A method and device for fault diagnosis of drive motor based on artificial intelligence technology

Abstract

The application discloses a kind of based on artificial intelligence technology's drive motor fault diagnosis method and device, it is related to fault diagnosis technical field, the method includes: respectively in the preset part of drive motor layout multimodal sensor, and obtain the multidimensional data signal that multimodal sensor is collected;Respectively to multidimensional data signal carries out real-time filtering, amplification and noise reduction processing, and then utilizes feature extraction algorithm to multidimensional data signal carries out feature extraction and obtains multiple key characteristic parameters;Based on the dynamic weighted fusion algorithm of preestablished, a plurality of key characteristic parameters are fused to obtain fusion key characteristic parameters;Establish the hybrid deep network architecture of long short-term memory network and convolutional neural network, and the fusion key characteristic parameters are input into the hybrid deep network architecture of pre-trained and obtain the corresponding drive motor fault information.The application solves the problem of low accuracy of drive motor fault diagnosis in the prior art.

Description

Technical Field

[0001] This invention relates to the field of fault diagnosis technology, and in particular to a method and apparatus for fault diagnosis of drive motors based on artificial intelligence technology. Background Technology

[0002] As the core component of the entire powertrain in new energy vehicles, the drive motor faces higher requirements for safe, efficient, and long-life operation. Currently, many technologies in the field of motor fault diagnosis and prevention employ ultrasonic testing technology, which uses data modeling to predict faults by collecting parameters such as structural vibration and noise during motor operation in real time.

[0003] However, existing technologies only use limited ultrasonic or other single sensors for fault detection, which makes it difficult to fully capture localized minute cracks or early abnormal signals in the complex structure of the motor, resulting in low accuracy of drive motor fault diagnosis. Summary of the Invention

[0004] In view of this, the purpose of the present invention is to provide a drive motor fault diagnosis method and device based on artificial intelligence technology, which aims to solve the problem of low accuracy in drive motor fault diagnosis in the prior art.

[0005] This invention proposes a method for fault diagnosis of drive motors based on artificial intelligence technology, the method comprising:

[0006] Multimodal sensors are deployed at predetermined locations on the drive motor, and multidimensional data signals related to drive motor fault diagnosis are acquired from the multimodal sensors.

[0007] The multidimensional data signals are filtered, amplified, and denoised in real time. Then, a feature extraction algorithm is used to extract features from the multidimensional data signals to obtain multiple key feature parameters.

[0008] Based on a preset dynamic weighted fusion algorithm, according to the signal quality and current operating condition information collected by the multimodal sensors, different weights are assigned to the key feature parameters acquired by different sensor channels in the multimodal sensors, so as to fuse multiple key feature parameters to obtain fused key feature parameters.

[0009] A hybrid deep network architecture consisting of a long short-term memory network and a convolutional neural network is established, and the key feature parameters are input into the pre-trained hybrid deep network architecture to obtain the corresponding fault information of the drive motor.

[0010] Furthermore, in the aforementioned drive motor fault diagnosis method based on artificial intelligence technology, the step of extracting multiple key feature parameters from multidimensional data signals using a feature extraction algorithm includes:

[0011] Statistical analysis based on multidimensional data signals extracts time-domain features from multidimensional data signals;

[0012] Frequency domain features are extracted from multidimensional data signals based on the Fourier transform of the multidimensional data signals.

[0013] Time-frequency domain features are extracted from multidimensional data signals based on wavelet transform of multidimensional data signals;

[0014] The transform domain features of multidimensional data signals can be extracted from them using the Hilbert-Huang transform.

[0015] Furthermore, in the aforementioned drive motor fault diagnosis method based on artificial intelligence technology, the step of fusing multiple key feature parameters to obtain fused key feature parameters by assigning different weights to key feature parameters acquired by different sensor channels in the multimodal sensors according to the signal quality and current operating condition information of the multimodal sensors based on a preset dynamic weighted fusion algorithm includes:

[0016] The signal-to-noise ratio (SNR) and signal variance of the data signals collected by the multimodal sensors are obtained respectively. The SNR and signal variance are linearly mapped to a preset interval and then weighted and fused to obtain the corresponding sensor single-channel signal quality score.

