Motor residual life prediction model and motor residual life prediction method

Abstract

The invention relates to the technical field of motor state monitoring, in particular to a motor residual life prediction model and a motor residual life prediction method. The preset to-be-trained model is trained through the following method: obtaining a motor operation data set, wherein the motor operation data set comprises a vibration sample data sequence, a current sample data sequence and a temperature sample data sequence collected by a motor within a preset time interval; performing preprocessing and feature extraction on the vibration sample data sequence, the current sample data sequence and the temperature sample data sequence to obtain a mechanical feature sequence, an electrical feature sequence and a temperature feature sequence which respectively comprise a three-dimensional feature channel; carrying out feature channel splicing and layer normalization processing to obtain motor degradation fusion features; and training a preset to-be-trained model according to the plurality of motor degradation fusion features to obtain a motor residual life prediction model. The method can effectively fuse the multi-source monitoring data, extracts the complex degradation characteristics of the motor, and improves the prediction accuracy of the residual life of the motor.

Description

Technical Field

[0001] This application relates to the field of motor condition monitoring technology, and in particular to a motor remaining life prediction model and a motor remaining life prediction method. Background Technology

[0002] As a core power source in modern industry, the operating status of electric motors directly determines the continuity and safety of production systems. Accurately predicting their remaining service life is of significant engineering value for implementing predictive maintenance and avoiding unplanned downtime.

[0003] Existing prediction technologies have two main limitations: Physical model-based methods rely on the failure mechanisms of motor components. However, since motors involve strong coupling of multiple physical fields, establishing accurate models is highly complex and requires significant computational resources, making it difficult to meet the real-time requirements of online monitoring.

[0004] While data-driven methods can adaptively learn degradation patterns using operational data, they have significant drawbacks: on the one hand, limited by industrial site conditions, although they can detect mechanical faults, they are not sensitive enough to non-mechanical faults such as insulation aging and electrical faults; on the other hand, the feature extraction capabilities of the commonly used basic network architecture are limited, making it difficult to capture complex degradation patterns and resulting in low long-term prediction accuracy. Summary of the Invention

[0005] This application provides a motor remaining life prediction model and method. It can effectively integrate multi-source monitoring data, possesses stronger feature extraction capabilities, extracts the complex degradation characteristics of the motor throughout its entire life cycle, and improves the accuracy of motor remaining life prediction and the engineering practicality of the motor remaining life prediction model.

[0006] Firstly, this application provides a motor remaining life prediction model, including a preset training model, which is trained using the following method: A motor operation dataset is acquired, comprising multiple motor operation data sets corresponding to different remaining lifespans of the motor. Each motor operation data set includes vibration sample data sequences, current sample data sequences, and temperature sample data sequences collected within a preset time interval. The vibration sample data sequences, current sample data sequences, and temperature sample data sequences are preprocessed to obtain standardized motor degradation data. Feature extraction is performed on the standardized motor degradation data, yielding mechanical feature sequences, electrical feature sequences, and temperature feature sequences, each including three-dimensional feature channels. The feature channels of the mechanical feature sequences, electrical feature sequences, and temperature feature sequences are concatenated and layer normalization is performed to obtain motor degradation fusion features. The preset training model is trained based on the multiple motor degradation fusion features corresponding to different remaining lifespans of the motor to obtain the motor remaining life prediction model.

[0007] In some implementations, the vibration sample data sequence, current sample data sequence, and temperature sample data sequence are preprocessed to obtain standardized motor degradation data. This includes: applying the least squares method to perform detrending processing on the vibration sample data sequence, current sample data sequence, and temperature sample data sequence respectively, to obtain detrended vibration data sequence, detrended current data sequence, and detrended temperature data sequence. Applying a convolutional moving average filtering algorithm to perform denoising processing on the detrended vibration data sequence, detrended current data sequence, and detrended temperature data sequence respectively, to obtain denoised vibration data sequence, denoised current data sequence, and denoised temperature data sequence. Applying standard scores to perform standardization processing on the denoised vibration data sequence, denoised current data sequence, and denoised temperature data sequence respectively, to obtain standardized motor degradation data, which includes degraded vibration data sequence, degraded current data sequence, and degraded temperature data sequence.

[0008] In some implementations, before feature extraction from standardized motor degradation data, the method further includes: constructing a motor degradation feature extraction network. The motor degradation feature extraction network structure includes a spatial attention feature extraction unit, a first convolutional layer, and a first activation function. The spatial attention feature extraction unit includes a second convolutional layer, a batch normalization layer, a spatial attention mechanism layer, a second activation function, and a pooling layer.

[0009] In some implementations, feature extraction is performed on standardized motor degradation data to obtain mechanical feature sequences, electrical feature sequences, and temperature feature sequences, all of which include three-dimensional feature channels. This includes: inputting the standardized motor degradation data into a motor degradation feature extraction network; generating an attention weight matrix consistent with the spatial size of the input data; weighting and fusing the features of the input data; focusing on key regions in the features; and performing dimensionality reduction processing on the extracted features in terms of channel dimensions to remove redundant information, thereby obtaining mechanical feature sequences, electrical feature sequences, and temperature feature sequences, all of which include three-dimensional feature channels.

[0010] In some implementations, a preset training model is trained based on multiple motor degradation fusion features corresponding to different remaining lifespans of the motor to obtain a motor remaining lifespan prediction model. This includes: using a depthwise separable convolution module to sequentially perform depthwise convolution, pointwise convolution, and pooling operations on the multiple motor degradation fusion features, reshaping the multiple motor degradation fusion features into multiple model features conforming to the input format of the encoder module. The encoder module performs position encoding, multi-head attention mechanism, residual connection and layer normalization, and feedforward neural network processing on the multiple model features, and mines the complex nonlinear mapping relationship between the multiple model features and the remaining lifespan of the motor based on the self-attention mechanism. A fully connected fitting layer module transforms and fits the nonlinear mapping relationship features output by the encoder to obtain the motor remaining lifespan prediction model.

[0011] Secondly, this application provides a method for predicting the remaining life of a motor. The method includes: running the motor and obtaining the predicted motor degradation fusion characteristics corresponding to the motor's operation. The predicted motor degradation fusion characteristics are then input into the motor remaining life prediction model in any one of the first aspects to obtain the motor remaining life prediction result.

[0012] In some implementations, the predicted motor degradation fusion features are input into the motor remaining life prediction model in any of the first aspects to obtain the motor remaining life prediction result. This includes: using a depthwise separable convolution module to sequentially perform depthwise convolution, pointwise convolution, and pooling operations on the predicted motor degradation fusion features, reshaping the predicted motor degradation fusion features into predicted features that conform to the input format of the encoder module. The encoder module performs position encoding, multi-head attention mechanism, residual connection and layer normalization, and feedforward neural network processing on the predicted features, and outputs the predicted nonlinear mapping relationship features between the predicted features and the motor remaining life based on the model's nonlinear mapping relationship features; a fully connected fitting layer module transforms and fits the predicted nonlinear mapping relationship features output by the encoder to generate the final motor remaining life prediction result.

