Fusion fault diagnosis method and system for special release motor

Abstract

The invention discloses a fusion fault diagnosis method and system for a special release motor, and relates to the technical field of fault diagnosis. The method comprises the steps of performing multi-dimensional sensing acquisition on a special release motor through a multi-source sensor array, constructing a multi-dimensional space-time sequence data set, mapping the multi-dimensional space-time sequence data set to a three-dimensional space rectangular coordinate system, and generating a motor operation state matrix; performing multi-feature analysis to obtain a fused fault feature map; inputting the fusion fault feature map into a spatio-temporal feature fusion channel for spatio-temporal analysis, and extracting spatial fault features and time fault features; and performing weighted fusion on the space fault features and the time fault features, generating a fusion fault feature vector for fault diagnosis, identifying a target fault type, and dynamically triggering a generator maintenance strategy based on the target fault type. The technical problem of inaccurate and incomplete fault diagnosis of the special release motor in the prior art is solved, and the technical effect of improving the accuracy and comprehensiveness of fault diagnosis of the motor is achieved.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of fault diagnosis, in particular to a fusion fault diagnosis method and system for special release motors. BACKGROUND

[0002] A special release motor is a high-response motor used for executing fast start-stop, directional release or locking action under specific working conditions, and is widely applied in aviation equipment, intelligent manufacturing equipment and precision actuating mechanisms. Due to the characteristics of frequent start-stop, high load impact and complex electromagnetic coupling during the operation of the special release motor, the internal structure and operating state of the motor are often highly nonlinear and time-varying, which easily leads to various compound faults such as winding short circuit, rotor eccentricity, bearing wear and drive module instability. The existing motor fault diagnosis methods are mostly based on single sensor signal or single-dimensional feature analysis, which cannot effectively reflect the spatio-temporal correlation characteristics between different dimensional operating parameters, resulting in inaccurate and incomplete diagnosis results. In addition, the traditional methods mostly rely on manual threshold setting or experience model, lack fusion and adaptive analysis capability for multi-source data, and are difficult to meet the real-time and accurate fault identification requirements of the special release motor in complex environments. SUMMARY

[0003] The application provides a fusion fault diagnosis method and system for a special release motor, which solves the technical problems of inaccurate and incomplete fault diagnosis of the special release motor in the prior art.

[0004] In a first aspect, the application provides a fusion fault diagnosis method for a special release motor, which comprises: A multi-dimensional sensing acquisition is performed on the special release motor by a multi-source sensor array, a multi-dimensional spatio-temporal sequence dataset is constructed, the multi-dimensional spatio-temporal sequence dataset is mapped to a three-dimensional space rectangular coordinate system, and a motor operating state matrix is generated; multi-feature analysis is performed based on the motor operating state matrix, a fusion fault feature map is obtained; the fusion fault feature map is input into a spatio-temporal feature fusion channel for spatio-temporal analysis, spatial fault features and temporal fault features are extracted; the spatial fault features and the temporal fault features are weighted and fused to generate a fusion fault feature vector for fault diagnosis, a target fault type is identified, and a motor maintenance strategy is dynamically triggered based on the target fault type.

[0005] In a second aspect, the application provides a fusion fault diagnosis system for a special release motor, which comprises: Data acquisition component: Performs multi-dimensional sensing and acquisition of data from a special release motor using a multi-source sensor array, constructs a multi-dimensional spatiotemporal sequence dataset, maps the multi-dimensional spatiotemporal sequence dataset to a three-dimensional Cartesian coordinate system, and generates a motor operating state matrix; Feature analysis component: Performs multi-feature analysis based on the motor operating state matrix to obtain a fused fault feature map; Spatiotemporal analysis component: Inputs the fused fault feature map into a spatiotemporal feature fusion channel for spatiotemporal analysis, extracting spatial fault features and temporal fault features; Fault diagnosis component: Performs weighted fusion of the spatial fault features and the temporal fault features to generate a fused fault feature vector for fault diagnosis, identifies the target fault type, and dynamically triggers motor maintenance strategies based on the target fault type.

[0006] One or more technical solutions provided in this application have at least the following technical effects or advantages: First, a multi-source sensor array is used to collect multi-dimensional data from a special release motor, constructing a multi-dimensional spatiotemporal sequence dataset. This dataset is then mapped to a three-dimensional Cartesian coordinate system to generate a motor operating state matrix. Next, multi-feature analysis is performed based on the motor operating state matrix to obtain a fused fault feature map. Then, the fused fault feature map is input into a spatiotemporal feature fusion channel for spatiotemporal analysis, extracting spatial and temporal fault features. Finally, the spatial and temporal fault features are weighted and fused to generate a fused fault feature vector for fault diagnosis, identifying the target fault type and dynamically triggering motor maintenance strategies based on that type. This approach solves the technical problems of inaccurate and incomplete fault diagnosis for special release motors in existing technologies, achieving a significant improvement in the accuracy and comprehensiveness of motor fault diagnosis. Attached Figure Description

[0007] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0008] Figure 1 A schematic flowchart of a fusion fault diagnosis method for a special release motor provided in an embodiment of this application; Figure 2 This is a schematic diagram of the fusion fault diagnosis system for special release motors provided in an embodiment of this application.

[0009] Explanation of reference numerals in the attached diagram: Data acquisition component 11, Feature analysis component 12, Spatiotemporal analysis component 13, Fault diagnosis component 14. Detailed Implementation

[0010] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0011] Example 1, as Figure 1 As shown, this application provides a fusion fault diagnosis method for special release motors, wherein the method includes: Multi-source sensor arrays are used to collect multi-dimensional data from a special release motor, constructing a multi-dimensional spatiotemporal sequence dataset. The multi-dimensional spatiotemporal sequence dataset is then mapped to a three-dimensional Cartesian coordinate system to generate a motor operating state matrix.

