulation accuracy of the stress concentration area.

[0149] In a specific example, testing organization A established a digital twin model for a ceramic-coated pin produced by company B:

[0150] The geometry reconstruction module imports the wear depth distribution map extracted in step 2.3 and generates a refined surface containing 372 concave features at the root of the pin;

[0151] The material library loads the three-point bending fatigue test data of the pin (the point where the slope of the SN curve changes when the number of cycles is greater than 10^6 times);

[0152] The real-time interface receives the compensation displacement and hydraulic pressure value fed back in step 6.3 every 5 seconds and dynamically updates the clamping force load distribution.

[0153] The first round of simulations running the model showed that the maximum von Mises stress was concentrated at the edge of the wear pit at 3.2 mm axially, with a position deviation of less than 0.15 mm from the field failure case.

[0154] Step 7.2: Inject historical fault data into the finite element digital twin model, simulate the stress distribution and wear evolution trend under different working conditions, and evaluate the effectiveness of the compensation strategy.

[0155] Among them, "Historical fault data injection" includes the following data types:

[0156] Failure mode library: Contains microscopic morphology scanning data of 12 typical types of damage, including abrasive wear, fatigue spalling, and corrosion pits;

[0157] Working condition parameter set: covers three-dimensional combination conditions of speed (200-8,000rpm), clamping force (50-500N), and temperature (-20℃ to 150℃);

[0158] Compensation strategy case: Stores 100 sets of successful compensation records and 32 sets of failure intervention cases in step 6.

[0159] An explicit dynamics solver is used in the simulation to implant equivalent damaged geometry in the worn area and calculate the residual strength decay rate and crack propagation path.

[0160] Continuing with the above embodiment, test organization A verifies the compensation strategy:

[0161] The historically recorded level III wear characteristics (Sa = 1.2 μm, wear area 9.8%) were embedded in the digital twin model.

[0162] The amplitude of the alternating stress caused by the clamping force fluctuation under the simulated uncompensated state reaches 850 MPa, which exceeds the fatigue limit of the material;

[0163] After enabling the displacement compensation in step 6.1, the stress amplitude dropped to 620 MPa, and the estimated life was extended by 2.7 times.

[0164] The model compared eight compensation solutions and recommended a 9.3μm compensation to reduce the stress concentration factor from 2.8 to 1.9.

[0165] Step 7.3: When the remaining life predicted by the finite element digital twin model is lower than the preset threshold, the maintenance system is triggered to automatically generate a spare part replacement work order and push it to the maintenance terminal.

[0166] Among them, the "preset threshold" is divided into three levels based on engineering experience:

[0167] Warning threshold (remaining life 30%): Push lubrication maintenance recommendations to mobile terminals;

[0168] Intervention threshold (15% remaining life): Generate a preventive replacement work order and schedule a maintenance window;

[0169] Emergency threshold (5% remaining life): Triggers an emergency stop of the equipment and issues an expedited replacement order.

[0170] The work order content includes the spare part batch number, installation torque standard and image guidance of historical similar replacement operations.

[0171] In a specific example, after the embodiment of step 7.1 has been run for three months:

[0172] The digital twin model predicted a remaining life of 14.7% based on real-time wear rates (below the 15% intervention threshold);

[0173] The maintenance system automatically generates work order C-2023-079, specifying the use of 5th-generation reinforced tungsten carbide pins (batch number WN5-2237);

[0174] The work order is simultaneously pushed to the workshop manager's terminal and the warehouse management system, triggering the spare parts delivery and equipment downtime window appointment.

[0175] The system shows that the maintenance priority is P1, and the associated historical data indicates that the average replacement time for similar pins is 28±5 minutes.

[0176] The above is an embodiment of the method proposed in this application. Based on the same inventive concept, this application embodiment also provides a chuck pin intelligent detection and warning device, the structure of which is as follows: Figure 2 shown.

[0177] Figure 2 This is a schematic diagram of the internal structure of a chuck pin intelligent detection and warning device provided in an embodiment of the present application. Figure 2 As shown, the equipment includes:

[0178] at least one processor 201;

[0179] and, a memory 202 communicatively coupled to the at least one processor;

[0180] The memory 202 stores instructions that can be executed by at least one processor, and the instructions are executed by the at least one processor 201 to enable the at least one processor 201 to:

[0181] The inner wall of the hole of the chuck pin is scanned in all directions based on a laser interferometer to obtain the three-dimensional gap distribution data of the inner wall of the hole; the pin surface of the chuck pin is subjected to multi-spectral illumination and confocal microscopy imaging based on a microscopic imaging device to extract surface wear feature data; the dynamic clamping force curve of the chuck pin during the clamping process is collected in real time based on a pressure sensing network, and the temperature drift and vibration interference are separated by a decoupling algorithm to generate a dynamic clamping force curve; the three-dimensional gap distribution data, surface wear feature data and dynamic clamping force curve are integrated, and a comprehensive state matrix is ​​constructed through a data fusion algorithm, and the matrix is ​​subjected to dimensionality reduction processing to extract key features; the comprehensive state matrix is ​​input into the The preset deep learning model outputs the wear level, remaining life and failure risk label of the chuck pin based on a multimodal fusion training strategy; wherein, the deep learning model optimizes the generalization ability of unknown wear patterns through transfer learning; according to the wear level, remaining life and failure risk label, the piezoelectric driver calculates the nanometer-level displacement compensation amount to correct the position deviation, and the hydraulic system pressure is adjusted through closed-loop PID control to stabilize the clamping force output; a digital twin model of the chuck pin is established, and the stress distribution and wear evolution are simulated based on real-time detection data. The compensation strategy is verified through historical failure data, and a maintenance plan is automatically pushed when the remaining life is lower than the threshold.

[0182] Some embodiments of the present application provide corresponding Figure 1 A non-volatile computer storage medium for intelligent detection and early warning of a chuck pin stores computer executable instructions, wherein the computer executable instructions are set to:

[0183] The inner wall of the hole of the chuck pin is scanned in all directions based on a laser interferometer to obtain the three-dimensional gap distribution data of the inner wall of the hole; the pin surface of the chuck pin is subjected to multi-spectral illumination and confocal microscopy imaging based on a microscopic imaging device to extract surface wear feature data; the dynamic clamping force curve of the chuck pin during the clamping process is collected in real time based on a pressure sensing network, and the temperature drift and vibration interference are separated by a decoupling algorithm to generate a dynamic clamping force curve; the three-dimensional gap distribution data, surface wear feature data and dynamic clamping force curve are integrated, and a comprehensive state matrix is ​​constructed through a data fusion algorithm, and the matrix is ​​subjected to dimensionality reduction processing to extract key features; the comprehensive state matrix is ​​input into the The preset deep learning model outputs the wear level, remaining life and failure risk label of the chuck pin based on a multimodal fusion training strategy; wherein, the deep learning model optimizes the generalization ability of unknown wear patterns through transfer learning; according to the wear level, remaining life and failure risk label, the piezoelectric driver calculates the nanometer-level displacement compensation amount to correct the position deviation, and the hydraulic system pressure is adjusted through closed-loop PID control to stabilize the clamping force output; a digital twin model of the chuck pin is established, and the stress distribution and wear evolution are simulated based on real-time detection data. The compensation strategy is verified through historical failure data, and a maintenance plan is automatically pushed when the remaining life is lower than the threshold.

[0184] The various embodiments in this application are described in a progressive manner. Similar portions between the various embodiments can be referenced to each other. Each embodiment focuses on the differences from the other embodiments. In particular, the IoT device and media embodiments are generally similar to the method embodiments, so their description is relatively simple. For relevant portions, refer to the description of the method embodiments.

[0185] The system and medium provided in the embodiments of the present application correspond one-to-one to the method. Therefore, the system and medium also have similar beneficial technical effects to their corresponding methods. Since the beneficial technical effects of the method have been described in detail above, the beneficial technical effects of the system and medium will not be repeated here.

[0186] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0187] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the steps in the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0188] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0189] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 The steps for the function specified in one or more boxes.

[0190] In a typical configuration, a computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory.

[0191] Memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. Memory is an example of a computer-readable medium.

[0192] Computer-readable media includes permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. The information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer-readable media does not include transitory computer-readable media (transitory media), such as modulated data signals and carrier waves.

[0193] It should also be noted that the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, commodity, or apparatus that includes a series of elements includes not only those elements but also other elements not explicitly listed, or includes elements inherent to such process, method, commodity, or apparatus. In the absence of further limitations, an element defined by the phrase "comprises a ..." does not exclude the presence of other identical elements in the process, method, commodity, or apparatus that includes the element.

[0194] The foregoing is merely an embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application should all be included within the scope of the claims of the present application.