[0013] In some implementations, obtaining the predicted motor degradation fusion features corresponding to motor operation includes: acquiring a predicted motor operation data set, which includes predicted vibration sample data sequences, predicted current sample data sequences, and predicted temperature sample data sequences collected within a preset time interval; preprocessing the predicted vibration sample data sequences, predicted current sample data sequences, and predicted temperature sample data sequences to obtain predicted standardized motor degradation data; extracting features from the predicted standardized motor degradation data to obtain predicted mechanical feature sequences, predicted electrical feature sequences, and predicted temperature feature sequences, all of which include three-dimensional feature channels; concatenating the feature channels of the predicted mechanical feature sequences, predicted electrical feature sequences, and predicted temperature feature sequences, and performing layer normalization processing to obtain the predicted motor degradation fusion features.

[0014] In some implementations, the predicted vibration sample data sequence, predicted current sample data sequence, and predicted temperature sample data sequence are preprocessed to obtain predicted standardized motor degradation data. Specifically, this includes: applying the least squares method to perform detrending processing on the predicted vibration sample data sequence, predicted current sample data sequence, and predicted temperature sample data sequence respectively, to obtain predicted detrended vibration data sequence, predicted detrended current data sequence, and predicted detrended temperature data sequence. Then, applying a convolutional moving average filtering algorithm to perform denoising processing on the predicted detrended vibration data sequence, predicted detrended current data sequence, and predicted detrended temperature data sequence respectively, to obtain predicted denoised vibration data sequence, predicted denoised current data sequence, and predicted denoised temperature data sequence. Finally, applying standard scores to perform standardization processing on the predicted denoised vibration data sequence, predicted denoised current data sequence, and predicted denoised temperature data sequence respectively, to obtain predicted degradation vibration data sequence, predicted degradation current data sequence, and predicted degradation temperature data sequence. Feature extraction is performed on the predicted standardized motor degradation data to obtain predicted mechanical feature sequences, predicted electrical feature sequences, and predicted temperature feature sequences, all of which include three-dimensional feature channels. Specifically, the predicted standardized motor degradation data are input into the motor degradation feature extraction network. By generating an attention weight matrix with the same spatial size as the input data, the features of the input data are weighted and fused, focusing on the key regions in the features. The extracted features are then subjected to dimensionality reduction processing of the channel dimension to remove redundant information, resulting in predicted mechanical feature sequences, predicted electrical feature sequences, and predicted temperature feature sequences, all of which include three-dimensional feature channels.

[0015] Thirdly, this application provides a chip for performing the method as described in any of the second aspects above.

[0016] Fourthly, this application provides an electronic device including a processor and a memory, the memory for storing a motor remaining life prediction model as described in any of the first aspects above, and the processor for executing a computer program stored in the memory to implement the method as described in any of the second aspects above. Alternatively, Electronic devices include chips, as described in the third aspect.

[0017] Fifthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method as described in any of the second aspects above.

[0018] In a sixth aspect, this application provides a computer program product storing a computer program that, when executed by a processor, implements the method as described in any of the second aspects above.

[0019] The technical solution provided in this application allows for the training of a pre-defined training model on a motor's remaining lifespan using the following method: First, a motor operation dataset is acquired, comprising multiple sets of motor operation data corresponding to different remaining lifespans. Each set includes vibration sample data sequences, current sample data sequences, and temperature sample data sequences collected within a preset time interval. These sequences are then preprocessed to obtain standardized motor degradation data. Next, feature extraction is performed on the standardized degradation data, yielding mechanical, electrical, and temperature feature sequences, each containing three-dimensional feature channels. These feature channels are then concatenated and layer normalization is applied to obtain fused motor degradation features. Finally, the pre-defined training model is trained based on these fused motor degradation features corresponding to different remaining lifespans to obtain the motor's remaining lifespan prediction model. The technical solution provided in this application can fuse multi-source data during motor operation through vibration sample data sequence, current sample data sequence and temperature sample data sequence, and train a preset training model through multiple motor degradation fusion features corresponding to different remaining lifespans of the motor. This enhances the model's feature extraction capability and long-term prediction accuracy for complex degradation modes throughout the motor's lifespan, thereby improving the accuracy of motor remaining lifespan prediction and the engineering practicality of the motor remaining lifespan prediction model. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a schematic diagram of the training process of a motor remaining life prediction model provided in an embodiment of this application; Figure 2 This is a schematic diagram illustrating the process of obtaining a training dataset for a motor remaining life prediction model provided in this application embodiment; Figure 3 A schematic diagram of a data preprocessing process for a motor remaining life prediction model provided in this application embodiment; Figure 4 This is a schematic diagram of the network structure of a preset motor degradation feature extraction network for a motor remaining life prediction model provided in this application embodiment; Figure 5 This application provides a schematic diagram of the network structure of a preset training model for a motor remaining life prediction model. Figure 6 A schematic diagram of a motor remaining life prediction process provided in this application embodiment; Figure 7 This is a comparative diagram of test results for a motor remaining life prediction model provided in an embodiment of this application; Figure 8 This is a schematic diagram of an electronic device provided in an embodiment of this application. Detailed Implementation

[0022] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application can also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0023] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0024] It should also be understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0025] As used in this application specification and the appended claims, the term "if" may be interpreted, depending on the context, as "when," "once," "in response to determination," or "in response to detection." Similarly, the phrase "if determined" or "if detected [the described condition or event]" may be interpreted, depending on the context, as meaning "once determined," "in response to determination," "once detected [the described condition or event]," or "in response to detection [the described condition or event]."

[0026] Furthermore, in the description of this application and the appended claims, the terms "first," "second," "third," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0027] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0028] As a core power source in modern industrial systems, the operating status of electric motors directly affects the continuity and safety of production systems. According to relevant standards from the International Electrotechnical Commission (IEC), the reliability of electric motors directly impacts the operational efficiency of the entire industrial system, and sudden failures can lead to serious production disruptions. Therefore, accurately predicting the remaining service life of electric motors is of significant engineering value for implementing predictive maintenance strategies, avoiding unplanned downtime, and ensuring production safety.

[0029] In traditional technical solutions for predicting the remaining service life of motors, physical model-based prediction methods were the earliest and most widely studied solution. This method establishes mathematical models based on the physical failure mechanisms of various motor components, as well as equations describing the thermal aging process of insulation materials. However, because a motor is a complex system involving strong coupling of multiple physical fields such as electromagnetic fields, temperature fields, and mechanical stress fields, establishing an accurate physical model requires considering the interaction of multiple physical processes simultaneously, leading to an exponential increase in model complexity. Especially in industrial scenarios requiring online applications, solving the physical model consumes significant computational resources, making it difficult to meet the engineering requirements of real-time monitoring and prediction.