[0012] In this embodiment, a multi-source sensor array is arranged at key locations of the special release motor, including the motor stator housing, both ends of the shaft, and the heat dissipation surface of the drive module. The multi-source sensor array includes at least a triaxial vibration sensor, a current sensor, and a distributed temperature sensor. The triaxial vibration sensor is fixedly installed on the motor bearing housing and end cover to collect radial and axial vibration acceleration signals of the motor shaft in real time. The current sensor is connected in parallel to the motor drive circuit to collect three-phase current waveform data of the motor. The distributed temperature sensor array is evenly distributed on the surface of the motor stator, windings, and housing to collect temperature distribution data during motor operation. The data collected by each type of sensor is timestamped, forming a raw multi-source data stream with time-stamped information. The system segments and synchronizes the multi-source data according to a preset sampling frequency and time window to construct a multi-dimensional spatiotemporal sequence dataset, where each data dimension corresponds to a sensing channel, sampling position, and time series. A three-dimensional Cartesian coordinate system for the special release motor is constructed using the motor axis direction as the Z-axis and the motor end face direction as the X-axis and Y-axis planes. The system maps data points in each dimension to their corresponding spatial coordinates based on sensor installation location information and time synchronization sequences, generating a three-dimensional spatially distributed data matrix. Furthermore, by overlaying continuous sampling windows in the time dimension to form a spatiotemporally coupled data cube, the system performs interpolation, smoothing, and normalization on the sensor data at different spatial locations and time points, thereby generating a motor operating state matrix reflecting the overall operating state of the motor.

[0013] Furthermore, a multi-dimensional spatiotemporal sequence dataset is constructed by using a multi-source sensor array to perform multi-dimensional sensing and data acquisition on a special release motor. The methods include: Multi-directional vibration acceleration signals are obtained by acquiring data from the shaft of a special release motor using a triaxial vibration sensor; three-phase current waveform data is obtained by sensing the drive circuit of the special release motor using a current sensor; and temperature distribution data is obtained by sensing the temperature of the special release motor using a distributed temperature sensor array. The multi-directional vibration acceleration signals, the three-phase current waveform data, and the temperature distribution data are traversed according to the sensing time sequence, and timestamps are added to generate an original dataset with time stamp information. A time window is constructed based on the time stamp information, and the original dataset is segmented according to the time window to obtain multiple data segments for time-domain analysis and combination, thus constructing the multi-dimensional spatiotemporal sequence dataset.

[0014] The triaxial vibration sensors are fixedly installed on key parts such as the motor bearing housing, end cover, and housing to collect multi-directional vibration acceleration signals of the motor shaft in real time in the X, Y, and Z directions, reflecting the dynamic characteristics of the rotating parts in the radial, tangential, and axial directions. The current sensor is connected to the three-phase power supply terminal of the motor drive circuit to synchronously collect the three-phase current waveform data of the motor, reflecting the electromagnetic drive characteristics of the motor. The distributed temperature sensor array is evenly distributed in the stator core, winding, and housing areas to collect temperature distribution data during motor operation, reflecting local heating and insulation aging trends.

[0015] The system iterates through multi-directional vibration acceleration signals, three-phase current waveform data, and temperature distribution data, and automatically adds a unified timestamp according to the sampling sequence to form a raw dataset with time stamp information.

[0016] To ensure timing consistency of data across different sampling channels, the system uses the master clock synchronization signal as the time reference and performs time alignment processing on various sensor signals. Specifically, a sliding time window is constructed based on time stamp information, with the window length automatically set according to the motor rotation cycle or sampling frequency, for example, ranging from 0.1s to 1s. The original dataset is segmented according to the time window to obtain multiple time-continuous data segments. Time-domain feature combination and channel normalization operations are performed on each data segment to ensure comparability of data across different dimensions. Finally, various sensor signals are combined in three dimensions—spatial location, time series, and signal type—to construct a multidimensional spatiotemporal sequence dataset. This multidimensional spatiotemporal sequence dataset, with time as the primary dimension and sensor channels as the secondary dimension, can comprehensively characterize the dynamic operating characteristics of the special release motor at different times and spatial measurement points.

[0017] Furthermore, the method for mapping the multidimensional spatiotemporal sequence dataset to a three-dimensional Cartesian coordinate system to generate a motor operating state matrix includes: Using the motor axis of the special release motor as the Z-axis and the motor end face of the special release motor as the X-axis and Y-axis planes, the intersection points of the Z-axis and the X-axis and Y-axis planes are perpendicularly combined to construct the three-dimensional Cartesian coordinate system of the special release motor. The multi-dimensional spatiotemporal sequence dataset is sensor-read to obtain data position information. The data position information is mapped to the three-dimensional Cartesian coordinate system for matching to determine the position coordinate correspondence. Based on the position coordinate correspondence, multiple data coordinates are determined. According to the multiple data sampling time sequences, the multi-dimensional spatiotemporal sequence dataset is arranged according to the multiple data coordinates to construct a four-dimensional data array. Based on the four-dimensional data array, the operating state of the special release motor is analyzed, and interpolation is performed based on the analysis results to construct the motor operating state matrix.

[0018] A three-dimensional spatial reference system is established based on the geometry of a special release motor. The motor's axial direction is defined as the Z-axis, and the motor's end face is defined as the X-axis and Y-axis planes. The intersection of the Z-axis and the X-axis and Y-axis planes is the origin, used to identify the center position of the motor. By perpendicularly combining the intersection points of the Z-axis and the X-axis and Y-axis planes, a complete three-dimensional rectangular coordinate system can be formed to characterize the spatial distribution relationship of internal and external sensors of the motor.

[0019] Based on the sensor's installation coordinates on and inside the motor surface, the system reads the physical position information corresponding to each sensing channel in the multidimensional spatiotemporal sequence dataset, including axial, radial, and angular coordinates. Subsequently, the spatial coordinate values ​​of each sensing position are mapped to a three-dimensional Cartesian coordinate system for matching, determining the positional coordinate correspondence between the spatial point and the sensor channel.

