Intelligent fault diagnosis method for numerical control machine tool

Abstract

The invention provides an intelligent fault diagnosis method for a numerical control machine tool, which is applied to a control system of the numerical control machine tool. The method comprises the steps of collecting multi-source sensor data in the operation process of the numerical control machine tool, wherein the multi-source sensor data comprises at least two of a vibration signal, a temperature signal, a current signal and a sound signal; performing preprocessing and feature extraction on the multi-source sensor data to obtain preprocessed data; based on the preprocessed data, constructing and training a machine tool dynamic model to obtain a target machine tool dynamic model; when abnormal operation of the numerical control machine tool is detected, fault state estimation is carried out based on the target machine tool dynamic model and a model prediction control method, and a fault estimation result is obtained; and based on the fault estimation result, constructing a fault evolution path by adopting a Monte Carlo tree search algorithm, and outputting a fault diagnosis result. The accuracy and efficiency of fault diagnosis of the numerical control machine tool can be improved.

Description

Technical Field

[0001] The present application relates to the technical field of numerically controlled machine tools, and in particular to an intelligent fault diagnosis method for numerically controlled machine tools. Background Art

[0002] CNC machine tools are key equipment in modern industrial production, and their operating status directly affects production efficiency and processing quality. With the improvement of industrial automation level, higher requirements are placed on the intelligence and accuracy of CNC machine tool fault diagnosis.

[0003] At present, common fault diagnosis methods for CNC machine tools mainly include threshold monitoring based on a single sensor and rule judgment based on expert experience. For example, the vibration amplitude of the bearing is monitored by a vibration sensor, and it is judged as a fault when it exceeds the preset threshold; or the temperature of key components is monitored by a temperature sensor, and fault judgment is performed in combination with expert experience.

[0004] The more advanced diagnostic method uses traditional machine learning algorithms, such as support vector machines (SVM), to classify fault features. This method learns from historical fault data and establishes a mapping relationship between fault features and fault types, thereby achieving fault identification.

[0005] However, the existing technology has the following problems: on the one hand, a single sensor or a single algorithm is difficult to fully reflect the complex operating status of the machine tool, resulting in insufficient accuracy of the diagnostic results; on the other hand, traditional methods lack the ability to predict the development trend of faults and cannot timely warn of potential fault risks. Summary of the invention

[0006] In view of this, the present application provides an intelligent fault diagnosis method and device for CNC machine tools, which solves the problem in the prior art that a single sensor or a single algorithm is difficult to fully reflect the complex operating status of the machine tool and lacks the ability to predict the development trend of faults.

[0007] The embodiment of the present application provides an intelligent fault diagnosis method for a CNC machine tool, which is applied to a control system of the CNC machine tool, including: Collecting multi-source sensor data during the operation of the CNC machine tool, wherein the multi-source sensor data includes at least two of a vibration signal, a temperature signal, a current signal, and a sound signal; Preprocessing and feature extraction are performed on the multi-source sensor data to obtain preprocessed data; Based on the preprocessed data, a machine tool dynamic model is constructed and trained to obtain a target machine tool dynamic model; When an abnormal operation of a CNC machine tool is detected, a fault state prediction is performed based on the target machine tool dynamic model and the model predictive control method to obtain a fault prediction result; Based on the fault prediction result, a Monte Carlo tree search algorithm is used to construct a fault evolution path, and a fault diagnosis result is output.

[0008] The method of constructing and training a machine tool dynamic model based on the preprocessed data to obtain a target machine tool dynamic model includes: Constructing a deep learning model architecture, wherein the deep learning model architecture includes a convolutional neural network (CNN) for spatial feature extraction and a bidirectional long short-term memory (LSTM) network for temporal feature processing, and a fully connected layer for fusing the features extracted by the CNN and the LSTM; Based on the outputs of the CNN and the LSTM, multiple groups of loss functions are designed, wherein the loss functions include: a prediction error loss for measuring prediction accuracy and an L2 regularization constraint for preventing model overfitting, wherein different loss weights are set for different types of fault samples; According to the calculation result of the loss function, the Adam optimizer is used for model training, and the learning rate is dynamically adjusted based on the performance index of the validation set. When the loss value of the validation set does not decrease for three consecutive training cycles, the learning rate is adjusted to 0.8 times the original value; And adopt a model evaluation mechanism based on K-fold cross-validation to determine the dynamic model of the target machine tool, including: in each fold training, use a stratified sampling strategy to divide the data set to ensure the balanced distribution of samples of different fault types, and evaluate the model performance by counting the accuracy, precision and recall rate on each fold validation set, and select the optimal model parameter configuration.

[0009] In the deep learning model architecture: The CNN includes a three-layer convolutional layer structure connected in sequence. The first layer uses 32 3×3 convolution kernels for feature extraction and normalization through the BatchNormalization layer. The second layer uses 64 5×5 convolution kernels to expand the receptive field and enhance the nonlinear expression ability through the ReLU activation function. The third layer reduces the dimension through a 2×2 maximum pooling window. The LSTM includes two bidirectional LSTM layers, each layer contains 128 neurons, and a Dropout layer with a dropout rate of 0.3 is set between the two layers; The fully connected layer includes three layers, and the corresponding numbers of nodes are 512, 256 and 128 respectively, which are used to fuse spatial features and temporal features, and a BatchNormalization layer is added before the last layer to ensure stable feature distribution.

[0010] When the CNC machine tool is detected to be operating abnormally, a fault state prediction is performed based on the target machine tool dynamic model and the model prediction control method to obtain a fault prediction result, including: Based on the target machine tool dynamic model, a state predictor of a recursive prediction structure is constructed, wherein the state predictor uses 20 sampling points as the time window length of a prediction unit; By means of the state predictor, an adaptive rolling time domain prediction strategy is designed, wherein the rolling time domain prediction strategy includes short-term prediction of 3 sampling points, medium-term prediction of 6 sampling points and long-term prediction of 10 sampling points, and the prediction step size is dynamically adjusted according to the working condition of the CNC machine tool; Based on the rolling time domain prediction strategy, a multi-step prediction is performed on the future state of the CNC machine tool, the output result of the prediction strategy is obtained, and the deviation between the predicted state and the normal operating state of the CNC machine tool is calculated in real time to evaluate the fault risk level.

[0011] The deviation between the real-time calculation prediction state and the normal operation state of the CNC machine tool includes: Construct a CNC machine tool health status benchmark library containing the normal operating characteristic distribution under different cutting parameters; Calculate the Mahalanobis distance between the predicted state sequence and the corresponding working condition reference state in the reference library to obtain a state deviation value; Performing linear regression analysis on the state deviation value to obtain deviation change trend characteristics; Based on the historical fault data, the state transition matrix is ​​trained, and the fault development trend is predicted by combining the current state deviation value and deviation change trend; According to the state deviation value, deviation change trend and fault development trend, the fuzzy comprehensive evaluation method is adopted to divide the operation state of the CNC machine tool into four levels: normal, slight abnormality, moderate fault and severe fault.

