Numerical control machining equipment state monitoring system based on big data analysis

Abstract

The invention discloses a numerical control machining equipment state monitoring system based on big data analysis, and relates to the technical field of intelligent manufacturing. The method is used for solving the problems of correlation analysis and real-time feedback between tool wear and machining quality in the machining process. A time sequence correlation feature vector is extracted through a vibration signal, an acoustic emission signal and main shaft axial micro-displacement data, a machining quality degradation index is calculated in combination with a nonlinear regression model, and the machining quality is monitored in real time. And on the basis of adaptive feature weight adjustment of the types of the processing materials, feature fusion during processing of different materials is enhanced, and the precision of quality evaluation is improved. And establishing a nonlinear mapping relation between the tool wear and the surface roughness by utilizing a gradient lifting tree model, and realizing accurate prediction of the tool wear and the machining quality. And finally, through a closed-loop dynamic adjustment mechanism, the cutting speed and the feeding amount are automatically adjusted according to the machining quality degradation index and the fault positioning result, and the stability of the machining process and the product quality are ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent manufacturing, and specifically to a numerical control processing equipment status monitoring system based on big data analysis. Background Technique

[0002] With the continuous development of the manufacturing industry, numerical control processing equipment plays an increasingly important role in modern production. With the increasing complexity of manufacturing requirements and the improvement of high-precision requirements, the stability, efficiency, and processing quality of numerical control processing equipment directly affect production efficiency and product quality. Therefore, real-time monitoring and predictive maintenance of the status of numerical control processing equipment can not only improve the operational reliability of the equipment, but also reduce equipment failure rates, reduce production downtime, and improve overall production efficiency. Combining big data analysis and intelligent monitoring technology, through real-time monitoring and analysis of the operating status of numerical control equipment, it can provide a scientific basis for equipment maintenance decisions and further promote the development of intelligent manufacturing.

[0003] Currently, the monitoring systems of most numerical control processing equipment still rely on traditional monitoring means, usually performing fault diagnosis through single-sensor data acquisition and static threshold judgment. However, these methods have obvious deficiencies. First, traditional methods usually only rely on a single type of data, such as vibration signals or temperature signals, and cannot comprehensively reflect the complexity of the equipment operating status. Second, these systems often fail to consider the changes in various working conditions during the processing process, resulting in poor data fusion and analysis effects. In addition, due to the installation position and type limitations of sensors, the existing systems have a low prediction accuracy of the equipment health status and are difficult to adapt to the changing processing environment. Moreover, the existing methods lack in-depth analysis of the relationship between tool wear and processing quality, and fail to effectively establish an accurate mapping relationship between tool wear and quality degradation, affecting the accuracy of equipment maintenance and processing quality optimization. Therefore, there is an urgent need for a technical solution that integrates multi-modal perception, big data analysis, and deep learning to improve the monitoring effect of numerical control processing equipment and overcome the limitations of existing methods. Summary of the Invention

[0004] Aiming at the deficiencies of the prior art, the present invention provides a numerical control processing equipment status monitoring system based on big data analysis, which solves the problems in the above background technique.

[0005] To achieve the above objectives, the present invention is realized through the following technical solutions: A state monitoring system for CNC machining equipment based on big data analysis, including the following modules: a multi-modal perception fusion module, a latent machining quality correlation analysis module, a small-sample federated residual graph network module, and a machining parameter execution module; the multi-modal perception fusion module is used to collect vibration, temperature, acoustic emission signals and spindle axial micro-displacement data of the CNC equipment in real time, combine the working condition labels obtained by G-code parsing, construct a dynamic weight matrix according to the machining stage recognition result, perform adaptive fusion processing on multi-source heterogeneous data, and output a multi-dimensional time series feature vector representing the equipment operation state; the latent machining quality correlation analysis module is used to receive the multi-dimensional time series feature vector and the workpiece surface roughness and dimension data detected online, extract the feature sub-bands highly related to the roughness in the acoustic emission signal through the time-frequency domain attention mechanism, combine the non-linear fusion strategy to perform two-channel time series feature alignment on the non-stationary features of the vibration signal, the high-frequency pulse features of the acoustic emission, and the thermal deformation trend quantity in the spindle axial micro-displacement, construct a time series mapping model of the machining state and quality degradation, and then generate a machining quality degradation index and establish a non-linear mapping relationship between tool wear and machining quality; the small-sample federated residual graph network module is used to construct a graph structure according to the physical connection relationship of the equipment components, train the residual parameters of multiple factories in cooperation according to the federated learning framework, initialize the graph edge weights in combination with the spindle-bearing stiffness matrix, and output the equipment remaining life prediction and latent fault location results; the machining parameter execution module is used to receive the machining quality degradation index and the fault location result, dynamically adjust the cutting speed and feed rate parameters through a feedback closed-loop, and generate a preventive maintenance work order.

[0006] Further, the specific process of collecting vibration, temperature, acoustic emission signals and spindle axial micro-displacement data of the CNC equipment, combining the working condition labels obtained by G-code parsing, and constructing a dynamic weight matrix according to the machining stage recognition result is as follows: Parse the G-code of the CNC equipment, extract process parameters such as cutting speed and feed rate, and identify the current machining stage as rough machining, semi-finishing or finishing; and according to the different machining stages, adjust the weights of the vibration signal, acoustic emission signal and spindle axial micro-displacement data respectively; in the finishing stage, by increasing the weight of the vibration signal, enhance the sensitivity to minute vibrations; at the end of the tool change cycle, by enhancing the weight of the acoustic emission signal, improve the ability to capture high-frequency pulses; in the continuous machining mode, by gradually increasing the weight of the spindle axial micro-displacement data, to capture the subtle displacement changes caused by thermal deformation.

[0007] Furthermore, the specific process of adaptively fusing multi-source heterogeneous data and outputting a multi-dimensional time-series feature vector representing the operating state of the device is as follows: perform wavelet packet transform on the vibration signal to extract non-stationary features, perform peak counting on the acoustic emission signal to capture high-frequency pulse features, and perform integral operation on the spindle axial micro-displacement data to quantify the thermal deformation trend; perform weighted fusion of the dynamic weight matrix with the non-stationary features of the vibration signal, the high-frequency pulse features of the acoustic emission, and the thermal deformation trend in the spindle axial micro-displacement to generate a multi-dimensional time-series feature vector containing time-domain statistics, frequency-domain energy distribution, and trend cumulative quantity.

[0008] Furthermore, the specific process of extracting the feature sub-bands highly correlated with roughness from the acoustic emission signal through the time-frequency domain attention mechanism and performing two-channel time-series feature alignment on the non-stationary features of the vibration signal, the high-frequency pulse features of the acoustic emission, and the thermal deformation trend in the spindle axial micro-displacement by combining the non-linear fusion strategy is as follows: perform short-time Fourier transform on the acoustic emission signal to generate a time-frequency diagram, divide it into multiple frequency sub-bands, and through the attention weight calculation module, screen the sub-bands significantly related to the change in the online detected surface roughness, and focus on the frequency band with dense burst pulses in the high-frequency band; use wavelet packet decomposition to extract the energy distribution of the non-stationary frequency band related to bearing wear in the vibration signal, and calculate the energy entropy as the vibration feature degradation index; based on the time-series integral operation of the micro-displacement data, quantify the cumulative effect of spindle thermal deformation, and dynamically correct the thermal deformation trend by combining the environmental temperature change rate; construct a two-channel LSTM network, with the first channel inputting the non-stationary features of the vibration and the thermal deformation trend, and the second channel inputting the energy distribution of the acoustic emission feature sub-bands; align the high-frequency device state data and the low-frequency quality detection data through the sliding window algorithm, and achieve cross-rate feature fusion through timestamp matching, and output the time-series correlation feature vector of the processing state and quality degradation.