[0020] After establishing spatial matching relationships, the system arranges the multidimensional spatiotemporal sequence dataset according to spatial coordinate indices and time sequence order based on the sampled time series, forming a four-dimensional data array indexed by (X, Y, Z, T). The three-dimensional spatial coordinates reflect the physical location distribution, and the time dimension T reflects the sampling time or window number. To eliminate spatial gaps caused by uneven sampling intervals and locations, the system performs multi-point interpolation on the four-dimensional data array, using cubic splines or inverse distance weighted algorithms to interpolate spatial gaps, and combines this with smoothing filtering within a time sliding window to correct the time dimension.

[0021] After interpolation, the resulting four-dimensional array is normalized and spatially reconstructed. The comprehensive feature values ​​of each spatial point within the time window (such as root mean square vibration, mean current, temperature gradient, etc.) are extracted and formed into a physically meaningful motor operating state matrix in a three-dimensional coordinate system. Each element of the motor operating state matrix corresponds to a multi-source signal fusion state value at a spatial location, which can intuitively reflect the motor's operating health status at different locations and times.

[0022] Multi-feature analysis is performed based on the motor operating state matrix to obtain a fused fault feature map.

[0023] Furthermore, based on the motor operating state matrix, multi-feature analysis is performed to obtain a fused fault feature map. The method includes: Based on the motor operating state matrix, operational analysis is performed to obtain vibration signals, current signals, and temperature distribution signals. Time-domain analysis is performed on the vibration signals to extract time-domain feature sets, which are then integrated according to the multiple data coordinates to construct a time-domain feature matrix. Frequency-domain analysis is performed on the current signals to extract frequency-domain feature sets, which are then correlated with the multiple data coordinates to construct a frequency-domain feature distribution map. Time-frequency analysis is performed on the temperature distribution signals to obtain temperature time-frequency analysis results, which are then integrated with the multiple data coordinates to construct a time-frequency domain feature set. The time-domain feature set, the frequency-domain feature set, and the time-frequency domain feature set are normalized, and the normalization results are then fused for feature visualization to construct a two-dimensional image. Image enhancement is performed on the two-dimensional image to construct the fused fault feature map.

[0024] The system performs operational analysis on the motor operating state matrix, separating the multi-dimensional signal channels to obtain vibration signals, current signals, and temperature distribution signals corresponding to the motor's operating state. The vibration signal characterizes the motor's mechanical vibration characteristics, the current signal reflects the motor's electromagnetic drive characteristics, and the temperature signal describes the motor's heat conduction and cooling characteristics.

[0025] After acquiring the vibration signal, the system performs time-domain analysis, including calculating statistical indicators such as root mean square value, peak value, variance, kurtosis, and waveform factor, to capture characteristics such as bearing wear, rotor eccentricity, and mechanical shock. By integrating the time-domain feature set according to multi-dimensional data coordinates, a time-domain feature matrix is ​​constructed to describe the time-domain feature distribution at different spatial locations. Subsequently, frequency-domain analysis is performed on the current signal. Frequency-domain feature sets, including fundamental amplitude, harmonic distribution, spectral energy concentration, and characteristic frequency components, are extracted through Fast Fourier Transform or power spectral density analysis. These features are then correlated with the corresponding spatial coordinates to form a frequency-domain feature distribution map, which reflects stator current imbalance or short-circuit characteristics.

[0026] For the temperature distribution signal, the system employs short-time Fourier transform or continuous wavelet transform for time-frequency analysis to obtain the frequency distribution characteristics of temperature changes over time, thereby characterizing the dynamics of heat conduction and abnormal heating regions during motor operation. The obtained temperature time-frequency characteristics are integrated with spatial coordinates to construct a time-frequency domain feature set.

[0027] The system normalizes the time-domain, frequency-domain, and time-frequency-domain feature sets respectively to ensure consistent numerical ranges for each feature, thereby eliminating the influence of different dimensions. The normalized multidimensional features are then combined and mapped according to spatial coordinates. A feature visualization method projects the high-dimensional feature data onto a two-dimensional plane, generating a two-dimensional feature image. Each pixel in this image corresponds to a comprehensive state feature value at a spatial location, visually reflecting the multidimensional operating feature distribution of the motor. Finally, image enhancement processing is performed on the two-dimensional feature image, including contrast enhancement, smoothing filtering, and edge sharpening, to improve the clarity and distinguishability of the feature regions. The enhanced image is the fused fault feature map, which integrates the temporal, frequency, and spatial features of multiple sources such as vibration, temperature, and current, and can intuitively represent abnormal patterns in the motor's operating state in image form.

[0028] Furthermore, the time-domain feature set, the frequency-domain feature set, and the time-frequency-domain feature set are normalized, and the normalization results are then fused for feature visualization to construct a two-dimensional image. The method includes: Based on the normalization result, the time-domain feature set, frequency-domain feature set, and time-frequency-domain feature set are dimensionality reduced to obtain principal component feature analysis results. Based on the principal component feature analysis results, symmetry point analysis is performed to determine multiple symmetry points. The principal component feature analysis results are then projected onto a two-dimensional plane according to these multiple symmetry points to obtain a two-dimensional feature map. Based on the two-dimensional feature map, bilinear interpolation is performed to set the target image resolution. Finally, the two-dimensional feature maps are integrated according to the target image resolution to construct the two-dimensional image.

[0029] For each feature dimension in the time-domain, frequency-domain, and time-frequency-domain feature sets, the system employs min-max normalization or Z-score standardization to transform the feature values ​​to a standardized interval of [0, 1] or a mean of 0 and a variance of 1, thereby eliminating the influence of dimensional differences on subsequent analysis results. The normalized feature data are then used to construct a high-dimensional feature matrix through multi-dimensional vector combinations, where each row represents a fused feature vector of a spatial sampling point under different feature dimensions.

[0030] After normalization, the system performs dimensionality reduction on the high-dimensional feature matrix and uses principal component analysis (PCA) to extract the main feature components. PCA decomposes the feature covariance matrix and selects principal components with a cumulative variance contribution rate of no less than 90%, forming the PCA feature analysis results. This approach reduces redundant data dimensionality while retaining key feature information.