[0030] To overcome the limitations of physical model methods, data-driven prediction methods have gradually become an important technological approach. Based on the IEEE standard for motor condition monitoring, this method uses condition monitoring data collected during equipment operation to automatically learn performance degradation patterns through machine learning algorithms. The advantage of data-driven methods lies in their ability to directly extract degradation features from historical operating data and real-time monitoring data, demonstrating better adaptability to individual motor differences and changes in the operating environment.

[0031] However, existing data-driven methods still have significant shortcomings in engineering practice. First, at the data acquisition level, due to limitations in industrial site conditions, most current applications rely solely on vibration signals as a single data source. While vibration signals can effectively identify mechanical faults such as bearing wear and rotor imbalance, their sensitivity to electrical faults such as insulation aging and inter-turn short circuits is clearly insufficient. Second, regarding model architecture, existing technologies generally employ basic CNN (Convolutional Neural Networks) and RNN (Recurrent Neural Network) model structures. These basic networks suffer from limited feature extraction capabilities and difficulties in modeling long-term sequence dependencies when dealing with the complex degradation patterns exhibited throughout the motor's lifespan, resulting in long-term prediction accuracy for the motor's remaining lifespan failing to meet engineering requirements.

[0032] In summary, the existing technology has the following main technical defects: First: The monitoring data lacks sufficient dimensions, making it difficult to comprehensively characterize the overall degradation state of the motor under multiple failure modes.

[0033] Second: The feature extraction capability of the prediction model is limited, and it cannot effectively capture the key features in the complex degradation process of the motor.

[0034] Third: The long-term prediction accuracy is low and cannot meet the actual needs of predictive maintenance in industrial sites.

[0035] In view of this, this application provides a motor remaining life prediction model that can effectively integrate multi-source monitoring data, has stronger feature extraction capabilities, extracts the complex degradation characteristics of the motor throughout its entire life cycle, and improves the accuracy of motor remaining life prediction and the engineering practicality of the motor remaining life prediction model.

[0036] The technical solutions of the embodiments of this application are described below with reference to the examples in the accompanying drawings.

[0037] Figure 1 This is a schematic diagram illustrating the training process of a motor remaining life prediction model provided in an embodiment of this application. Figure 1 As shown, the training process for the motor remaining life prediction model provided in this application embodiment may include the following steps: Step S101: Obtain the motor operation dataset. The motor operation dataset includes multiple motor operation data groups corresponding to different remaining lifespans of the motor. Each motor operation data group includes vibration sample data sequence, current sample data sequence and temperature sample data sequence collected within a preset time interval.

[0038] The model training device can acquire a motor operation dataset and train a preset model to be trained based on the motor operation dataset to obtain a motor remaining life prediction model. Figure 2 This is a schematic diagram illustrating the process of obtaining a motor operation dataset for a motor remaining life prediction model provided in this application embodiment. Figure 2 As shown, the method for the model training device to obtain the motor operation dataset may include: Step S101-1: Throughout the entire life cycle of the motor, continuously collect vibration data, current data, and temperature data according to the preset sampling rate, and record the sampling start time and sampling end time to obtain the full-cycle vibration data sequence, full-cycle current data sequence, and full-cycle temperature data sequence.

[0039] It should be understood that during motor operation, different sensor data can be used to characterize the motor's health status in different dimensions. Among these, vibration data can be used to characterize the motor's mechanical state and dynamic imbalance. For example, when a motor bearing experiences rolling failure in the inner or outer race, the bearing rotation will generate high-frequency impacts, resulting in specific fault frequencies on the vibration spectrum. Vibration data can then characterize this type of motor fault condition.

[0040] Current data can be used to characterize the electrical state and load conditions of a motor. For example, when the stator coil insulation ages or the rotor bars break, the current data collected by the model training device will generate specific harmonic components.

[0041] Temperature data can be used to characterize the thermal management status and overall degradation of a motor. For example, poor bearing lubrication or bearing wear can lead to increased bearing friction, resulting in higher temperatures.

[0042] Step S101-2: Based on the sampling start time and sampling end time, divide the full-cycle vibration data sequence, full-cycle current data sequence and full-cycle temperature data sequence according to the preset time interval to obtain multiple motor operation data groups.

[0043] It should be understood that the model training device can collect full-cycle vibration data sequences, full-cycle current data sequences, and full-cycle temperature data sequences corresponding to multiple test prototypes (test motors) to obtain a large training dataset, which facilitates the training of the preset training model and improves the accuracy of model training. For any one of the multiple test prototypes (motor A), the method by which the model training device obtains its corresponding multiple sets of training data may include: First, the model training device records the start time of motor A as the sampling start time and collects vibration, current, and temperature data corresponding to motor A at a preset sampling frequency (e.g., 1 kHz), obtaining a full-cycle vibration data sequence, a full-cycle current data sequence, and a full-cycle temperature data sequence for motor A. When the motor finishes running (until it is scrapped), the sampling end time is recorded as the end time of motor A. Then, the collected full-cycle vibration, current, and temperature data sequences are divided into multiple motor operation data groups according to preset time intervals (e.g., 1 minute), each containing continuously collected vibration, current, and temperature data within the preset time interval.

[0044] For example, if the preset sampling frequency is 1kHz and the preset time interval is 1 minute, then each set of training data includes 6000 vibration data, 6000 current data and 6000 temperature data.

[0045] Step S101-3: Based on the entire life cycle of motor operation, the sampling start time, and the sampling end time, determine the multiple remaining life cycles corresponding to the multiple motor operation data groups, and obtain the motor operation dataset.

[0046] It should be understood that for any motor test prototype, among its multiple motor operation data sets, the first motor operation data set has the longest remaining lifespan (which is the full operating lifespan of the test prototype, or close to the full operating lifespan of the test prototype), while the last motor operation data set has the shortest remaining lifespan (which is 0 or close to 0).

[0047] Step S102: Preprocess the vibration sample data sequence, current sample data sequence, and temperature sample data sequence to obtain standardized motor degradation data.

[0048] In the technical solutions provided in the embodiments of this application, the preprocessing process may include, but is not limited to, trend removal processing, noise reduction processing, and / or standardization processing.

[0049] Figure 3 This is a schematic diagram of a data preprocessing flow for a motor remaining life prediction model training method provided in an embodiment of this application. Figure 3 As shown, the data preprocessing performed by the model training device may include the following steps: Step S102-1: Using the least squares method, perform detrending processing on the vibration sample data sequence, current sample data sequence, and temperature sample data sequence respectively to obtain the detrended vibration data sequence, the detrended current data sequence, and the detrended temperature data sequence.

[0050] The least squares method can remove interference components from the data (vibration sample data sequence, current sample data sequence, and temperature sample data sequence) that are not related to the degradation of the equipment itself but mask the degradation trend, so as to highlight the true degradation status, eliminate interference from slowly changing non-equipment factors (such as ambient temperature), and retain the true degradation information and random noise of the motor in the data.

