Numerical control machine tool fault prediction and diagnosis method and system based on AI model

Abstract

The invention discloses a numerical control machine tool fault prediction and diagnosis method and system based on an AI model, and the method comprises the steps: building a vibration-temperature correlation curve with time sequence characteristics according to a vibration signal and a temperature signal, and building a current-rotating speed correlation curve with time sequence characteristics according to a current signal and a rotating speed signal; inter-axis space linkage coding is carried out on the motion axes to generate inter-axis interaction response characteristics, and the inter-axis space linkage coding comprises analysis of signal mutual influence relations among different motion axes; performing time scale linkage coding on the vibration-temperature correlation curve and the current-rotating speed correlation curve to generate time interaction response characteristics; and fusing the inter-axis interaction response features and the time interaction response features, and generating a fault prediction and diagnosis result through a pre-trained AI model. The problems of single evaluation data and insufficient multi-source signal correlation analysis in the prior art are solved, and the accuracy of the fault prediction diagnosis result of the numerical control machine tool is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of machine tool fault diagnosis, and in particular to a numerical control machine tool fault prediction and diagnosis method and system based on an AI model. BACKGROUND

[0002] In the field of intelligent manufacturing, numerical control machine tools as core production equipment, the stability of its running state directly affects the production efficiency and product quality. Traditional numerical control machine tool fault prediction and diagnosis technology mostly relies on single signal analysis or simple threshold judgment, such as only judging fault according to vibration signal amplitude exceeding limit, or isolated analysis of temperature, current and other data. This way is difficult to capture the complex space-time correlation between multi-source signals, and cannot effectively analyze the coupling relationship between different motion axes, resulting in fault prediction lag, low diagnosis accuracy, and often false positives and false negatives, which cannot meet the needs of modern high-precision and high-efficiency production for equipment health management. SUMMARY

[0003] In order to solve at least one of the above technical problems, the present application provides a numerical control machine tool fault prediction and diagnosis method and system based on an AI model.

[0004] In a first aspect, the present application provides a numerical control machine tool fault prediction and diagnosis method based on an AI model, the method comprising:

[0005] Collecting multi-source monitoring data of the numerical control machine tool, including historical signals collected from multiple motion axes of the numerical control machine tool, the historical signals including vibration signals, temperature signals, current signals and speed signals;

[0006] Establishing a vibration-temperature correlation curve with time sequence characteristics according to the vibration signals and the temperature signals, and establishing a current-speed correlation curve with time sequence characteristics according to the current signals and the speed signals;

[0007] Inter-axis space linkage coding is performed on the multiple motion axes to generate inter-axis interaction response features, the inter-axis space linkage coding including analyzing the signal mutual influence relationship between different motion axes;

[0008] Time scale linkage coding is performed on the vibration-temperature correlation curve and the current-speed correlation curve to generate time interaction response features, the time scale linkage coding including analyzing the signal trend correlation at different time points;

[0009] Fusing the inter-axis interaction response features and the time interaction response features, and generating a fault prediction and diagnosis result through a pre-trained AI model.

[0010] Preferably, the time scale linkage coding includes:

[0011] encoding the vibration-temperature correlation curve through a time series encoder to obtain a vibration-temperature time series feature vector; and encoding the current-speed correlation curve through a time series encoder to obtain a current-speed time series feature vector;

[0012] inputting the vibration-temperature time series feature vector into a first LSTM network to extract vibration-temperature long and short term trend features; and inputting the current-speed time series feature vector into a second LSTM network to extract current-speed long and short term trend features;

[0013] fusing the vibration-temperature long and short term trend features and the current-speed long and short term trend features to generate the time interaction response feature.

[0014] Preferably, the inter-axis spatial linkage encoding comprises:

[0015] splitting the signal of each motion axis into a local feature vector sequence;

[0016] performing cross-attention mechanism analysis on the local feature vector sequences of the plurality of motion axes to quantify the linkage response strength between different motion axes;

[0017] generating the inter-axis interaction response feature based on the linkage response strength.

[0018] Preferably, the plurality of motion axes comprises linear motion axes and rotary motion axes, and the inter-axis spatial linkage encoding further comprises analyzing the dynamic coupling relationship between linear axes and rotary axes.

[0019] Preferably, after collecting the multi-source monitoring data of the numerical control machine tool, the method further comprises:

[0020] performing adaptive sampling frequency fusion processing on the historical signals, including identifying the inherent change frequency of each signal, adaptively adjusting the sampling frequency of each signal based on the inherent change frequency and the current machine tool operating condition; oversampling and smoothing low-frequency signals, and reducing the sampling rate of high-frequency signals to retain features, to generate a unified time reference fusion sampling data stream;

[0021] performing dynamic weighted noise reduction on the fusion sampling data stream, including dynamically calculating the noise reduction weight of each signal channel based on the signal-to-noise ratio and the relevance to the target fault; and performing channel-adaptive noise reduction processing based on the weight according to wavelet transform or empirical mode decomposition.