[0009] Furthermore, generating the machining quality degradation index and establishing the non-linear mapping relationship between tool wear and machining quality includes the following steps: input the time-series correlation feature vector into the non-linear regression model, fuse the acoustic emission high-frequency pulse density, vibration energy entropy, and thermal deformation cumulative quantity, and calculate the machining quality degradation index; adaptively adjust the feature weights according to the machining material type, strengthening the vibration and thermal deformation features during metal machining, and strengthening the acoustic emission features during composite material machining; establish the corresponding relationship between different tool wear stages and the acoustic emission high-frequency pulse features according to the tool wear experiment data; fit the non-linear influence of the tool wear degree on the surface roughness through the gradient boosting tree, quantify the wear amount-roughness transfer coefficient, and establish the non-linear mapping relationship between tool wear and machining quality.

[0010] Furthermore, the specific process of constructing a graph structure according to the physical connection relationship of device components is as follows: abstract the spindle, bearings, guide rails, and cutting tools as graph nodes, and define the mechanical transmission path between nodes as graph edges according to the mechanical assembly relationship; initialize the graph edge weights using the spindle-bearing stiffness matrix to characterize the propagation intensity of force and thermal loads between components.

[0011] Furthermore, the specific process of training the residual parameters of multiple factories collaboratively according to the federated learning framework, initializing the graph edge weights in combination with the spindle-bearing stiffness matrix, and outputting the remaining life prediction and hidden fault location results of the device is as follows: each factory locally trains a residual graph convolutional network, calculates the parameter difference between the local model and the global model as the residual amount, and only uploads the residual parameters to the cloud for weighted aggregation to generate a global federated model and distribute it to each edge node; during the federated training process, constrain the adjustment range of the graph edge weights according to the spindle-bearing stiffness matrix to avoid violating the physical logic caused by pure data-driven; define the fault propagation path for insufficient bearing preload and guide rail lubrication failure, and locate abnormal nodes through the graph attention mechanism; based on the vibration energy entropy and thermal deformation trend of the graph nodes, evaluate the health of the device components, infer the remaining life through a time series prediction model, and output the fault location result.

[0012] Furthermore, the specific process of receiving the machining quality degradation index and the fault location result and dynamically adjusting the cutting speed and feed rate parameters through a feedback closed-loop is as follows: when the machining quality degradation index breaks through the primary threshold, linearly reduce the cutting speed according to the exponential increase ratio and reduce the feed rate to suppress vibration; if the hidden fault location result shows insufficient bearing preload, dynamically adjust the speed according to the spindle load current to avoid the resonance frequency range.

[0013] Furthermore, the specific process of generating a preventive maintenance work order is as follows: divide the early warning interval according to the quality degradation index, and define the rising rate threshold and absolute value threshold of the index; when the change rate of the quality degradation index exceeds the rising rate threshold or the quality degradation index exceeds the absolute value threshold, trigger a shutdown instruction and push a tool replacement work order.

[0014] The present invention has the following beneficial effects:

[0015] (1) The numerical control machining equipment status monitoring system based on big data analysis can comprehensively monitor the operating status of the numerical control equipment through the real-time data collection of the multi-modal perception fusion module and the accurate analysis of the hidden machining quality correlation analysis module. The adaptive fusion processing of multi-source heterogeneous data makes the characterization of the equipment status more accurate, can timely capture the tiny fault signs of the equipment, and thus effectively improve the fault early warning ability of the equipment. By extracting features highly related to the machining quality and combining with the time-frequency domain attention mechanism, it can accurately evaluate the machining quality degradation situation and provide real-time quality control and decision support for the production process.

[0016] (2) The CNC machining equipment status monitoring system based on big data analysis. The small-sample federated residual graph network module combines the physical connection relationships of the equipment and realizes cross-factory collaboration through the federated learning framework. It can accurately predict the remaining life of the equipment and locate potential hidden faults. The advantage of this module is that it can perform efficient model training and optimization without exposing the local data of the equipment, effectively protecting industrial data privacy. In addition, the machining parameter execution module is based on a feedback closed-loop control mechanism and can dynamically adjust machining parameters (such as cutting speed, feed rate) according to the real-time monitoring results, thereby optimizing machining accuracy and extending the service life of the equipment. By generating preventive maintenance work orders, potential risks can be eliminated in advance, the unplanned downtime of the equipment can be reduced, and production efficiency can be improved.

[0017] Of course, it is not necessary for any product implementing the present invention to achieve all the above-mentioned advantages simultaneously. Description of the Drawings

[0018] Figure 1 It is a flow chart of the CNC machining equipment status monitoring system based on big data analysis of the present invention.

[0019] Figure 2 It is a flow chart of the machining quality degradation index of the present invention. Detailed Embodiment

[0020] The embodiment of the present application solves the problems of equipment fault prediction, tool wear monitoring, and machining quality control in the CNC machining process through a CNC machining equipment status monitoring system based on big data analysis. The system collects various sensor data of the machining equipment in real time, including vibration signals, acoustic emission signals, spindle speed, temperature, load current, etc., and uses big data analysis technology to process and mine these data, thereby realizing precise monitoring and fault warning of the equipment status. Therefore, the embodiment of the present application can provide scientific and effective equipment monitoring, fault prediction, and machining quality control solutions during the CNC machining process, greatly improving production efficiency and equipment reliability.

[0021] The general idea of the solution in the embodiment of the present application is as follows:

[0022] Collect the vibration, temperature, acoustic emission signals, and spindle axial micro-displacement data of the CNC equipment in real time, combine the working condition labels obtained by G-code parsing, construct a dynamic weight matrix according to the machining stage recognition results, perform adaptive fusion processing on multi-source heterogeneous data, and output a multi-dimensional time series feature vector representing the equipment operation status.

[0023] Receive multi-dimensional time series feature vectors, as well as the measured surface roughness and dimensional data of the workpiece during online inspection. Extract the feature sub-bands highly related to roughness from the acoustic emission signals through the time-frequency domain attention mechanism. Combine the non-linear fusion strategy to perform two-channel time series feature alignment on the non-stationary features of the vibration signals, the high-frequency pulse features of the acoustic emission, and the thermal deformation trend in the axial micro-displacement of the spindle. Construct a time series mapping model of the machining state and quality degradation, and then generate a machining quality degradation index and establish a non-linear mapping relationship between tool wear and machining quality.

[0024] Construct a graph structure based on the physical connection relationships of the equipment components. Train the residual parameters of multiple factories collaboratively according to the federated learning framework. Initialize the graph edge weights in combination with the spindle-bearing stiffness matrix, and output the prediction of the remaining equipment life and the location result of latent faults.

[0025] Receive the machining quality degradation index and the fault location result, dynamically adjust the cutting speed and feed rate parameters through a feedback closed-loop, and generate a preventive maintenance work order.

