Machine tool multi-source sensing method and system based on dynamic characteristic coupling analysis

Abstract

The invention discloses a machine tool multi-source sensing method and system based on dynamic characteristic coupling analysis, and belongs to the technical field of machine tool monitoring. A polynomial regression model is used for evaluating and grading the accuracy of a coupling relation between node pairs, self-adaptive edge weight correction is carried out on incomplete accurate edges, and finally an optimized coupling network diagram and key network indexes are output, so that the authenticity and the stability of coupling recognition of the machine tool structure are effectively improved, and the accuracy and the reliability of the machine tool structure are improved. And a more reliable structure sensing basis is provided for subsequent operation state monitoring and fault prediction.

Description

Technical Field

[0001] The present invention relates to the technical field of machine tool monitoring, and in particular to a machine tool multi-source perception method and system based on dynamic characteristic coupling analysis. Background Art

[0002] Dynamic coupling analysis, in systems engineering or structural mechanics, is the process of studying the mutual influence and coupling between different subsystems or structural components in terms of dynamic behavior (such as vibration and response frequency). This analysis reveals how the dynamic characteristics of one component are affected by the motion or vibration of another, enabling more accurate predictions of overall system performance and optimized design.

[0003] The existing technology has the following shortcomings:

[0004] In a machine tool's dynamic coupling network built using DCCA, high correlations caused by multiple components being subject to the same external disturbances (such as foundation vibrations or temperature control systems) can be mistakenly interpreted as direct coupling. This problem stems from the fact that DCCA can only measure trend correlations and cannot distinguish between true physical coupling and pseudo-coupling caused by common driving factors. Furthermore, ignoring this issue during machine tool status perception and fault location can lead to misjudgment of coupling paths and misidentification of key components, potentially causing system compensation failures or wasted maintenance resources. Summary of the Invention

[0005] The purpose of the present invention is to provide a machine tool multi-source perception method and system based on dynamic characteristic coupling analysis to address the shortcomings of the background technology.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a multi-source perception method for machine tools based on dynamic characteristic coupling analysis, comprising:

[0007] Obtain multi-source dynamic characteristic data of key structural components of machine tools under different working conditions;

[0008] The structural components are used as network nodes, and the detrended cross-correlation coefficients between each node pair are calculated based on the multi-source dynamic characteristic data, and a weighted undirected coupled network graph is constructed with the detrended cross-correlation coefficients as edge weights;

[0009] Auxiliary characteristic parameters for evaluating the accuracy of coupling relationships are extracted from multi-source dynamic characteristic data, including delayed cointegration disturbance residuals and environmental variable driven deviation index;

[0010] Inputting the auxiliary feature parameters into a preset coupling relationship anomaly prediction model, evaluating the accuracy of the coupling relationship between the node pairs, and classifying each edge into three types: accurate, partially accurate, and inaccurate according to the evaluation results;

[0011] For node pairs whose evaluation results are not completely accurate, based on the accuracy evaluation results and the correlation correction rule, the edge weights thereof are adaptively adjusted, and the weighted coupling network is reconstructed;

[0012] The corrected coupling network diagram and its network indicators are output for subsequent machine tool operation status analysis and fault prediction.

[0013] Preferably, the key structural components of the machine tool include a spindle unit, a worktable and pallet system, a feed system, a bed and column structure, cooling and lubrication system components, and electronic control and drive components.

[0014] Preferably, the multi-source dynamic characteristic data includes vibration acceleration, displacement change, strain signal, temperature, heat flux density, motor current, voltage, spindle load, servo control error, tool cutting force, workpiece processing error and environmental variable data.

[0015] Preferably, the calculation of the detrended cross-correlation coefficient between each pair of nodes includes: converting the time series data of each node into a cumulative deviation sequence; performing polynomial trend fitting in multiple sliding windows; calculating the detrended cross covariance and normalizing it to obtain the detrended cross correlation coefficient; the detrended cross correlation coefficient is used to construct the edge weights between nodes to form a weighted undirected graph.

[0016] Preferably, the delayed cointegration disturbance residual is obtained by taking the pre-processed dynamic characteristic time series of two structural components i and j: ; Establish a cointegration regression model: ; Where τ is the delay, ϵ(t) is the residual term, α and β are the regression coefficients of the cointegration model, and the disturbance residual sequence is extracted: ; Calculate the delayed cointegration disturbance residuals , the expression is: ; N is the total number of time points.