[0051] Step S102-2: Using the convolutional moving average filtering algorithm, denoising is performed on the detrended vibration data sequence, the detrended current data sequence, and the detrended temperature data sequence to obtain the denoised vibration data sequence, the denoised current data sequence, and the denoised temperature data sequence.

[0052] Convolutional moving average filtering algorithms can be used to suppress random / high-frequency noise in data and reduce interference signals in the data, so as to improve the convergence speed and stability of the model during the training process.

[0053] Step S102-3: Standardize the noise-reduced vibration data sequence, noise-reduced current data sequence, and noise-reduced temperature data sequence using standard fractions to obtain standardized motor degradation data, which includes the degradation vibration data sequence, degradation current data sequence, and degradation temperature data sequence.

[0054] Standard scores (Z-scores) can be used to standardize the scale range of data, ensuring the robustness and generalization of model training.

[0055] Step S103: Extract features from the standardized motor degradation data to obtain mechanical feature sequences, electrical feature sequences, and temperature feature sequences, all of which include three-dimensional feature channels.

[0056] In the technical solution provided in this application embodiment, before performing feature extraction on standardized motor deterioration data, a motor deterioration feature extraction network can be constructed. The motor deterioration feature extraction network structure includes a spatial attention feature extraction unit, a first convolutional layer, and a first activation function. The spatial attention feature extraction unit includes a second convolutional layer, a batch normalization layer, a spatial attention mechanism layer, a second activation function, and a pooling layer.

[0057] The model training device extracts features from standardized motor degradation data, obtaining mechanical feature sequences, electrical feature sequences, and temperature feature sequences, all of which include three-dimensional feature channels. This process includes: inputting standardized motor degradation data into a motor degradation feature extraction network; generating an attention weight matrix with the same spatial size as the input data; weighting and fusing the features of the input data; focusing on key regions in the features; and performing dimensionality reduction processing on the extracted features to remove redundant information, resulting in mechanical feature sequences, electrical feature sequences, and temperature feature sequences, all of which include three-dimensional feature channels.

[0058] For example, Figure 4 This is a schematic diagram of the network structure of a preset motor degradation feature extraction network for a motor remaining life prediction model provided in this application embodiment.

[0059] like Figure 4 As shown, the network structure of the preset motor degradation feature extraction network may include a spatial attention feature extraction unit, a convolutional layer with a 1×1 kernel (first convolutional layer), and ReLU (Rectified Linear Unit, an activation function). The spatial attention feature extraction unit may include a convolutional layer with a 1×3 kernel (second convolutional layer), a BN (Batch Normalization) layer, a SAM (Spatial Attention Module) layer, a ReLU layer (second activation function), and a pooling layer.

[0060] Spatial attention feature extraction (SAM) units can be implemented by embedding them into convolutional neural networks. SAMs generate attention weight matrices with the same spatial dimensions as the input data, weighting and fusing input features to enable the model to focus on key regions within the input features. 1×1 convolutional layers can be used to reduce the channel dimension of the features extracted by the SAM units, removing redundant information.

[0061] Specifically, the model training device can input motor degradation data (including degradation vibration data sequence, degradation current data sequence and degradation temperature data sequence) into a preset motor degradation feature extraction network and output motor degradation features (including mechanical feature sequence, electrical feature sequence and temperature feature sequence).

[0062] For example, for any one of the data sequences (X) in the deterioration vibration data sequence, deterioration current data sequence, and deterioration temperature data sequence, the extraction process of its corresponding data features (mechanical feature sequence, electrical feature sequence, or temperature feature sequence) can be as shown in the following formulas (1) to (4): Formula (1) Where ReLU represents the activation function, BN represents the batch normalization process, and Conv... 1×3 This indicates that a convolution process with a kernel of 1×3 is being performed.

[0063] Formula (2) Where SAM represents the spatial attention module, σ represents the activation function, and Conv 7×7 This indicates that a 7×7 convolution kernel is being performed, MaxPool indicates that the pooling operation with the largest channel dimension is being performed, MinPool indicates that the pooling operation with the smallest channel dimension is being performed, and ";" indicates a feature concatenation operation.

[0064] Formula (3) Where ReLu represents the activation function, SAM represents the spatial attention module, and ⊙ represents element-wise multiplication.

[0065] Formula (4) Where ReLU represents the activation function, Conv 1×1 This indicates that a convolution process with a 1×1 kernel is being performed.

[0066] Step S104: The mechanical feature sequence, electrical feature sequence and temperature feature sequence are spliced ​​together by feature channels and subjected to layer normalization to obtain the motor degradation fusion feature.

[0067] In the technical solutions provided in this application embodiment, the number of features in the mechanical feature sequence, electrical feature sequence and temperature feature sequence are all the same, and they correspond one-to-one with the sampling time corresponding to the preset sampling rate.

[0068] The model training device can perform splicing and layer normalization processing on the mechanical feature sequence, electrical feature sequence, and temperature feature sequence according to the feature channel to obtain the motor degradation fusion feature. Specifically, it includes: performing splicing processing on the target first mechanical feature, target first electrical feature, and target first temperature feature along the dimension of the feature channel to obtain the target first spliced ​​feature. The target first mechanical feature is any mechanical feature in the mechanical feature sequence, the target first electrical feature is an electrical feature in the electrical feature sequence corresponding to the target first mechanical feature, and the target first temperature feature is a temperature feature in the temperature feature sequence corresponding to the target first mechanical feature. Layer normalization processing is performed on the target first spliced ​​feature to obtain the target first motor hybrid feature, thereby obtaining the first motor hybrid feature sequence, and thus obtaining multiple sets of motor hybrid feature sequences corresponding to multiple sets of degradation data.

[0069] For example, the dimensions of the first mechanical feature, the first electrical feature, and the first temperature feature of the target can all be 1×W×D. Then, the model training device performs splicing and layer normalization processing on the first mechanical feature, the first electrical feature, and the first temperature feature of the target, and the resulting first motor hybrid feature has a dimension of 1×W×3D.

[0070] For example, for the target first mechanical feature Fv, the target first electrical feature Fi, and the target first temperature feature Ft, the process of performing splicing and layer normalization on the target first mechanical feature, the target first electrical feature, and the target first temperature feature to obtain the target first motor hybrid feature can be shown in the following formulas (5) to (6): Formula (5) Where || represents channel splicing, F cat This represents the first splicing feature of the target.

[0071] Formula (6) Here, LayerNorm represents the execution layer regularization process. fused This indicates the hybrid characteristics of the first motor in the target system.

[0072] Step S105: Train the preset training model based on the fusion features of multiple motor degradation corresponding to different remaining lifespans of the motor to obtain the motor remaining lifespan prediction model.

[0073] The remaining life prediction model of a motor is obtained by training a pre-set training model based on the fusion characteristics of multiple motor degradations corresponding to different remaining lifespans of the motor, including the following steps: Step-1: The depthwise separable convolution module is used to perform depthwise convolution, pointwise convolution and pooling operations on multiple motor degradation fusion features in sequence, so as to reshape multiple motor degradation fusion features into multiple model features that conform to the input format of the encoder module.