[0022] In a second aspect, the present application also provides a numerical control machine tool fault prediction and diagnosis system based on an AI model, which comprises:

[0023] The signal acquisition unit is configured to acquire multi-source monitoring data of the numerical control machine tool, including historical signals acquired from a plurality of motion axes of the numerical control machine tool, wherein the historical signals include vibration signals, temperature signals, current signals and rotation speed signals.

[0024] The curve construction unit is configured to construct a vibration-temperature correlation curve with time sequence characteristics according to the vibration signals and the temperature signals, and construct a current-rotation speed correlation curve with time sequence characteristics according to the current signals and the rotation speed signals.

[0025] The space coding unit is configured to code the plurality of motion axes in an inter-axis space linkage manner to generate an inter-axis interaction response feature, wherein the inter-axis space linkage coding includes analyzing a signal mutual influence relationship between different motion axes.

[0026] The time coding unit is configured to code the vibration-temperature correlation curve and the current-rotation speed correlation curve in a time scale linkage manner to generate a time interaction response feature, wherein the time scale linkage coding includes analyzing a signal trend correlation at different time points.

[0027] The fault prediction unit is configured to fuse the inter-axis interaction response feature and the time interaction response feature, and generate a fault prediction and diagnosis result through a pre-trained AI model.

[0028] Preferably, the time coding unit is further configured to:

[0029] encode the vibration-temperature correlation curve through a time series encoder to obtain a vibration-temperature time sequence feature vector, and encode the current-rotation speed correlation curve through a time series encoder to obtain a current-rotation speed time sequence feature vector.

[0030] input the vibration-temperature time sequence feature vector into a first LSTM network to extract vibration-temperature long and short term trend features, and input the current-rotation speed time sequence feature vector into a second LSTM network to extract current-rotation speed long and short term trend features.

[0031] fuse the vibration-temperature long and short term trend features and the current-rotation speed long and short term trend features to generate the time interaction response feature.

[0032] Preferably, the space coding unit is further configured to:

[0033] split the signals of each motion axis into a local feature vector sequence;

[0034] perform cross-attention mechanism analysis on the local feature vector sequences of the plurality of motion axes to quantify linkage response strengths between different motion axes;

[0035] generate the inter-axis interaction response feature based on the linkage response strengths.

[0036] In a third aspect, the present application also provides an electronic device, comprising a processor and a memory, the memory being configured to store computer program code, the computer program code comprising computer instructions, when the processor executes the computer instructions, the electronic device executes the method of the first aspect and any possible implementation manner thereof.

[0037] In a fourth aspect, the present application also provides a computer readable storage medium, the computer readable storage medium storing a computer program, the computer program comprising program instructions, when the program instructions are executed by a processor of an electronic device, the processor executes the method of the first aspect and any possible implementation manner thereof.

[0038] Compared with the prior art, the present application has the following beneficial effects:

[0039] The present application breaks through the limitations of single signal analysis, and excavates the internal relationship between signals by collecting multi-source monitoring data and establishing vibration-temperature correlation curve and current-speed correlation curve. The present application breaks the barrier of isolated analysis of each axis data by analyzing the mutual influence of different motion axis signals through inter-axis space linkage coding. The present application strengthens the ability to capture the development trend of faults by analyzing the trend correlation of signals at different time points through time scale linkage coding. Finally, the present application fuses the inter-axis interactive response characteristics and time interactive response characteristics, and realizes accurate diagnosis with the help of a pre-trained AI model. The present application effectively solves the problems of insufficient multi-source signal correlation analysis, difficulty in analyzing inter-axis coupling relationship, and untimely fault trend prediction in the prior art, and significantly improves the foresight of numerical control machine tool fault prediction and the accuracy of diagnosis results, thereby providing a powerful guarantee for preventive maintenance and stable operation of numerical control machine tools.

[0040] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, but not limiting the present disclosure. BRIEF DESCRIPTION OF DRAWINGS

[0041] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the background art, the drawings needed to be used in the embodiments of the present application or the background art will be described below.

[0042] The drawings herein are incorporated into the specification and form part of the specification, which show embodiments consistent with the present disclosure, and together with the specification, serve to explain the technical solutions of the present disclosure.