[0012] The state predictor includes the following feedback correction mechanism: The error tracker computes the sequence of deviations between the predicted values ​​and the actual observed values; An adaptive weight regulator dynamically adjusts the correction weight in the range of 0.6 to 1.0 according to the prediction accuracy, where the prediction accuracy is calculated by the mean square error between the predicted value and the actual value; The state corrector performs real-time correction on the prediction result based on the correction weight.

[0013] The method of constructing a fault evolution path based on the fault prediction result by using a Monte Carlo tree search algorithm includes: According to the fault prediction result, an initial state node is established, wherein the attribute information of the initial state node represents the feature vector, fault risk level and confidence of the current state of the CNC machine tool; A state transfer rule base is established based on the node, wherein the state transfer rule base includes: component-level fault transfer rules, system-level fault transfer rules, and environmental factor influence rules, wherein each rule is configured with a corresponding transfer probability and time constraint; Based on the state transition rule base, a heuristic method is used to dynamically determine the expansion depth, wherein a maximum expansion depth of 8 layers is used for important fault paths, and a maximum expansion depth of 4 layers is used for minor fault paths; The improved UCT algorithm is used to perform node selection and expansion to determine the fault evolution path.

[0014] The improved UCT algorithm is used to select and expand nodes to determine the fault evolution path, including: Calculate the node score: Score = Q / N + c*sqrt(ln(N_parent) / N) + w*R, Where Q represents the cumulative value of the node, N represents the number of node visits, N_parent represents the number of parent node visits, c represents the exploration coefficient, R represents the node risk coefficient, and w represents the risk weight; Generate child nodes based on the rules in the state transition rule base, and dynamically adjust the transition probability according to historical fault data during the expansion process; Perform node simulation using fast stochastic simulation method and record the cumulative risk value and time delay on the path; Starting from the leaf node, update the visit count and cumulative value of each node along the path upward.

[0015] The application also performs at least one of the following operations: Counting the number of visits to each node in the fault evolution path, and identifying a path whose number of visits exceeds 30% of the total simulation times as a high-probability fault evolution path; Generate a hierarchical fault diagnosis report based on the high-probability fault evolution path, including a description of the current fault state, a potential fault evolution warning, and corresponding maintenance suggestions; The fault evolution tree structure is displayed in a visual way. Different colors are used to identify the fault level. The node size indicates the state transition probability, and the thickness of the connecting edge indicates the path access frequency.

[0016] The embodiment of the present application also provides an intelligent fault diagnosis device for a CNC machine tool, comprising: A data acquisition unit, used to acquire multi-source sensor data during the operation of the CNC machine tool, wherein the multi-source sensor data includes at least two of a vibration signal, a temperature signal, a current signal and a sound signal; A preprocessing unit, connected to the data acquisition unit by signal, configured to receive the multi-source sensor data and perform preprocessing and feature extraction to obtain preprocessed data; A model training unit is connected to the preprocessing unit signal, and is used to receive the preprocessing data, and build and train the machine tool dynamic model based on the preprocessing data to obtain a target machine tool dynamic model; A state prediction unit is connected to the model training unit signal, and is used to predict the fault state based on the target machine tool dynamic model and the model prediction control method to obtain a fault prediction result when the CNC machine tool is detected to be operating abnormally; The diagnosis unit is connected to the state prediction unit signal, and is used to construct a fault evolution path based on the fault prediction result by using a Monte Carlo tree search algorithm, and output a fault diagnosis result.

[0017] The present application also provides a computer device, the computer device comprising: at least one processor; and, a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the above-mentioned intelligent fault diagnosis method for CNC machine tools.

[0018] An embodiment of the present application also provides a computer-readable storage medium, which stores computer instructions, and the computer instructions are used to enable a computer to execute the above-mentioned intelligent fault diagnosis method for CNC machine tools.

[0019] An embodiment of the present application also provides a computer program product, including computer instructions, which, when executed by a processor, implement the steps of the above-mentioned intelligent fault diagnosis method for CNC machine tools.

[0020] This application has the following technical effects: 1. By collecting and fusing multi-source sensor data, the perception capability of machine tool operation status and information utilization efficiency are improved.

[0021] 2. The CNN-LSTM hybrid deep learning architecture is adopted to effectively combine spatial features and temporal features, improving the accuracy of fault feature extraction and classification.

[0022] 3. The introduction of model predictive control method and adaptive feedback mechanism enables accurate prediction and dynamic correction of fault status.

[0023] 4. The innovative application of Monte Carlo tree search to construct the fault evolution path improves the interpretability and early warning capabilities of fault diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for describing the embodiments are briefly introduced below.

[0025] Figure 1 It is a flowchart of the intelligent fault diagnosis method for CNC machine tools provided in the embodiment of the present application; Figure 2 It is a schematic diagram of the multi-source sensor data collection and preprocessing process provided by an embodiment of the present application; Figure 3 It is a schematic diagram of the deep learning model architecture provided in the embodiment of the present application; Figure 4 It is a schematic diagram of the state predictor structure provided in an embodiment of the present application. DETAILED DESCRIPTION

[0026] The present application provides an intelligent fault diagnosis method and device for a CNC machine tool, and its preliminary technical solution is as follows: S1: Collecting multi-source sensor data during the operation of the CNC machine tool, wherein the multi-source sensor data includes at least two of a vibration signal, a temperature signal, a current signal and a sound signal.

[0027] This application collects machine tool operating status data through multi-source sensors deployed in key positions of CNC machine tools.

[0028] Specifically, combined Figure 2 Vibration sensors are installed on key functional components such as the machine tool spindle, feed system, and tool system to collect vibration signals during the operation of the machine tool; temperature sensors are installed on the spindle box, motor, bearing seat and other heat-prone parts to monitor temperature changes in real time; current sensors are installed on the spindle motor and feed motor to collect the motor operating current; sound sensors are installed around the machine tool workspace to collect acoustic signals during the processing process.

[0029] It should be noted that vibration signals are mainly used to reflect the dynamic characteristics and mechanical failures of machine tools, such as bearing failures, gear failures, etc.; temperature signals can reflect the thermal state of machine tool components, which helps to identify failures such as overload and poor lubrication; current signals can reflect motor load status and electrical system failures; and sound signals can assist in determining processing status and mechanical failures.