[0074] In the technical solution provided in the embodiments of this application, the depth-separable convolutional module in the preset training model may include a depthwise convolutional layer, a pointwise convolutional layer and a pooling layer, and the encoder module includes a positional encoding embedding layer, a multi-head attention calculation layer, a residual connection and normalization layer and a feedforward neural network layer.

[0075] The encoder module can be used to mine the complex nonlinear mapping relationship between input features and the remaining life of the motor through a self-attention mechanism. The fully connected fitting module is used to transform and fit the features output by the encoder module, and output the predicted remaining life of the motor.

[0076] For example, Figure 5 This is a schematic diagram of the network structure of a preset model to be trained, provided in an embodiment of this application for a method of training a motor remaining life prediction model. Figure 5 As shown, the network structure of the pre-set model to be trained may include a depthwise separable convolutional module, a Transformer encoder (an encoder based on natural language processing and computer vision) module, and a fully connected fitting module (fully connected layer).

[0077] The depthwise separable convolutional module may include depthwise convolutional layers, pointwise convolutional layers, and pooling layers. The Transformer encoder module includes a positional encoding embedding layer, a multi-head attention computation layer, a first residual connection and normalization layer, a feedforward neural network layer, and a second residual connection and normalization layer.

[0078] Among them, the depthwise separable convolutional module's depthwise convolutional layer can be used to perform spatial convolution independently on each channel of the input features, extracting the spatial features of each channel, so as to reduce the amount of computation and parameters, and the number of channels output by the depthwise convolutional layer remains unchanged.

[0079] Pointwise convolutional layers can be used to fuse and combine information between channels of the output of depthwise convolutional layers using 1x1 convolutions, and improve feature representation capabilities through weighted summation. Furthermore, the number of channels in pointwise convolutional layers can be flexibly adjusted.

[0080] Pooling layers can be used to downsample feature maps to reduce spatial size, lower computational complexity, expand the receptive field, and enhance the robustness of features to positional changes.

[0081] The model training device can further extract motor degradation features through depthwise convolution, pointwise convolution, and pooling, and reshape the input motor degradation mixed features (including multiple sets of motor mixed feature sequences and multiple remaining lifetimes) into an input format that conforms to the Transformer encoder.

[0082] Step-2: The encoder module is used to perform position encoding, multi-head attention mechanism, residual connection and layer normalization and feedforward neural network processing on multiple model features respectively, and the complex nonlinear mapping relationship between multiple model features and motor remaining life is mined based on the self-attention mechanism.

[0083] The first and second residual connection and normalization layers in the Transformer encoder module are used to mitigate gradients, prevent information loss, and provide the next layer with data with a stable distribution.

[0084] In this network structure, the first residual connection and normalization layer are located after the multi-head attention computation layer. The first residual connection and normalization layer can perform residual connection processing and normalization processing respectively. The residual connection processing in the first residual connection and normalization layer can prevent network degradation and alleviate gradient vanishing, allowing the model to retain original input information. The normalization processing in the first residual connection and normalization layer can stabilize the output of the self-attention layer, accelerating model training convergence.

[0085] The second residual connection and normalization layer is located after the feedforward neural network layer in the network structure. This layer can be used to perform residual connection processing and normalization processing respectively. The residual connection processing in the second residual connection and normalization layer can mitigate gradients and prevent information loss, allowing the network to learn effectively even as it deepens. The normalization processing in the second residual connection and normalization layer can stabilize the output of the feedforward neural network layer and provide stable data distribution for the next encoder or output layer.

[0086] For example, the calculation process corresponding to the motor remaining life prediction model can be shown in the following formulas (7) to (10) in sequence: Formula (7) Among them, F des-fused This represents the output feature after the input motor degradation mixture features are processed by a depthwise separable convolution module; "pool" indicates pooling. W represents a depthwise convolutional layer. dw W represents the depthwise convolution kernel of the depthwise convolutional layer, and the solid black pentagram represents the pointwise convolutional layer. pw This represents the pointwise convolution kernel of a pointwise convolutional layer. "Pointwise" indicates the pointwise convolution operation. Ffused This represents a sequence of mixed features from multiple motors, and Depthwise represents a depthwise convolution operation.

[0087] Formula (8) Where LayerNorm represents layer normalization operation, MultHesdAttention represents multi-head attention computation, and p represents the position encoding vector.

[0088] Formula (9) Here, LayerNorm represents the layer normalization operation, and FFN represents the computation of the feedforward neural network layer.

[0089] Formula (10) in, This represents the predicted remaining lifespan of the motor, where σ represents the activation function, and W... T denoted as the weight matrix, Fltten indicates the flattening operation, and b indicates the weight bias.

[0090] Step-3: The fully connected fitting layer module is used to transform and fit the nonlinear mapping relationship features of the encoder output model to obtain the motor remaining life prediction model.

[0091] After the model training device trains the preset model to be trained based on the combined degradation characteristics of multiple motors corresponding to different remaining lifespans of the motor, and obtains the motor remaining life prediction model, the model training device can also calculate and output the motor remaining life prediction result based on the motor remaining life prediction model and the test data corresponding to the motor to be tested.

[0092] Specifically, Figure 6 A schematic diagram of a motor remaining life prediction process is provided as an embodiment of this application for a method of training a motor remaining life prediction model. Figure 6 As shown, the method for predicting the remaining life of a motor based on a motor remaining life prediction model may include the following steps: Step S601: Acquire test data. The test data includes vibration test data sequence, current test data sequence and temperature test data sequence collected within a preset time period. The end time corresponding to the preset time period is the current time, and the start time corresponding to the preset time period is the time before the current time and the time interval between the current time and the current time is a preset time interval.

[0093] For example, if the preset time interval is 1 minute and the preset sampling rate is 1kHz, then the preset time period is the period from 1 minute ago to the current moment, and the test data includes 60,000 vibration test data, 60,000 current test data, and 60,000 temperature test data corresponding to 60,000 sampling moments.

[0094] Step S602: After preprocessing the vibration test data sequence, current test data sequence and temperature test data sequence respectively, input them into the preset motor degradation feature extraction network to obtain the second mechanical feature sequence corresponding to the vibration test data sequence, the second electrical feature sequence corresponding to the current test data sequence and the second temperature feature sequence corresponding to the temperature test data sequence.

[0095] The model training device performs preprocessing procedures on the vibration test data sequence, current test data sequence, and temperature test data sequence, respectively, as described above. Figure 1 The data preprocessing process in the corresponding step S102 has the same or similar calculation principle, which will not be elaborated here.

[0096] Step S603: Perform splicing and layer normalization processing on the second mechanical feature sequence, the second electrical feature sequence and the second temperature feature sequence according to the feature channels to obtain the second motor degradation fusion feature.