[0043] Figure 1 A flowchart of a numerical control machine tool fault prediction and diagnosis method based on an AI model provided by the present application is shown in the figure.

[0044] Figure 2A structure schematic diagram of an AI model-based numerical control machine tool fault prediction and diagnosis system is provided for an embodiment of the present application. DETAILED DESCRIPTION

[0045] In order for those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0046] The terms "first", "second", and the like in the specification and claims of the present application and the above-described drawings are used to distinguish different objects, rather than to describe a specific order. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not limited to the listed steps or units, but can optionally include steps or units not listed, or can optionally include other steps or units inherent to the process, method, product, or device.

[0047] Reference herein to "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment can be included in at least one embodiment of the present application. The appearance of the phrase in various places in the specification does not necessarily all refer to the same embodiment, nor is it necessarily mutually exclusive of other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0048] Please refer to Figure 1 , Figure 1 A flowchart of an AI model-based numerical control machine tool fault prediction and diagnosis method is provided for an embodiment of the present application. As shown in Figure 1 , the method comprises:

[0049] S10, collecting multi-source monitoring data of the numerical control machine tool, including historical signals collected from a plurality of motion axes of the numerical control machine tool, the historical signals including vibration signals, temperature signals, current signals, and rotation speed signals;

[0050] Different component faults of the numerical control machine tool will be reflected in different types of signals. For example, bearing wear will cause vibration abnormalities, motor failure will be reflected in current changes, increased friction will cause temperature rise, and rotation speed fluctuations are related to transmission system problems. A single type of signal is difficult to fully reflect the device state, and multi-source data fusion can provide more rich fault feature information.

[0051] In order to collect various signals, usually need to help sensors. Collecting vibration signals can be installed in the bearings of each motion axis, guide rail and other key parts of acceleration sensor, sampling frequency according to the device running speed is determined, generally for 10kHz~50kHz, to capture high frequency vibration components. Temperature signal acquisition can use thermocouple or infrared temperature sensor, measuring the temperature of the motor, bearing and other heat components, sampling period is 1 second~10 seconds, taking into account the real-time and data volume. Current signal acquisition through current transformer acquisition motor current, sampling frequency and vibration signal matching, in order to analyze the relevance of current and mechanical vibration. Speed signal can use encoder or hall sensor to obtain the speed of the main shaft and each feed shaft, sampling period is 10 milliseconds~100 milliseconds, to ensure that the instantaneous change of speed can be captured.

[0052] S20, according to the vibration signal and temperature signal to establish the vibration-temperature correlation curve with time sequence characteristics, according to the current signal and speed signal to establish the current-speed correlation curve with time sequence characteristics;

[0053] Modern numerical control system has self-diagnosis function, can be divided into start self-diagnosis, online self-diagnosis and offline self-diagnosis. Start self-diagnosis is automatically executed when the system is powered on, checks the key hardware and control software. Online self-diagnosis is real-time monitoring when the system is running, through CRT display alarm information or use light emitting diode flicker to indicate fault. Offline self-diagnosis is to use special software and test device to check after system failure, fault positioning is more accurate. But the existing fault diagnosis is usually based on single signal and their own threshold value comparison, once trigger threshold value after the alarm prompt, however, this way ignores the internal relationship between different signals, vibration signal can reflect the running state of machine tool moving parts, temperature signal can reflect the heating of components, there is a correlation between the two. For example, bearing wear will make the vibration intensified, at the same time, friction heating leads to temperature rise. Establishing vibration-temperature correlation curve can capture the time sequence characteristics in the development process of the fault, which helps to find early signs of failure. Similarly, current signal and speed signal are related, motor load change will cause current change, and then affect the speed, current-speed correlation curve can reflect the working state of motor and transmission system, help to identify problems such as motor failure, transmission chain abnormality.

[0054] Specifically, when constructing the vibration-temperature correlation curve, the vibration signal is decomposed by wavelet packet, and the energy features of specific frequency bands are extracted. The temperature signal is filtered by moving average to eliminate random fluctuations. A three-dimensional scatter plot is drawn with time as the horizontal axis, vibration features as the vertical axis, and temperature values as color mapping. The correlation curve is generated by surface fitting. When constructing the current-speed correlation curve, the current signal is Fourier transformed, and the fundamental and harmonic components are extracted. The speed fluctuation coefficient (standard deviation of speed / average speed) is calculated. The parameterized correlation curve is constructed with speed as the horizontal axis, current harmonic amplitude as the vertical axis, and speed fluctuation coefficient as the curve parameter.

