Fault prediction method and device for numerical control machine tool

Abstract

The invention discloses a fault prediction method and device for a numerical control machine tool, and the method and device have the following advantages: collecting various types of data, including multi-source process parameter data and processing image data, comprehensively covering all aspects of the operation of the numerical control machine tool, fully mining the data value through an innovative data preprocessing and feature extraction method, and improving the fault prediction efficiency of the numerical control machine tool. The operation state of the machine tool can be described more accurately, and fault prediction accuracy is improved; a unique dynamic space-time attention network architecture can adaptively focus key space-time characteristics, capture dynamic association between data and early-stage weak symptoms of faults, and is more sensitive to perception of fault characteristics, so that the prediction performance is improved; fusing fault probabilities: fusing the two output probabilities to obtain a result, and keeping stable prediction precision under different machine tools and working conditions; and multi-stage early warning and decision support: multi-stage fault early warning and detailed fault reason analysis reports and maintenance suggestions provide all-around decision support for enterprises, and reduce production loss and maintenance cost.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of numerical control machine tool maintenance, and particularly relates to a fault prediction method and device for a numerical control machine tool. BACKGROUND

[0002] In modern manufacturing, numerical control machine tools, as core production equipment, directly affect production efficiency and product quality in terms of running stability. The existing fault prediction methods mainly have the following deficiencies: firstly, most of them rely on single type sensor data, such as only using vibration signals or temperature data for fault judgment, which cannot comprehensively reflect the complex running state of numerical control machine tools, resulting in poor accuracy of fault prediction; secondly, some methods use machine learning algorithms, which need a large amount of manual feature extraction, and are not only time-consuming and laborious, but also have weak model generalization ability and are difficult to adapt to different working conditions and machine tool types; thirdly, the existing methods cannot fully mine the dynamic correlation between multi-source data, cannot timely capture the weak signs of early faults, and are not timely enough in prediction, which makes it difficult to effectively avoid production losses caused by sudden failures. SUMMARY

[0003] Therefore, it is necessary to provide a fault prediction method and device for a numerical control machine tool, which fuses multi-modal data and uses an innovative deep learning model to deeply mine data features and internal relations, realizes high-precision and forward-looking prediction of faults of the numerical control machine tool, improves equipment running reliability, and reduces maintenance cost and production interruption risk.

[0004] In a first aspect, an embodiment of the present application provides a fault prediction method for a numerical control machine tool, comprising the following steps: S1: acquiring multi-source modal data of a monitored numerical control machine tool in a running process, wherein the multi-source modal data comprises multi-source running data and machining image data of the monitored numerical control machine tool in a monitoring period; the multi-source running data at least comprises one or more of running signal data, working parameter data and historical maintenance data; S2: pre-processing the multi-source running data and the machining image data respectively to remove noise data and abnormal data therefrom; S3: normalizing and extracting features from the pre-processed multi-source running data to obtain multi-source modal feature data; performing image recognition and feature extraction on the pre-processed machining image data to obtain machining image feature data; S4: inputting the multi-source modal feature data into a pre-trained multi-source fault prediction model to output a first fault prediction probability distribution; inputting the machining image feature data into a pre-trained image fault prediction model to output a second fault prediction probability distribution; S5: input the first fault prediction probability distribution and the second fault prediction probability distribution into a fault prediction fusion model to obtain a fusion fault prediction probability of the numerical control machine tool; S6: if the fusion fault prediction probability exceeds a fault warning threshold, issue a fault warning signal, and provide a fault cause and a maintenance suggestion according to historical maintenance data and a current running state of the numerical control machine tool; wherein the fault prediction fusion model is: (1); is a fusion fault prediction probability, is a first fault prediction average probability of the first fault prediction probability distribution, is a second fault prediction average probability of the second fault prediction probability distribution, is a weight coefficient, and .

[0005] Preferably, the processing image data includes multiple processing region images obtained by multi-angle acquisition of the monitored numerical control machine tool.

[0006] Preferably, the normalization and feature extraction of the preprocessed multi-source running data to obtain multi-source modal feature data include: S311: normalizing the preprocessed multi-source running data to obtain multi-source normalized data; S312: performing feature extraction on the multi-source normalized data to obtain multi-source modal feature data.

[0007] Preferably, the image enhancement and normalization of the preprocessed processing image data to obtain processing image feature data include: S321: image enhancement and splicing of the preprocessed processing image data to obtain intermediate image data; S322: normalizing pixel values of the intermediate image data to obtain processing image feature data.