[0097] The model training device performs splicing and layer normalization processes on the second mechanical feature sequence, the second electrical feature sequence, and the second temperature feature sequence according to feature channels, similar to the above. Figure 1 The splicing and layer normalization processes in the corresponding step S104 have the same or similar calculation principles, which will not be elaborated here.

[0098] Step S604: Input the second motor degradation fusion features into the motor remaining life prediction model and output the motor remaining life prediction result. The motor remaining life prediction result shall include at least the remaining life cycle corresponding to the test data.

[0099] The process for calculating the remaining life prediction results of the motor using the motor remaining life prediction model is the same as described above. Figure 1 The process of calculating the remaining life prediction result of the motor in the corresponding step S105 has the same or similar calculation principle, and will not be described in detail here.

[0100] In the technical solution provided in this application embodiment, the model training device can train a preset training model using the following method: First, a motor operation dataset is acquired. The motor operation dataset includes multiple motor operation data groups corresponding to different remaining lifespans of the motor. Each motor operation data group includes vibration sample data sequences, current sample data sequences, and temperature sample data sequences collected within a preset time interval. The vibration sample data sequences, current sample data sequences, and temperature sample data sequences are preprocessed to obtain standardized motor degradation data. Then, feature extraction is performed on the standardized motor degradation data to obtain mechanical feature sequences, electrical feature sequences, and temperature feature sequences, each including a three-dimensional feature channel. The feature channels of the mechanical feature sequences, electrical feature sequences, and temperature feature sequences are concatenated and layer normalization is performed to obtain motor degradation fusion features. Finally, the preset training model is trained based on the multiple motor degradation fusion features corresponding to different remaining lifespans of the motor to obtain a motor remaining life prediction model. The technical solution provided in this application can fuse multi-source data during motor operation through vibration sample data sequence, current sample data sequence and temperature sample data sequence, and train a preset training model through multiple motor degradation fusion features corresponding to different remaining lifespans of the motor. This enhances the model's feature extraction capability and long-term prediction accuracy for complex degradation modes throughout the motor's lifespan, thereby improving the accuracy of motor remaining lifespan prediction and the engineering practicality of the motor remaining lifespan prediction model.

[0101] In some embodiments, after the model training device trains a preset model to be trained based on multiple motor degradation fusion characteristics corresponding to different remaining lifespans of the motor to obtain a motor remaining lifespan prediction model, the accuracy of the motor remaining lifespan prediction model can be verified. If the accuracy of the motor remaining lifespan prediction model meets the preset accuracy requirements, it can be put into use; if the motor remaining lifespan prediction model does not meet the preset accuracy requirements, the present application can also be used. Figure 1 The technical solutions provided in steps S101 to S105 perform secondary training on the model until the accuracy of the motor remaining life prediction model meets the preset accuracy requirements. The method by which the model training device verifies the accuracy of the motor remaining life prediction model (DSC-Transformer model) is as follows: The model training device can collect vibration, current, and temperature verification data sequences throughout the entire lifecycle of multiple motors to be verified. Specifically, the sampling frequency can be 1 kHz. The device can collect temperature verification data sequences using K-type thermocouples, current verification data sequences using Hall effect sensors, and vibration verification data sequences using accelerometers. The collected vibration, current, and temperature verification data sequences are preprocessed and segmented into 30-minute time windows to obtain multiple motor degradation samples. Each motor degradation sample covers the complete lifecycle from normal operation to final failure, and each sample contains approximately 220 million raw data points. The model training device can then input the preprocessed and feature-fusing motor degradation samples into the DSC-Transformer model.

[0102] The network structure corresponding to the DSC-Transformer model is as follows: Figure 5 As shown, the network structure of the DSC-Transformer model is shown below. Figure 5 Corresponding embodiments are not detailed here. For example, the AdamW optimizer (an optimizer) can be used during the training of the DSC-Transformer model, with an initial learning rate of 0.001, a batch size of 32, and 50 training epochs, or other values, which are not limited here. The model training device can use an early stopping strategy to prevent overfitting of the DSC-Transformer model.

[0103] After the DSC-Transformer model is trained, the model training device can use at least one motor degradation sample (test data) to evaluate the predictive performance of the DSC-Transformer model. The test data covers degradation samples throughout the entire life cycle of the motor.

[0104] Figure 7 This is a schematic diagram comparing test results of a training method for predicting the remaining life of an electric motor provided in an embodiment of this application. Figure 7 Includes the actual lifetime and model-predicted lifetime of the test samples, such as Figure 7 As shown, the DSC-Transformer model can predict the remaining life of the motor relatively accurately.

[0105] To comprehensively evaluate the performance advantages of this method, this application also selected four prediction models as benchmarks: PSF (Point Spread Function) prediction model, DNN (Deep Neural Network) prediction model, TCN (Temporal Convolutional Network) prediction model, and TransRUL (a remaining lifetime prediction model based on the Transformer model architecture). Evaluation metrics may include, but are not limited to, mean absolute error (MAE), normalized root mean square error (NRMSE), mean absolute percentage error (MAPE), root mean square error (RMSE), and a comprehensive score. Experimental results are shown in Table 1. Table 1

[0106] As shown in Table 1, the DSC-Transformer model provided in this application significantly outperforms existing technologies in all metrics, especially in MAE and MAPE, reducing them by approximately 25.0% and 31.8 percentage points compared to the optimal baseline TransRUL, respectively. Therefore, the DSC-Transformer model provided in this application can more accurately capture the nonlinear evolution of motor degradation. Furthermore, the DSC-Transformer model provided in this application achieves a Score of 0.382, demonstrating superior overall prediction accuracy and stability compared to other comparative models.

[0107] The motor remaining life prediction model provided in this application addresses the low prediction accuracy issues caused by the single input signal and difficulty in feature extraction from long-sequence data in traditional solutions by designing a channel splicing feature fusion layer and a DSC-Transformer attention mechanism. This further improves prediction accuracy and robustness. Furthermore, the motor remaining life prediction model provided in this application is applicable to the intelligent assessment and life prediction of motor health status under complex operating conditions, and has a wide range of applications.

[0108] It should be understood that, provided there are no logical conflicts, the above-described embodiments can be combined and implemented to adapt to actual application needs. These combined embodiments or implementation schemes are still within the scope of protection of this application.

[0109] Corresponding to the motor remaining life prediction model in the above embodiments, this application provides a method for predicting the remaining life of a motor: the method includes: Run the motor and obtain the predicted motor degradation fusion features corresponding to the motor's operation.

[0110] The predicted motor degradation fusion characteristics are input into the above motor remaining life prediction model to obtain the motor remaining life prediction results.

[0111] In some embodiments, the predicted motor degradation fusion features are input into the aforementioned motor remaining life prediction model to obtain the motor remaining life prediction result. This includes: using a depthwise separable convolution module to sequentially perform depthwise convolution, pointwise convolution, and pooling operations on the predicted motor degradation fusion features, reshaping the predicted motor degradation fusion features into predicted features that conform to the input format of the encoder module. The encoder module performs position encoding, multi-head attention mechanism, residual connection and layer normalization, and feedforward neural network processing on the predicted features, and outputs the predicted nonlinear mapping relationship features between the predicted features and the motor remaining life based on the model's nonlinear mapping relationship features. A fully connected fitting layer module transforms and fits the predicted nonlinear mapping relationship features output by the encoder to generate the final motor remaining life prediction result.