[0055] A single signal may be disturbed by multiple factors, leading to inaccurate fault judgment. By establishing a correlation curve, multiple signals are considered comprehensively, which can reduce false positives and false negatives. For example, vibration abnormalities may be normal phenomena caused by heavy cutting. Combined with the temperature signal, if there is no significant change in temperature, it may not be a fault, thereby improving the reliability of diagnosis.

[0056] S30, inter-axis space linkage coding is performed on the plurality of motion axes to generate inter-axis interaction response features, the inter-axis space linkage coding comprising analyzing signal mutual influence relationship between different motion axes;

[0057] There is a coupling relationship between each motion axis of the numerical control machine tool, and an abnormality of one axis may cause a response change of other axes. For example, wear of the X-axis guide rail may increase the load of the Y-axis motor, which is reflected in the current and vibration signals. Analyzing the inter-axis linkage relationship can find hidden faults. The same type of signals (such as X, Y, and Z axis vibration signals) of each axis are preprocessed to eliminate trend items and noise. A vector autoregressive model (VAR) is established to determine whether the signals of one axis can predict the signal changes of another axis through F test. Finally, a causal relationship diagram is constructed, with nodes representing axes and edge weights representing causal strength. The continuous signals are discretized, and the mutual information between different axis signals is calculated. The greater the mutual information value, the stronger the correlation between the two axis signals. Preferably, a graph neural network (GNN) can also be used, regarding each axis as a graph node and regarding signal correlation relationship as an edge, automatically learning inter-axis interaction features through a message passing mechanism, or using a dynamic Bayesian network to consider the influence of time factors on inter-axis relationship. By quantifying the mutual influence between each axis, the key fault propagation path can be identified.

[0058] S40, time scale linkage coding is performed on the vibration-temperature correlation curve and the current-speed correlation curve to generate time interaction response features, the time scale linkage coding comprising analyzing signal trend correlation at different time points;

[0059] The development of a fault usually has a stage from early minor damage to functional failure, which takes a certain time. The signal changes at different time scales contain different fault information, such as short-term fluctuations that may reflect random interference, and long-term trends that are more likely to be related to the development of a fault. When encoding, the correlation curve is decomposed into components of different scales, such as approximation coefficients (low frequency) and detail coefficients (high frequency) using wavelet transform, and features such as energy, variance, skewness, etc. are extracted for each scale component. In order to extract time-varying features, a fixed length window can be slid on the correlation curve by using a sliding window method, and the statistical features in each window are calculated. The rate of change of the features between adjacent windows reflects the evolution trend of the features over time.

[0060] S50, fuse the inter-shaft interaction response features and the time interaction response features, and generate a fault prediction and diagnosis result through a pre-trained AI model.

[0061] The inter-shaft causal relationship matrix and the time scale feature vector are spliced into a high-dimensional feature vector. A plurality of sub-models are trained using different features, and the output results of the models are integrated by voting or weighted averaging. A deep neural network (DNN) is trained using historical fault data, and a transfer learning technique is used to migrate model parameters pre-trained on large-scale industrial equipment data to the system. When performing fault diagnosis, only the fusion feature vector needs to be input, and the fault type, severity and remaining life prediction can be output.

[0062] In one embodiment, the time scale linkage encoding includes:

[0063] The vibration-temperature correlation curve is encoded by a time series encoder to obtain a vibration-temperature time series feature vector, and the current-speed correlation curve is encoded by a time series encoder to obtain a current-speed time series feature vector.

[0064] The vibration-temperature time series feature vector is input into a first LSTM network to extract vibration-temperature long and short term trend features, and the current-speed time series feature vector is input into a second LSTM network to extract current-speed long and short term trend features.

[0065] The vibration-temperature long and short term trend features and the current-speed long and short term trend features are fused to generate the time interaction response features.

[0066] The embodiment specifically includes the following steps:

[0067] Data preprocessing: The vibration-temperature correlation curve and the current-speed correlation curve are standardized and reshaped to adapt to the input requirements of the LSTM network.

[0068] Feature extraction network: two independent LSTM networks process vibration-temperature and current-speed time series features respectively, and LSTM networks can effectively capture long and short term dependencies at different time scales;

[0069] Feature fusion: the outputs of the two LSTM networks are spliced, and the final time interaction response features are generated through a fully connected layer.

[0070] In practical applications, the following parameters can be adjusted according to the specific CNC machine tool system and data characteristics: LSTM unit number: affects the ability of the model to capture time series features; fusion layer dimension: controls the expression ability of the final features; sequence length: determined according to the sampling frequency of the monitoring data and the fault development period; training rounds and batch size: affect the efficiency and generalization ability of the model training; by extracting time series features of different types of signals and fusing them, the dynamic changes of the CNC machine tool running state can be effectively captured, providing strong support for subsequent fault prediction and diagnosis.