[0030] In practical applications, at least two types of sensor signals can be selected for combination according to the machine tool type and monitoring requirements. For example, for precision machining centers, vibration and temperature signals can be collected in particular; for heavy-duty cutting machines, vibration, temperature and current signals can be collected simultaneously. This multi-source data collection method can fully reflect the operating status of the machine tool from different angles and provide a reliable data basis for subsequent fault diagnosis.

[0031] In addition, the sampling frequency of various sensors needs to be reasonably configured according to the signal characteristics. For example, the sampling frequency of vibration signals is usually set to above 10kHz to capture high-frequency fault characteristics; the sampling frequency of temperature signals can be relatively low, set to around 1Hz; the sampling frequency of current signals needs to be determined according to the motor control cycle, generally around 1kHz.

[0032] In summary, by rationally configuring the multi-source sensor network, comprehensive perception of the machine tool operating status can be achieved, laying a solid data foundation for subsequent intelligent fault diagnosis.

[0033] It should be emphasized that the sensor's installation position, sampling parameters and other configurations directly affect the data quality. Therefore, in practical applications, it is necessary to combine the machine tool characteristics and professional experience for optimized design.

[0034] S2: Preprocessing and feature extraction are performed on the multi-source sensor data to obtain preprocessed data.

[0035] Preprocessing and feature extraction of multi-source sensor data are key steps in achieving accurate fault diagnosis.

[0036] Combination Figure 2 First, the collected raw data is cleaned, mainly including outlier detection and noise removal. Specifically, the 3σ criterion based on statistical characteristics is used to detect outliers, and the data points that exceed the normal range are marked; then the median filter method is used to eliminate the impulse noise and random interference in the signal, where the filter window size is dynamically adjusted according to the signal characteristics, such as a 5-9 point window for vibration signals and a 3-5 point window for temperature signals.

[0037] Secondly, in order to solve the problem of missing values ​​that may occur during data collection, a piecewise linear interpolation method is used to repair data. In practical applications, data may be missing due to temporary sensor failure or communication interruption. At this time, by analyzing the time span of the missing data and the change trend of adjacent data, an adaptive interpolation strategy is adopted: for short-term missing data (less than 3 sampling cycles), linear interpolation is used; for longer missing data (3-10 sampling cycles), cubic spline interpolation is used; for longer missing data, it needs to be marked and specially processed in subsequent analysis.

[0038] Thirdly, considering the significant differences in data dimensions and distribution characteristics among different types of sensors, data standardization is required.

[0039] This application adopts an improved Min-Max normalization method to uniformly map various sensor data to the [0,1] interval. In particular, when calculating the mapping range, not only the statistical characteristics of historical data are considered, but also a certain margin is reserved to accommodate possible data fluctuations, usually expanding the maximum and minimum values ​​by 10% respectively.

[0040] After completing data preprocessing, time synchronization and feature extraction are required.

[0041] First, align the multi-source sensor data based on the timestamp to ensure the consistency of data from different sources in the time dimension. Then, extract the corresponding feature parameters for different types of signals: for vibration signals, extract time domain features (mean, variance, peak, kurtosis, etc.) and frequency domain features (power spectrum density, main frequency component, etc.); for temperature signals, mainly extract statistical features (mean, rate of change, etc.) and trend features; for current signals, in addition to basic statistical features, it is also necessary to extract spectral features to reflect the operating status of the motor; for sound signals, focus on extracting energy features and frequency band features.

[0042] Finally, the extracted features are combined to form a feature vector.

[0043] It should be noted that the feature extraction process adopts a sliding window mechanism. The window length is determined according to the signal characteristics and the fault feature time scale. Usually, the window length of the vibration signal is 1024 points, and the window length of the temperature signal is 100 points. The window overlap rate is set to 50% to ensure the continuity of the features.

[0044] In summary, through systematic data preprocessing and feature extraction, data quality can be significantly improved, providing reliable input features for subsequent model training. In addition, the preprocessing scheme of this application has strong adaptability and robustness, and can effectively handle various data problems encountered in actual engineering.

[0045] S3: Based on the preprocessed data, a machine tool dynamic model is constructed and trained to obtain a target machine tool dynamic model.

[0046] Based on the preprocessed data, the process of constructing and training the machine tool dynamic model in this application includes multiple key links.

[0047] Firstly, an innovative deep learning model architecture is designed, which combines the advantages of CNN and LSTM and can process both spatial and temporal features.

[0048] Specifically, the CNN part is mainly used to extract the local features and spatial correlations of various sensor signals, while the LSTM part is responsible for capturing long-term temporal dependencies. The outputs of the two networks are fused through the fully connected layer to achieve comprehensive modeling of the dynamic characteristics of the machine tool.

[0049] During the model training process, multiple sets of loss functions are designed to guide the optimization of the model.

[0050] The prediction error loss is used to measure the degree of match between the model output and the actual fault type, and the cross entropy loss function is used; the L2 regularization constraint is used to prevent the model from overfitting, which is achieved by adding the sum of squares of the model parameters to the loss function.

[0051] In particular, different loss weights are set for different types of fault samples, and higher weights are given to severe faults and key component failure samples to improve the model's ability to identify important faults.

[0052] The Adam optimizer is used to update parameters during the training process. Compared with the traditional stochastic gradient descent method, the Adam optimizer has the characteristics of adaptive learning rate adjustment and can better handle the problem of gradient sparsity. At the same time, a dynamic learning rate adjustment mechanism based on the validation set performance is introduced: when the validation set loss value does not decrease for three consecutive training cycles, the learning rate is adjusted to 0.8 times the original value. This strategy can help the model escape the local optimal solution and obtain better generalization performance.

[0053] In order to ensure the reliability of the model, the K-fold cross-validation evaluation mechanism is adopted.

[0054] First, the data set is divided using a stratified sampling strategy to ensure that the distribution of various fault samples in the training set and the validation set is basically the same. In each fold of training, the model performance is evaluated by statistically analyzing indicators such as accuracy, precision, and recall on the validation set. Finally, based on the results of multiple rounds of cross-validation, the model parameter configuration with the best comprehensive performance is selected as the target machine tool dynamic model.

[0055] See also Figure 3 In the specific network structure design, the CNN part adopts a three-layer convolution structure. The first layer uses 32 3×3 convolution kernels for preliminary feature extraction, and the feature distribution is normalized through the BatchNormalization layer; the second layer uses 64 5×5 convolution kernels to expand the receptive field, and the nonlinear expression ability of the model is enhanced through the ReLU activation function; the third layer uses a 2×2 maximum pooling window for feature dimensionality reduction and extracts the most significant feature information.