[0112] In some embodiments, obtaining the predicted motor degradation fusion features corresponding to motor operation includes: obtaining a predicted motor operation data set, which includes a predicted vibration sample data sequence, a predicted current sample data sequence, and a predicted temperature sample data sequence collected within a preset time interval; preprocessing the predicted vibration sample data sequence, the predicted current sample data sequence, and the predicted temperature sample data sequence to obtain predicted standardized motor degradation data; extracting features from the predicted standardized motor degradation data to obtain predicted mechanical feature sequences, predicted electrical feature sequences, and predicted temperature feature sequences, each including a three-dimensional feature channel; concatenating the feature channels of the predicted mechanical feature sequences, the predicted electrical feature sequences, and the predicted temperature feature sequences, and performing layer normalization processing to obtain the predicted motor degradation fusion features.

[0113] In some embodiments, the predicted vibration sample data sequence, predicted current sample data sequence, and predicted temperature sample data sequence are preprocessed to obtain predicted standardized motor degradation data. Specifically, this includes: using the least squares method to perform detrending processing on the predicted vibration sample data sequence, predicted current sample data sequence, and predicted temperature sample data sequence respectively, to obtain predicted detrended vibration data sequence, predicted detrended current data sequence, and predicted detrended temperature data sequence. Using a convolutional moving average filtering algorithm to perform denoising processing on the predicted detrended vibration data sequence, predicted detrended current data sequence, and predicted detrended temperature data sequence respectively, to obtain predicted denoised vibration data sequence, predicted denoised current data sequence, and predicted denoised temperature data sequence. Using standard scores to perform standardization processing on the predicted denoised vibration data sequence, predicted denoised current data sequence, and predicted denoised temperature data sequence respectively, to obtain predicted degradation vibration data sequence, predicted degradation current data sequence, and predicted degradation temperature data sequence.

[0114] Feature extraction is performed on the predicted standardized motor degradation data to obtain predicted mechanical feature sequences, predicted electrical feature sequences, and predicted temperature feature sequences, all of which include three-dimensional feature channels. Specifically, the predicted standardized motor degradation data are input into the motor degradation feature extraction network. By generating an attention weight matrix with the same spatial size as the input data, the features of the input data are weighted and fused, focusing on the key regions in the features. The extracted features are then subjected to dimensionality reduction processing of the channel dimension to remove redundant information, resulting in predicted mechanical feature sequences, predicted electrical feature sequences, and predicted temperature feature sequences, all of which include three-dimensional feature channels.

[0115] It should be noted that the information exchange and execution process between the above methods are different from those in this application. Figure 1 The implementation examples of the motor remaining life prediction model are based on the same concept. For details regarding their specific functions and technical effects, please refer to [link / reference needed]. Figure 1 The corresponding implementation details will not be repeated here.

[0116] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0117] Based on the same inventive concept, embodiments of this application also provide an electronic device.

[0118] Figure 8 This is a schematic diagram of the structure of the electronic device provided in an embodiment of this application. For example... Figure 8 As shown, the electronic device 80 of this embodiment includes: at least one processor 810 ( Figure 8 Only one is shown in the diagram), memory 820, and communication module 840. Memory 820 stores a computer program 830 that may run on processor 810. When processor 810 executes computer program 830, it implements the steps in the above-described motor remaining life prediction model embodiment, for example... Figure 1 Steps S101 to S105 are shown. The communication module 840 can be a separate communication unit used to communicate with an external server or terminal device.

[0119] Electronic device 80 may include, but is not limited to, a processor 810 and a memory 820. Those skilled in the art will understand that... Figure 8 This is merely an example of electronic device 80 and does not constitute a limitation on electronic device 80. It may include more or fewer components than shown, or combine certain components, or different components. For example, electronic device 80 may also include input transmitting devices, network access devices, buses, etc.

[0120] The processor 810 can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor.

[0121] In some embodiments, memory 820 may be an internal storage unit of electronic device 80, such as a hard disk or memory of electronic device 80. Memory 820 may also be an external storage device of electronic device 80, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., provided on electronic device 80. Memory 820 may also include both internal and external storage units of electronic device 80. Memory 820 is used to store operating system, applications, bootloader, data, and other programs, such as the program code of computer program 830. Memory 820 can also be used to temporarily store data that has been sent or will be sent.

[0122] Furthermore, those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. In the various embodiments of this application, each functional unit can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0123] This application provides a computer-readable storage medium storing a computer program that, when run on an electronic device, causes the electronic device to perform the steps described in the various method embodiments above.

[0124] This application provides a chip, which includes a processor and a memory. The memory stores a computer program, which, when executed by the processor, implements the steps in the various method embodiments described above.

[0125] This application provides a computer program product that, when run on an electronic device, causes the electronic device to execute the steps described in the various method embodiments above.

[0126] It should be understood that the processor mentioned in the embodiments of this application can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor.

[0127] It should also be understood that the memory mentioned in the embodiments of this application can be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. Non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. Volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as Static RAM (SRAM), Dynamic RAM (DRAM), Synchronous DRAM (SDRAM), Double Data Rate SDRAM (DDR SDRAM), Enhanced SDRAM (ESDRAM), Synchlink DRAM (SLDRAM), and Direct Rambus RAM (DR RAM).

[0128] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0129] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0130] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0131] In the embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the system embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, apparatuses, or units, and may be electrical, mechanical, or other forms.

[0132] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0133] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0134] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include at least: any entity or device capable of carrying computer program code to a large-screen device, a recording medium, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium. Examples include USB flash drives, portable hard drives, magnetic disks, or optical disks. In some jurisdictions, according to legislation and patent practice, computer-readable media cannot be electrical carrier signals or telecommunication signals.

[0135] Finally, it should be noted that the above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A model for predicting the remaining life of an electric motor, characterized in that, This includes a preset model to be trained, which is trained using the following methods: Obtain a motor operation dataset, which includes multiple motor operation data sets corresponding to different remaining lifespans of the motor. Each motor operation data set includes a vibration sample data sequence, a current sample data sequence, and a temperature sample data sequence collected within a preset time interval. The vibration sample data sequence, current sample data sequence, and temperature sample data sequence are preprocessed to obtain standardized motor degradation data. Feature extraction was performed on the standardized motor degradation data to obtain mechanical feature sequences, electrical feature sequences, and temperature feature sequences, all of which include three-dimensional feature channels. The mechanical feature sequence, the electrical feature sequence, and the temperature feature sequence are spliced ​​together using feature channels and then subjected to layer normalization to obtain motor degradation fusion features. The preset training model is trained based on the combined characteristics of multiple motor degradations corresponding to different remaining lifespans of the motor to obtain the motor remaining lifespan prediction model.