[0008] Preferably, the multi-source fault prediction model is a dynamic spatio-temporal attention network model, including a multi-modal data input layer, a parallel feature extraction layer, a dynamic spatio-temporal attention layer, a time series fusion layer, and a fault prediction output layer; the inputting of the multi-source modal feature data into the pre-trained multi-source fault prediction model to output a first fault prediction probability distribution includes: S411: the multi-modal data input layer includes multiple multi-modal data input sub-layers, and multiple multi-modal data input sub-layers receive different types of data sources of the multi-modal feature data; S412: the parallel feature extraction layer includes a plurality of feature extraction sub-networks, and the plurality of feature extraction sub-networks receive corresponding types of feature data and perform feature extraction to obtain a plurality of preliminary feature vectors; S413: the dynamic spatiotemporal attention layer first splices the input plurality of preliminary feature vectors to obtain a fusion feature vector; then calculates a time attention weight matrix and a space attention weight matrix through a time attention module and a space attention module, respectively; and performs weighted fusion on the fusion feature vector based on the time attention weight matrix and the space attention weight matrix to obtain a spatiotemporal attention feature; S414: the time sequence fusion layer performs time sequence modeling on the spatiotemporal attention feature and inputs a fault prediction output layer; S415: the spatiotemporal attention feature after time sequence modeling is input into a full connection layer and a softmax function of the fault prediction output layer to obtain a first fault prediction probability distribution of each type of fault of the numerical control machine tool occurring in the monitoring period.

[0009] Preferably, the processing image feature data is input into a pre-trained image fault prediction model to output a second fault prediction probability distribution, including: S421: the processing image feature data is formatted according to the input requirement of the image fault prediction model; S422: the processed processing image feature data is input into the pre-trained image fault prediction model to obtain a second fault prediction probability distribution of each type of fault of the numerical control machine tool occurring in the monitoring period.

[0010] Preferably, the multi-source fault prediction model is obtained by pre-training in the following manner: The historical multi-source modal data of the numerical control machine tool in the historical running process is collected, and the historical multi-source modal data is divided into a training set and a test set according to a preset proportion; The training set data is used to train the dynamic spatiotemporal attention network model to obtain a trained multi-source fault prediction model; The test set data is used to test the multi-source fault prediction model to obtain a test result, and the multi-source fault prediction model is optimized and adjusted by using the test result.

[0011] Preferably, the image fault prediction model is a deep learning architecture for image fault prediction, and is obtained by pre-training in the following manner: The training parameters and loss function of the image fault prediction model are determined; The historical processing image data of the numerical control machine tool in the historical running process is collected, and the historical processing image data is preprocessed; The preprocessed historical machining image data is divided into a training set and a test set according to a preset ratio; The training set data is used to train a deep learning architecture of image fault prediction, and a trained image fault prediction model is obtained; The test set data is used to test the image fault prediction model to obtain a test result, and the image fault prediction model is optimized and adjusted using the test result.

[0012] In a second aspect, the embodiments of the present application provide a fault prediction device of a numerical control machine tool, comprising: A data acquisition module is configured to acquire multi-source modal data of a monitored numerical control machine tool during operation, wherein the multi-source modal data comprises multi-source operation data and machining image data of the monitored numerical control machine tool within a monitoring period; and the multi-source operation data comprises one or more of operation signal data, working parameter data and historical maintenance data. A data preprocessing module is configured to preprocess the multi-source operation data and the machining image data respectively to remove noise data and abnormal data therefrom. A feature data acquisition module is configured to normalize and extract features from the preprocessed multi-source operation data to obtain multi-source modal feature data, and perform image enhancement and normalization on the preprocessed machining image data to obtain machining image feature data. A probability distribution calculation module is configured to input the multi-source modal feature data into a pre-trained multi-source fault prediction model to output a first fault prediction probability distribution, and input the machining image feature data into a pre-trained image fault prediction model to output a second fault prediction probability distribution. A probability fusion calculation module is configured to input the first fault prediction probability distribution and the second fault prediction probability distribution into a fault prediction fusion model to obtain a fusion fault prediction probability of the numerical control machine tool. A fault warning module is configured to output a fault warning signal if the fusion fault prediction probability exceeds a fault warning threshold, and provide fault causes and repair suggestions according to historical maintenance data and a current operating state of the numerical control machine tool. The fault prediction fusion model is as follows: (1) ; is the fusion fault prediction probability, is a first fault prediction average probability of the first fault prediction probability distribution, is a second fault prediction average probability of the second fault prediction probability distribution, is a weight coefficient, and .

[0013] Compared with the prior art, the present application has the following beneficial effects: (1) Multi-modal data deep fusion: The present application collects multiple types of data, including multi-source process parameter data (numerical data) and machining image data (non-numerical data), which comprehensively covers all aspects of the operation of a numerical control machine tool. Through innovative data preprocessing and feature extraction methods, the data value is fully tapped, and compared with traditional single data or simple data fusion methods, the operation state of the machine tool can be more accurately described, and the accuracy of fault prediction is greatly improved.