[0026] Please refer to Figure 1 As shown in [reference], an embodiment of the present invention provides a technical solution: a numerical control machining equipment status monitoring system based on big data analysis, including the following modules: a multi-modal perception fusion module, a latent machining quality correlation analysis module, a small sample federated residual graph network module, and a machining parameter execution module; the multi-modal perception fusion module is used to collect vibration, temperature, acoustic emission signals and axial micro-displacement data of the spindle of the numerical control equipment in real time, combine the working condition labels obtained by G-code parsing, construct a dynamic weight matrix according to the machining stage recognition result, perform adaptive fusion processing on multi-source heterogeneous data, and output a multi-dimensional time series feature vector representing the equipment operation state; the latent machining quality correlation analysis module is used to receive the multi-dimensional time series feature vector, as well as the measured surface roughness and dimensional data of the workpiece during online inspection. Extract the feature sub-bands highly related to roughness from the acoustic emission signals through the time-frequency domain attention mechanism. Combine the non-linear fusion strategy to perform two-channel time series feature alignment on the non-stationary features of the vibration signals, the high-frequency pulse features of the acoustic emission, and the thermal deformation trend in the axial micro-displacement of the spindle. Construct a time series mapping model of the machining state and quality degradation, and then generate a machining quality degradation index and establish a non-linear mapping relationship between tool wear and machining quality; the small sample federated residual graph network module is used to construct a graph structure based on the physical connection relationships of the equipment components. Train the residual parameters of multiple factories collaboratively according to the federated learning framework. Initialize the graph edge weights in combination with the spindle-bearing stiffness matrix, and output the prediction of the remaining equipment life and the location result of latent faults; the machining parameter execution module is used to receive the machining quality degradation index and the fault location result, dynamically adjust the cutting speed and feed rate parameters through a feedback closed-loop, and generate a preventive maintenance work order.

[0027] In this implementation plan, the functions of the multi-modal perception fusion module: This module collects multiple signals of CNC machining equipment in real time, including vibration, temperature, acoustic emission signals (AE), and spindle axial micro-displacement data. Vibration and temperature reflect the mechanical state and temperature changes of the equipment. Acoustic emission signals provide clues about material or tool wear, while spindle axial micro-displacement data helps monitor the accuracy and stability of the equipment. G-code parsing: G-code is the instruction code of CNC machine tools, which provides information about machining conditions (such as cutting speed, feed rate, etc.). Combining this information, the system can identify different machining stages and construct a dynamic weight matrix according to the stage identification results to adjust the data fusion strategy. Adaptive fusion: The key of this module is to adaptively fuse heterogeneous data from different sources and of different types (such as vibration, temperature, etc.) to output a comprehensive and multi-dimensional time series feature vector. These feature vectors accurately reflect the operating state of the equipment and can provide basic data for subsequent analysis. The function of the implicit machining quality correlation analysis module: This module is used to analyze the machining quality of CNC equipment. The input data includes multi-dimensional time series feature vectors (output by the first module) and the workpiece surface roughness and dimension data detected online. Time-frequency domain attention mechanism: The time-frequency domain attention mechanism is a technology that allows the system to dynamically focus on important feature sub-bands in the time-frequency domain of the signal. For example, through this mechanism, the system can identify the acoustic emission signal features most relevant to the change in workpiece surface roughness. Non-linear fusion strategy: This strategy combines the complex relationships between different signals (such as the non-stationary characteristics of vibration signals, the high-frequency pulse characteristics of acoustic emission, and the thermal deformation trend of spindle micro-displacement) to ensure that the features of multiple signals can be effectively aligned and fused. These fused features are used to construct a time series mapping model to reflect the relationship between the machining state and quality degradation, and finally generate a machining quality degradation index. Tool wear and quality mapping relationship: Through this analysis, the impact of tool wear on machining quality can be identified and a non-linear mapping relationship between the two can be established, which is of great significance for predicting and managing machining quality degradation in advance. The function of the small sample federated residual graph network module: The core of this module is to use the physical connection relationship of equipment components to construct a graph structure. Equipment components (such as spindles, bearings, etc.) are connected by graph edges to form a complex network, reflecting the interaction between components. Federated learning framework: Federated learning is a distributed machine learning technology that allows multiple factories or devices to share models without exchanging data and jointly train a global model. This framework is particularly suitable for data privacy protection to ensure that the local data of devices and factories will not be leaked. Residual parameter training: By training residual parameters, the system can capture the abnormal behavior or hidden faults of the equipment. These parameters reflect the "residual" part of the equipment during operation, that is, the deviation from the expected normal state.Spindle - Bearing Stiffness Matrix Initialization: The spindle and bearings are key moving components in a CNC machine tool, and their stiffness relationship affects the accuracy and lifespan of the equipment. The initialization of the stiffness matrix helps assign appropriate weights to the edges of the graph structure, thereby improving the accuracy of fault location. Output: This module outputs the remaining lifespan prediction and latent fault location results of the equipment, which helps predict the future performance of the equipment, locate potential faults, and reduce unexpected downtimes. Machining Parameter Execution Module Function: This module receives the machining quality degradation index and fault location results from the latent machining quality correlation analysis module and feeds them back to the machining process in real - time. The system dynamically adjusts machining parameters such as cutting speed and feed rate based on these feedbacks to optimize the machining process. Closed - loop Control: Through the feedback closed - loop mechanism, the system can adjust parameters in real - time during the machining process, ensuring machining quality and reducing quality fluctuations caused by equipment failures or improper operations. Preventive Maintenance: Through fault prediction and quality degradation analysis, the system can generate preventive maintenance work orders, conduct equipment inspections and maintenance in advance, thereby reducing unplanned downtime and extending the equipment lifespan. Multi - modal Sensing Fusion: Combining data from different types of sensors (such as vibration, temperature, acoustic emission, etc.) for fusion provides more comprehensive and accurate equipment status monitoring. Time - frequency Domain Attention Mechanism: During the signal processing process, the system can focus on the most important frequency or time - domain features in the signal, improving the accuracy of analysis. Non - linear Fusion Strategy: When fusing different signal features, instead of using a simple linear weighting method, the complex non - linear relationship between signals is considered. Federated Learning Framework: A distributed machine learning method that supports multi - party collaboration without exchanging local data, protecting data privacy. Residual Graph Network: Capturing abnormal features between components of the equipment through a graph - structured network model helps identify potential faults.

[0028] Specifically, the specific process of collecting vibration, temperature, acoustic emission signals and spindle axial micro - displacement data of the CNC equipment, combining with the working condition labels obtained by G - code parsing, and constructing a dynamic weight matrix according to the machining stage recognition result is as follows: Parse the G - code of the CNC equipment, extract process parameters such as cutting speed and feed rate, and identify the current machining stage as rough machining, semi - finishing machining or finishing machining; and according to the different machining stages, adjust the weights of vibration signals, acoustic emission signals and spindle axial micro - displacement data respectively; in the finishing machining stage, by increasing the weight of vibration signals, enhance the sensitivity to micro - vibrations; at the end of the tool change cycle, by enhancing the weight of acoustic emission signals, improve the ability to capture high - frequency pulses; in the continuous machining mode, by gradually increasing the weight of spindle axial micro - displacement data to capture the subtle displacement changes caused by thermal deformation.