[0017] Preferably, the method for obtaining the environmental variable driven deviation index is: input node pair time series: ; Multiple environmental variable time series: ; K is the total number of environmental variables; create the data matrix X(t): ; Input it into the PCMCI model for causal structure learning; Output the causal graph G(S,G), where: S is the variable node; G represents the significant time-delay causal connection; Determine whether two causal paths exist at the same time: ; If the same environment variable exists right The simultaneous generation of causal driving indicates the existence of a common cause driving path. For each pair of nodes (i, j), the environmental variable driving deviation index is calculated, and the expression is: ; is an indicator variable: if If it is a significant causal path, it is 1, otherwise it is 0; is an indicator variable: if If it is a significant causal path, it is 1, otherwise it is 0; is a small constant; For environment variables right The significance of causal connection; For environment variables right The significance of causal connection.

[0018] Preferably, the auxiliary feature parameters are input into a preset coupling relationship anomaly prediction model to evaluate the accuracy of the coupling relationship between the node pairs, and the edges are divided into three types: accurate, partially accurate, and inaccurate according to the evaluation results, specifically including:

[0019] The delayed cointegration disturbance residuals and the environmental variable-driven deviation index are converted into comprehensive feature vectors, which are used as inputs of the machine learning model. The machine learning model uses the accuracy score value label of the coupling relationship between node pairs predicted by each set of comprehensive feature vectors as the prediction target, and minimizes the sum of the prediction errors of the accuracy score value labels of the coupling relationships between all node pairs as the training target. The machine learning model is trained until the sum of the prediction errors reaches convergence, and the model training is stopped. The accuracy score value of the coupling relationship between node pairs is determined according to the model output results, wherein the machine learning model is a polynomial regression model.

[0020] Preferably, the obtained accuracy score value of the coupling relationship between the node pairs is compared with a gradient standard threshold, the gradient standard threshold includes a first standard threshold and a second standard threshold, and the first standard threshold is less than the second standard threshold, and the accuracy score value of the coupling relationship between the node pairs is compared with the first standard threshold and the second standard threshold respectively;

[0021] If the accuracy score value of the coupling relationship between the node pairs is greater than the second standard threshold, it means that the coupling relationship between the node pairs is stable and they are classified as accurate node pairs; if the accuracy score value of the coupling relationship between the node pairs is greater than or equal to the first standard threshold and less than or equal to the second standard threshold, it means that there is partial uncertainty in the coupling relationship and they are classified as incompletely accurate node pairs; if the accuracy score value of the coupling relationship between the node pairs is less than the first standard threshold, they are classified as inaccurate node pairs.

[0022] Preferably, for node pairs whose evaluation results are not completely accurate, based on the accuracy evaluation results and the correlation correction rules, the edge weights are adaptively adjusted, and the weighted coupling network is reconstructed, specifically including:

[0023] Extract the set of all node pairs classified as not completely accurate from the accuracy score: ; is the accuracy score of the coupling relationship between node pair i and j, is the first standard threshold, is the second standard threshold;

[0024] For each edge , extract its comprehensive feature vector: ; It and the edges adjacent to node i or j in the local neighborhood constitute a local feature sample set;

[0025] For each sample edge (m,n) in the neighborhood, calculate the Euclidean distance to the edge to be adjusted (i,j) , and set the Gaussian weight function , the expression is: ; where σ is the kernel width hyperparameter that adjusts the local range;

[0026] In the current local sample set, a weighted linear regression model is used to fit the modified coupling weights: ; is the adjusted edge weight;

[0027] The edge weight of the (i,j) edge in the original coupled network Replace with adjusted value , complete the adaptive weight update of the network; and update the network graph.

[0028] The present invention also provides a machine tool multi-source perception system based on dynamic characteristic coupling analysis, including a multi-source perception data acquisition module, a structural coupling modeling module, an auxiliary feature extraction module, an accuracy assessment module, an adaptive edge weight adjustment module, and a coupling network output module;

[0029] Multi-source perception data acquisition module: acquires multi-source dynamic characteristic data of key structural components of machine tools under different working conditions;

[0030] Structural coupling modeling module: taking structural components as network nodes, and calculating the detrended cross-correlation coefficients between each node pair based on the multi-source dynamic characteristic data, and constructing a weighted undirected coupling network graph with the detrended cross-correlation coefficients as edge weights;

[0031] Auxiliary feature extraction module: extracts auxiliary feature parameters for evaluating the accuracy of coupling relationships from multi-source dynamic characteristic data, including delayed cointegration disturbance residuals and environmental variable driven deviation index;

[0032] Accuracy evaluation module: inputs the auxiliary feature parameters into the preset coupling relationship anomaly prediction model, evaluates the accuracy of the coupling relationship between the node pairs, and classifies each edge into three types: accurate, partially accurate, and inaccurate based on the evaluation results;

[0033] Adaptive edge weight adjustment module: for node pairs with inaccurate evaluation results, adaptively adjusts their edge weights based on the accuracy evaluation results and correlation correction rules, and reconstructs the weighted coupling network;

[0034] Coupling network output module: outputs the corrected coupling network diagram and its network indicators for subsequent machine tool operation status analysis and fault prediction.