[0031] The system performs symmetry point analysis based on principal component feature analysis results, identifying the symmetric structure of feature points in the dimensionality-reduced feature distribution space. By calculating the Euclidean distance and distribution angle of feature points relative to the data center, multiple sets of symmetric points are determined to reflect the similarity and symmetric distribution characteristics between different types of fault features. Based on the spatial distribution of multiple symmetric points, the system uses a projection mapping algorithm to project the principal component feature analysis results onto a two-dimensional plane, obtaining a two-dimensional feature map. This two-dimensional feature map uses coordinate axes to represent principal component 1 and principal component 2, and the brightness or color of pixels in the map represents feature intensity, visually displaying the feature differences between different spatial points.

[0032] The system performs bilinear interpolation on the two-dimensional feature maps. By calculating linear weights between adjacent pixels, it generates smoothly transitioning pixel values ​​and sets the target image resolution, such as 256×256 or 512×512 pixels, to meet the input requirements for subsequent convolutional feature extraction. Finally, the system spatially integrates all two-dimensional feature maps according to the target image resolution to obtain a two-dimensional image that incorporates multi-dimensional feature information.

[0033] The fused fault feature map is input into the spatiotemporal feature fusion channel for spatiotemporal analysis to extract spatial fault features and temporal fault features.

[0034] Furthermore, the fused fault feature map is input into the spatiotemporal feature fusion channel for spatiotemporal analysis to extract spatial fault features and temporal fault features. The method includes: The fused fault feature map is input into a spatiotemporal feature fusion channel, which includes a spatial feature extraction branch, a temporal feature extraction branch, and a feature fusion module. Based on the spatial feature extraction branch, multi-scale convolutional analysis is performed on the fused fault feature map to obtain spatial structural features. Based on the temporal feature extraction branch, cyclic temporal dimension analysis is performed on the fused fault feature map to obtain temporal dimension features. The spatial structural features and the temporal dimension features are synchronized to the feature fusion module for feature interaction enhancement to generate spatiotemporally enhanced fault features, which include the spatial fault features and the temporal fault features.

[0035] The enhanced and fused fault feature map is input into the spatiotemporal feature fusion channel, which consists of a spatial feature extraction branch, a temporal feature extraction branch, and a feature fusion module. The spatial feature extraction branch extracts the structural features of the feature map in the spatial distribution dimension, the temporal feature extraction branch extracts the dynamic patterns of feature changes over time, and the feature fusion module performs information interaction and enhanced fusion of spatial and temporal features to generate a comprehensive spatiotemporal feature representation.

[0036] In the spatial feature extraction branch, the system performs multi-scale convolutional analysis on the fused fault feature map based on a convolutional neural network structure. Specifically, the spatial feature extraction branch employs a feature pyramid network or a multi-resolution convolutional architecture, using convolutional kernels of different sizes at different scales to extract global structural information, local region features, and edge detail features. Shallow convolutional layers extract local energy changes and edge texture features, while deep convolutional layers extract high-level abstract structural features. The multi-layer output features are fused through skip connections and channel concatenation to construct a multi-scale spatial structural feature map, thus fully preserving the spatial information in the fused fault feature map.

[0037] In the time feature extraction branch, the system employs a recurrent neural network or a long short-term memory network to perform cyclic time dimension analysis on the fused fault feature map within a continuous time window. The time feature extraction branch takes the feature frame sequence of the fused fault feature map as input, extracts features frame by frame, and transmits hidden state information along the time axis to model the dependency relationship of fault features changing over time. Through forward and backward propagation, the system can capture the feature evolution patterns of the motor's operating state over short and long time spans, thereby obtaining time dimension features that reflect the time response patterns and dynamic stability of different types of faults.

[0038] Spatial structural features and temporal features are simultaneously input into the feature fusion module for interactive feature enhancement. The feature fusion module employs a weighted attention mechanism and a feature channel matching strategy, calculating the spatial feature weight matrix and the temporal feature correlation matrix to achieve interactive learning and joint optimization between spatial and temporal features. Through the feature fusion module, the system can dynamically adjust the contribution ratio of spatial and temporal features in the overall representation, enhancing the weight of important feature regions and suppressing redundant feature information. The final output spatiotemporally enhanced fault features include two parts: spatial fault features and temporal fault features. The former is used to identify structural anomalies in the spatial distribution of the motor, while the latter reflects the temporal evolution characteristics of fault occurrence and development.

[0039] Furthermore, based on the spatial feature extraction branch, multi-scale convolutional analysis is performed on the fused fault feature map to obtain spatial structural features. The method includes: A feature pyramid network structure is adopted to construct a multi-scale feature extraction path, which includes a first-scale path, a second-scale path, and a third-scale path. Based on the first-scale path, a first-size convolutional kernel is used to extract global context information of the fused fault feature map. Based on the second-scale path, a first-size convolutional kernel is used to extract local region features of the fused fault feature map. Based on the third-scale path, a first-size convolutional kernel is used to extract detailed features of the fused fault feature map. The global context information, the local region features, and the detailed features are fused by skip connections to obtain the spatial structure features.

[0040] The system employs a feature pyramid network structure to construct multi-scale feature extraction paths, enabling the simultaneous capture of global structural features, local texture features, and detailed edge features at different spatial resolutions. These multi-scale feature extraction paths include a first-scale path, a second-scale path, and a third-scale path, each corresponding to a combination of convolutional layers with different receptive fields, to achieve multi-level spatial information extraction.

[0041] In the first-scale path, the system uses relatively large convolutional kernels (such as 7×7 or 5×5) to perform convolution operations on the fused fault feature map to extract global contextual information. This path can capture the overall energy distribution, structural contour, and regional trends of the feature map, which can be used to characterize the spatial variation characteristics of the motor's operating state at the macroscopic level.

[0042] In the second-scale path, the system uses medium-sized convolutional kernels (e.g., 3×3) for convolutional analysis, focusing on extracting features from local regions. This path can identify abnormal patterns in local areas on or inside the motor surface, such as areas of concentrated vibration, areas of current density distortion, or areas of localized temperature rise, thereby revealing fault-sensitive features in spatially localized areas.