[0029] In this implementation scheme, G code parsing: G code (or CNC code) is a set of instructions for controlling the machining process of CNC machine tools. Each instruction corresponds to a specific operation in the machining process, such as cutting speed, feed rate, tool path, etc. In this step, the G code must first be parsed to extract process parameters such as cutting speed, feed rate, tool diameter, etc. These parameters determine the specific requirements of the machining process and provide basic data for the subsequent construction of the dynamic weight matrix. Processing stage identification: According to the parameters extracted from the G code (such as cutting speed, feed rate, etc.), the system determines whether the current CNC machining stage is rough machining, semi-finishing machining or finishing machining. The requirements and characteristics of each machining stage are different: Rough machining stage: usually large cutting volume, high cutting speed, mainly removing materials. Semi-finishing stage: moderate cutting volume, mainly contour finishing, high precision requirements. Finishing stage: small cutting volume, low cutting speed, mainly fine machining, high precision and low surface roughness are required. Different machining stages have different signal characteristics, so it is necessary to adjust the data weight according to the characteristics of different stages during the signal fusion process. Weight adjustment: Based on the different processing stages, the system needs to adjust the weight of each signal source to ensure that it can sensitively capture the slight changes in the processing state. Finishing stage: Since finishing requires higher processing accuracy, the system needs to increase the weight of the vibration signal at this stage. This is because the vibration signal can help monitor tiny mechanical vibrations. The finishing stage is very sensitive to these tiny changes. The enhancement of the vibration signal can help the system detect fine processing errors. At the end of the tool change cycle: At the end of the tool change cycle, tool wear may cause problems with the surface quality of the workpiece. Therefore, it is important to enhance the weight of the acoustic emission signal. The acoustic emission signal can reflect the high-frequency pulses generated during the contact between the tool and the workpiece. Increasing its weight helps to better capture the abnormalities caused by tool wear or breakage. Continuous processing mode: In the continuous processing mode, the temperature and mechanical vibration of the spindle may cause thermal deformation due to the continuous processing process. In order to monitor the displacement changes caused by thermal deformation, it is necessary to gradually increase the weight of the spindle axial micro-displacement data. These tiny axial displacement changes may directly affect the processing accuracy, so weighting this signal in this mode helps to capture the impact of thermal deformation. Dynamic weight matrix: This matrix dynamically adjusts the weight of each signal according to different processing stages and real-time collected signal data. Through the working condition labels and processing stage information obtained by G-code analysis, the system can generate a dynamic weight matrix, which reflects the importance of each signal (vibration, acoustic emission, micro-displacement) in the current processing stage in real time. The construction of the weight matrix enables the system to intelligently adjust the data fusion strategy according to the processing stage and signal characteristics, thereby improving the accuracy and sensitivity of equipment status monitoring.

[0030] Specifically, the specific process of adaptively fusing multi-source heterogeneous data and outputting a multi-dimensional time-series feature vector characterizing the device operation state is as follows: perform wavelet packet transform on the vibration signal to extract non-stationary features, perform peak counting on the acoustic emission signal to capture high-frequency pulse features, and perform integral operation on the spindle axial micro-displacement data to quantify the thermal deformation trend; perform weighted fusion of the dynamic weight matrix with the non-stationary features of the vibration signal, the high-frequency pulse features of the acoustic emission, and the thermal deformation trend in the spindle axial micro-displacement to generate a multi-dimensional time-series feature vector containing time-domain statistics, frequency-domain energy distribution, and trend cumulative quantity.

[0031] In this implementation, wavelet packet transform is used to analyze non-stationary signals and can effectively extract time-frequency local features from vibration signals. Signal decomposition: Decompose the vibration signal into multiple frequency bands through wavelet packet transform to represent the signal features of different frequency bands respectively. Extract non-stationary features: Perform time-domain analysis on each decomposed frequency band and calculate features such as root mean square value (RMS), peak value, and mean value to capture the non-stationarity of the vibration signal. Formula representation: Among them: x(t) is the vibration signal, and T is the total time of the signal. RMS is the root mean square value of the signal, which is used to describe the energy of the signal. Acoustic emission signals usually reflect high-frequency vibrations and pulse activities during the machining process, especially during tool wear or failure, which are manifested as high-frequency pulses. Peak counting: Analyze the high-frequency part of the acoustic emission signal and count the number of peak values exceeding a certain threshold in the signal. These peak values correspond to key events during the machining process, such as tool contact and wear. High-frequency pulse features: Capture the rapidly changing features in the signal through peak counting to understand the abnormalities and problems during the machining process. Formula representation: Among them: x(t) is the acoustic emission signal, and N is the number of signal sampling points. θ is the set threshold, defined as the minimum value of the peak, and values exceeding this value are considered valid pulses. 1 is the indicator function, which is 1 when |x i |>θ and 0 otherwise. The spindle axial micro-displacement data reflects the axial displacement caused by thermal deformation during the cutting process. This data usually needs to be integrated to quantify the thermal deformation trend. Integral operation: Integrate the spindle axial micro-displacement to obtain the cumulative effect of thermal deformation. The integral operation can reflect the trend of small displacement changes caused by temperature rise and thermal expansion. Formula representation: Among them: v axial (t) is the time-series data of the spindle axial micro-displacement. D thermalIt is the amount of thermal deformation trend, representing the cumulative change of the spindle axial micro-displacement. The non-stationary characteristics of the vibration signal, the high-frequency pulse characteristics of the acoustic emission signal, and the thermal deformation trend amount of the spindle axial micro-displacement are weighted and fused. The weighted fusion adjusts the contribution of different characteristics to the final fusion result through a dynamic weight matrix. Dynamic weight matrix: Generated from the G-code parsing and machining stage recognition results mentioned above, it dynamically adjusts the weights of each signal characteristic to adapt to different machining stages. Weighted fusion: The characteristics of each signal are weighted according to their weights, and finally a multi-dimensional time series feature vector containing time-domain statistics, frequency-domain energy distribution, and trend accumulation amount is obtained. Multi-dimensional time series feature vector: Through the weighted fused characteristics, the system finally generates a multi-dimensional time series feature vector containing various characteristics (such as time-domain statistics, frequency-domain energy distribution, trend accumulation amount), characterizing the operating state of the device. Time-domain statistics: Include mean, variance, peak value, root mean square, etc., reflecting the basic characteristics of the signal. Frequency-domain energy distribution: Reflects the energy distribution of the signal in the frequency domain and is used to identify the signal characteristics of different frequency components. Trend accumulation amount: Such as the thermal deformation trend amount, representing the cumulative effect of the signal over time. This feature vector can comprehensively reflect the operating state of the CNC device under different working conditions and provide a basis for equipment fault prediction and maintenance decision-making.

[0032] Specifically, the specific process of two-channel time series feature alignment for the non-stationary characteristics of the vibration signal, the high-frequency pulse characteristics of the acoustic emission, and the thermal deformation trend amount in the spindle axial micro-displacement by extracting the feature sub-bands highly correlated with roughness from the acoustic emission signal through the time-frequency domain attention mechanism and combining the non-linear fusion strategy is as follows: Perform short-time Fourier transform on the acoustic emission signal to generate a time-frequency diagram, divide it into multiple frequency sub-bands, and through the attention weight calculation module, screen the sub-bands significantly correlated with the online detected surface roughness change, focusing on the frequency band with dense burst pulses in the high-frequency band; Use wavelet packet decomposition to extract the non-stationary frequency band energy distribution related to bearing wear in the vibration signal, and calculate the energy entropy as the vibration feature degradation index; Based on the time series integration operation of the micro-displacement data, quantify the cumulative effect of the spindle thermal deformation, and dynamically correct the thermal deformation trend amount in combination with the environmental temperature change rate; Construct a two-channel LSTM network, with the first channel inputting the non-stationary characteristics of the vibration and the thermal deformation trend amount, and the second channel inputting the energy distribution of the acoustic emission feature sub-bands; Align the high-frequency equipment state data and the low-frequency quality inspection data through the sliding window algorithm, and achieve cross-rate feature fusion through timestamp matching, and output the time series correlation feature vector of the machining state and quality degradation.