[0035] In the above technical solution, the technical effects and advantages provided by the present invention are:

[0036] 1. This paper constructs a machine tool dynamic coupling network based on DCCA and integrates auxiliary features such as delayed cointegration disturbance residuals and environmental variable drive deviation index to effectively address the problem of false coupling misjudgment caused by common-cause interference in traditional methods. By introducing a coupling relationship accuracy score and a three-category grading mechanism, it not only enhances the credibility of coupling identification but also significantly improves the accuracy of identifying critical coupling paths, providing more reliable data support for machine tool structural status perception.

[0037] 2. This invention uses an adaptive correction method based on local weighted regression to dynamically optimize edge weights for inaccurate node pairs, constructing a more realistic and robust coupled network structure. The resulting corrected network and its key indicators can be directly used for machine tool operating status trend monitoring, fault tracing, and intelligent prediction. The overall solution offers high accuracy, high interpretability, and high adaptability, making it suitable for intelligent perception and operational optimization scenarios for complex manufacturing equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments described in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0039] Figure 1 This is a mind map of the method of the present invention.

[0040] Figure 2 This is a mind map of the system modules of the present invention. DETAILED DESCRIPTION

[0041] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0042] Example 1, please refer to Figure 1 As shown, the multi-source perception method of a machine tool based on dynamic characteristic coupling analysis described in this embodiment includes:

[0043] Obtain multi-source dynamic characteristic data of key structural components of machine tools under different working conditions;

[0044] The structural components are used as network nodes, and the detrended cross-correlation coefficients between each node pair are calculated based on the multi-source dynamic characteristic data, and a weighted undirected coupled network graph is constructed with the detrended cross-correlation coefficients as edge weights;

[0045] Auxiliary characteristic parameters for evaluating the accuracy of coupling relationships are extracted from multi-source dynamic characteristic data, including delayed cointegration disturbance residuals and environmental variable driven deviation index;

[0046] Inputting the auxiliary feature parameters into a preset coupling relationship anomaly prediction model, evaluating the accuracy of the coupling relationship between the node pairs, and classifying each edge into three types: accurate, partially accurate, and inaccurate according to the evaluation results;

[0047] For node pairs whose evaluation results are not completely accurate, based on the accuracy evaluation results and the correlation correction rule, the edge weights thereof are adaptively adjusted, and the weighted coupling network is reconstructed;

[0048] The corrected coupling network diagram and its network indicators are output for subsequent machine tool operation status analysis and fault prediction.

[0049] In the machine tool multi-source perception system, obtaining multi-source dynamic characteristic data of each key structural component under different working conditions is the core premise. The target components (key structural units) for collection include but are not limited to the following core components:

[0050] Spindle unit: spindle box, bearings, tool interface, etc.

[0051] Workbench and pallet system: slide table, guide rail pair and drive motor that carry the workpiece.

[0052] Feed system: ball screw, servo motor, coupling.

[0053] Bed and column structure: Main load-bearing components, used to evaluate overall stiffness and coupling characteristics.

[0054] Cooling and lubrication system components: affect thermal deformation and friction characteristics.

[0055] Electronic control and drive components: CNC unit, driver, sensor module.

[0056] The data types collected (multi-source dynamic characteristics) include: mechanical dynamic signals: vibration acceleration (3-axis); displacement changes (laser interferometer, LVDT); stress / strain (chip strain gauge); stiffness changes (through modal excitation inversion); thermodynamic signals: temperature (component surface, coolant temperature); heat flux distribution (infrared temperature measurement module); control and electrical signals: current, voltage (motor side feedback); spindle torque and load changes; servo delay, following error (recorded internally in the control system); process signals: tool cutting force (Kistler force sensor); workpiece processing error feedback (online measurement system); environmental variables: workshop ambient temperature and humidity; ground vibration (foundation influence).