[0043] In the third-scale path, the system uses a smaller convolutional kernel (e.g., 1×1) to perform high-resolution convolutional analysis, extracting detailed information and edge structure features from the feature map to detect small-scale fault signals, such as microscopic anomalies like bearing cracks, rotor eccentric boundaries, or stator coil hot spot edges.

[0044] The system uses skip connections to fusion feature maps from different scale paths, performing dimensional alignment and feature concatenation. Specifically, the detailed features output from the third-scale path are upsampled to the same resolution as the second-scale path output, and then fused with the second-scale features at the channel level. Subsequently, the fused result is upsampled again and integrated with the global context features of the first-scale path to form a multi-scale feature fusion tensor. The fused features simultaneously contain macroscopic trend information and detailed anomaly features, and the output serves as a spatial structure feature to reflect the operating status distribution and potential fault areas of the special release motor in three-dimensional space.

[0045] The spatial fault features and the temporal fault features are weighted and fused to generate a fused fault feature vector for fault diagnosis, identifying the target fault type, and dynamically triggering motor maintenance strategies based on the target fault type.

[0046] Furthermore, the spatial fault features and the temporal fault features are weighted and fused to generate a fused fault feature vector. The method includes: Information entropy is evaluated for the spatial fault features based on multiple spatial feature dimensions, and the information entropy value of the spatial fault features is calculated. Feature stability is evaluated for the temporal fault features based on temporal correlation, and the temporal correlation coefficient of the temporal fault features is calculated. The spatial fault feature weight coefficient and the temporal fault feature weight coefficient are obtained by weighted summation based on the information entropy value of the spatial fault features and the temporal correlation coefficient of the temporal fault features. The spatial fault features and the temporal fault features are then weighted and fused according to the spatial fault feature weight coefficient and the temporal fault feature weight coefficient to generate the fused fault feature vector.

[0047] The system evaluates the information entropy of the spatial fault features based on multiple spatial feature dimensions. Specifically, the feature distribution of each dimension in the spatial feature matrix is ​​treated as a random variable, and its probability distribution is calculated. Obtain the information entropy value The information entropy value is used to measure the uncertainty and dispersion of spatial features. A higher information entropy indicates that the spatial features are significantly different and contribute more to fault location. By normalizing the entropy values ​​of different spatial channels, a basic spatial weight index reflecting the importance of features can be obtained.

[0048] The system assesses the stability of the time-related fault features based on time-domain correlation. The time-related fault feature sequence typically contains dynamic changes in the motor's operating state across different time segments. The system evaluates the stability of the time series by calculating the autocorrelation function or Pearson correlation coefficient.

[0049] Let the time feature sequence be Then its time series correlation coefficient is defined as: Where Cov represents covariance. Indicates standard deviation, This is the time delay step. If A higher value indicates that the feature is highly stable over time and can effectively reflect the evolution of faults over time. The system averages and normalizes the correlation coefficients of features across different time dimensions to obtain a time feature stability weight index.

[0050] The system performs a weighted summation calculation based on the information entropy value of the spatial fault features and the temporal correlation coefficient of the temporal fault features to determine the weight coefficient of the spatial fault features. Weighting coefficients of time-related fault characteristics The specific calculation formula is as follows: , .

[0051] The system performs linear weighted fusion of the two types of features based on the calculated weight coefficients for spatial and temporal fault features. The weighted fusion process can be represented as: Where F represents the fused fault feature vector, S represents the spatial fault feature matrix, and T represents the temporal fault feature matrix. The system normalizes and rearranges the weighted results to obtain the final fused fault feature vector.

[0052] Furthermore, the method involves generating a fused fault feature vector for fault diagnosis, identifying the target fault type, and dynamically triggering a motor maintenance strategy based on the target fault type. The method includes: A random forest algorithm is used to train a multi-class support vector machine as the base classifier to construct a fault classifier. The fused fault feature vector is synchronized to the fault classifier for fault classification, identifying multiple fault types. Confidence analysis is performed on the multiple fault types to obtain multiple confidence scores. The multiple fault types are sorted in descending order according to the multiple confidence scores, and the fault type with the highest order is selected as the target fault type. Maintenance analysis is performed on the multiple fault types to obtain multiple motor maintenance strategies. The multiple fault types are matched and associated with the multiple motor maintenance strategies to construct a fault-maintenance mapping relationship network. The target fault type is used as an index to retrieve the fault-maintenance mapping relationship network to generate an initial motor maintenance strategy. Maintenance evaluation is performed on special release motors based on the initial motor maintenance strategy, and a maintenance effect score is generated and fed back to the initial motor maintenance strategy for updating, thus generating the final motor maintenance strategy.

[0053] The system constructs a fault classifier based on historical operational sample data. Training samples include fused fault feature vectors and their corresponding known fault labels. To improve the discriminative performance and generalization ability of the classification model, the system employs a random forest algorithm to combine multi-class support vector machines as base classifiers for training. Each support vector machine classifier learns independently using different feature subsets and sample subsets, achieving multi-model complementarity through bootstrapping and random feature selection mechanisms. The ensemble output of the random forest determines the final classification result through majority voting or confidence weighting, thus constructing a robust motor fault classifier.

[0054] During the fault diagnosis phase, the system inputs the fused fault feature vectors extracted in real time into the fault classifier for parallel inference and fault identification. The fault classifier performs multidimensional feature mapping and hyperplane discrimination on the input features, outputting multiple candidate fault types and their corresponding classification confidence scores. These fault types include, but are not limited to, bearing wear faults, rotor eccentricity faults, winding short-circuit faults, and heat dissipation failure faults. The system sorts the classification results according to their confidence scores, obtaining multiple confidence scores, and then selects the type with the highest confidence score in descending order as the target fault type, thereby achieving automatic identification of the target fault.

[0055] After identifying the target fault type, the system enters the maintenance strategy generation phase. The system invokes a pre-defined fault-maintenance mapping network, which consists of a multi-dimensional mapping matrix and records the association rules and priority information between various faults and their corresponding maintenance measures. Using historical maintenance records and expert rules from the maintenance knowledge base, the system configures a set of candidate maintenance strategies for each type of fault, such as: changing lubricating oil, correcting shaft balance, adjusting excitation current, cleaning heat dissipation channels, and updating control parameters.