[0017] Based on the current operating condition information, find the basic weight coefficient of each sensor channel corresponding to the current operating condition information in the preset basic weight mapping table;

[0018] The initial weights of different sensor channels are calculated by combining the single-channel signal quality score of the sensor with the basic weight coefficients under the corresponding operating conditions.

[0019] The fused key feature parameters are obtained by weighting and summing the corresponding key feature parameters according to the initial weights of different sensor channels.

[0020] Furthermore, in the aforementioned drive motor fault diagnosis method based on artificial intelligence technology, before the step of obtaining the fused key feature parameters by weighted summation of the corresponding key feature parameters according to the initial weights of different sensor channels, the method further includes:

[0021] A penalty factor is introduced into the initial weights of sensor channels with transmission lag, and an exponential moving average filter is used for smoothing.

[0022] The expression for the initial weighting process is as follows:

[0023] ;

[0024] ;

[0025] ;

[0026] in, No. The initial weights of each sensor channel are adjusted by introducing a penalty factor. For the first The initial weights of each sensor channel, As a penalty factor, It is the natural logarithm. The attenuation coefficient controls the strength of the effect of delay on the weights. For the first Transmission delay time of each sensor channel For the first The final smoothed weights of each sensor channel. For smoothing coefficients, For the previous moment The final smoothing weights for each sensor channel.

[0027] Furthermore, in the aforementioned drive motor fault diagnosis method based on artificial intelligence technology, the step of establishing a hybrid deep network architecture composed of a long short-term memory network and a convolutional neural network, and inputting the fused key feature parameters into the pre-trained hybrid deep network architecture to obtain the corresponding drive motor fault information includes:

[0028] Long Short-Term Memory (LSTM) networks are used to learn long-term dependencies in time-series data from key feature parameters through gating mechanisms, and local structured features are extracted by sliding convolutional kernels in convolutional neural networks.

[0029] By fusing the temporal global features output by the Long Short-Term Memory network with the local structured features output by the convolutional neural network, and mapping them to the classification / regression target through a fully connected layer, fault information of the drive motor can be obtained.

[0030] Furthermore, the aforementioned drive motor fault diagnosis method based on artificial intelligence technology further includes:

[0031] The corresponding abnormal thresholds for different operating conditions are determined based on the failure probability distribution of normal samples under different operating conditions.

[0032] During the inference phase of the hybrid deep network architecture, the corresponding abnormal threshold is loaded according to the identified operating conditions, and the abnormal probability output by the hybrid deep network architecture is compared with the abnormal threshold to determine whether the drive motor has a fault.

[0033] Furthermore, in the aforementioned drive motor fault diagnosis method based on artificial intelligence technology, the step of deploying multimodal sensors at predetermined locations on the drive motor includes:

[0034] Ultrasonic probes are placed on the stator, rotor, and bearings of the drive motor, and vibration and temperature sensors are installed both inside and outside the drive motor.

[0035] Another object of the present invention is to provide a drive motor fault diagnosis device based on artificial intelligence technology, the device comprising:

[0036] The acquisition module is used to deploy multimodal sensors at preset locations on the drive motor and acquire multidimensional data signals related to drive motor fault diagnosis collected by the multimodal sensors.

[0037] The extraction module is used to perform real-time filtering, amplification and noise reduction on multidimensional data signals, and then use feature extraction algorithms to extract features from the multidimensional data signals to obtain multiple key feature parameters.

[0038] The fusion module is used to assign different weights to the key feature parameters acquired by different sensor channels in the multimodal sensors based on the signal quality and current operating conditions of the multimodal sensors, according to a preset dynamic weighted fusion algorithm, so as to fuse multiple key feature parameters to obtain fused key feature parameters.

[0039] The diagnostic module is used to establish a hybrid deep network architecture composed of a long short-term memory network and a convolutional neural network, and input the fused key feature parameters into the pre-trained hybrid deep network architecture to obtain the corresponding fault information of the drive motor.