[0057] Sensor placement strategy: Sensors should be placed in a targeted manner based on modal analysis and critical coupling paths. For example: triaxial accelerometers and thermocouples on the spindle end; strain gauges and displacement gauges on the slide base and both ends of the leadscrew; and embedded temperature probes and structural strain gauges within the machine bed. All sensors should be connected to the same data acquisition card or distributed synchronous clock system. Data timestamps should be aligned to ensure timing consistency during dynamic analysis. Typical machining conditions should be set, such as high-speed cutting, step feed, and stop-start cycles. The control system should be integrated with external loading mechanisms to simulate shock loads or thermal drift conditions.

[0058] Data acquisition cycle and accuracy control include: sampling frequency: vibration and current data: above 10 kHz; temperature and strain data: 1–10 Hz; resolution and accuracy: vibration / displacement accuracy is controlled within ±0.01g / ±1μm; temperature accuracy is controlled within ±0.5°C; acquisition time: set according to the modal change cycle or thermal equilibrium time, such as each working condition lasting 2–5 minutes, to ensure coverage of the complete dynamic response cycle.

[0059] The raw data is time synchronized, interpolated and aligned, band-pass filtered, and subjected to wavelet packet denoising. Dimensional compression is performed using principal component analysis (PCA) or mutual information method to retain key dynamic characteristic variables. The corresponding operating condition type of the data is marked to form structured data labels to support subsequent analysis and modeling.

[0060] The structural components are used as network nodes, and the detrended cross-correlation coefficients between each node pair are calculated based on the multi-source dynamic characteristic data, and a weighted undirected coupled network graph with the detrended cross-correlation coefficients as edge weights is constructed, specifically including:

[0061] Mark the key structural components of the machine tool (such as spindle unit, worktable, bed, slide, screw, cooling system, etc.) as nodes in the network , n is the total number of key structural components. Each node corresponds to a specific physical component and is associated with its collected multi-source dynamic characteristic time series data.

[0062] Synchronous preprocessing of the time series data corresponding to each node: time alignment and interpolation filling; denoising (such as wavelet transform, low-pass filtering); normalization to eliminate dimension effects; ensuring that the data format of each node is a time series of uniform length .

[0063] For each pair of nodes Time series , calculate the cumulative deviation sequence: Where, is the mean (i.e., average) of node i in the entire time series, is the dynamic characteristic data of the jth structural component at time point t, is the time series average of node j, k is the index of the current time step, representing the cumulative process from t=1 to t=k, 、 are the cumulative deviation sequences of nodes i and j, respectively, indicating the overall trend of the data of the node deviating from its mean before each moment k.

[0064] Divide the accumulated deviation sequence into multiple non-overlapping windows, each with a length of s;

[0065] Perform polynomial fitting (usually first-order or second-order) within each segment to remove the trend and obtain the residual;

[0066] Calculate the detrended cross covariance for each segment , the expression is: ; Take the average of the cross covariance of all windows to get the overall DCCA cross volatility function ; 、 Represents the local trend term obtained by polynomial (usually linear or quadratic) regression fitting in the window v;

[0067] For each pair of nodes , the DCCA correlation coefficient is calculated using the following formula : ;in is the fluctuation intensity (autocorrelation variance) obtained by using DFA on a single node. ∈[−1,1], the closer the value is to ±1, the stronger the coupling is.

[0068] Create a graph G=(V,E,W), where: V represents the set of structural component nodes; E represents the edges with coupling relationships between node pairs; W represents the weight set of edges, corresponding to . Remove or ignore edges with low coupling strength (e.g., set a threshold The edges are not added to the graph) to improve network sparsity and interpretability.

[0069] Use graph theory tools (such as NetworkX, Gephi, and Matlab) to draw coupled network diagrams; analyze network metrics such as node degree, average clustering coefficient, and network density; identify key nodes (such as betweenness centrality); and provide network diagram output for scenarios such as condition monitoring and fault prediction.

[0070] Auxiliary characteristic parameters for evaluating the accuracy of coupling relationships are extracted from multi-source dynamic characteristic data, including delayed cointegration disturbance residuals and environmental variable driven deviation index, specifically including:

[0071] The delayed cointegration disturbance residual (LCDR) is used to measure whether the long-term cointegration relationship between the dynamic signals of two components is stable and to identify disturbance residuals caused by coupling delay or structural misalignment.

[0072] Take the preprocessed dynamic characteristic time series of two structural components i and j: ;

[0073] Establish a cointegration regression model: ; where τ is the delay (which can be optimized by AIC or BIC), ϵ(t) is the residual term, α and β are the regression coefficients of the cointegration model, and the disturbance residual sequence is extracted: ; Calculate the delayed cointegration disturbance residuals , the expression is: ; N is the total number of time points.