[0056] The LSTM part adopts a bidirectional structure, including two LSTM layers, each containing 128 neurons. The bidirectional structure can simultaneously consider the forward and backward dependencies of the sequence, improving the model's ability to understand the temporal pattern. A Dropout layer with a dropout rate of 0.3 is set between the two LSTM layers to prevent overfitting by randomly inactivating some neurons.

[0057] The feature fusion part uses a three-layer fully connected network with 512, 256 and 128 nodes, which realizes the gradual dimension reduction and high-level semantic extraction of features. A BatchNormalization layer is added before the last fully connected layer to ensure the stability of feature distribution, which is conducive to the convergence and generalization of the model.

[0058] In summary, through a carefully designed deep learning architecture and systematic training strategy, this application has built an intelligent model that can accurately reflect the dynamic characteristics of machine tools. This model has strong feature extraction and fault identification capabilities, providing a reliable basis for subsequent fault state prediction.

[0059] S3 specifically includes: S3.1: Construct a deep learning model architecture, which includes a convolutional neural network (CNN) for spatial feature extraction and a bidirectional long short-term memory (LSTM) network for temporal feature processing, as well as a fully connected layer for fusing the features extracted by the CNN and the LSTM.

[0060] This application designs a deep learning model architecture that integrates CNN and LSTM, which can simultaneously process the spatial features and temporal features of machine tool operation data. Among them, the CNN part is mainly responsible for extracting the spatial features of multi-source sensor data, including the correlation and local feature patterns between different sensor signals. The LSTM part focuses on processing the temporal features of the data, capturing long-term dependencies through a bidirectional structure, and can understand the evolution of the machine tool status over time.

[0061] The feature extraction results of the two networks are fused through the fully connected layer. This fusion mechanism allows the model to learn the optimal combination of spatial features and temporal features, thereby forming a comprehensive representation of the machine tool status. The fully connected layer maps the fused features to the final fault diagnosis space through layer-by-layer dimensionality reduction and nonlinear transformation.

[0062] S3.2: Based on the outputs of the CNN and the LSTM, multiple sets of loss functions are designed, wherein the loss functions include: a prediction error loss for measuring prediction accuracy and an L2 regularization constraint for preventing model overfitting, wherein different loss weights are set for different types of fault samples.

[0063] During the model training process, two key loss functions were designed. The prediction error loss uses the cross entropy loss function to measure the difference between the model prediction result and the actual fault type. This loss function sets differentiated weights for different types of fault samples. For example, it gives higher weights (usually 1.5-2.0) to severe fault samples such as bearing faults and spindle faults, and standard weights (1.0) to general fault samples. This weighting strategy ensures that the model can pay more attention to the identification of critical faults.

[0064] The L2 regularization constraint is implemented by adding the square sum of model parameters to the loss function, and its coefficient is usually set between 0.001 and 0.01. This constraint mechanism can suppress the excessive growth of model parameters, avoid overfitting of the model to the training data, and improve its generalization ability in practical applications.

[0065] S3.3: According to the calculation result of the loss function, the Adam optimizer is used for model training, and the learning rate is dynamically adjusted based on the performance index of the validation set. When the loss value of the validation set does not decrease for three consecutive training cycles, the learning rate is adjusted to 0.8 times the original value.

[0066] The model training uses the Adam optimizer, which is an adaptive moment estimation optimization algorithm. This optimizer combines the advantages of the momentum method and RMSprop and can adaptively adjust the learning rate of each parameter. In the actual training, the initial learning rate of the Adam optimizer is set to 0.001, the β1 parameter is set to 0.9, and the β2 parameter is set to 0.999.

[0067] At the same time, a dynamic learning rate adjustment mechanism based on the validation set performance is introduced.

[0068] Specifically, after each training cycle, the loss value of the model on the validation set is evaluated. If the validation set loss value does not decrease after three consecutive training cycles, the current learning rate is reduced to 0.8 times the original. This strategy can help the model converge better in the later stages of training and avoid oscillation near the local optimal solution.

[0069] S3.4: Adopt a model evaluation mechanism based on K-fold cross-validation to determine the dynamic model of the target machine tool, including: in each fold of training, use a stratified sampling strategy to divide the data set to ensure a balanced distribution of samples of different fault types, and evaluate the model performance and select the optimal model parameter configuration by statistically analyzing the accuracy, precision and recall rate on each fold of the validation set.

[0070] This application uses a K-fold cross-validation mechanism to evaluate model performance, and the K value is usually set to 5 or 10. When dividing the data set, a stratified sampling strategy is used to ensure that the proportion of each type of fault sample in each subset is basically consistent with the original data set. For example, if a certain type of fault sample accounts for 15% in the original data set, then in each divided subset, the proportion of this type of fault sample should also be maintained at around 15%.

[0071] In each fold of training, the model is evaluated by calculating multiple performance indicators on the validation set. These indicators include accuracy (overall prediction accuracy), precision (the proportion of true positive examples predicted as positive), and recall (the proportion of true positive examples predicted correctly). Based on the comprehensive score of these indicators, the model parameter configuration with the best performance is selected as the final model.

[0072] Among them, in the deep learning model architecture: The CNN includes a three-layer convolutional layer structure connected in sequence. The first layer uses 32 3×3 convolution kernels for feature extraction and normalizes through the BatchNormalization layer. The second layer uses 64 5×5 convolution kernels to expand the receptive field and enhance the nonlinear expression ability through the ReLU activation function. The third layer reduces the dimension through a 2×2 maximum pooling window.

[0073] The LSTM includes two bidirectional LSTM layers, each layer includes 128 neurons, and a Dropout layer with a dropout rate of 0.3 is set between the two layers.

[0074] The fully connected layer includes three layers, and the corresponding numbers of nodes are 512, 256 and 128 respectively, which are used to fuse spatial features and temporal features, and a BatchNormalization layer is added before the last layer to ensure stable feature distribution.

[0075] The CNN part uses a three-layer convolution structure for spatial feature extraction. The first layer uses 32 3×3 convolution kernels for preliminary feature extraction, with a convolution step of 1, and performs feature normalization through the BatchNormalization layer, which helps to accelerate network training and improve model stability. The second layer uses 64 5×5 convolution kernels. This larger convolution kernel can capture a wider range of spatial correlations. At the same time, the ReLU activation function introduces nonlinear transformations to enhance the model's expressiveness. The third layer uses a 2×2 maximum pooling window for dimensionality reduction to retain the most significant feature information.

[0076] The LSTM part contains two layers of bidirectional LSTM structures, each layer contains 128 neurons. The bidirectional structure can consider both the forward and backward information of the sequence, providing a more complete representation of temporal features. The Dropout layer (dropout rate 0.3) set between the two layers of LSTM effectively prevents overfitting by randomly inactivating some neurons.