[0033] In this implementation scheme, short-time Fourier transform (STFT): First, perform short-time Fourier transform on the acoustic emission signal to generate a time-frequency diagram. The short-time Fourier transform can combine the information of the signal in the time domain and the frequency domain, enabling us to observe the energy distribution of each frequency band at different time points. The formula is expressed as: Among them: x(τ) is the original acoustic emission signal. h(t - τ) is the window function, which controls the locality in the time domain. X(t, f) is the time-frequency diagram, representing the distribution of the signal in the time-frequency domain. Sub-band division and attention weight calculation: The time-frequency diagram is divided into multiple frequency sub-bands. Through the attention weight calculation module, the frequency bands significantly related to the surface roughness change are screened out. Particular focus is placed on the frequency bands with burst pulses in the high-frequency band, and these frequency bands are often closely related to the changes in the machining quality (such as roughness changes). Through the attention mechanism, the attention weight a i is calculated to focus on the relevant frequency bands: Among them: f i is the eigenvalue of the i-th frequency band. a i is the attention weight of frequency band i, which determines the importance of this frequency band in the model. Extraction of non-stationary features of vibration signals, wavelet packet decomposition: The vibration signal is decomposed by wavelet packet to extract the energy distribution of non-stationary frequency bands related to bearing wear. The wavelet packet transform decomposes the signal into different frequency bands by selecting an appropriate mother wavelet and frequency bandwidth and analyzes the energy distribution of each frequency band. The formula for wavelet packet transform is: Among them: ψ(t) is the mother wavelet, α is the scale parameter, and β is the translation parameter. W ψ (α, β) is the wavelet packet coefficient, representing the distribution of the signal at different scales and positions. Energy entropy: By calculating the energy distribution of different frequency bands, the energy entropy is obtained as a degradation index of vibration characteristics. The energy entropy reflects the complexity and uncertainty of the energy distribution in the vibration signal and can effectively represent the degradation trend of the signal. The calculation formula for energy entropy is: Among them: p i is the probability distribution of the i-th energy frequency band, representing the proportion of this frequency band in the overall signal energy. N is the total number of frequency bands. Time series integration: Integral operation is performed on the spindle axial micro-displacement data to quantify the cumulative effect of thermal deformation. The thermal deformation trend quantity represents the cumulative effect of micro-displacement caused by temperature changes during the machining process. Through integration, the cumulative change of the spindle axial micro-displacement can be obtained. The formula is expressed as: Among them: v axial (t) is the instantaneous value of the spindle axial micro-displacement. D thermal (t) is the thermal deformation trend quantity, representing the cumulative effect of spindle thermal deformation. Environmental temperature correction: The thermal deformation trend quantity is dynamically corrected in combination with the environmental temperature change rate. The environmental temperature change will affect the thermal expansion and contraction of the spindle, so correction needs to be made according to the actual environmental temperature change. The correction formula is: D thermal (t) = D thermal(t) + λ·ΔT(t); where: ΔT(t) is the environmental temperature change. λ is the coefficient of the influence of temperature on thermal deformation. LSTM network: Construct a dual-channel long short-term memory (LSTM) network for processing time-series data. The first channel inputs the non-stationary features of the vibration signal and the spindle thermal deformation trend, and the second channel inputs the energy distribution of the acoustic emission feature sub-bands. The basic formula of the LSTM network is as follows: h t = f(W h h t-1 + W x x t + b); where: h t is the hidden state at the current moment. W h and W x are weight matrices, which are used to connect the hidden state of the previous moment and the current input respectively. b is the bias term, and x t is the input at the current moment. Dual-channel design: The first channel processes vibration and thermal deformation data, and the second channel processes acoustic emission features. The high-frequency device status data and low-frequency quality inspection data are aligned through the sliding window algorithm. Feature vector output: Output the time-series correlation feature vector of the processing state and quality degradation through the dual-channel LSTM network. This feature vector synthesizes signal information from different sources and is used to reflect the working state of the device and the processing quality. The finally output time-series feature vector is a high-dimensional vector, which contains the correlation information of signals such as vibration, acoustic emission, and spindle micro-displacement at different time points and can effectively reflect the health state of the device and the quality degradation during the processing.

[0034] Please refer to Figure 2 , specifically, generating the processing quality degradation index and establishing the non-linear mapping relationship between tool wear and processing quality includes the following steps: Input the time-series correlation feature vector into the non-linear regression model, fuse the acoustic emission high-frequency pulse density, vibration energy entropy, and thermal deformation cumulative amount, and calculate the processing quality degradation index; Adaptively adjust the feature weights according to the processing material type, strengthen the vibration and thermal deformation features during metal processing, and strengthen the acoustic emission features during composite material processing; According to the tool wear experiment data, establish the corresponding relationship between different tool wear stages and the acoustic emission high-frequency pulse characteristics; Fit the non-linear influence of tool wear degree on surface roughness through the gradient boosting tree, quantify the wear amount-roughness transfer coefficient, and establish the non-linear mapping relationship between tool wear and processing quality.