[0056] The system uses the target fault type as an index to search the fault-maintenance mapping network and generate an initial motor maintenance strategy. This initial strategy includes maintenance steps, execution priorities, and expected maintenance results. After performing maintenance actions, the system collects and evaluates the motor's operating status data in real time, calculates maintenance effectiveness scoring indicators (including vibration attenuation rate, temperature stabilization rate, and current waveform recovery rate), and feeds the scoring results back to the maintenance strategy module. When the maintenance effectiveness score is lower than a preset threshold, the system automatically updates the parameters or reselects the strategy based on the feedback results. For example, it may adjust the maintenance execution order, add auxiliary calibration steps, or replace alternative measures to form an optimized motor maintenance strategy. After optimization, the system sends the updated maintenance strategy to the execution layer, realizing a dynamic adaptive maintenance closed loop.

[0057] In summary, the embodiments of this application have at least the following technical effects: First, a multi-source sensor array is used to collect multi-dimensional data from a special release motor, constructing a multi-dimensional spatiotemporal sequence dataset. This dataset is then mapped to a three-dimensional Cartesian coordinate system to generate a motor operating state matrix. Next, multi-feature analysis is performed based on the motor operating state matrix to obtain a fused fault feature map. Then, the fused fault feature map is input into a spatiotemporal feature fusion channel for spatiotemporal analysis, extracting spatial and temporal fault features. Finally, the spatial and temporal fault features are weighted and fused to generate a fused fault feature vector for fault diagnosis, identifying the target fault type and dynamically triggering motor maintenance strategies based on that type. This approach solves the technical problems of inaccurate and incomplete fault diagnosis for special release motors in existing technologies, achieving a significant improvement in the accuracy and comprehensiveness of motor fault diagnosis.

[0058] Example 2, based on the same inventive concept as the fusion fault diagnosis method for special release motors in the foregoing examples, such as... Figure 2 As shown, this application provides a fusion fault diagnosis system for special release motors, wherein the system includes: Data acquisition component 11: Performs multi-dimensional sensing and acquisition of data from a special release motor using a multi-source sensor array, constructs a multi-dimensional spatiotemporal sequence dataset, maps the multi-dimensional spatiotemporal sequence dataset to a three-dimensional Cartesian coordinate system, and generates a motor operating state matrix; Feature analysis component 12: Performs multi-feature analysis based on the motor operating state matrix to obtain a fused fault feature map; Spatiotemporal analysis component 13: Inputs the fused fault feature map into a spatiotemporal feature fusion channel for spatiotemporal analysis, extracting spatial fault features and temporal fault features; Fault diagnosis component 14: Performs weighted fusion of the spatial fault features and the temporal fault features to generate a fused fault feature vector for fault diagnosis, identifies the target fault type, and dynamically triggers motor maintenance strategies based on the target fault type.

[0059] Furthermore, the data acquisition component 11 is used to perform the following methods: Multi-directional vibration acceleration signals are obtained by acquiring data from the shaft of a special release motor using a triaxial vibration sensor; three-phase current waveform data is obtained by sensing the drive circuit of the special release motor using a current sensor; and temperature distribution data is obtained by sensing the temperature of the special release motor using a distributed temperature sensor array. The multi-directional vibration acceleration signals, the three-phase current waveform data, and the temperature distribution data are traversed according to the sensing time sequence, and timestamps are added to generate an original dataset with time stamp information. A time window is constructed based on the time stamp information, and the original dataset is segmented according to the time window to obtain multiple data segments for time-domain analysis and combination, thus constructing the multi-dimensional spatiotemporal sequence dataset.

[0060] Furthermore, the data acquisition component 11 is used to perform the following methods: Using the motor axis of the special release motor as the Z-axis and the motor end face of the special release motor as the X-axis and Y-axis planes, the intersection points of the Z-axis and the X-axis and Y-axis planes are perpendicularly combined to construct the three-dimensional Cartesian coordinate system of the special release motor. The multi-dimensional spatiotemporal sequence dataset is sensor-read to obtain data position information. The data position information is mapped to the three-dimensional Cartesian coordinate system for matching to determine the position coordinate correspondence. Based on the position coordinate correspondence, multiple data coordinates are determined. According to the multiple data sampling time sequences, the multi-dimensional spatiotemporal sequence dataset is arranged according to the multiple data coordinates to construct a four-dimensional data array. Based on the four-dimensional data array, the operating state of the special release motor is analyzed, and interpolation is performed based on the analysis results to construct the motor operating state matrix.

[0061] Furthermore, the feature analysis component 12 is used to perform the following methods: Based on the motor operating state matrix, operational analysis is performed to obtain vibration signals, current signals, and temperature distribution signals. Time-domain analysis is performed on the vibration signals to extract time-domain feature sets, which are then integrated according to the multiple data coordinates to construct a time-domain feature matrix. Frequency-domain analysis is performed on the current signals to extract frequency-domain feature sets, which are then correlated with the multiple data coordinates to construct a frequency-domain feature distribution map. Time-frequency analysis is performed on the temperature distribution signals to obtain temperature time-frequency analysis results, which are then integrated with the multiple data coordinates to construct a time-frequency domain feature set. The time-domain feature set, the frequency-domain feature set, and the time-frequency domain feature set are normalized, and the normalization results are then fused for feature visualization to construct a two-dimensional image. Image enhancement is performed on the two-dimensional image to construct the fused fault feature map.

[0062] Furthermore, the feature analysis component 12 is used to perform the following methods: Based on the normalization result, the time-domain feature set, frequency-domain feature set, and time-frequency-domain feature set are dimensionality reduced to obtain principal component feature analysis results. Based on the principal component feature analysis results, symmetry point analysis is performed to determine multiple symmetry points. The principal component feature analysis results are then projected onto a two-dimensional plane according to these multiple symmetry points to obtain a two-dimensional feature map. Based on the two-dimensional feature map, bilinear interpolation is performed to set the target image resolution. Finally, the two-dimensional feature maps are integrated according to the target image resolution to construct the two-dimensional image.