[0014] (2) Dynamic spatio-temporal attention network innovation: The unique dynamic spatio-temporal attention network architecture, especially the double-channel attention mechanism, can adaptively focus on key spatio-temporal features, effectively capture complex dynamic correlations between data and early weak signs of faults, make the model more sensitive to fault features, and improve the prediction performance and adaptability of the model to complex working conditions. (3) Fusion of fault probability results: The output probabilities of the two fault prediction models are fused to obtain a fused fault probability result, which can improve the prediction generalization ability and make the model maintain stable prediction accuracy under different machine tools and different working conditions.

[0015] (4) Multi-level early warning and decision support: Multi-level fault early warning and detailed fault cause analysis report and maintenance suggestion provide comprehensive decision support for enterprises. Enterprises can reasonably arrange production plans and maintenance work according to the early warning level, prevent faults from occurring in advance, reduce equipment downtime, reduce production losses and maintenance costs, and have significant economic benefits and practical value. BRIEF DESCRIPTION OF DRAWINGS

[0016] The exemplary embodiments of the present application can be more completely understood by reference to the following drawings. The drawings are included to provide a further understanding of the embodiments of the present application and are incorporated in and constitute a part of this specification, illustrate embodiments of the present application and serve to explain the present application, and do not limit the present application. In the drawings, the same reference numerals generally refer to the same parts or steps throughout.

[0017] Figure 1 A flowchart of a fault prediction method for a numerical control machine tool according to an exemplary embodiment of the present application is provided. Figure 2 A schematic diagram of a fault prediction device for a numerical control machine tool according to an exemplary embodiment of the present application is provided. DETAILED DESCRIPTION

[0018] Exemplary embodiments of the present disclosure will be described in greater detail below with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it is understood that the present disclosure can be embodied in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided so that the present disclosure can be more thoroughly understood, and so that the full scope of the present disclosure is conveyed to those skilled in the art.

[0019] In the description of the present application, it should be noted that the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, and are only for the purpose of facilitating the description of the present application and simplifying the description, and do not indicate or imply that the device or element referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as limiting the present application. In addition, the terms "first", "second", "third" are only for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0020] In addition, the technical features involved in the different embodiments of the present application described below can be combined with each other as long as there is no conflict between them.

[0021] Embodiment 1 With reference to Figure 1 The present embodiment discloses a fault prediction method of a numerical control machine tool, characterized in that it comprises the following steps: S1: acquiring multi-source modal data of the monitored numerical control machine tool in the running process, the multi-source modal data comprising multi-source running data and machining image data of the monitored numerical control machine tool in the monitoring period; the multi-source running data comprising one or more of running signal data, working parameter data and historical maintenance data; Specifically, at key positions of the numerical control machine tool, such as the main shaft, guide rail, servo motor, tool magazine, etc., multiple types of sensors are deployed, including vibration sensors, temperature sensors, current sensors, voltage sensors, sound sensors and strain gauge sensors, etc., to collect vibration signals, temperature data, current and voltage waveforms, running sound audio and mechanical structure strain data in real time. At the same time, the working parameters of the numerical control machine tool are acquired, such as machining process parameters (cutting speed, feed rate, cutting depth), machine tool running mode parameters (automatic / manual mode, coordinate axis linkage state), and the historical maintenance data of the equipment is integrated, covering fault occurrence time, fault type, repair record, component replacement information, etc.

[0022] Specifically, the image acquisition process can be: in the processing of the numerical control machine tool, the processing image is acquired in real time by using the industrial camera installed at a suitable position. The resolution, frame rate and shooting angle of the camera are ensured to meet the requirement of accurately obtaining the details of the processing image, so as to capture the key features related to the fault. The processing image data includes multiple processing region images obtained by multi-angle acquisition of the monitored numerical control machine tool.

[0023] S2: preprocessing the multi-source operation data and the processing image data respectively to remove noise data and abnormal data therefrom; Specifically, different preprocessing methods can be used for different data types, for example: an improved adaptive median filtering algorithm is used to denoise the vibration data and temperature data, which can dynamically adjust the filtering window size according to the local characteristics of the data, effectively removing noise while preserving signal details; an abnormal value detection algorithm based on density clustering is used to process abnormal values of current and voltage data.

[0024] The multi-modal data is normalized by using a standardization normalization formula: (2) ; wherein, is the data mean, is the standard deviation, is the normalized data, and formula (2) processes each type of data to make different types of data comparable.

[0025] In the preferred embodiment, the image preprocessing can include the following processes: (1) denoising: a filtering algorithm (such as median filtering, Gaussian filtering, etc.) is used to denoise the collected image, remove noise interference in the image, improve image quality, and make subsequent feature extraction more accurate.