[0035] In this implementation, the feature vector includes: Acoustic emission high-frequency pulse density: It reflects the instantaneous pulse information during the material cutting process, represents the high-frequency noise characteristics generated during the contact process between the tool and the workpiece, and is usually closely related to the tool wear. Vibration energy entropy: It can characterize the complexity of the vibration signal, and the change trend of the vibration energy is closely related to the tool wear and the stability of the machining process. Cumulative thermal deformation: Through the integration operation of the micro-displacement data, the thermal deformation trend of the spindle or the tool is quantified. Tool wear is usually accompanied by thermal deformation, which in turn affects the machining quality. Nonlinear regression model input: The time-series correlation feature vector generated in the previous step is input into the nonlinear regression model. The goal of this regression model is to calculate the dynamic degradation index of the machining quality based on multiple features (such as acoustic emission high-frequency pulse density, vibration energy amplitude, and cumulative thermal deformation). Nonlinear regression can handle the complex nonlinear relationships between features, thus more accurately describing the degradation process of the machining quality. Formula representation: D = σ(W2·ReLU(W1·[E AE ,E Vib ,E Thermal +b1)+b2); [E AE ,E Vib ,E Thermal is the input feature vector, representing the high-frequency pulse density E AE in the acoustic emission signal, the energy entropy E Vib of the vibration signal, and the cumulative thermal deformation E Thermal respectively. W1 and W2 are the weight matrices of the model, controlling the nonlinear relationship between the features and the degradation index. b1 and b2 are the bias terms. ReLU is the activation function, used to increase the nonlinear features. σ is the output activation function (sigmoid function), and finally the machining quality degradation index D is output. Adjusting feature weights according to the machining material type: In order to enable the regression model to adapt to different types of machining materials, in metal machining and composite material machining, the influence of different signals is strengthened by adaptively adjusting the feature weights. For example, in metal machining, the vibration and thermal deformation features contribute more to the quality degradation, while in composite material machining, the high-frequency pulse feature of the acoustic emission signal may be more representative. The adaptive adjustment formula is: w feature (M) = α M ·w vibration + β M ·w AE + γ M ·w thermal ; where: w feature (M) is the feature weight under the machining material M. α M , β M , γ M are the adjustment coefficients related to the material type M, corresponding to the vibration, acoustic emission, and thermal deformation features respectively. w vibration , w AE,w thermal is the initial weight of vibration, acoustic emission, and thermal deformation characteristics. During metal processing, the weights (α M and γ M ) of vibration and thermal deformation are relatively large; during composite material processing, the weight (β M ) of the acoustic emission characteristic is relatively large. Corresponding relationship between tool wear stage and acoustic emission characteristics: Through experimental data, establish the corresponding relationship between different tool stages (such as mild wear, moderate wear, severe wear) and the high-frequency pulse characteristics (such as pulse density, frequency, etc.) of acoustic emission signals. Generally, as tool wear intensifies, the pulse density and frequency of the acoustic emission signal change, reflecting the change in the contact state between the tool and the work surface and the cutting force. The modeling in this stage usually adopts empirical formulas or interpolation methods based on experimental data. The formula is as follows: Where: P AE (W) is the high-frequency pulse density of acoustic emission under the tool wear degree W. φ a (W) is the basis function (such as interpolation function) related to the wear stage W. ξ a is the weight of each basis function, representing the contribution of the acoustic emission characteristic. Through this formula, a non-linear relationship between the wear degree and the acoustic emission signal characteristics can be established. Gradient Boosting Tree fitting: Gradient Boosting Tree (GBDT) is a powerful regression method that can effectively capture the non-linear impact of tool wear degree on the surface roughness of the machining surface. By using tool wear data (such as wear amount) and surface roughness data as the training set, use the Gradient Boosting Tree model to fit the relationship between the two. The formula is expressed as: R surface = GBT(W tool ); Where: R surface is the surface roughness. W tool is the tool wear degree. GBT is the Gradient Boosting Tree model, representing the non-linear mapping of tool wear on the surface roughness. Wear amount roughness transfer coefficient: According to the output of the Gradient Boosting Tree, quantify the transfer coefficient between the tool wear amount and the surface roughness. This coefficient reflects the influence degree of the wear amount on the surface roughness. The transfer coefficient formula is: Where: K wear is the transfer coefficient between the wear amount and the roughness. dR surface is the small change in the surface roughness. dW tool is the small change in the tool wear degree. This coefficient is used to quantify the impact of wear on the machining quality and can be used as an important parameter in subsequent quality prediction. Establishment of non-linear mapping relationship: Combining the wear roughness transfer coefficient fitted by the Gradient Boosting Tree and the machining quality degradation index D, a non-linear mapping relationship between tool wear and machining quality is established. This mapping relationship can reflect the comprehensive impact of tool wear on machining quality degradation. The non-linear mapping formula is: Q quality = g(W tool,K wear ,D); Among them: Q quality is the final processing quality index. g(·) is a non-linear mapping function, and usually a neural network or support vector regression (SVR) is used to model it.

[0036] Specifically, the specific process of constructing a graph structure according to the physical connection relationship of device components is as follows: abstract the spindle, bearing, guide rail, and tool into graph nodes, and define the mechanical transmission path between nodes as graph edges according to the mechanical assembly relationship; initialize the graph edge weights using the spindle-bearing stiffness matrix to characterize the propagation intensity of force and thermal loads between components.

[0037] In this implementation, node definition: abstract each core component of the device (such as the spindle, bearing, guide rail, tool, etc.) into a node in the graph. Each node represents a component, and their physical connection relationship is represented by edges. Edge definition: Define the mechanical transmission path between nodes as the edge of the graph according to the mechanical assembly relationship. For example, the connection between the spindle and the bearing is represented by an edge, and this edge reflects the mechanical transmission between the spindle and the bearing. This transmission path not only includes the transmission of force but may also include the influence of thermal loads. Initialization of graph edge weights: The weight of each edge is used to describe the propagation intensity of force and thermal loads between components. The spindle-bearing stiffness matrix is an important basis for initializing the edge weights. It can reflect the stiffness characteristics between the bearing and the spindle, thus providing a quantitative basis for the intensity of the mechanical transmission process. The stiffness matrix usually contains information about stiffness coefficients, indicating the displacement response between components under specific load conditions.

[0038] Specifically, the specific process of training the residual parameters of multiple factories in collaboration according to the federated learning framework, combining with the initialization of graph edge weights using the spindle-bearing stiffness matrix, and outputting the remaining life prediction and hidden fault location results of the device is as follows: Each factory locally trains a residual graph convolutional network, calculates the parameter difference between the local model and the global model as the residual amount, and only uploads the residual parameters to the cloud for weighted aggregation to generate a global federated model and distribute it to each edge node; during the federated training process, constrain the adjustment range of graph edge weights according to the spindle-bearing stiffness matrix to avoid physical logic violations caused by pure data-driven; define the fault propagation path for insufficient bearing preload and guide rail lubrication failure, and locate abnormal nodes through the graph attention mechanism; based on the vibration energy entropy and thermal deformation trend of graph nodes, evaluate the health of device components, infer the remaining life through a time series prediction model, and output the fault location result.

[0039] In this implementation plan, the federated learning framework conducts residual parameter training and local training: Each factory deploys a graph convolutional network (GCN) model locally for training. Each factory uses its own data for residual training by calculating the parameter difference (i.e., the residual amount) between the local model and the global model. This residual training method can effectively preserve the characteristics of the local model while ensuring the coordination of the global model. Upload residual parameters: Each factory only uploads the residual parameters generated during the training process to the cloud, rather than directly uploading the complete data. The advantage of this is to protect the data privacy of each factory while avoiding the bandwidth and storage burden caused by a large amount of data transmission. Global model aggregation: The cloud performs weighted aggregation on all uploaded residual parameters to generate a global federated model. This global model will contain the fusion information of the data of each factory and can reflect the production situation of all factories. When aggregating, the weighted average method is usually used to ensure that the contributions of each factory are reasonably adjusted according to their data volume. The model is sent to the edge nodes: The aggregated global model is sent to each edge node (i.e., each factory) for local prediction and fault detection. Stiffness matrix constraint: During the federated training process, to avoid violating the physical logic caused by pure data-driven (i.e., only relying on data training and ignoring the physical characteristics of the equipment), the spindle-bearing stiffness matrix is used as a constraint condition to limit the adjustment range of the graph edge weights. This ensures that when adjusting the model, it will not exceed the physical characteristic boundaries of the equipment. Significance of physical constraints: The transmission of mechanical and thermal loads is not only affected by data characteristics but also must conform to actual physical laws. Therefore, by using the stiffness matrix for constraint, the model training can be made more in line with physical reality, avoiding the model overfitting the data and losing the ability to predict the true state of the equipment. Definition of fault propagation path: For common faults of equipment components (such as insufficient bearing preload or guide rail lubrication failure), it is necessary to define the fault propagation path. By analyzing the mechanical connection and heat conduction relationship between each node, determine the propagation path of the fault from one node to other nodes. Insufficient bearing preload may cause an increase in vibration, which in turn affects the stability of the entire spindle system; guide rail lubrication failure may cause an increase in friction, which in turn affects the machining accuracy of the tool. Graph attention mechanism: Use the graph attention mechanism (GAT) to identify and locate abnormal nodes in the graph. In the graph structure, some nodes (such as the components where faults occur) may have a greater impact on the overall system. The graph attention mechanism highlights important nodes by calculating the weights and correlation degrees between each node, thereby effectively locating the source of the fault. For example, if the vibration energy of a certain node is abnormal or the thermal deformation trend is abnormal, it may affect other nodes through the propagation mechanism of the graph, indicating that this node may be the fault source. Equipment health assessment: Based on the vibration energy entropy and thermal deformation trend of the graph nodes, the health status of each component can be evaluated. The vibration energy entropy can reflect the complexity of the vibration signal and thus reflect the health status of the equipment; the thermal deformation trend reflects the cumulative damage caused by temperature changes in the equipment.Through these two metrics, the health of the equipment can be comprehensively evaluated. Remaining life prediction: Using a time series prediction model (such as LSTM) combined with the above health assessment metrics, the remaining life of the equipment is predicted. These metrics, as inputs, can dynamically predict possible failures or performance degradations of the equipment over a period of time in the future through the time series model, thus providing a reference for the maintenance and replacement of the equipment. Fault location: Through the definition of graph attention mechanism and fault propagation path, combined with equipment health assessment and remaining life prediction, the model can output fault location results, indicating which components may have latent faults or are about to fail. These results provide guidance for maintenance personnel to help them intervene before problems occur in the equipment, avoiding downtime and production interruptions.