[0063] Furthermore, the spatiotemporal analysis component 13 is used to perform the following methods: The fused fault feature map is input into a spatiotemporal feature fusion channel, which includes a spatial feature extraction branch, a temporal feature extraction branch, and a feature fusion module. Based on the spatial feature extraction branch, multi-scale convolutional analysis is performed on the fused fault feature map to obtain spatial structural features. Based on the temporal feature extraction branch, cyclic temporal dimension analysis is performed on the fused fault feature map to obtain temporal dimension features. The spatial structural features and the temporal dimension features are synchronized to the feature fusion module for feature interaction enhancement to generate spatiotemporally enhanced fault features, which include the spatial fault features and the temporal fault features.

[0064] Furthermore, the spatiotemporal analysis component 13 is used to perform the following methods: A feature pyramid network structure is adopted to construct a multi-scale feature extraction path, which includes a first-scale path, a second-scale path, and a third-scale path. Based on the first-scale path, a first-size convolutional kernel is used to extract global context information of the fused fault feature map. Based on the second-scale path, a first-size convolutional kernel is used to extract local region features of the fused fault feature map. Based on the third-scale path, a first-size convolutional kernel is used to extract detailed features of the fused fault feature map. The global context information, the local region features, and the detailed features are fused by skip connections to obtain the spatial structure features.

[0065] Furthermore, the fault diagnosis component 14 is used to perform the following method: Information entropy is evaluated for the spatial fault features based on multiple spatial feature dimensions, and the information entropy value of the spatial fault features is calculated. Feature stability is evaluated for the temporal fault features based on temporal correlation, and the temporal correlation coefficient of the temporal fault features is calculated. The spatial fault feature weight coefficient and the temporal fault feature weight coefficient are obtained by weighted summation based on the information entropy value of the spatial fault features and the temporal correlation coefficient of the temporal fault features. The spatial fault features and the temporal fault features are then weighted and fused according to the spatial fault feature weight coefficient and the temporal fault feature weight coefficient to generate the fused fault feature vector.

[0066] Furthermore, the fault diagnosis component 14 is used to perform the following method: A random forest algorithm is used to train a multi-class support vector machine as the base classifier to construct a fault classifier. The fused fault feature vector is synchronized to the fault classifier for fault classification, identifying multiple fault types. Confidence analysis is performed on the multiple fault types to obtain multiple confidence scores. The multiple fault types are sorted in descending order according to the multiple confidence scores, and the fault type with the highest order is selected as the target fault type. Maintenance analysis is performed on the multiple fault types to obtain multiple motor maintenance strategies. The multiple fault types are matched and associated with the multiple motor maintenance strategies to construct a fault-maintenance mapping relationship network. The target fault type is used as an index to retrieve the fault-maintenance mapping relationship network to generate an initial motor maintenance strategy. Maintenance evaluation is performed on special release motors based on the initial motor maintenance strategy, and a maintenance effect score is generated and fed back to the initial motor maintenance strategy for updating, thus generating the final motor maintenance strategy.

[0067] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A fusion fault diagnosis method for special release motors, characterized in that, The method includes: A multi-source sensor array is used to perform multi-dimensional sensing and data acquisition on a special release motor, and a multi-dimensional spatiotemporal sequence dataset is constructed. The multi-dimensional spatiotemporal sequence dataset is then mapped to a three-dimensional Cartesian coordinate system to generate a motor operating state matrix. Multi-feature analysis is performed based on the motor operating state matrix to obtain a fused fault feature map; The fused fault feature map is input into the spatiotemporal feature fusion channel for spatiotemporal analysis to extract spatial fault features and temporal fault features. The spatial fault features and the temporal fault features are weighted and fused to generate a fused fault feature vector for fault diagnosis, identifying the target fault type, and dynamically triggering motor maintenance strategies based on the target fault type.

2. The fusion fault diagnosis method for special release motors as described in claim 1, characterized in that, A multidimensional spatiotemporal sequence dataset is constructed by using a multi-source sensor array to perform multidimensional sensing and data acquisition on a special release motor. The method includes: Multi-directional vibration acceleration signals are obtained by collecting data from the shaft of a special release motor using a triaxial vibration sensor. The drive circuit of the special release motor is current-sensing by a current sensor to obtain three-phase current waveform data. Temperature distribution data is obtained by sensing the temperature of a special release motor using a distributed temperature sensor array. The multi-directional vibration acceleration signal, the three-phase current waveform data, and the temperature distribution data are traversed according to the sensing time sequence, and timestamps are added to generate an original dataset, which has time stamp information. A time window is constructed based on the time stamp information. The original dataset is segmented according to the time window to obtain multiple data segments for time-domain analysis and combination to construct the multidimensional spatiotemporal sequence dataset.

3. The fusion fault diagnosis method for special release motors as described in claim 1, characterized in that, The method for mapping the multidimensional spatiotemporal sequence dataset to a three-dimensional Cartesian coordinate system to generate a motor operating state matrix includes: Using the motor axis of the special release motor as the Z-axis and the motor end face of the special release motor as the X-axis and Y-axis planes, the intersection points of the Z-axis and the X-axis and Y-axis planes are perpendicularly combined to construct the three-dimensional rectangular coordinate system of the special release motor. The multidimensional spatiotemporal sequence dataset is sensor-based to read the data location information. The data location information is mapped to the three-dimensional Cartesian coordinate system for matching to determine the correspondence of position coordinates; Based on the location coordinate correspondence, multiple data coordinates are determined, and the multidimensional spatiotemporal sequence dataset is arranged according to the multiple data coordinates according to the multiple data sampling time sequence to construct a four-dimensional data array; The operating status of the special release motor is analyzed based on the four-dimensional data array, and the motor operating status matrix is ​​constructed by interpolation based on the analysis results.