[0071] In one embodiment, the inter-axis space linkage coding includes:

[0072] Splitting the signal of each motion axis into a local feature vector sequence;

[0073] Cross-attention mechanism analysis is performed on the local feature vector sequences of the plurality of motion axes to quantify the linkage response strength between different motion axes;

[0074] Based on the linkage response strength, the inter-axis interaction response features are generated.

[0075] In this embodiment, in order to perform local feature extraction, an independent feature extraction network is created for each motion axis to convert the original signal into a local feature vector sequence with semantic information; a multi-head attention mechanism is used to calculate the correlation between different axes, with the feature sequence of each axis as the query and the features of all axes as the key and value, and the attention mechanism can automatically learn the dynamic dependency between axes; the linkage strength is calculated through the interaction of the attention output and the target axis feature, and the sigmoid activation function is used to normalize the strength to the [0, 1] interval, and the strength value reflects the influence degree between different axes; the linkage response strength between all axes is integrated, and the final inter-axis interaction response features are generated through a multi-layer perception.

[0076] For ease of understanding, assume that the CNC machine tool has motion axes, and the local feature vector sequence extracted by each axis within a time window is: , represents a real matrix with dimensions rows columns;

[0077] Step 1: Generate Query, Key, Value vector ;

[0078] Local feature sequence of each axis Linear transformation is performed:

[0079] ;

[0080] ;

[0081] ;

[0082] wherein, , , is a trainable parameter matrix,

[0083] Step 2: Calculate the cross-axis attention weight:

[0084] For any two axes (target axis) and (correlation axis), the linkage response strength is calculated:

[0085] ;

[0086] wherein, represents the time step linkage strength matrix of axis to axis , represents the vector obtained by transposing the linearly transformed feature sequence of axis , represents the projection dimension of the query vector (Query) and the key vector (Key);

[0087] Step 3: Generate cross-axis interaction response features:

[0088] ;

[0089] ;

[0090] wherein, represents the normalized time step linkage strength matrix, represents the cross-axis interaction response features of axis , represents the value vector of axis ;

[0091] Preferably, the plurality of motion axes includes linear motion axes and rotary motion axes, and the cross-axis spatial linkage encoding further includes analyzing the dynamic coupling relationship between the linear axes and the rotary axes.

[0092] Specifically, the linear-axis encoder and the rotary-axis encoder are designed respectively, considering the difference in physical characteristics of the two axes, using different neural network branches for processing, for each pair of linear and rotary axes, using the outer product operation to capture the nonlinear relationship between them, the outer product can represent the interaction between the two axes, which corresponds to the tensor product operation in linear algebra; for example, the vibration of the X-axis may have a certain relationship with the rotation speed of the C-axis, which can be effectively captured by the outer product. The generated outer product matrix is reduced and features are extracted, meaningful coupling features are extracted using multiple nonlinear transformations, the dependence strength between different axes is implicitly learned through the attention mechanism, and a special network is designed to evaluate the coupling strength between linear and rotary axes. The strength value ranges between 0-1, indicating the degree of association between the two axes, these strength values can be used for weight allocation in subsequent fault diagnosis, so as to capture the complex physical relationship between linear and rotary motion, and detect the small changes in the coupling relationship between the axes. The coupling strength can be interpreted as the degree of dynamic association between different axes, providing more comprehensive state information for intelligent operation and maintenance.

[0093] In summary, the method provided by the embodiment breaks through the limitations of single signal analysis by collecting multi-source monitoring data and establishing vibration-temperature correlation curves and current-speed correlation curves, and excavates the inherent relationship between signals; the space linkage coding between the axes analyzes the mutual influence of different motion axis signals, breaking the barrier of isolated analysis of each axis data; the time scale linkage coding analyzes the signal trend correlation at different time points, and strengthens the ability to capture the fault development trend. Finally, the inter-axle interaction response features and time interaction response features are fused, and the pre-trained AI model is used to realize accurate diagnosis. This scheme effectively solves the problems of insufficient multi-source signal correlation analysis, difficulty in analyzing the coupling relationship between the axes, and untimely fault trend prediction in the prior art, significantly improves the foresight of numerical control machine tool fault prediction and the accuracy of diagnosis results, and provides a strong guarantee for preventive maintenance and stable operation of numerical control machine tools.