[0040] Specifically, the specific process of receiving the machining quality degradation index and the fault location result and dynamically adjusting the cutting speed and feed rate parameters through the feedback closed loop is as follows: When the machining quality degradation index breaks through the primary threshold, the cutting speed is linearly reduced according to the exponential increase rate, and the feed rate is reduced to suppress vibration; if the latent fault location result shows insufficient bearing preload, the rotational speed is dynamically adjusted according to the spindle load current to avoid the resonance frequency range.

[0041] In this implementation plan, the specific process of dynamically adjusting the cutting speed and feed rate parameters through the feedback closed loop

[0042] After receiving the machining quality degradation index and the fault location result, the system will dynamically adjust the cutting speed and feed rate according to this information to optimize the machining process and reduce the risk of failure. The specific steps are as follows: The quality degradation index breaks through the primary threshold: When the machining quality degradation index (such as surface roughness, vibration energy, etc.) reaches the set primary threshold, it indicates an obvious downward trend in machining quality. At this time, the system will linearly reduce the cutting speed according to the ratio of the exponential increase. At the same time, to reduce vibration, the system will reduce the feed rate. This process helps to slow down the tool wear, improve machining accuracy, and avoid further quality decline. The latent fault location result shows insufficient bearing preload: When the fault location result indicates insufficient bearing preload, the system will dynamically adjust the rotational speed according to the spindle load current information. By adjusting the rotational speed to avoid entering the resonance frequency range, the vibration of the equipment is prevented from deteriorating and the safety of the equipment is protected.

[0043] Specifically, the specific process of generating a preventive maintenance work order is as follows: The early warning interval is divided according to the quality degradation index, and the threshold of the exponential rising rate and the absolute value threshold are defined; when the change rate of the quality degradation index exceeds the rising rate threshold or the quality degradation index exceeds the absolute value threshold, a shutdown instruction is triggered and a tool replacement work order is pushed.

[0044] In this implementation, according to the change of the quality degradation index, the system can automatically generate preventive maintenance work orders to avoid further damage to the equipment and ensure production stability. The specific steps are as follows: Divide the warning range for the quality degradation index: First, the system divides the quality degradation index into different warning ranges. Specifically, a rising rate threshold and an absolute value threshold are defined to trigger warnings. Rising rate threshold: When the change rate of the quality degradation index (i.e., the time change rate of the degradation index) exceeds the set rising rate threshold, the system will issue a warning signal. Absolute value threshold: When the quality degradation index itself exceeds the set absolute value threshold, the system will also trigger a warning. Trigger the shutdown instruction and push the tool replacement work order: Once the quality degradation index exceeds the warning threshold, the system will automatically trigger the shutdown instruction and simultaneously generate a tool replacement work order. This preventive maintenance helps ensure that the equipment is not irreversibly damaged due to excessive wear.

[0045] In summary, this application has at least the following effects:

[0046] The CNC machining equipment status monitoring system based on big data analysis can, through the extraction of time-series correlation feature vectors and the calculation of the machining quality degradation index, monitor the changes in tool wear and machining quality during the machining process in real time, give early warnings in a timely manner and adjust the machining parameters to ensure machining accuracy and stability. According to the machining quality degradation index and the fault location result, the system can dynamically adjust the cutting speed and feed rate, optimize the machining process, reduce vibration and improve machining stability, thereby improving machining efficiency and product quality. By generating the quality degradation index and comparing it with the warning range, the system can trigger preventive maintenance measures in a timely manner, such as shutdown instructions and tool replacement work orders, to avoid equipment failures and production stagnation and improve equipment utilization. By establishing a non-linear mapping relationship between tool wear and machining quality, the influence of tool wear on surface roughness can be accurately quantified, providing a scientific wear assessment and control scheme, thereby improving the controllability and accuracy of the production process. Using the federated learning framework and combining the data of multiple factories for collaborative training can improve the generalization ability and accuracy of the model, ensure that each factory can share and optimize the global model while making personalized adjustments, and improve the overall machining quality and efficiency of the production line.

[0047] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the present invention can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0048] The present invention is described with reference to the flowcharts and / or block diagrams of systems, apparatus (systems), and computer program products according to embodiments of the invention. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and combinations of flows and / or blocks in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing device to produce a machine, such that the instructions executed by the processor of the computer or other programmable data processing device produce means for implementing the functions specified in the flow Figure 1 one or more flows and / or blocks Figure 1 or means for implementing the functions specified in one or more blocks.

[0049] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to operate in a particular manner, such that the instructions stored in the computer-readable memory produce a manufacture including instruction means for implementing the functions specified in the flow Figure 1 one or more flows and / or blocks Figure 1 or means for implementing the functions specified in one or more blocks.

[0050] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operational steps are performed on the computer or other programmable device to produce a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in the flow Figure 1 one or more flows and / or blocks Figure 1 or means for implementing the functions specified in one or more blocks.

[0051] Although the preferred embodiments of the present invention have been described, additional changes and modifications can be made by those skilled in the art once they learn of the basic inventive concept. Therefore, the appended claims are intended to be construed to include the preferred embodiments as well as all changes and modifications that fall within the scope of the present invention.

[0052] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention is also intended to include these modifications and variations.