[0074] A larger delayed cointegration perturbation residual (LCDR) indicates that the cointegration relationship between the two nodes is unstable or significantly deviates after accounting for time delay, and the residual fluctuates significantly, indicating that the dynamic response relationship between them may be weakly coupled, mismatched, or even pseudo-coupled. In this case, the accuracy of the coupling relationship is low and may be affected by structural response delays, signal offsets, or external interference.

[0075] Conversely, when the delayed cointegration perturbation residuals are smaller, it indicates that the two structural components maintain a high degree of long-term cointegration after delayed compensation, and the residuals are stable and convergent, indicating a more realistic and stable dynamic coupling relationship between them. In this case, the identification of coupling paths is more accurate, making it suitable for building reliable coupling networks and state-aware models.

[0076] The environmental variable driven deviation index is used to determine whether the high coupling between two components is driven by common environmental factors rather than the actual physical structural connection.

[0077] Input node pair time series: ; Multiple environmental variable time series: ; K is the total number of environment variables;

[0078] Create the data matrix X(t): ; Input it into the PCMCI model for causal structure learning; Use toolkits such as Tigramite to run PCMCI and analyze the conditional independence and causal connection between variables within a certain lag range (such as 1 to 5 steps).

[0079] Output causal graph G(S,G), where: S is the variable node (structural variables and environmental variables); G represents the significant time-delay causal connection (for example ).

[0080] Determine whether two causal paths exist at the same time: ; If the same environment variable exists right The simultaneous generation of causal driving indicates the existence of a common cause driving path. For each pair of nodes (i, j), the environmental variable driving deviation index is calculated, and the expression is: ; is an indicator variable: if If it is a significant causal path, it is 1, otherwise it is 0; is an indicator variable: if If it is a significant causal path, it is 1, otherwise it is 0; is a small constant to prevent the denominator from dividing by zero; For environment variables right The significance of the causal connection (p-value); For environment variables right The significance of the causal connection (p-value).

[0081] When the environmental variable driven deviation index (CDCI) is larger, it means that the dynamic response between the node pair is more likely to be driven by the same external environmental variable at the same time, that is, there is an obvious "common cause path". This coupling relationship may be pseudo-coupling and not caused by real physical structure or functional connection. Therefore, the accuracy of the coupling relationship is low and needs to be carefully identified and handled.

[0082] On the contrary, when the CDCI is smaller, it means that the dynamic changes of the two nodes are less driven by the same environmental factors, and their coupling relationship is more likely to be generated by direct physical connection or real functional association. The coupling identification result is more reliable and accurate, and is suitable for constructing key structural paths and intelligent diagnostic networks.

[0083] The auxiliary feature parameters are input into the preset coupling relationship anomaly prediction model to evaluate the accuracy of the coupling relationship between the node pairs. Based on the evaluation results, each edge is divided into three types: accurate, partially accurate, and inaccurate. Specifically,

[0084] The delayed cointegration disturbance residuals and the environmental variable-driven deviation index are converted into comprehensive feature vectors, which are used as inputs of the machine learning model. The machine learning model uses the accuracy score value label of the coupling relationship between node pairs predicted by each set of comprehensive feature vectors as the prediction target, and minimizes the sum of the prediction errors of the accuracy score value labels of the coupling relationships between all node pairs as the training target. The machine learning model is trained until the sum of the prediction errors reaches convergence, and the model training is stopped. The accuracy score value of the coupling relationship between node pairs is determined according to the model output results, wherein the machine learning model is a polynomial regression model.

[0085] Comparing the obtained accuracy score of the coupling relationship between the node pairs with the gradient standard threshold, the gradient standard threshold includes a first standard threshold and a second standard threshold, and the first standard threshold is less than the second standard threshold, and comparing the accuracy score of the coupling relationship between the node pairs with the first standard threshold and the second standard threshold respectively;

[0086] If the accuracy score of the coupling relationship between the node pairs is greater than the second standard threshold, it means that the coupling relationship between the node pairs is stable and they are classified as accurate node pairs;

[0087] If the accuracy score of the coupling relationship between the node pairs is greater than or equal to the first standard threshold and less than or equal to the second standard threshold, it means that there is some uncertainty in the coupling relationship and it may be affected by disturbances or indirect driving. It is classified as an incompletely accurate node pair.

[0088] If the accuracy score of the coupling relationship between the node pairs is less than the first standard threshold, it means that the coupling relationship is likely to be caused by pseudo-correlation or common cause factors, and the reliability is poor, and it is classified as an inaccurate node pair.