[0026] (2) image enhancement: necessary image enhancement operations are performed again to further highlight the key information in the image and enhance the contrast and clarity of the image.

[0027] (3) cropping and scaling: according to the requirements of the model input, the image is cropped and scaled to adjust the image to a fixed size, ensuring that the image format input into the model is uniform.

[0028] S3: normalizing and feature extracting the preprocessed multi-source operation data to obtain multi-source modal feature data; image recognizing and feature extracting the preprocessed processing image data to obtain processing image feature data; Specifically, the preprocessed multi-source operation data is feature extracted to obtain multi-source modal feature data, including: S311: Normalize the pre-processed multi-source operation data to obtain multi-source normalized data; Specifically, this step is to eliminate the dimensional differences of various data, such as current data, voltage data, temperature data, and different physical quantities, so that the data are unified in scale, and the optimization deviation caused by the difference in feature value domain during model training is avoided. The normalization method adopted in this embodiment includes minimum-maximum normalization and Z-score standardization, wherein the minimum-maximum normalization is suitable for the scene where the data distribution has no obvious abnormal value, such as the steady-state data of motor operating current and voltage; the Z-score standardization is suitable for data with sudden interference or noise (such as machining vibration signals), which can convert the data into a standard normal distribution with a mean of 0 and a variance of 1. When used, normalization can be realized by selecting different data classes.

[0029] S312: Feature extraction is performed on the multi-source normalized data to obtain multi-source modal feature data.

[0030] Specifically, the feature extraction process for various data can be: using short-time Fourier transform to perform time-frequency analysis on vibration signals and sound audio to obtain time-frequency feature maps; for temperature, current and voltage data, trend features, fluctuation features and periodic features are extracted; for strain data, strain gradient and strain concentration coefficient are calculated; working parameter data and historical maintenance data are processed by one-hot encoding and embedding encoding to convert into numerical feature vectors.

[0031] Specifically, the pre-processed machining image data is subjected to image enhancement and normalization to obtain machining image feature data, including: S321: The pre-processed machining image data is subjected to image enhancement and splicing to obtain intermediate image data; S322: The pixel values of the intermediate image data are normalized to obtain machining image feature data.

[0032] In a preferred embodiment, the feature extraction of the image can include the following processes: (1) Preliminary feature extraction Using a pre-trained model to extract features: input the pre-processed machining image into an image feature extraction network that has been trained (which can be part of the pre-trained image fault prediction model described above, such as removing the last classification layer) to obtain feature data of the image. These feature data contain various semantic information and detailed features of the image, which can effectively represent the state of the machining image.

[0033] (2) Feature screening and dimension reduction (optional): If the extracted feature dimension is high, feature selection algorithms (such as chi-square test, mutual information, etc.) can be used to screen the features and remove redundant and irrelevant features; or use dimension reduction algorithms (such as principal component analysis PCA, linear discriminant analysis LDA, etc.) to reduce the feature dimension, reduce the amount of calculation, and improve the running efficiency and prediction performance of the model.

[0034] S4: input the multi-source modal feature data into the pre-trained multi-source fault prediction model to output a first fault prediction probability distribution; input the machining image feature data into the pre-trained image fault prediction model to output a second fault prediction probability distribution; Specifically, the multi-source fault prediction model is obtained by pre-training in the following manner: Collect historical multi-source modal data of the numerical control machine tool in the historical running process, and divide the historical multi-source modal data into a training set and a test set according to a preset ratio; Train the dynamic spatio-temporal attention network model using the training set data to obtain a trained multi-source fault prediction model; Test the multi-source fault prediction model using the test set data to obtain a test result, and optimize and adjust the multi-source fault prediction model using the test result.

[0035] In the preferred embodiment, the training process of the multi-source fault prediction model can also be as follows: Collect a large amount of numerical control machine tool running data, and divide them into a training set, a validation set, and a test set according to a ratio of 8:1:1.

[0036] An improved focal loss function (Focal Loss) is used as the optimization target, which can automatically adjust the loss weight according to the difficulty of the sample, effectively solving the problem of unbalanced fault data samples. The AdamW optimizer is used to train the model, the learning rate is dynamically adjusted during the training process, and the cosine annealing learning rate adjustment strategy is adopted to avoid the model falling into local optimum.

[0037] Periodically evaluate the model performance on the validation set, optimize the model structure by adjusting the hyperparameters such as network layer number, node number, and attention mechanism parameters, and improve the model generalization ability.