[0077] The fully connected layer adopts a three-layer structure with 512, 256, and 128 nodes, respectively, to achieve gradual dimensionality reduction and abstraction of features. The BatchNormalization layer added before the last layer ensures stable feature distribution, which is beneficial for subsequent fault diagnosis. This progressive dimensionality reduction design not only retains important feature information, but also reduces computational complexity.

[0078] S4: When an abnormal operation of the CNC machine tool is detected, a fault state prediction is performed based on the target machine tool dynamic model and the model prediction control method to obtain a fault prediction result.

[0079] This step mainly solves the problem of fault prediction under abnormal machine tool operation. When the monitoring system detects that the machine tool operating parameters deviate from the normal range, the system immediately starts the fault state prediction process based on the dynamic model of the target machine tool. This process combines the prediction ability of the deep learning model with the optimization characteristics of the model predictive control, which can not only evaluate the current fault state, but also predict the development trend of the fault. The prediction process makes full use of historical data and real-time monitoring data, and realizes continuous state evaluation and prediction update through the sliding window mechanism.

[0080] Combination Figure 1 and Figure 4 , S4 specifically includes the following steps: S4.1: Based on the target machine tool dynamic model, a state predictor of a recursive prediction structure is constructed, wherein the state predictor uses 20 sampling points as the time window length of a prediction unit.

[0081] Based on the previously trained target machine tool dynamic model, a state predictor with a recursive prediction structure was constructed. The predictor uses 20 sampling points as the time window length of the basic prediction unit. The selection of this length is based on a comprehensive consideration of multiple factors: First, considering the characteristic evolution cycle of common machine tool faults, the time span of 20 sampling points is sufficient to capture the initial characteristics of most fault modes; second, the window length achieves a good balance between computational complexity and prediction accuracy, which can ensure the timeliness of prediction without introducing too much historical noise due to the long window. The recursive structure of the predictor allows new observation data to continuously update the prediction basis, thereby achieving continuous tracking of the machine tool status.

[0082] S4.2: Through the state predictor, an adaptive rolling time domain prediction strategy is designed, wherein the rolling time domain prediction strategy includes short-term prediction of 3 sampling points, medium-term prediction of 6 sampling points and long-term prediction of 10 sampling points, and the prediction step size is dynamically adjusted according to the working condition of the CNC machine tool.

[0083] The rolling time domain prediction strategy adopts a multi-scale prediction scheme and designs three prediction levels for prediction needs at different time scales. Short-term prediction (3 sampling points) focuses on the immediate evolution of the fault, with a short sampling interval of usually 50-100ms, which can capture sudden anomalies in a timely manner; medium-term prediction (6 sampling points) focuses on the development trend of the fault, and the sampling interval can be appropriately extended to 200-300ms, which helps to identify gradual faults; long-term prediction (10 sampling points) is used to evaluate the potential impact of the fault, and the sampling interval can reach more than 500ms. The prediction step will be dynamically adjusted according to the working conditions of the machine tool. For example, the prediction step can be shortened to 70% of the baseline value during high-speed cutting, and can be extended to 130% of the baseline value during low-speed stable processing to adapt to the evolution speed of fault characteristics under different working conditions.

[0084] S4.3: Based on the rolling time domain prediction strategy, a multi-step prediction is performed on the future state of the CNC machine tool, the output result of the prediction strategy is obtained, and the deviation between the predicted state and the normal operating state of the CNC machine tool is calculated in real time to evaluate the fault risk level.

[0085] Based on the rolling time domain prediction strategy, the system makes continuous multi-step predictions of the future state of the machine tool. The prediction process not only outputs the predicted values ​​of the state parameters at each time point, but also calculates the deviation between these predicted states and the normal operating state of the machine tool in real time. The deviation calculation adopts weighted Euclidean distance, and different weights are set for different parameters. For example, the weight of vibration characteristics is 0.4, temperature characteristics is 0.3, and current characteristics is 0.3. The system divides the fault risk into four levels according to the size of the deviation: when the deviation is less than the threshold T1 (usually set to 1.2 times the normal value), it is a safe state, between T1 and T2 (1.5 times) is a warning state, between T2 and T3 (2 times) is a warning state, and greater than T3 is a dangerous state. This real-time evaluation mechanism can timely discover potential fault risks and provide decision support for preventive maintenance.

[0086] In addition, in one of the embodiments, the deviation between the real-time calculation prediction state and the normal operation state of the CNC machine tool in S4.3 includes: A1. Build a CNC machine tool health status benchmark library containing the normal operating characteristic distribution under different cutting parameters.

[0087] The construction of the health status benchmark library is a systematic process that requires the collection and organization of normal operation data of machine tools under different cutting conditions. The benchmark library is classified according to different combinations of cutting parameters, mainly including: speed range (usually divided into low speed 800-2000rpm, medium speed 2000-4000rpm, high speed above 4000rpm), feed speed (5 levels within the range of 0.1-0.5mm / rev), cutting depth (4 levels within the range of 0.5-2.5mm), etc.

[0088] For each parameter combination, the characteristic distribution during normal operation is collected and statistically analyzed, including vibration characteristics (mean, standard deviation, peak value, etc.), temperature characteristics (temperature at each measuring point and its rate of change), and current characteristics (effective value, harmonic components, etc.). These characteristics are modeled through probability density functions to form a standardized reference benchmark. The benchmark library is updated using a sliding window mechanism and is updated every fixed time (such as one month) to adapt to the slow changes in machine tool performance.

[0089] A2: Calculate the Mahalanobis distance between the predicted state sequence and the corresponding working condition reference state in the reference library to obtain the state deviation value.

[0090] This step uses the Mahalanobis distance to calculate the deviation between the predicted state sequence and the reference state. Compared with the Euclidean distance, the Mahalanobis distance takes into account the correlation between features and is more suitable for processing distance metrics in multidimensional feature spaces. In the specific calculation process, the reference state distribution corresponding to the current working condition is first extracted from the reference library, and then the standardized distance of the predicted state sequence relative to the reference distribution is calculated. The calculation formula is: D = sqrt((x-μ)^T Σ^(-1) (x-μ)), where x is the predicted state vector, μ is the reference state mean vector, and Σ is the reference state covariance matrix. The state deviation value obtained in this way can more accurately reflect the degree of difference between the current state and the normal state.

[0091] A3: Perform linear regression analysis on the state deviation value to obtain deviation change trend characteristics.

[0092] Linear regression analysis is performed on the state deviation value sequence, using a sliding window mechanism, and the window size is usually set to 100 sampling points. The regression analysis uses the least squares method to calculate the regression coefficient β and the intercept α. The sign and size of the regression coefficient β reflect the trend of the deviation: a positive value indicates that the fault may be getting worse, and a negative value indicates that the state may be improving; the absolute value of the coefficient reflects the speed of the change. At the same time, the goodness of fit R² of the regression equation is calculated to evaluate the reliability of trend prediction. When R² is lower than 0.6, the system will reduce the window size and re-perform regression analysis to improve the accuracy of trend characteristics.