[0089] For accurate node pairs, their original edge weights in the coupling network should be retained and incorporated into the machine tool operation status monitoring and fault prediction model as key coupling paths. They can be further used to identify the core transmission chain or main response chain of the structural nodes, supporting steady-state operating condition modeling and diagnostic feature extraction.

[0090] For inaccurate node pairs, consideration should be given to removing them from the coupling network or imposing a penalty factor on their edge weights to prevent misjudgment transmission caused by false coupling. If necessary, additional data sources or causal verification mechanisms can be introduced to re-evaluate the edge to avoid misleading the structural state perception results.

[0091] For node pairs whose evaluation results are not completely accurate, based on the accuracy evaluation results and the correlation correction rules, the edge weights are adaptively adjusted, and the weighted coupling network is reconstructed, specifically including:

[0092] Extract the set of all node pairs classified as not completely accurate from the accuracy score: ; is the accuracy score of the coupling relationship between node pair i and j, is the first standard threshold, is the second standard threshold;

[0093] For each edge , extract its comprehensive feature vector: ; It and other edges in the local neighborhood (such as edges adjacent to node i or j) constitute a local feature sample set for fitting the local coupling correction model.

[0094] For each sample edge (m,n) in the neighborhood, calculate the Euclidean distance to the edge to be adjusted (i,j) , and set the Gaussian weight function , the expression is: ; where σ is the kernel width hyperparameter that adjusts the local range.

[0095] In the current local sample set, a weighted linear regression model is used to fit the modified coupling weights: ; is the adjusted edge weight.

[0096] The edge weight of the (i,j) edge in the original coupled network Replace with adjusted value , complete the adaptive weight update of the network; and update the network graph G=(V,E,Wnew); Wnew represents the weight set of the updated edges.

[0097] After adaptively adjusting the weights of inaccurate edges, the system generates a revised coupling network diagram. This diagram uses machine tool structural components as nodes and updated coupling strengths (edge ​​weights) as edge weights, fully reflecting the true coupling relationships between components at the dynamic response level. This diagram not only overcomes the problems of false coupling and misjudged paths, but also more accurately reflects the physical coupling logic and the degree of mutual influence between states between structures.

[0098] On this basis, the system further calculates several key network structure indicators, including but not limited to: node degree (Degree): used to identify the core components with the most active coupling; edge weight mean and distribution: measure the overall coupling strength level of the system; clustering coefficient (Clustering Coefficient): reflects the coupling tightness between local structures; betweenness centrality (Betweenness Centrality): locates the key nodes that act as bridges in the coupling transmission chain; coupling path connectivity and fracture rate: used to monitor potential structural failures or signal transmission interruption risks.

[0099] These indicators enable global awareness of the machine tool's operating status, rapid location of local anomalies, and feedforward prediction of fault evolution paths, significantly improving the accuracy and real-time performance of system diagnostics. The revised coupling network diagram and its indicator analysis results can also serve as an important basis for historical operating condition comparison, trend tracking, and structural optimization recommendations.

[0100] Example 2, please refer to Figure 2 As shown, the multi-source perception system for machine tools based on dynamic characteristic coupling analysis described in this embodiment includes a multi-source perception data acquisition module, a structural coupling modeling module, an auxiliary feature extraction module, an accuracy assessment module, an adaptive edge weight adjustment module, and a coupling network output module;

[0101] Multi-source perception data acquisition module: acquires multi-source dynamic characteristic data of key structural components of machine tools under different working conditions;

[0102] Structural coupling modeling module: taking structural components as network nodes, and calculating the detrended cross-correlation coefficients between each node pair based on the multi-source dynamic characteristic data, and constructing a weighted undirected coupling network graph with the detrended cross-correlation coefficients as edge weights;

[0103] Auxiliary feature extraction module: extracts auxiliary feature parameters for evaluating the accuracy of coupling relationships from multi-source dynamic characteristic data, including delayed cointegration disturbance residuals and environmental variable driven deviation index;

[0104] Accuracy evaluation module: inputs the auxiliary feature parameters into the preset coupling relationship anomaly prediction model, evaluates the accuracy of the coupling relationship between the node pairs, and classifies each edge into three types: accurate, partially accurate, and inaccurate based on the evaluation results;

[0105] Adaptive edge weight adjustment module: for node pairs with inaccurate evaluation results, adaptively adjusts their edge weights based on the accuracy evaluation results and correlation correction rules, and reconstructs the weighted coupling network;

[0106] Coupling network output module: outputs the corrected coupling network diagram and its network indicators for subsequent machine tool operation status analysis and fault prediction.