4. The fusion fault diagnosis method for special release motors as described in claim 3, characterized in that, Based on the motor operating state matrix, multi-feature analysis is performed to obtain a fused fault feature map. The method includes: Based on the motor operating state matrix, operation analysis is performed to obtain vibration signals, current signals, and temperature distribution signals; Based on the vibration signal, time-domain analysis is performed to extract a time-domain feature set. The time-domain feature set is then integrated according to the multiple data coordinates to construct a time-domain feature matrix. Frequency domain analysis is performed based on the current signal to extract a frequency domain feature set. The frequency domain feature set is then associated with the multiple data coordinates to construct a frequency domain feature distribution map. Time-frequency analysis is performed based on the temperature distribution signal to obtain temperature time-frequency analysis results. The temperature time-frequency analysis results are then integrated with the multiple data coordinates to construct a time-frequency domain feature set. The time-domain feature set, the frequency-domain feature set, and the time-frequency-domain feature set are normalized, and the normalization results are fused for feature visualization to construct a two-dimensional image. Image enhancement is performed on the two-dimensional image to construct the fused fault feature map.

5. The fusion fault diagnosis method for special release motors as described in claim 4, characterized in that, The time-domain feature set, the frequency-domain feature set, and the time-frequency-domain feature set are normalized, and the normalization results are then fused for feature visualization to construct a two-dimensional image. The method includes: Based on the normalization results, the time-domain feature set, frequency-domain feature set, and time-frequency-domain feature set are dimensionality-reduced to obtain the principal component feature analysis results. Based on the principal component feature analysis results, symmetry point analysis is performed to determine multiple symmetry points. The principal component feature analysis results are then projected onto a two-dimensional plane according to the multiple symmetry points to obtain a two-dimensional feature map. Bilinear interpolation is performed based on the two-dimensional feature map to set the target map resolution; The two-dimensional feature maps are integrated according to the target image resolution to construct the two-dimensional image.

6. The fusion fault diagnosis method for special release motors as described in claim 1, characterized in that, The fused fault feature map is input into the spatiotemporal feature fusion channel for spatiotemporal analysis to extract spatial fault features and temporal fault features. The method includes: The fused fault feature map is input into the spatiotemporal feature fusion channel, which includes a spatial feature extraction branch, a temporal feature extraction branch, and a feature fusion module. Based on the spatial feature extraction branch, multi-scale convolutional analysis is performed on the fused fault feature map to obtain spatial structural features; Based on the time feature extraction branch, the fused fault feature map is subjected to cyclic time dimension analysis to obtain time dimension features; The spatial structure features and the temporal dimension features are synchronized to the feature fusion module for feature interaction enhancement to generate spatiotemporal enhanced fault features, which include the spatial fault features and the temporal fault features.

7. The fusion fault diagnosis method for special release motors as described in claim 6, characterized in that, Based on the spatial feature extraction branch, multi-scale convolutional analysis is performed on the fused fault feature map to obtain spatial structural features. The method includes: A feature pyramid network structure is adopted to construct a multi-scale feature extraction path, which includes a first-scale path, a second-scale path, and a third-scale path. Global context information of the fused fault feature map is extracted using a first-size convolutional kernel based on the first-scale path; Based on the second-scale path, local region features of the fused fault feature map are extracted using a first-size convolutional kernel; Based on the third-scale path, detailed features of the fused fault feature map are extracted using a first-size convolutional kernel. The spatial structure features are obtained by skip-connecting and fusing the global context information, the local region features, and the detailed features.

8. The fusion fault diagnosis method for special release motors as described in claim 1, characterized in that, The method involves weighted fusion of the spatial fault features and the temporal fault features to generate a fused fault feature vector, comprising: The spatial fault features are evaluated for information entropy based on multiple spatial feature dimensions, and the information entropy value of the spatial fault features is calculated. The stability of the time fault features is evaluated based on their temporal correlation, and the temporal correlation coefficient of the time fault features is calculated. The spatial fault feature weight coefficient and the temporal fault feature weight coefficient are obtained by weighted summation based on the information entropy value of the spatial fault feature and the temporal correlation coefficient of the temporal fault feature. The spatial fault features and the temporal fault features are weighted and fused according to the spatial fault feature weight coefficient and the temporal fault feature weight coefficient to generate the fused fault feature vector.

9. The fusion fault diagnosis method for special release motors as described in claim 8, characterized in that, The method includes generating a fused fault feature vector for fault diagnosis, identifying the target fault type, and dynamically triggering a motor maintenance strategy based on the target fault type. A fault classifier is constructed by training a multi-class support vector machine as the base classifier using the random forest algorithm; The fused fault feature vector is synchronized to the fault classifier for fault classification, identifying multiple fault types; Confidence analysis is performed on the multiple fault types to obtain multiple confidence scores. The multiple fault types are then sorted in descending order according to the multiple confidence scores, and the fault type with the highest order is selected as the target fault type. Based on the multiple fault types, maintenance analysis is performed to obtain multiple motor maintenance strategies. The multiple fault types and the multiple motor maintenance strategies are matched and associated to construct a fault-maintenance mapping relationship network. The target fault type is used as an index to retrieve the fault-maintenance mapping network and generate an initial motor maintenance strategy. Based on the initial motor maintenance strategy, a maintenance assessment is performed on the special release motor, a maintenance effect score is generated and fed back to the initial motor maintenance strategy for updating, and the motor maintenance strategy is generated.

10. A fusion fault diagnosis system for special release motors, characterized in that, For implementing the fusion fault diagnosis method for special release motors according to any one of claims 1-9, the system comprises: Data acquisition component: Multi-source sensor array performs multi-dimensional sensing and acquisition of special release motor, constructs multi-dimensional spatiotemporal sequence dataset, maps the multi-dimensional spatiotemporal sequence dataset to a three-dimensional rectangular coordinate system, and generates motor operation state matrix; Feature analysis component: Performs multi-feature analysis based on the motor operating state matrix to obtain a fused fault feature map; Spatiotemporal analysis component: Input the fused fault feature map into the spatiotemporal feature fusion channel for spatiotemporal analysis to extract spatial fault features and temporal fault features; Fault diagnosis component: The spatial fault features and the temporal fault features are weighted and fused to generate a fused fault feature vector for fault diagnosis, identify the target fault type, and dynamically trigger motor maintenance strategies based on the target fault type.

Images