[0093] A4: Train the state transfer matrix based on historical fault data, and predict the fault development trend by combining the current state deviation value and deviation change trend.

[0094] The training of the state transfer matrix is ​​based on a large amount of historical fault data, and the Markov chain model is used to describe the transition probability between different fault states.

[0095] The training process first classifies the historical fault data according to the severity, counts the conversion frequencies between various fault states, and then obtains the transition probability matrix by frequency normalization. For example, for a certain fault state i, the probability of transitioning to state j is Pij = Nij / Ni, where Nij is the number of samples transferred from state i to state j, and Ni is the total number of samples in state i. Combined with the currently observed state deviation value and deviation change trend, the state transition matrix can be used to predict possible future fault states and their probabilities.

[0096] A5: According to the state deviation value, deviation change trend and fault development trend, the fuzzy comprehensive evaluation method is used to divide the operation status of the CNC machine tool into four levels: normal, slight abnormality, moderate fault and severe fault.

[0097] The fuzzy comprehensive evaluation method is based on three evaluation factors: state deviation value U1, deviation change trend U2 and fault development trend U3. First, the evaluation factor set U={U1,U2,U3} and rating set V={normal, slightly abnormal, moderate fault, severe fault}.

[0098] Then, the fuzzy relationship matrix R is constructed. Each element rij in the matrix represents the membership of the i-th factor to the j-th level. The weight vector A is determined by the hierarchical analysis method. Usually, the weight of the state deviation value is large (0.4-0.5), and the weight of the deviation change trend and the fault development trend is relatively small (0.2-0.3). Finally, the comprehensive evaluation result is obtained through the fuzzy synthesis operation B=A○R, and the level with the largest membership is selected as the evaluation result of the current state.

[0099] In addition, the state predictor includes the following feedback correction mechanism: The error tracker computes the sequence of deviations between the predicted values ​​and the actual observed values.

[0100] The adaptive weight regulator dynamically adjusts the correction weight in the range of 0.6 to 1.0 according to the prediction accuracy, where the prediction accuracy is calculated by the mean square error between the predicted value and the actual value.

[0101] The state corrector performs real-time correction on the prediction result based on the correction weight.

[0102] Specifically, the feedback correction mechanism of the state predictor contains three core components. The error tracker calculates the deviation sequence between the predicted value and the actual observed value through a sliding window. The window size is usually set to 20 sampling points. The deviation sequence includes not only the absolute error, but also the relative error and trend error. These error information are used to evaluate the accuracy of the prediction. The adaptive weight regulator dynamically adjusts the correction weight according to the prediction accuracy. ACC = 1 - RMSE / Range,

[0103] Where RMSE is the root mean square error, and Range is the range of the observed value. When ACC>0.9, the correction weight is close to 1.0; when ACC<0.7, the correction weight is close to 0.6. The state corrector makes real-time corrections to the prediction results according to the correction weight w. The corrected prediction value is equal to the weighted average of the original prediction value and the actual observation value: x_corrected = w*x_predicted + (1-w)*x_observed.

[0104] This adaptive correction mechanism can dynamically adjust the prediction strategy according to changes in prediction performance and improve the stability and reliability of the prediction.

[0105] S5: Based on the fault prediction result, a Monte Carlo tree search algorithm is used to construct a fault evolution path, and a fault diagnosis result is output.

[0106] In this step, the Monte Carlo Tree Search (MCTS) algorithm is used to construct the evolution path of machine tool faults.

[0107] MCTS is a sampling-based search algorithm that is particularly suitable for handling decision-making problems with large state spaces. In machine tool fault diagnosis, the algorithm can effectively identify the most likely fault path by repeatedly simulating different fault evolution possibilities. The search process includes four main steps: selection, expansion, simulation, and backtracking. Each iteration can improve the evaluation of the fault evolution path.

[0108] S5 specifically includes: S5.1: Establishing an initial state node according to the fault prediction result, wherein the attribute information of the initial state node represents the feature vector, fault risk level and confidence of the current state of the CNC machine tool.

[0109] An initial state node is established based on the fault prediction results. The node contains three types of key attribute information: feature vector, fault risk level and confidence.

[0110] The feature vector is composed of the current monitoring indicators of the machine tool, including vibration characteristics (such as root mean square value, peak factor, spectrum characteristics, etc.), temperature characteristics (such as temperature values ​​at key points and their rate of change), current characteristics (such as effective value, harmonic content, etc.) and other multi-dimensional data. The fault risk level is divided into four levels (normal, slight, moderate, severe), with a specific numerical score (0-100 points). The confidence level reflects the credibility of the current state judgment, which is comprehensively evaluated through historical data verification and expert experience, and the value range is 0-1.

[0111] S5.2: Establishing a state transfer rule base based on the node, the state transfer rule base includes: component-level fault transfer rules, system-level fault transfer rules and environmental factor influence rules, wherein each rule is configured with a corresponding transfer probability and time constraint.

[0112] The state transition rule base contains three types of rules: component-level fault transition rules describe the fault evolution law within a single component, such as the evolution process of a bearing from normal to slight wear, moderate wear and severe wear. Each stage has a corresponding characteristic indicator range and transition conditions; system-level fault transition rules describe the fault association and propagation mechanism between multiple components, such as the association rule that abnormal spindle vibration leads to increased tool wear; environmental factor influence rules consider the influence of external factors such as temperature, humidity, and dust on fault evolution. Each rule is configured with a transition probability (based on historical data statistics) and a time constraint (including the shortest transition time and the longest duration). For example, the transition probability of a bearing from slight wear to moderate wear is 0.3, the shortest time is 24 hours, and the longest time is no more than 72 hours.

[0113] S5.3: Based on the state transition rule base, a heuristic method is used to dynamically determine the expansion depth, wherein a maximum expansion depth of 8 layers is used for important fault paths, and a maximum expansion depth of 4 layers is used for minor fault paths.

[0114] A heuristic method is used to dynamically determine the expansion depth of the search tree.

[0115] For important fault paths that may cause machine tool downtime or major losses (such as spindle system failure, transmission system failure, etc.), the maximum expansion depth of 8 layers is used, each layer represents a key state transition point, which can predict the fault evolution process over a long period of time. For minor fault paths with relatively small impacts (such as minor abnormalities in the lubrication system, reduced efficiency of the cooling system, etc.), the maximum expansion depth of 4 layers is used to balance the computing efficiency. The dynamic adjustment of the expansion depth is based on the fault level of the current node (8 layers are used for severity greater than 70 points), risk assessment results (8 layers are used for potential losses exceeding 30% of the equipment value) and computing resource limitations.