[0107] The above formulas are all dimensionless and numerical calculations. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The preset parameters in the formulas are set by technicians in this field according to actual conditions.

[0108] It should be understood that the term "and / or" as used herein simply describes a relationship between associated objects, indicating that three possible relationships exist. For example, "A and / or B" can represent: A alone, A and B together, or B alone. A and B can be singular or plural. Furthermore, the character " / " as used herein generally indicates an "or" relationship between the associated objects, but it may also indicate an "and / or" relationship. For specific understanding, please refer to the context.

[0109] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0110] The above is only a specific implementation method of the present application, but the scope of protection of the present application is not limited thereto. Any technician familiar with this technical field can easily think of changes or replacements within the technical scope disclosed in this application, which should be covered by the scope of protection of the present application.

Claims

1. A multi-source perception method for machine tools based on dynamic characteristic coupling analysis, characterized by: include: Obtain multi-source dynamic characteristic data of key structural components of machine tools under different working conditions; The structural components are used as network nodes, and the detrended cross-correlation coefficients between each node pair are calculated based on the multi-source dynamic characteristic data, and a weighted undirected coupled network graph is constructed with the detrended cross-correlation coefficients as edge weights; Auxiliary characteristic parameters for evaluating the accuracy of coupling relationships are extracted from multi-source dynamic characteristic data, including delayed cointegration disturbance residuals and environmental variable driven deviation index; Inputting the auxiliary feature parameters into a preset coupling relationship anomaly prediction model, evaluating the accuracy of the coupling relationship between the node pairs, and classifying each edge into three types: accurate, partially accurate, and inaccurate according to the evaluation results; For node pairs whose evaluation results are not completely accurate, based on the accuracy evaluation results and the correlation correction rule, the edge weights thereof are adaptively adjusted, and the weighted coupling network is reconstructed; The corrected coupling network diagram and its network indicators are output for subsequent machine tool operation status analysis and fault prediction.

2. The multi-source perception method for machine tools based on dynamic characteristic coupling analysis according to claim 1, characterized in that: The key structural components of the machine tool include the spindle unit, worktable and pallet system, feed system, bed and column structure, cooling and lubrication system components, and electronic control and drive components.

3. The multi-source perception method for machine tools based on dynamic characteristic coupling analysis according to claim 2, characterized in that: The multi-source dynamic characteristic data includes vibration acceleration, displacement change, strain signal, temperature, heat flux density, motor current, voltage, spindle load, servo control error, tool cutting force, workpiece processing error and environmental variable data.

4. The multi-source perception method for machine tools based on dynamic characteristic coupling analysis according to claim 3 is characterized by: The method for calculating the detrended cross-correlation coefficient between each pair of nodes includes: converting the time series data of each node into a cumulative deviation sequence; performing polynomial trend fitting in multiple sliding windows; calculating the detrended cross covariance and normalizing it to obtain the detrended cross-correlation coefficient; and using the detrended cross-correlation coefficient to construct edge weights between nodes to form a weighted undirected graph.

5. The multi-source perception method for machine tools based on dynamic characteristic coupling analysis according to claim 1, characterized in that: The method for obtaining the delayed cointegration disturbance residual is to take the preprocessed dynamic characteristic time series of two structural components i and j: ; Establish a cointegration regression model: ; Where τ is the delay, ϵ(t) is the residual term, α and β are the regression coefficients of the cointegration model, and the disturbance residual sequence is extracted: ; Calculate the delayed cointegration disturbance residuals , the expression is: ; N is the total number of time points.

6. The multi-source perception method for machine tools based on dynamic characteristic coupling analysis according to claim 5, characterized in that: The method to obtain the environmental variable driven deviation index is as follows: Input node pair time series: ; Multiple environmental variable time series: ; K is the total number of environmental variables; create the data matrix X(t): ; Input it into the PCMCI model for causal structure learning; output the causal graph G(S,G), where S is the variable node; G represents the significant time-delay causal connection; determine whether the two causal paths exist at the same time: ; If the same environment variable exists right The simultaneous generation of causal driving indicates the existence of a common cause driving path. For each pair of nodes (i, j), the environmental variable driving deviation index is calculated, and the expression is: ; is an indicator variable: if If it is a significant causal path, it is 1, otherwise it is 0; is an indicator variable: if If it is a significant causal path, it is 1, otherwise it is 0; is a small constant; For environment variables right The significance of causal connection; For environment variables right The significance of causal connection.