[0038] Specifically, the multi-source fault prediction model is a dynamic spatio-temporal attention network model, which includes a multi-modal data input layer, a parallel feature extraction layer, a dynamic spatio-temporal attention layer, a time series fusion layer, and a fault prediction output layer; the multi-source modal feature data is input into the pre-trained multi-source fault prediction model to output a first fault prediction probability distribution, which includes: S411: The multi-modal data input layer includes a plurality of multi-modal data input sub-layers, and the plurality of multi-modal data input sub-layers receive different types of data sources of the multi-modal feature data; S412: The parallel feature extraction layer includes a plurality of feature extraction sub-networks, which receive corresponding types of feature data and perform feature extraction to obtain a plurality of preliminary feature vectors; S413: The dynamic spatio-temporal attention layer first splices the plurality of preliminary feature vectors inputted to obtain a fusion feature vector; then calculates a time attention weight matrix and a space attention weight matrix through a time attention module and a space attention module respectively; and performs weighted fusion on the fusion feature vector based on the time attention weight matrix and the space attention weight matrix to obtain a spatio-temporal attention feature; S414: The time sequence fusion layer performs time sequence modeling on the spatio-temporal attention feature and inputs a fault prediction output layer; S415: The spatio-temporal attention feature after time sequence modeling is inputted into a full connection layer and a softmax function of the fault prediction output layer to obtain a first fault prediction probability distribution of the numerical control machine tool in the monitoring period.

[0039] Specifically, a new multi-level network architecture is designed in the embodiment, which includes a multi-modal data input layer, a parallel feature extraction layer, a dynamic spatio-temporal attention layer, a time sequence fusion layer, and a fault prediction output layer.

[0040] The multi-modal data input layer receives different types of data respectively, and the parallel feature extraction layer adopts a customized sub-network for feature extraction for each type of data, such as a two-dimensional convolutional neural network (2D-CNN) for extracting spatial features of a time-frequency spectrum of vibration data and a long short-term memory network (LSTM) for extracting time sequence features of time sequence data.

[0041] The dynamic spatio-temporal attention layer is the core innovative part of the model. This layer designs a double-channel attention mechanism to assign weights to features from the time dimension and the space dimension respectively. In the time dimension, the time attention weight is calculated according to the time sequence and the correlation between the data before and after, highlighting the features of key time points; in the space dimension, the space attention weight is determined according to the correlation between different modal data, strengthening the feature expression of important modal data. The specific calculation process is as follows: First, the features outputted by each parallel feature extraction layer are spliced to obtain initial fusion features ; then, a time attention weight matrix and a space attention weight matrix are calculated through a time attention module and a space attention module respectively; finally, the initial fusion features are weighted and fused to obtain a feature representation containing spatio-temporal attention: (3) ; The temporal fusion layer uses a Gated Recurrent Unit (GRU) to model the temporal characteristics of the features processed by the spatio-temporal attention, capturing the dynamic trends of the data over time. The final output layer uses a fully connected layer and a softmax function to output the probability distribution of each type of fault occurring in the CNC machine at different time intervals in the future.

[0042] Specifically, the machining image feature data is input into the pre-trained image fault prediction model, and a second fault prediction probability distribution is output, including: S421: Format the machining feature image data according to the input requirements of the image fault prediction model; S422: Input the processed machining feature image data into the pre-trained image fault prediction model to obtain a second fault prediction probability distribution of each type of fault occurring in the CNC machine within the monitoring period.

[0043] In a preferred embodiment, the process of obtaining the second fault prediction probability distribution can be: Input data: Format the processed and extracted machining image feature data according to the input requirements of the model and input it into the pre-trained image fault prediction model.

[0044] Model inference: After receiving the feature data, the model performs layer-by-layer calculation and analysis on the features through its internal network structure, and uses the learned fault feature patterns and rules to predict faults in the input machining image.

[0045] Output result: The model finally outputs a second fault prediction probability distribution, which represents the probability values of the input machining image belonging to each fault category. For example, the output probability distribution may be [0.1, 0.7, 0.2], corresponding to the probabilities of normal state, fault type A and fault type B, indicating that there is a 70% chance of fault type A in the machining image.

[0046] Specifically, the image fault prediction model is a deep learning architecture for image fault prediction, which is pre-trained in the following way: Determine the training parameters and loss function of the image fault prediction model; Collect historical machining image data of CNC machines during their historical operation, and pre-process the historical machining image data; Divide the pre-processed historical machining image data into a training set and a test set according to a pre-set ratio; Use the training set data to train the deep learning architecture for image fault prediction, and obtain the trained image fault prediction model; Test the image fault prediction model using the test set data to obtain test results, and optimize and adjust the image fault prediction model using the test results.

[0047] In a preferred embodiment, the training process of the image fault prediction model can be specifically as follows: Set training parameters: determine appropriate training parameters such as learning rate (dynamic learning rate adjustment strategies such as cosine annealing learning rate can be used), batch size, number of training rounds (Epoch), etc. At the same time, select a suitable loss function. For multi-class fault prediction problems, the cross-entropy loss function is usually used.