[0116] S5.4: Use the improved UCT algorithm to perform node selection and expansion to determine the fault evolution path.

[0117] The improved UCT (Upper Confidence Bound for Trees) algorithm introduces a fault risk weight factor in the node selection and expansion process. When selecting nodes, the number of node visits, cumulative return value, and risk weight are comprehensively considered, and the node with the highest score is selected for expansion. Specifically, the node selection score calculation formula is: Score = Q / N + c*sqrt(ln(N_parent) / N) + w*R,

[0118] Where Q represents the cumulative value of the node, N represents the number of visits, c is the exploration coefficient (usually set to 1.414), w is the risk weight (dynamically adjusted according to the fault level, ranging from 0.1-0.5), and R is the node risk coefficient. The node expansion process strictly follows the rules in the state transition rule base, generates child nodes through probability sampling, and dynamically adjusts the transition probability based on historical data.

[0119] In one embodiment, S5.4 specifically includes: B1. Calculate the node score: Score = Q / N + c*sqrt(ln(N_parent) / N) + w*R Where Q represents the cumulative value of the node, N represents the number of node visits, N_parent represents the number of parent node visits, c represents the exploration coefficient, R represents the node risk coefficient, and w represents the risk weight.

[0120] The node score calculation adopts the improved UCB formula, which contains three key components: the value utilization term (Q / N) reflects the average benefit of the node, the exploration term (csqrt(ln(N_parent) / N)) ensures that opportunities are given to nodes that are not fully explored, and the risk weighting term (wR) reflects the impact of the severity of the fault. Specifically, the Q value is accumulated through node simulation and ranges from 0 to 100; the N value records the access frequency of the node; the exploration coefficient c is usually set between 0.7 and 1.4, and is dynamically adjusted according to the search stage (larger in the early stage to promote exploration, and smaller in the later stage to enhance utilization); the node risk coefficient R is determined according to the fault level, ranging from 0 to 1; the risk weight w is set according to the severity of the fault, usually between 0.3 and 0.8.

[0121] B2: Generate child nodes based on the rules in the state transition rule base, and dynamically adjust the transition probability according to historical fault data during the expansion process.

[0122] The child node generation process strictly follows the rules in the state transition rule base and adopts a probabilistic sampling method. First, all possible transition rules are screened from the rule base based on the state characteristics of the current node. Then, sampling is performed based on the transition probabilities of these rules to generate multiple candidate child nodes. During the expansion process, the system dynamically adjusts the transition probability based on the recent fault data. For example, if a certain fault mode occurs frequently in the near future, the probability of the relevant transition rule will increase by 10%-30%. In addition, the system also considers time constraints to ensure that the generated child node state transition is within a reasonable time range.

[0123] B3: Use fast random simulation method to perform node simulation and record the cumulative risk value and time delay on the path; Fast random simulation uses a lightweight Monte Carlo method, starting from the current node, and randomly selects the successor state according to the state transition rule until the termination condition is reached (such as reaching the maximum depth or the termination state appears). During the simulation, the system records two key indicators: the cumulative risk value (the weighted sum of the risk coefficients of each node) and the time delay (the sum of the time required for state transition). To improve the efficiency of the simulation, a fast evaluation strategy is adopted, focusing only on the state transition of key nodes, and simplified calculations are used for intermediate states. Each node usually performs 100-200 simulations to obtain statistically reliable results.

[0124] B4: Starting from the leaf node, update the visit count and cumulative value of each node along the path upward.

[0125] The back-propagation process adopts a bottom-up update strategy. Starting from the leaf node, the statistical information of each node is updated upward along the search path. The number of visits N is simply increased, while the update of the cumulative value Q takes into account the risk level and time delay of the node. The specific update formula is: Q_new = Q_old + (R_value * T_discount), where R_value is the reward value obtained by simulation, and T_discount is the discount factor based on time delay. This update mechanism ensures that the search tree can gradually converge to the most likely and dangerous fault evolution direction.

[0126] The embodiment of the present application also includes: C1: Count the number of visits to each node in the fault evolution path, and identify the path whose visit number exceeds 30% of the total simulation times as a high-probability fault evolution path.

[0127] The system performs statistical analysis on all explored fault evolution paths and calculates the access frequency of each node. Paths with access times exceeding 30% of the total simulation times are identified as high-probability fault evolution paths. The selection of this threshold is based on a large amount of experimental data and can effectively filter out the most likely fault evolution sequence. At the same time, the system also considers the risk of the path. For paths with access frequencies close to the threshold (25%-30%) but with higher risks, they will also be included in the high-probability path set.

[0128] C2: Generate a hierarchical fault diagnosis report based on the high-probability fault evolution path, including a description of the current fault status, a potential fault evolution warning, and corresponding maintenance suggestions.

[0129] The hierarchical fault diagnosis report adopts a three-layer structure: the first layer is the description of the current fault status, which lists the abnormal indicators, fault levels and confidence levels in detail; the second layer is the early warning of potential fault evolution, which predicts the possible fault development in the next 24 hours, 72 hours and one week based on the high-probability fault path; the third layer is maintenance suggestions, which provide specific maintenance plans for different fault evolution situations, including recommended maintenance time, required spare parts and operation steps. The report adopts a structured format to facilitate system analysis and manual reading.

[0130] C3: The fault evolution tree structure is displayed in a visual way. Different colors are used to identify the fault level. The node size indicates the state transition probability, and the thickness of the connecting edge indicates the path access frequency.

[0131] The visualization of the fault evolution tree uses a force-directed graph layout algorithm to ensure the spatial efficiency and visual beauty of node distribution. Nodes use different colors to represent fault levels (green for normal, yellow for minor faults, orange for moderate faults, and red for serious faults), and the size of the node is proportional to the probability of state transition (the greater the probability, the larger the node). The thickness of the connecting edge indicates the frequency of path visits. The more times the path is visited, the thicker the edge. In addition, the system also provides interactive operation functions, allowing users to click on nodes to view detailed information, or adjust the view layout by dragging and dropping.

Claims

1. An intelligent fault diagnosis method for a CNC machine tool, applied to the control system of the CNC machine tool, characterized in that the method include: Collecting multi-source sensor data during the operation of the CNC machine tool, wherein the multi-source sensor data includes at least two of a vibration signal, a temperature signal, a current signal, and a sound signal; Preprocessing and feature extraction are performed on the multi-source sensor data to obtain preprocessed data; Based on the preprocessed data, a machine tool dynamic model is constructed and trained to obtain a target machine tool dynamic model; When an abnormal operation of a CNC machine tool is detected, a fault state prediction is performed based on the target machine tool dynamic model and the model predictive control method to obtain a fault prediction result; Based on the fault prediction result, a Monte Carlo tree search algorithm is used to construct a fault evolution path, and a fault diagnosis result is output.