Claims

1. A chuck pin intelligent detection and early warning method, applied to a chuck pin intelligent detection system, wherein the chuck pin intelligent detection system includes a laser interferometer, a microscopic imaging device, and a pressure sensing network, and is characterized in that: The method comprises: The inner wall of the chuck pin hole is scanned in all directions using a laser interferometer to obtain three-dimensional gap distribution data on the inner wall of the hole; The surface of the chuck pin is imaged using a microscopic imaging device using multispectral illumination and confocal microscopy to extract surface wear characteristic data. The dynamic clamping force curve of the chuck pin clamping process is collected in real time based on the pressure sensing network, and the temperature drift and vibration interference are separated by the decoupling algorithm to generate the dynamic clamping force curve; Integrating the three-dimensional gap distribution data, surface wear characteristic data and dynamic clamping force curve, constructing a comprehensive state matrix through a data fusion algorithm, and performing dimensionality reduction processing on the matrix to extract key features; Inputting the comprehensive state matrix into a preset deep learning model, the wear level, remaining life, and failure risk label of the chuck pin are output based on a multimodal fusion training strategy; wherein the deep learning model optimizes the generalization ability of unknown wear patterns through transfer learning; Based on the wear level, remaining life, and failure risk label, the piezoelectric driver calculates the nanometer-level displacement compensation to correct the position deviation, and the hydraulic system pressure is adjusted through closed-loop PID control to stabilize the clamping force output; A digital twin model of the chuck pin is established to simulate stress distribution and wear evolution based on real-time detection data. The compensation strategy is verified through historical failure data, and a maintenance plan is automatically pushed when the remaining life falls below the threshold.

2. The intelligent detection and early warning method for chuck pins according to claim 1 is characterized in that: The inner wall of the chuck pin hole is scanned circumferentially using a laser interferometer to obtain three-dimensional gap distribution data on the inner wall of the hole, including: Control the rotating scanning probe to perform full-circumferential, multi-level coverage scanning along the axis of the pin hole, and drive the probe to rotate and move axially through a stepper motor or servo motor; The height difference of each point on the inner wall of the hole is analyzed based on the phase change of the interference fringes to generate the initial three-dimensional gap distribution data; Compensate for data from scanning blind spots using a pre-trained neural network model; wherein the deep learning model is trained based on historical complete scanning data, inputs blind spot location information, and outputs predicted gap values; The compensated data are normalized to eliminate noise interference and generate a complete three-dimensional gap distribution data set of the hole inner wall.

3. The intelligent detection and early warning method for chuck pins according to claim 1 is characterized in that: The surface of the chuck pin is subjected to multispectral illumination and confocal microscopy imaging using a microscopic imaging device to extract surface wear feature data, including: Switch the multi-spectral light source to scan the pin surface point by point to obtain surface topography images at different wavelengths; The three-dimensional topography of the pin surface was reconstructed by focusing and recording the reflected light intensity layer by layer through confocal microscopy. The 3D topography data is processed for denoising and contrast enhancement, and quantitative indicators of surface roughness, wear depth and wear area are extracted to generate surface wear feature data.

4. The intelligent detection and early warning method for chuck pins according to claim 1 is characterized in that: The dynamic clamping force curve of the chuck pin clamping process is collected in real time based on the pressure sensing network. The decoupling algorithm is used to separate temperature drift and vibration interference to generate the dynamic clamping force curve. The specific features include: A distributed pressure sensor array is placed inside the pin and at key locations on the chuck to collect force signals from multiple nodes in real time. Construct mathematical models of temperature drift and vibration interference, and separate environmental interference components from the original force signal through decoupling algorithms; The denoised force signal is subjected to time series analysis to generate a dynamic curve of the clamping force changing with time, and the location and intensity of abnormal fluctuation events are marked.

5. The intelligent detection and early warning method for chuck pins according to claim 1 is characterized in that: The three-dimensional gap distribution data, surface wear characteristic data and dynamic clamping force curve are integrated, and a comprehensive state matrix is ​​constructed through a data fusion algorithm. The matrix is ​​then subjected to dimensionality reduction processing to extract key features, including: Standardize and preprocess the three-dimensional gap distribution data, surface wear characteristic data and clamping force timing signals to unify the data format and dimension; The principal component analysis algorithm is used to fuse multi-source data to construct a comprehensive state matrix including gap distribution, wear characteristics and dynamic changes of clamping force; The dimension of the comprehensive state matrix is ​​reduced based on linear discriminant analysis, and key eigenvectors that are strongly correlated with wear level and failure risk are extracted.

6. The intelligent detection and early warning method for chuck pins according to claim 1 is characterized in that: The comprehensive state matrix is ​​input into a preset deep learning model, and the wear level, remaining life, and failure risk label of the chuck pin are output based on a multimodal fusion training strategy, specifically including: A multimodal fusion convolutional neural network model is constructed, where the input layer receives gap distribution data, three-dimensional wear characteristic map, and clamping force time series signal respectively; Load pre-trained mechanical parts wear prediction model parameters through transfer learning to optimize the network's ability to recognize unknown wear patterns; The output layer generates a visual warning map containing the coordinates of high-risk areas and risk levels, and associates the remaining life prediction value and failure risk probability label.