2. The motor remaining life prediction model according to claim 1, characterized in that, The preprocessing of vibration sample data sequences, current sample data sequences, and temperature sample data sequences yields standardized motor degradation data, including: The least squares method is used to perform detrending processing on the vibration sample data sequence, the current sample data sequence, and the temperature sample data sequence respectively to obtain the detrended vibration data sequence, the detrended current data sequence, and the detrended temperature data sequence. The detrended vibration data sequence, the detrended current data sequence, and the detrended temperature data sequence are denoised using a convolutional moving average filtering algorithm to obtain a denoised vibration data sequence, a denoised current data sequence, and a denoised temperature data sequence. The noise-reduced vibration data sequence, the noise-reduced current data sequence, and the noise-reduced temperature data sequence are standardized using standard scores to obtain the standardized motor degradation data, which includes the degradation vibration data sequence, the degradation current data sequence, and the degradation temperature data sequence.

3. The motor remaining life prediction model according to claim 1, characterized in that, Before performing feature extraction on the standardized motor degradation data, the following steps are also included: A motor degradation feature extraction network is constructed. The structure of the motor degradation feature extraction network includes a spatial attention feature extraction unit, a first convolutional layer and a first activation function. The spatial attention feature extraction unit includes a second convolutional layer, a batch normalization layer, a spatial attention mechanism layer, a second activation function and a pooling layer.

4. The motor remaining life prediction model according to claim 3, characterized in that, Feature extraction was performed on the standardized motor degradation data, yielding mechanical feature sequences, electrical feature sequences, and temperature feature sequences, each including a three-dimensional feature channel; including: Standardized motor degradation data are input into the motor degradation feature extraction network. By generating an attention weight matrix with the same spatial size as the input data, the features of the input data are weighted and fused, focusing on the key regions in the features. The extracted features are then subjected to channel dimension reduction processing to remove redundant information, resulting in mechanical feature sequences, electrical feature sequences, and temperature feature sequences, each including three-dimensional feature channels.

5. The motor remaining life prediction model according to claim 4, characterized in that, The preset training model is trained based on multiple motor degradation fusion features corresponding to different remaining lifespans of the motor to obtain the motor remaining lifespan prediction model, including: A depthwise separable convolution module is used to sequentially perform depthwise convolution, pointwise convolution, and pooling operations on multiple motor degradation fusion features, thereby reshaping the multiple motor degradation fusion features into multiple model features that conform to the input format of the encoder module; An encoder module is used to perform position encoding, multi-head attention mechanism, residual connection and layer normalization and feedforward neural network processing on multiple model features respectively, and the complex nonlinear mapping relationship between multiple model features and motor remaining life is mined based on the self-attention mechanism. A fully connected fitting layer module is used to transform and fit the nonlinear mapping relationship features of the encoder output model to obtain the motor remaining life prediction model.

6. A method for predicting the remaining life of an electric motor, characterized in that, The method includes: Run the motor and obtain the predicted motor degradation fusion features corresponding to the motor's operation; The predicted motor degradation fusion features are input into any one of the motor remaining life prediction models described in 1-5 to obtain the motor remaining life prediction results.

7. The method for predicting the remaining life of a motor according to claim 6, characterized in that, The predicted motor degradation fusion features are input into any one of the motor remaining life prediction models described in 1-5 to obtain the motor remaining life prediction results, including: A depthwise separable convolution module is used to sequentially perform depthwise convolution, pointwise convolution, and pooling operations on the predicted motor degradation fusion features, thereby reshaping the predicted motor degradation fusion features into predicted features that conform to the input format of the encoder module. The predicted features are processed by an encoder module, which performs position encoding, multi-head attention mechanism, residual connection and layer normalization, and feedforward neural network processing. Based on the nonlinear mapping relationship features of the model, the predicted nonlinear mapping relationship features between the predicted features and the remaining life of the motor are output. A fully connected fitting layer module is used to transform and fit the predicted nonlinear mapping relationship features output by the encoder to generate the final motor remaining life prediction result.

8. The method for predicting the remaining life of a motor according to claim 6, characterized in that, Obtain the predicted motor degradation fusion features corresponding to the motor's operation, including: Acquire a set of predicted motor operation data, which includes a sequence of predicted vibration sample data, a sequence of predicted current sample data, and a sequence of predicted temperature sample data collected within a preset time interval. Preprocessing is performed on the predicted vibration sample data sequence, the predicted current sample data sequence, and the predicted temperature sample data sequence to obtain the predicted standardized motor degradation data. Feature extraction is performed on the predicted standardized motor degradation data to obtain predicted mechanical feature sequences, predicted electrical feature sequences, and predicted temperature feature sequences, all of which include three-dimensional feature channels. The predicted mechanical feature sequence, the predicted electrical feature sequence, and the predicted temperature feature sequence are spliced ​​together using feature channels and then subjected to layer normalization to obtain the predicted motor degradation fusion feature.

9. The method for predicting the remaining life of an electric motor according to claim 8, characterized in that, Preprocessing of the predicted vibration sample data sequence, predicted current sample data sequence, and predicted temperature sample data sequence yields predicted standardized motor degradation data; specifically including: The least squares method is used to perform detrending processing on the predicted vibration sample data sequence, the predicted current sample data sequence, and the predicted temperature sample data sequence, respectively, to obtain the predicted detrended vibration data sequence, the predicted detrended current data sequence, and the predicted detrended temperature data sequence. The convolutional moving average filtering algorithm is used to perform denoising processing on the predicted detrended vibration data sequence, the predicted detrended current data sequence, and the predicted detrended temperature data sequence, respectively, to obtain the predicted denoised vibration data sequence, the predicted denoised current data sequence, and the predicted denoised temperature data sequence. The predicted noise-reduced vibration data sequence, the predicted noise-reduced current data sequence, and the predicted noise-reduced temperature data sequence are standardized using standard scores to obtain the predicted deterioration vibration data sequence, the predicted deterioration current data sequence, and the predicted deterioration temperature data sequence. Feature extraction is performed on the predicted standardized motor degradation data to obtain predicted mechanical feature sequences, predicted electrical feature sequences, and predicted temperature feature sequences, all of which include three-dimensional feature channels. Specifically, these include: The predicted standardized motor degradation data are input into the motor degradation feature extraction network. By generating an attention weight matrix with the same spatial size as the input data, the features of the input data are weighted and fused, focusing on the key regions in the features. The extracted features are then subjected to channel dimension reduction processing to remove redundant information, resulting in a predicted mechanical feature sequence, a predicted electrical feature sequence, and a predicted temperature feature sequence, all of which include three-dimensional feature channels.

10. An electronic device, characterized in that, The device includes a processor and a memory, the memory being used to store the motor remaining life prediction model as described in any one of claims 1 to 5, and the processor being used to execute the computer program stored in the memory to implement the motor remaining life prediction method as described in any one of claims 6 to 9.

Images