[0048] Training process: input the preprocessed image data and corresponding labels into the model for training, calculate the gradient of the loss function on the model parameters through the back propagation algorithm, and update the model parameters using an optimizer (such as Adam, SGD, etc.), continuously adjust the weights of the model, and continuously improve the prediction accuracy of the model on the training data.

[0049] Verification and tuning: during the training process, the model is evaluated on the validation set regularly to monitor the performance indicators (such as accuracy, precision, recall, F1 value, etc.) of the model. According to the verification result, adjust the hyperparameters or training strategy of the model to prevent overfitting or underfitting of the model, until the model reaches good performance on the validation set.

[0050] S5: input the first fault prediction probability distribution and the second fault prediction probability distribution into the fault prediction fusion model to obtain the fused fault prediction probability of the numerical control machine tool; wherein the fault prediction fusion model is: (1); to fuse the fault prediction probability, is the first fault prediction average probability of the first fault prediction probability distribution, is the second fault prediction average probability of the second fault prediction probability distribution, and are obtained by general matrix calculation method, which will not be repeated here; is a weight coefficient, and .

[0051] Specifically, in this embodiment, various types of sensor data are collected, camera image data of the numerical control machine tool is collected through the camera, first fault prediction probability and second fault prediction probability are obtained through numerical data and non-numerical data respectively, and the two prediction probabilities are fused through formula (1) to obtain the final fault prediction probability. In formula (1), the weights of the two types of prediction probabilities are limited. Since the number of sensors is set to be relatively large in actual application, the weight of the first fault prediction probability is set to a relatively large proportion during fusion, so as to ensure the accuracy of the final result.

[0052] S6: If the fusion failure prediction probability exceeds the failure warning threshold, a failure warning signal is sent out, and based on the historical maintenance data and the current running state of the numerical control machine tool, a failure cause and a maintenance suggestion are provided.

[0053] The multi-level failure warning threshold is set, and when the prediction probability exceeds the corresponding threshold, a warning signal of different levels is triggered. At the same time, combined with historical failure data and current features, an explanatory algorithm (such as SHAP value analysis) is used to generate a detailed failure cause analysis report and targeted maintenance suggestions to assist maintenance personnel in quickly locating the failure root cause and developing a maintenance plan.

[0054] Compared with the prior art, the present application has the following beneficial effects: (1) Multi-modal data deep fusion: The present application collects multiple types of data, including multi-source process parameter data (numerical data) and machining image data (non-numerical data), which comprehensively covers all aspects of the operation of the numerical control machine tool. Through innovative data preprocessing and feature extraction methods, the data value is fully tapped, and compared with traditional single data or simple data fusion methods, the machine tool running state can be more accurately described, and the accuracy of failure prediction can be greatly improved.

[0055] (2) Dynamic spatio-temporal attention network innovation: The unique dynamic spatio-temporal attention network architecture, especially the dual-channel attention mechanism, can adaptively focus on key spatio-temporal features, effectively capture complex dynamic correlations between data and early weak signs of failure, making the model more sensitive to failure features and improving the prediction performance and adaptability of the model to complex working conditions. (3) Fusion failure probability result: The output probabilities of the two failure prediction models are fused to obtain a fusion failure probability result, which can improve the prediction generalization ability and enable the model to maintain stable prediction accuracy under different machine tools and different working conditions.

[0056] (4) Multi-level warning and decision support: Multi-level failure warning and detailed failure cause analysis report and maintenance suggestion provide comprehensive decision support for enterprises. Enterprises can reasonably arrange production plans and maintenance work according to the warning level, prevent failure from occurring in advance, reduce equipment downtime, reduce production loss and maintenance cost, and have significant economic benefits and practical value.

[0057] In another embodiment, the present application also provides a failure prediction device 20 for a numerical control machine tool, comprising: A data acquisition module 201 is configured to acquire multi-source modal data of a monitored numerical control machine tool during operation, wherein the multi-source modal data includes multi-source running data and machining image data of the monitored numerical control machine tool within a monitoring period; and the multi-source running data includes at least one or more of running signal data, working parameter data and historical maintenance data. The data preprocessing module 202 is configured to preprocess the multi-source operation data and the processing image data respectively to remove noise data and abnormal data therefrom. The feature data acquisition module 203 is configured to normalize and extract features from the preprocessed multi-source operation data to obtain multi-source modal feature data, and perform image enhancement and normalization on the preprocessed processing image data to obtain processing image feature data. The probability distribution calculation module 204 is configured to input the multi-source modal feature data into a pre-trained multi-source fault prediction model to output a first fault prediction probability distribution, and input the processing image feature data into a pre-trained image fault prediction model to output a second fault prediction probability distribution. The probability fusion calculation module 205 is configured to input the first fault prediction probability distribution and the second fault prediction probability distribution into a fault prediction fusion model to obtain a fusion fault prediction probability of the CNC machine tool. The fault early warning module 206 is configured to output a fault early warning signal if the fusion fault prediction probability exceeds a fault early warning threshold, and provide a fault cause and a repair suggestion according to historical maintenance data and a current operation state of the CNC machine tool. The fault prediction fusion model is as follows: (1) ; is the fusion fault prediction probability, is a first fault prediction average probability of the first fault prediction probability distribution, is a second fault prediction average probability of the second fault prediction probability distribution, is a weight coefficient, and .