7. The intelligent detection and early warning method for chuck pins according to claim 1 is characterized in that: Based on the wear level, remaining life, and failure risk label, the piezoelectric driver calculates the nanometer-level displacement compensation to correct the position deviation, and the hydraulic system pressure is adjusted through closed-loop PID control to stabilize the clamping force output. Specifically, the following steps are performed: The nanometer-level displacement compensation of the piezoelectric actuator is calculated based on the position error, and a real-time control signal is generated to drive the piezoelectric element to adjust the pin position. Dynamically adjust the hydraulic system pressure through the PID controller, and adjust the proportional, integral and differential parameters according to the clamping force deviation value to stabilize the output; The displacement adjustment and pressure change data during the compensation process are recorded and fed back to the deep learning model to optimize the prediction accuracy and compensation strategy.

8. The intelligent detection and early warning method for chuck pins according to claim 1 is characterized in that: A digital twin model of the chuck pin is established to simulate stress distribution and wear evolution based on real-time detection data. Compensation strategies are verified using historical failure data, and maintenance plans are automatically implemented when the remaining life falls below a threshold. Specifically, the model includes: Build a finite element digital twin model of the pin and synchronize 3D gap distribution, surface wear characteristics, and clamping force data to the model in real time; Injecting historical fault data into the finite element digital twin model, simulating stress distribution and wear evolution trends under different working conditions, and evaluating the effectiveness of the compensation strategy; When the finite element digital twin model predicts that the remaining life is lower than the preset threshold, the maintenance system is triggered to automatically generate a spare parts replacement work order and push it to the maintenance terminal.

9. A chuck pin intelligent detection and early warning device, characterized in that: The device comprises: at least one processor; and, a memory communicatively coupled to the at least one processor; The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to: The inner wall of the chuck pin hole is scanned in all directions using a laser interferometer to obtain three-dimensional gap distribution data on the inner wall of the hole; The surface of the chuck pin is imaged using a microscopic imaging device using multispectral illumination and confocal microscopy to extract surface wear characteristic data. The dynamic clamping force curve of the chuck pin clamping process is collected in real time based on the pressure sensing network, and the temperature drift and vibration interference are separated by the decoupling algorithm to generate the dynamic clamping force curve; Integrating the three-dimensional gap distribution data, surface wear characteristic data and dynamic clamping force curve, constructing a comprehensive state matrix through a data fusion algorithm, and performing dimensionality reduction processing on the matrix to extract key features; Inputting the comprehensive state matrix into a preset deep learning model, the wear level, remaining life, and failure risk label of the chuck pin are output based on a multimodal fusion training strategy; wherein the deep learning model optimizes the generalization ability of unknown wear patterns through transfer learning; Based on the wear level, remaining life, and failure risk label, the piezoelectric driver calculates the nanometer-level displacement compensation to correct the position deviation, and the hydraulic system pressure is adjusted through closed-loop PID control to stabilize the clamping force output; A digital twin model of the chuck pin is established to simulate stress distribution and wear evolution based on real-time detection data. The compensation strategy is verified through historical failure data, and a maintenance plan is automatically pushed when the remaining life falls below the threshold.

10. A non-volatile computer storage medium for intelligent detection and early warning of a chuck pin, storing computer-executable instructions, characterized in that: The computer executable instructions are configured to: The inner wall of the chuck pin hole is scanned in all directions using a laser interferometer to obtain three-dimensional gap distribution data on the inner wall of the hole; The surface of the chuck pin is imaged using a microscopic imaging device using multispectral illumination and confocal microscopy to extract surface wear characteristic data. The dynamic clamping force curve of the chuck pin clamping process is collected in real time based on the pressure sensing network, and the temperature drift and vibration interference are separated by the decoupling algorithm to generate the dynamic clamping force curve; Integrating the three-dimensional gap distribution data, surface wear characteristic data and dynamic clamping force curve, constructing a comprehensive state matrix through a data fusion algorithm, and performing dimensionality reduction processing on the matrix to extract key features; Inputting the comprehensive state matrix into a preset deep learning model, the wear level, remaining life, and failure risk label of the chuck pin are output based on a multimodal fusion training strategy; wherein the deep learning model optimizes the generalization ability of unknown wear patterns through transfer learning; Based on the wear level, remaining life, and failure risk label, the piezoelectric driver calculates the nanometer-level displacement compensation to correct the position deviation, and the hydraulic system pressure is adjusted through closed-loop PID control to stabilize the clamping force output; A digital twin model of the chuck pin is established to simulate stress distribution and wear evolution based on real-time detection data. The compensation strategy is verified through historical failure data, and a maintenance plan is automatically pushed when the remaining life falls below the threshold.

Images