[0094] In one embodiment, after collecting the multi-source monitoring data of the numerical control machine tool, it further includes:

[0095] The adaptive sampling frequency fusion processing of the historical signals includes identifying the inherent change frequency of each signal, and adaptively adjusting the sampling frequency of each signal based on the inherent change frequency and the current machine tool operating condition; oversampling smoothing is performed on low-frequency signals, and feature preservation is performed on high-frequency signals, generating a unified time reference fusion sampling data stream;

[0096] The dynamic weighted noise reduction of the fusion sampling data stream includes dynamically calculating the noise reduction weight of each signal channel based on the signal-to-noise ratio and the relevance to the target fault; based on the weight, the channel adaptive noise reduction processing is performed according to the wavelet transform or empirical mode decomposition.

[0097] In this embodiment, the inherent change frequency of each signal (vibration, temperature, current, and rotating speed) is identified; based on the inherent change frequency and the current machine tool operating condition (such as the machining mode and the load size), the sampling frequency of each signal is adaptively adjusted to meet the minimum requirement of feature extraction of each signal and to maintain the time sequence alignment between signals; the low-frequency signal (such as temperature) is oversampled and smoothed or the high-frequency signal (such as vibration) is downsampled and feature-preserved to generate a unified time reference fusion sampling data stream. Then, the fusion sampling data stream is dynamically weighted and denoised; specifically, based on the signal-to-noise ratio and the correlation with the target fault, the denoising weight of each signal channel is dynamically calculated; the weight is applied to perform channel-adaptive denoising processing in combination with wavelet transform or empirical mode decomposition (EMD). The establishment of the vibration-temperature correlation curve and the current-rotating speed correlation curve is based on the fusion sampling data stream and the denoised signal.

[0098] The above embodiment can solve the problem of multi-source and different frequency data fusion. Instead of simply unifying sampling or independent processing, the sampling rate is adaptively adjusted according to the signal characteristics and the operating condition, and then aligned and fused, and then combined with dynamic weight denoising, which significantly improves the accuracy and efficiency of subsequent correlation curve construction.

[0099] In one embodiment, the fusion of the inter-axis interaction response feature and the time interaction response feature specifically includes:

[0100] The inter-axis interaction response feature is regarded as a spatial dimension feature map;

[0101] The time interaction response feature is regarded as a time dimension feature sequence;

[0102] A space-time fusion module is constructed, which includes:

[0103] A spatial attention submodule: used to learn the importance weight of different motion axes in spatial linkage and to weightedly fuse the spatial dimension feature map;

[0104] A time attention submodule: used to learn the importance weight of different time points in trend correlation and to weightedly fuse the time dimension feature sequence;

[0105] A cross-attention interaction submodule: used to learn the mutual dependence relationship between the spatial feature map and the time feature sequence to generate enhanced space-time joint features.

[0106] The pre-trained AI model generates fault prediction and diagnosis results based on the enhanced space-time joint features.

[0107] By introducing a dual attention mechanism to achieve deep spatiotemporal interactive fusion, it not only focuses on the importance of spatial axes and time points respectively, but also explicitly models the cross-dependencies between spatiotemporal features. It goes beyond simple feature splicing or average fusion and can more accurately capture complex fault symptom patterns.

[0108] In one embodiment, the method further comprises:

[0109] Construct a fault knowledge graph, where nodes represent specific fault types, relevant motion axes, key signal characteristics (such as vibration energy in a specific frequency band, temperature anomaly slope, current ripple coefficient), and typical operating conditions. Edges represent causal, concurrent, or conditional dependencies between nodes. The preliminary fault prediction and diagnosis results generated by the AI ​​model (including predicted fault type, probability, relevant axes, and signal characteristics) are input into the fault knowledge graph for reasoning verification and interpretation enhancement.

[0110] Furthermore, based on the graph reasoning results, the preliminary diagnostic results are verified to ensure they conform to the fault logic relationships defined in the graph. Missing key feature evidence or associated potential concurrent faults in the preliminary results are identified and supplemented. An interpretable diagnostic report is generated, including the root cause of the fault, the impact path, relevant evidence features, and recommended maintenance measures. Knowledge graphs are integrated for interpretability verification and enhancement. Structured domain knowledge is leveraged to logically verify, supplement, and deeply interpret the "black box" output of the AI ​​model, significantly improving the credibility and operability of the diagnostic results and providing decision support for maintenance personnel.