[0058] The apparatus provided by the embodiments of the present application can implement the method described above, and the system can be implemented by software, hardware or a combination of software and hardware. For example, the system can include integrated or separated functional modules or units to perform corresponding steps in the above methods. In some embodiments of the present application, the system provided by the embodiments of the present application is based on the same inventive concept as the method provided by the foregoing embodiments of the present application, and has the same beneficial effects. Therefore, no further description is given here.

[0059] To illustrate, the herein disclosed system, method and computer program product can be implemented in a system that includes one or more electronic devices, such as a computer system. In this regard, the system can include a memory and a processor. The memory can include computer program code, which, when executed by the processor, causes the system to implement the steps or acts described herein. In other words, the system can include a memory and a processor, and the memory can include computer program code, which, when executed by the processor, causes the system to implement the steps or acts described herein. The system can also include a user interface, which can include a display and an input device, such as a keyboard or a pointing device. The system can also include a communication interface, which can be used to communicate with one or more other systems or devices.

[0060] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the system, device and unit described above can refer to the corresponding process in the foregoing method embodiments, which will not be described here.

[0061] In several embodiments provided in the present application, it should be understood that the disclosed apparatus and method can be implemented in other ways. The apparatus embodiments described above are only schematic. For example, the division of the units is only a logical function division, and there can be another division manner in actual implementation. For example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the units shown or discussed can be indirect coupling or communication connection through some communication interface, and can be electrical, mechanical or other forms.

[0062] The units described as separate components can or can not be physically separate, and the components shown as units can or can not be physical units, that is, they can be located in one place, or can be distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the embodiment scheme.

[0063] In addition, the functional units in each embodiment of the present application can be integrated in one processing unit, or each unit can be physically present separately, or two or more units can be integrated in one unit.

[0064] If the functions are implemented in the form of software function units and sold or used as independent products, they can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application essentially or the parts that contribute to the prior art or parts of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a number of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various media that can store program codes.

[0065] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application, and they should be covered in the scope of the claims and the specification of the present application.

Claims

1. A fault prediction method for CNC machine tools, characterized in that, Includes the following steps: S1: Acquire multi-source modal data of the monitored CNC machine tool during operation. The multi-source modal data includes multi-source operating data and machining image data of the monitored CNC machine tool during the monitoring period. The multi-source operating data includes at least one or more of the following: operating signal data, working parameter data, and historical maintenance data. S2: Preprocess the multi-source running data and the processed image data respectively to remove noise and abnormal data; S3: Normalize and extract features from the preprocessed multi-source running data to obtain multi-source modal feature data; The preprocessed image data is subjected to image recognition and feature extraction to obtain the processed image feature data; S4: Input the multi-source modal feature data into a pre-trained multi-source fault prediction model and output a first fault prediction probability distribution; input the processed image feature data into a pre-trained image fault prediction model and output a second fault prediction probability distribution; S5: Input the first fault prediction probability distribution and the second fault prediction probability distribution into the fault prediction fusion model to obtain the fusion fault prediction probability of the CNC machine tool; S6: If the fusion fault prediction probability exceeds the fault warning threshold, a fault warning signal is issued, and the cause of the fault and maintenance suggestions are provided based on the historical maintenance data and current operating status of the CNC machine tool. The fault prediction fusion model is as follows: （1）； To integrate fault prediction probabilities, The average probability of the first fault prediction is given by the first fault prediction probability distribution. The second fault prediction average probability is the second fault prediction probability distribution. These are the weighting coefficients, and .

2. The method according to claim 1, characterized in that, The processing image data includes multiple images of the processing area obtained by acquiring data from multiple angles of the monitored CNC machine tool.

3. The method according to claim 2, characterized in that, The step of normalizing and extracting features from the preprocessed multi-source operational data to obtain multi-source modal feature data includes: S311: Normalize the preprocessed multi-source running data to obtain multi-source normalized data; S312: Extract features from the multi-source normalized data to obtain multi-source modal feature data.

4. The method according to claim 2, characterized in that, The preprocessed image data is then enhanced and normalized to obtain processed image feature data, including: S321: Perform image enhancement and stitching on the preprocessed image data to obtain intermediate image data; S322: Normalize the pixel values ​​of the intermediate image data to obtain processed image feature data.