2. The diagnostic method according to claim 1, characterized in that The method of constructing and training a machine tool dynamic model based on the preprocessed data to obtain a target machine tool dynamic model includes: Constructing a deep learning model architecture, wherein the deep learning model architecture includes a convolutional neural network (CNN) for spatial feature extraction and a bidirectional long short-term memory (LSTM) network for temporal feature processing, and a fully connected layer for fusing the features extracted by the CNN and the LSTM; Based on the outputs of the CNN and the LSTM, multiple groups of loss functions are designed, wherein the loss functions include: a prediction error loss for measuring prediction accuracy and an L2 regularization constraint for preventing model overfitting, wherein different loss weights are set for different types of fault samples; According to the calculation result of the loss function, the Adam optimizer is used for model training, and the learning rate is dynamically adjusted based on the performance index of the validation set. When the loss value of the validation set does not decrease for three consecutive training cycles, the learning rate is adjusted to 0.8 times the original value; And adopt a model evaluation mechanism based on K-fold cross-validation to determine the dynamic model of the target machine tool, including: in each fold training, use a stratified sampling strategy to divide the data set to ensure the balanced distribution of samples of different fault types, and evaluate the model performance by counting the accuracy, precision and recall rate on each fold validation set, and select the optimal model parameter configuration.

3. The diagnostic method according to claim 2, characterized in that In the deep learning model architecture: The CNN includes a three-layer convolutional layer structure connected in sequence. The first layer uses 32 3×3 convolution kernels for feature extraction and normalization through the BatchNormalization layer. The second layer uses 64 5×5 convolution kernels to expand the receptive field and enhance the nonlinear expression ability through the ReLU activation function. The third layer reduces the dimension through a 2×2 maximum pooling window. The LSTM includes two bidirectional LSTM layers, each layer contains 128 neurons, and a Dropout layer with a dropout rate of 0.3 is set between the two layers; The fully connected layer includes three layers, and the corresponding numbers of nodes are 512, 256 and 128 respectively, which are used to fuse spatial features and temporal features, and a BatchNormalization layer is added before the last layer to ensure stable feature distribution.

4. The diagnostic method according to claim 1, characterized in that: When the CNC machine tool is detected to be operating abnormally, a fault state prediction is performed based on the target machine tool dynamic model and the model prediction control method to obtain a fault prediction result, including: Based on the target machine tool dynamic model, a state predictor of a recursive prediction structure is constructed, wherein the state predictor uses 20 sampling points as the time window length of a prediction unit; By means of the state predictor, an adaptive rolling time domain prediction strategy is designed, wherein the rolling time domain prediction strategy includes short-term prediction of 3 sampling points, medium-term prediction of 6 sampling points and long-term prediction of 10 sampling points, and the prediction step size is dynamically adjusted according to the working condition of the CNC machine tool; Based on the rolling time domain prediction strategy, a multi-step prediction is performed on the future state of the CNC machine tool, the output result of the prediction strategy is obtained, and the deviation between the predicted state and the normal operating state of the CNC machine tool is calculated in real time to evaluate the fault risk level.

5. The diagnostic method according to claim 4, characterized in that: The deviation between the real-time calculation prediction state and the normal operation state of the CNC machine tool includes: Construct a CNC machine tool health status benchmark library containing the normal operating characteristic distribution under different cutting parameters; Calculate the Mahalanobis distance between the predicted state sequence and the corresponding working condition reference state in the reference library to obtain a state deviation value; Performing linear regression analysis on the state deviation value to obtain deviation change trend characteristics; Based on the historical fault data, the state transition matrix is ​​trained, and the fault development trend is predicted by combining the current state deviation value and deviation change trend; According to the state deviation value, deviation change trend and fault development trend, the fuzzy comprehensive evaluation method is adopted to divide the operation state of the CNC machine tool into four levels: normal, slight abnormality, moderate fault and severe fault.

6. The diagnostic method according to claim 4, characterized in that: The state predictor includes the following feedback correction mechanism: The error tracker computes the sequence of deviations between the predicted values ​​and the actual observed values; An adaptive weight regulator dynamically adjusts the correction weight in the range of 0.6 to 1.0 according to the prediction accuracy, where the prediction accuracy is calculated by the mean square error between the predicted value and the actual value; The state corrector performs real-time correction on the prediction result based on the correction weight.

7. The diagnostic method according to claim 1, characterized in that: The method of constructing a fault evolution path based on the fault prediction result by using a Monte Carlo tree search algorithm includes: According to the fault prediction result, an initial state node is established, wherein the attribute information of the initial state node represents the feature vector, fault risk level and confidence of the current state of the CNC machine tool; A state transfer rule base is established based on the node, wherein the state transfer rule base includes: component-level fault transfer rules, system-level fault transfer rules, and environmental factor influence rules, wherein each rule is configured with a corresponding transfer probability and time constraint; Based on the state transition rule base, a heuristic method is used to dynamically determine the expansion depth, wherein a maximum expansion depth of 8 layers is used for important fault paths, and a maximum expansion depth of 4 layers is used for minor fault paths; The improved UCT algorithm is used to perform node selection and expansion to determine the fault evolution path.

8. The diagnostic method according to claim 7, characterized in that: The improved UCT algorithm is used to select and expand nodes to determine the fault evolution path, including: Calculate the node score: Score = Q / N + c*sqrt(ln(N_parent) / N) + w*R, Where Q represents the cumulative value of the node, N represents the number of node visits, N_parent represents the number of parent node visits, c represents the exploration coefficient, R represents the node risk coefficient, and w represents the risk weight; Generate child nodes based on the rules in the state transition rule base, and dynamically adjust the transition probability according to historical fault data during the expansion process; Perform node simulation using fast stochastic simulation method and record the cumulative risk value and time delay on the path; Starting from the leaf node, update the visit count and cumulative value of each node along the path upward.

9. The diagnostic method according to claim 7, characterized in that: Also do at least one of the following: Counting the number of visits to each node in the fault evolution path, and identifying a path whose number of visits exceeds 30% of the total simulation times as a high-probability fault evolution path; Generate a hierarchical fault diagnosis report based on the high-probability fault evolution path, including a description of the current fault state, a potential fault evolution warning, and corresponding maintenance suggestions; The fault evolution tree structure is displayed in a visual way. Different colors are used to identify the fault level. The node size indicates the state transition probability, and the thickness of the connecting edge indicates the path access frequency.

Images