[0111] See also Figure 2 In one embodiment, the present invention further provides a CNC machine tool fault prediction and diagnosis system based on an AI model, the system comprising:

[0112] The signal acquisition unit 100 is used to collect multi-source monitoring data of the CNC machine tool, including historical signals collected from multiple motion axes of the CNC machine tool, wherein the historical signals include vibration signals, temperature signals, current signals, and speed signals;

[0113] The curve construction unit 200 is used to establish a vibration-temperature correlation curve with time series characteristics based on the vibration signal and the temperature signal, and to establish a current-speed correlation curve with time series characteristics based on the current signal and the speed signal;

[0114] The spatial coding unit 300 is used to perform inter-axis spatial linkage coding on multiple motion axes to generate inter-axis interaction response features. The inter-axis spatial linkage coding includes analyzing the signal interaction relationship between different motion axes;

[0115] The time coding unit 400 is configured to perform time scale linkage coding on the vibration-temperature correlation curve and the current-speed correlation curve to generate a time interaction response feature, and the time scale linkage coding comprises analyzing signal trend correlation at different time points.

[0116] The fault prediction unit 500 is configured to fuse the inter-axis interaction response feature and the time interaction response feature to generate a fault prediction and diagnosis result through a pre-trained AI model.

[0117] In an embodiment, the time coding unit 400 is further configured to:

[0118] The vibration-temperature correlation curve is coded by a time series encoder to obtain a vibration-temperature time series feature vector, and the current-speed correlation curve is coded by a time series encoder to obtain a current-speed time series feature vector.

[0119] The vibration-temperature time series feature vector is input into a first LSTM network to extract vibration-temperature long and short term trend features, and the current-speed time series feature vector is input into a second LSTM network to extract current-speed long and short term trend features.

[0120] The vibration-temperature long and short term trend features and the current-speed long and short term trend features are fused to generate the time interaction response feature.

[0121] In an embodiment, the space coding unit 300 is further configured to:

[0122] The signals of each motion axis are split into local feature vector sequences.

[0123] The local feature vector sequences of the plurality of motion axes are analyzed through a cross attention mechanism to quantify linkage response strength between different motion axes.

[0124] The inter-axis interaction response feature is generated based on the linkage response strength.

[0125] It can be understood that the system provided by the embodiment has functions or contains modules that can be used to execute the method described in the above method embodiment, and the specific implementation can refer to the description of the above method embodiment. For the sake of brevity, it will not be repeated here.

[0126] The present application also provides an electronic device comprising a processor and a memory, wherein the memory is configured to store computer program code, and the computer program code comprises computer instructions, and when the processor executes the computer instructions, the electronic device executes the method of any one of the above possible implementation manners.

[0127] The application further provides a computer readable storage medium, wherein a computer program is stored in the computer readable storage medium, the computer program comprises program instructions, and the program instructions, when executed by a processor of an electronic device, cause the processor to perform the method of any one of the possible implementation manners described above.

[0128] Those skilled in the art can clearly understand that the units and algorithm steps of the examples described in combination with the embodiments disclosed herein can be realized by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized in hardware or software manner depends on the specific application and design constraints of the technical solution. The skilled person can use different methods to realize the described functions for each specific application, but such implementation should not be considered beyond the scope of the application.

[0129] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the foregoing method embodiments, which will not be described herein. Those skilled in the art can also clearly understand that each embodiment of the application describes each with emphasis, and for the convenience and brevity of description, the same or similar parts in different embodiments can not be described in detail, and therefore, the parts not described or not described in detail in a certain embodiment can be referred to in the description of other embodiments.

Claims

1. A method for fault prediction and diagnosis of CNC machine tools based on AI model, characterized in that: The method comprises: Collecting multi-source monitoring data of the CNC machine tool, including historical signals collected from multiple motion axes of the CNC machine tool, the historical signals including vibration signals, temperature signals, current signals and speed signals; A vibration-temperature correlation curve with time series characteristics is established based on the vibration signal and the temperature signal, and a current-speed correlation curve with time series characteristics is established based on the current signal and the speed signal; Performing inter-axis spatial linkage encoding on multiple motion axes to generate inter-axis interactive response features, wherein the inter-axis spatial linkage encoding includes analyzing the signal mutual influence relationship between different motion axes; Performing time-scale linkage coding on the vibration-temperature correlation curve and the current-speed correlation curve to generate a time interaction response feature, wherein the time-scale linkage coding includes analyzing the signal trend correlation at different time points; The inter-axis interaction response characteristics and the time interaction response characteristics are integrated to generate fault prediction and diagnosis results through a pre-trained AI model.

2. The AI ​​model-based CNC machine tool fault prediction and diagnosis method according to claim 1, characterized in that: The time scale linkage coding includes: The vibration-temperature correlation curve is encoded by a time series encoder to obtain a vibration-temperature time series feature vector; the current-speed correlation curve is encoded by a time series encoder to obtain a current-speed time series feature vector; Input the vibration-temperature time series feature vector into the first LSTM network to extract the long-term and short-term trend features of vibration-temperature; input the current-speed time series feature vector into the second LSTM network to extract the long-term and short-term trend features of current-speed; The time interaction response feature is generated by fusing the vibration-temperature long-term and short-term trend feature with the current-speed long-term and short-term trend feature.