7. The multi-source perception method for machine tools based on dynamic characteristic coupling analysis according to claim 6, characterized in that: The auxiliary feature parameters are input into the preset coupling relationship anomaly prediction model to evaluate the accuracy of the coupling relationship between the node pairs. Based on the evaluation results, each edge is divided into three types: accurate, partially accurate, and inaccurate. Specifically, The delayed cointegration disturbance residuals and the environmental variable-driven deviation index are converted into comprehensive feature vectors, which are used as inputs of the machine learning model. The machine learning model uses the accuracy score value label of the coupling relationship between node pairs predicted by each set of comprehensive feature vectors as the prediction target, and minimizes the sum of the prediction errors of the accuracy score value labels of the coupling relationships between all node pairs as the training target. The machine learning model is trained until the sum of the prediction errors reaches convergence, and the model training is stopped. The accuracy score value of the coupling relationship between node pairs is determined according to the model output results, wherein the machine learning model is a polynomial regression model.

8. The multi-source perception method for machine tools based on dynamic characteristic coupling analysis according to claim 7, characterized in that: Comparing the obtained accuracy score of the coupling relationship between the node pairs with the gradient standard threshold, the gradient standard threshold includes a first standard threshold and a second standard threshold, and the first standard threshold is less than the second standard threshold, and comparing the accuracy score of the coupling relationship between the node pairs with the first standard threshold and the second standard threshold respectively; If the accuracy score of the coupling relationship between the node pairs is greater than the second standard threshold, it means that the coupling relationship between the node pairs is stable and they are classified as accurate node pairs; If the accuracy score of the coupling relationship between the node pairs is greater than or equal to the first standard threshold and less than or equal to the second standard threshold, it indicates that there is partial uncertainty in the coupling relationship, and it is classified as an incompletely accurate node pair; if the accuracy score of the coupling relationship between the node pairs is less than the first standard threshold, it is classified as an inaccurate node pair.

9. The multi-source perception method for machine tools based on dynamic characteristic coupling analysis according to claim 8, characterized in that: For node pairs whose evaluation results are not completely accurate, based on the accuracy evaluation results and the correlation correction rules, the edge weights are adaptively adjusted, and the weighted coupling network is reconstructed, specifically including: Extract the set of all node pairs classified as not completely accurate from the accuracy score: ; is the accuracy score of the coupling relationship between node pair i and j, is the first standard threshold, is the second standard threshold; For each edge , extract its comprehensive feature vector: ; It and the edges adjacent to node i or j in the local neighborhood constitute a local feature sample set; For each sample edge (m,n) in the neighborhood, calculate the Euclidean distance to the edge to be adjusted (i,j) , and set the Gaussian weight function , the expression is: ; where σ is the kernel width hyperparameter that adjusts the local range; In the current local sample set, a weighted linear regression model is used to fit the modified coupling weights: ; is the adjusted edge weight; The edge weight of the (i,j) edge in the original coupled network Replace with adjusted value , complete the adaptive weight update of the network; and update the network graph.

10. A machine tool multi-source perception system based on dynamic characteristic coupling analysis, used to implement the machine tool multi-source perception method based on dynamic characteristic coupling analysis according to any one of claims 1 to 9, characterized in that: It includes multi-source perception data acquisition module, structural coupling modeling module, auxiliary feature extraction module, accuracy assessment module, adaptive edge weight adjustment module and coupling network output module; Multi-source perception data acquisition module: acquires multi-source dynamic characteristic data of key structural components of machine tools under different working conditions; Structural coupling modeling module: taking structural components as network nodes, and calculating the detrended cross-correlation coefficients between each node pair based on the multi-source dynamic characteristic data, and constructing a weighted undirected coupling network graph with the detrended cross-correlation coefficients as edge weights; Auxiliary feature extraction module: extracts auxiliary feature parameters for evaluating the accuracy of coupling relationships from multi-source dynamic characteristic data, including delayed cointegration disturbance residuals and environmental variable driven deviation index; Accuracy evaluation module: inputs the auxiliary feature parameters into the preset coupling relationship anomaly prediction model, evaluates the accuracy of the coupling relationship between the node pairs, and classifies each edge into three types: accurate, partially accurate, and inaccurate based on the evaluation results; Adaptive edge weight adjustment module: for node pairs with inaccurate evaluation results, adaptively adjusts their edge weights based on the accuracy evaluation results and correlation correction rules, and reconstructs the weighted coupling network; Coupling network output module: outputs the corrected coupling network diagram and its network indicators for subsequent machine tool operation status analysis and fault prediction.

Images