Claims

1. A numerical control machining equipment status monitoring system based on big data analysis, characterized in that, It includes the following modules: multimodal perception fusion module, implicit processing quality correlation analysis module, small-sample federated residual graph network module, and processing parameter execution module; The multimodal perception fusion module is used to collect vibration, temperature, acoustic emission signals and spindle axial micro-displacement data of the numerical control equipment in real time, combine the working condition labels obtained by G-code parsing, construct a dynamic weight matrix according to the processing stage recognition result, perform adaptive fusion processing on multi-source heterogeneous data, and output a multi-dimensional time series feature vector representing the equipment operation state; The implicit processing quality correlation analysis module is used to receive the multi-dimensional time series feature vector and the workpiece surface roughness and dimension data detected online, extract the feature sub-bands highly correlated with roughness in the acoustic emission signal through the time-frequency domain attention mechanism, combine the non-linear fusion strategy to perform two-channel time series feature alignment on the non-stationary features of the vibration signal, the high-frequency pulse features of the acoustic emission, and the thermal deformation trend in the spindle axial micro-displacement, construct a time series mapping model of the processing state and quality degradation, and then generate a processing quality degradation index and establish a non-linear mapping relationship between tool wear and processing quality; The small-sample federated residual graph network module is used to construct a graph structure according to the physical connection relationship of the equipment components, train the residual parameters of multiple factories in cooperation according to the federated learning framework, initialize the graph edge weights in combination with the spindle-bearing stiffness matrix, and output the equipment remaining life prediction and implicit fault location results; The processing parameter execution module is used to receive the processing quality degradation index and the fault location result, dynamically adjust the cutting speed and feed rate parameters through a feedback closed loop, and generate a preventive maintenance work order.

2. The state monitoring system for numerical control machining equipment based on big data analysis according to claim 1, wherein: The specific process of collecting vibration, temperature, acoustic emission signals and spindle axial micro-displacement data of the numerical control equipment, combining the working condition labels obtained by G-code parsing, and constructing a dynamic weight matrix according to the processing stage recognition result is as follows: Parse the G-code of the numerical control equipment, extract process parameters such as cutting speed and feed rate, and identify the current processing stage as rough machining, semi-finishing or finishing; And according to the different processing stages, adjust the weights of the vibration signal, acoustic emission signal and spindle axial micro-displacement data respectively; In the finishing stage, by increasing the weight of the vibration signal, enhance the sensitivity to micro-vibrations; At the end of the tool change cycle, by enhancing the weight of the acoustic emission signal, improve the ability to capture high-frequency pulses; In the continuous processing mode, by gradually increasing the weight of the spindle axial micro-displacement data, to capture the subtle displacement changes caused by thermal deformation.

3. The state monitoring system for CNC machining equipment based on big data analysis according to claim 2, characterized in that: The specific process of performing adaptive fusion processing on multi-source heterogeneous data and outputting a multi-dimensional time series feature vector representing the equipment operation state is as follows: Perform wavelet packet transform on the vibration signal to extract non-stationary features, perform peak counting on the acoustic emission signal to capture high-frequency pulse features, and perform integral operation on the spindle axial micro-displacement data to quantify the thermal deformation trend; Perform weighted fusion on the dynamic weight matrix and the non-stationary features of the vibration signal, the high-frequency pulse features of the acoustic emission, and the thermal deformation trend in the spindle axial micro-displacement to generate a multi-dimensional time series feature vector including time domain statistics, frequency domain energy distribution and trend cumulative amount.

4. The state monitoring system for CNC machining equipment based on big data analysis according to claim 3, wherein: Extract the feature sub - bands highly related to roughness height in the acoustic emission signal through the time - frequency domain attention mechanism. The specific process of two - channel time - series feature alignment for the non - stationary features of vibration signals, the high - frequency pulse features of acoustic emission, and the thermal deformation trend in the axial micro - displacement of the spindle is as follows: Perform short - time Fourier transform on the acoustic emission signal to generate a time - frequency map, divide it into multiple frequency sub - bands, and through the attention weight calculation module, select the sub - bands significantly related to the online detected surface roughness change, focusing on the frequency bands with dense burst pulses in the high - frequency band; Use wavelet packet decomposition to extract the energy distribution of the non - stationary frequency band related to bearing wear in the vibration signal, and calculate the energy entropy as the vibration feature degradation index; Based on the time - series integration operation of the micro - displacement data, quantify the cumulative effect of spindle thermal deformation, and dynamically correct the thermal deformation trend in combination with the environmental temperature change rate; Construct a two - channel LSTM network. The first channel inputs the non - stationary vibration features and the thermal deformation trend, and the second channel inputs the energy distribution of the acoustic emission feature sub - bands; Align the high - frequency equipment status data and the low - frequency quality inspection data through the sliding window algorithm, and achieve cross - rate feature fusion through timestamp matching, and output the time - series correlation feature vector of the processing status and quality degradation.

5. The state monitoring system for CNC machining equipment based on big data analysis according to claim 4, characterized in that: The process of generating the machining quality degradation index and establishing the non - linear mapping relationship between tool wear and machining quality includes the following steps: Input the time - series correlation feature vector into the non - linear regression model, fuse the acoustic emission high - frequency pulse density, vibration energy entropy, and thermal deformation cumulative amount, and calculate the machining quality degradation index; Adaptive adjust the feature weights according to the machining material type, strengthen the vibration and thermal deformation features during metal machining, and strengthen the acoustic emission features during composite material machining; According to the tool wear experiment data, establish the corresponding relationship between different tool wear stages and the high - frequency pulse features of acoustic emission; Fit the non - linear influence of tool wear degree on surface roughness through the gradient - boosting tree, quantify the wear amount - roughness transfer coefficient, and establish the non - linear mapping relationship between tool wear and machining quality.

6. The state monitoring system for CNC machining equipment based on big data analysis according to claim 5, wherein: The specific process of constructing a graph structure according to the physical connection relationship of equipment components is as follows: Abstract the spindle, bearing, guide rail, and tool as graph nodes, and define the mechanical transmission path between nodes as graph edges according to the mechanical assembly relationship; Initialize the graph edge weights using the spindle - bearing stiffness matrix to characterize the propagation intensity of force and thermal load between components.

7. The state monitoring system for CNC machining equipment based on big data analysis according to claim 6, characterized in that: The specific process of collaborative residual parameter training of multiple factories according to the federated learning framework, combined with initializing the graph edge weights using the spindle - bearing stiffness matrix, and outputting the equipment remaining life prediction and hidden fault location results is as follows: Each factory locally trains the residual graph convolutional network, calculates the parameter difference between the local model and the global model as the residual amount, and only uploads the residual parameters to the cloud for weighted aggregation to generate the global federated model and distribute it to each edge node; During the federated training process, constrain the adjustment range of the graph edge weights according to the spindle - bearing stiffness matrix; Define the fault propagation path for insufficient bearing pre - load and guide rail lubrication failure, and locate the abnormal nodes through the graph attention mechanism; Based on the vibration energy entropy and thermal deformation trend of graph nodes, evaluate the health of equipment components, infer the remaining life through a time series prediction model, and output the fault location result.

8. The state monitoring system for CNC machining equipment based on big data analysis according to claim 7, wherein: The specific process of receiving the machining quality degradation index and the fault location result and dynamically adjusting the cutting speed and feed rate parameters through a feedback closed-loop is as follows: When the machining quality degradation index breaks through the primary threshold, linearly reduce the cutting speed according to the exponential increase ratio, and reduce the feed rate to suppress vibration; If the hidden fault location result shows insufficient bearing preload, dynamically adjust the speed according to the spindle load current to avoid the resonance frequency range.

9. The state monitoring system for CNC machining equipment based on big data analysis according to claim 8, characterized in that: The specific process of generating a preventive maintenance work order is as follows: Divide the early warning interval according to the quality degradation index, and define the exponential rise rate threshold and the absolute value threshold; When the change rate of the quality degradation index exceeds the rise rate threshold or the quality degradation index exceeds the absolute value threshold, trigger a shutdown command and push a tool replacement work order.

Images