3. The AI ​​model-based fault prediction and diagnosis method for CNC machine tools according to claim 1, characterized in that: The inter-axis spatial linkage coding includes: Split the signal of each motion axis into a sequence of local feature vectors; performing a cross-attention mechanism analysis on the local feature vector sequences of the multiple motion axes to quantify the linkage response strength between different motion axes; Based on the linkage response strength, the inter-axis interaction response feature is generated.

4. The AI ​​model-based fault prediction and diagnosis method for CNC machine tools according to claim 1, characterized in that: The multiple motion axes include linear motion axes and rotational motion axes, and the inter-axis spatial linkage encoding further includes analyzing the dynamic coupling relationship between the linear axis and the rotational axis.

5. The AI ​​model-based fault prediction and diagnosis method for CNC machine tools according to claim 1, characterized in that: After collecting multi-source monitoring data of the CNC machine tool, the method further includes: Adaptively perform sampling frequency fusion processing on the historical signals, including identifying the inherent change frequency of each signal, and adaptively adjusting the sampling frequency of each signal based on the inherent change frequency and the current machine tool operating condition; oversampling and smoothing low-frequency signals, and downsampling and retaining the characteristics of high-frequency signals to generate a fused sampling data stream with a unified time base; Dynamically weighted denoising is performed on the fused sampled data stream, including dynamically calculating the denoising weight of each signal channel based on the signal-to-noise ratio and the correlation with the target fault; and based on the weight, performing channel-adaptive denoising processing according to wavelet transform or empirical mode decomposition.

6. A CNC machine tool fault prediction and diagnosis system based on AI model, characterized in that: The system comprises: A signal acquisition unit, configured to acquire multi-source monitoring data of a CNC machine tool, including historical signals acquired from multiple motion axes of the CNC machine tool, wherein the historical signals include vibration signals, temperature signals, current signals, and speed signals; A curve construction unit is used to establish a vibration-temperature correlation curve with time series characteristics based on the vibration signal and the temperature signal, and to establish a current-speed correlation curve with time series characteristics based on the current signal and the speed signal; A spatial encoding unit is used to perform inter-axis spatial linkage encoding on multiple motion axes to generate inter-axis interaction response features, wherein the inter-axis spatial linkage encoding includes analyzing the mutual influence relationship between signals of different motion axes; A time coding unit is used to perform time-scale linkage coding on the vibration-temperature correlation curve and the current-speed correlation curve to generate a time interaction response feature, wherein the time-scale linkage coding includes analyzing the signal trend correlation at different time points; A fault prediction unit is used to fuse the inter-axis interaction response characteristics and the time interaction response characteristics, and generate fault prediction and diagnosis results through a pre-trained AI model.

7. The AI ​​model-based CNC machine tool fault prediction and diagnosis system according to claim 6, characterized in that: The time coding unit is further configured to: The vibration-temperature correlation curve is encoded by a time series encoder to obtain a vibration-temperature time series feature vector; the current-speed correlation curve is encoded by a time series encoder to obtain a current-speed time series feature vector; Inputting the vibration-temperature time series feature vector into a first LSTM network to extract the long-term and short-term trend features of vibration-temperature; Input the current-speed time series feature vector into the second LSTM network to extract the long-term and short-term trend features of the current-speed; The time interaction response feature is generated by fusing the vibration-temperature long-term and short-term trend feature with the current-speed long-term and short-term trend feature.

8. The AI ​​model-based CNC machine tool fault prediction and diagnosis system according to claim 6, characterized in that: The spatial coding unit is further configured to: Split the signal of each motion axis into a sequence of local feature vectors; performing a cross-attention mechanism analysis on the local feature vector sequences of the multiple motion axes to quantify the linkage response strength between different motion axes; Based on the linkage response strength, the inter-axis interaction response feature is generated.

9. An electronic device, characterized in that: include: A processor and a memory, the memory being used to store computer program code, the computer program code including computer instructions, and when the processor executes the computer instructions, the electronic device executes the CNC machine tool fault prediction and diagnosis method based on the AI ​​model as described in any one of claims 1 to 5.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, which includes program instructions. When the program instructions are executed by a processor of an electronic device, the processor executes the AI ​​model-based CNC machine tool fault prediction and diagnosis method described in any one of claims 1 to 5.

Images