5. The method according to claim 3, characterized in that, The multi-source fault prediction model is a dynamic spatiotemporal attention network model, which includes a multimodal data input layer, a parallel feature extraction layer, a dynamic spatiotemporal attention layer, a temporal fusion layer, and a fault prediction output layer. The step of inputting the multi-source modal feature data into a pre-trained multi-source fault prediction model and outputting a first fault prediction probability distribution includes: S411: The multi-source modal data input layer includes multiple multi-modal data input sub-layers, and the multiple multi-modal data input sub-layers receive different types of data sources of the multi-modal feature data; S412: The parallel feature extraction layer includes multiple feature extraction sub-networks. The multiple feature extraction sub-networks receive feature data of the corresponding type and perform feature extraction to obtain multiple preliminary feature vectors. S413: The dynamic spatiotemporal attention layer first concatenates multiple initial feature vectors to obtain a fused feature vector; then calculates the temporal attention weight matrix and the spatial attention weight matrix through the temporal attention module and the spatial attention module respectively; and performs weighted fusion on the fused feature vector based on the temporal attention weight matrix and the spatial attention weight matrix to obtain spatiotemporal attention features. S414: The temporal fusion layer performs temporal modeling on the spatiotemporal attention features and inputs them into the fault prediction output layer; S415: The spatiotemporal attention features after time-series modeling are input into the fully connected layer and softmax function of the fault prediction output layer to obtain the first fault prediction probability distribution of various faults occurring in the CNC machine tool during the monitoring period.

6. The method according to claim 4, characterized in that, The processed image feature data is input into a pre-trained image fault prediction model, which outputs a second fault prediction probability distribution, including: S421: Format the processed image feature data according to the input requirements of the image fault prediction model; S422: Input the processed image feature data into a pre-trained image fault prediction model to obtain a second fault prediction probability distribution of various faults occurring in the CNC machine tool during the monitoring period.

7. The method according to claim 5, characterized in that, The multi-source fault prediction model is pre-trained in the following manner: Collect historical multi-source modal data of the CNC machine tool during its historical operation, and divide the historical multi-source modal data into training set and test set according to a preset ratio; The dynamic spatiotemporal attention network model is trained using the training set data to obtain a trained multi-source fault prediction model. The multi-source fault prediction model is tested using test set data to obtain test results, and the multi-source fault prediction model is optimized and adjusted using the test results.

8. The method according to claim 6, characterized in that, The image fault prediction model is a deep learning architecture for image fault prediction, which is pre-trained in the following manner: Determine the training parameters and loss function of the image fault prediction model; Collect historical machining image data of the CNC machine tool during its historical operation, and preprocess the historical machining image data; The preprocessed historical image data is divided into training and testing sets according to a preset ratio; The deep learning architecture for image fault prediction is trained using the training set data to obtain the trained image fault prediction model. The image fault prediction model is tested using test set data to obtain test results, and the image fault prediction model is then optimized and adjusted using the test results.

9. A fault prediction device for CNC machine tools, characterized in that, include: The data acquisition module is used to acquire multi-source modal data of the monitored CNC machine tool during operation. The multi-source modal data includes multi-source operation data and machining image data of the monitored CNC machine tool during the monitoring period. The multi-source operational data includes at least one or more of the following: operational signal data, operating parameter data, and historical maintenance data; The data preprocessing module is used to preprocess the multi-source running data and the processed image data respectively to remove noise data and abnormal data. The feature data acquisition module is used to normalize and extract features from the preprocessed multi-source running data to obtain multi-source modal feature data, and to enhance and normalize the preprocessed processed image data to obtain processed image feature data. The probability distribution calculation module is used to input the multi-source modal feature data into a pre-trained multi-source fault prediction model and output a first fault prediction probability distribution, and input the processed image feature data into a pre-trained image fault prediction model and output a second fault prediction probability distribution. The probability fusion calculation module is used to input the first fault prediction probability distribution and the second fault prediction probability distribution into the fault prediction fusion model to obtain the fusion fault prediction probability of the CNC machine tool. The fault early warning module is used to issue a fault early warning signal when the fused fault prediction probability exceeds the fault early warning threshold, and to provide fault causes and maintenance suggestions based on the historical maintenance data and current operating status of the CNC machine tool. The fault prediction fusion model is as follows: （1）； To integrate fault prediction probabilities, The average probability of the first fault prediction is given by the first fault prediction probability distribution. The second fault prediction average probability is the second fault prediction probability distribution. The model sets a weighting coefficient and a fault warning threshold. When the fault probability output by the model exceeds the threshold, a fault warning signal is issued, and possible causes of the fault and maintenance suggestions are provided based on historical maintenance data and the current operating status.

Images