[0040] Another object of the present invention is to provide a readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described above.

[0041] Another object of the present invention is to provide an electronic device including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the program to implement the steps of the method described above.

[0042] This invention collects multi-dimensional data signals by deploying multi-modal sensors at predetermined locations on the drive motor. After real-time filtering, amplification, and noise reduction, key feature parameters are extracted. A dynamic weighted fusion algorithm is used to assign weights to the parameters of different sensor channels based on signal quality and operating condition information, and then the data is fused. Finally, the fused key feature parameters are input into a hybrid deep network architecture composed of a long short-term memory network and a convolutional neural network for fault diagnosis. Multi-modal sensors overcome the limitations of single-sensor information dimensions, covering multiple types of fault features. Dynamic weight optimization improves data reliability and enhances adaptability to operating conditions. Deep feature learning enhances the distinguishability of complex fault modes, ultimately effectively improving the accuracy of fault diagnosis. This solves the problem of low accuracy in drive motor fault diagnosis in existing technologies. Attached Figure Description

[0043] Figure 1 This is a flowchart of the drive motor fault diagnosis method based on artificial intelligence technology in the first embodiment of the present invention;

[0044] Figure 2 This is a structural block diagram of a drive motor fault diagnosis device based on artificial intelligence technology in the third embodiment of the present invention.

[0045] The following detailed description, in conjunction with the accompanying drawings, will further illustrate the present invention. Detailed Implementation

[0046] To facilitate understanding of the present invention, a more complete description will be given below with reference to the accompanying drawings. Several embodiments of the invention are illustrated in the drawings. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete.

[0047] It should be noted that when a component is said to be "fixed to" another component, it can be directly on the other component or there may be an intervening component. When a component is said to be "connected to" another component, it can be directly connected to the other component or there may be an intervening component. The terms "vertical," "horizontal," "left," "right," and similar expressions used in this document are for illustrative purposes only.

[0048] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0049] Example 1

[0050] Please see Figure 1 The figure shows a drive motor fault diagnosis method based on artificial intelligence technology in the first embodiment of the present invention, the method including steps S10 to S12.

[0051] Step S10: Multimodal sensors are deployed at preset locations on the drive motor, and multidimensional data signals related to drive motor fault diagnosis are acquired by the multimodal sensors.

[0052] In order to obtain more comprehensive and accurate information for fault diagnosis during the operation of the drive motor, various types of sensors are installed at multiple preset key parts of the drive motor to form a multimodal sensor array. Specifically, ultrasonic probes can be arranged in the stator, rotor and bearings of the drive motor, while vibration and temperature sensors are installed both outside and inside the drive motor.

[0053] For example, during the operation of a drive motor, the stator, rotor, and bearings are core components, and their operating status directly affects the overall performance and stability of the motor. When these components malfunction, they generate specific ultrasonic signals. Installing an ultrasonic probe on the stator allows for real-time monitoring of changes in ultrasonic signals caused by faults such as partial discharge and insulation aging during stator winding operation. For instance, when the insulation layer of the stator winding is damaged, the high-frequency ultrasonic signals generated by the discharge can be sensitively captured by the probe. Placing an ultrasonic probe on the rotor allows for the detection of ultrasonic characteristics generated by abnormal friction or collisions caused by component loosening or structural defects during high-speed rotor rotation. For example, the ultrasonic fluctuations generated by the vibration transmitted to the surface due to broken rotor bars can be collected by the probe. Placing an ultrasonic probe on the bearing allows for the monitoring of ultrasonic signals generated by abnormal friction between the bearing rolling elements and the inner and outer rings due to wear, poor lubrication, and other problems. For instance, when the bearing balls wear more severely, the ultrasonic probe will record stronger and more complex ultrasonic signals.

[0054] Vibration sensors are installed at the bearing to capture the vibration signals generated by the bearing during operation. These signals contain key information about mechanical faults such as bearing wear and loosening.

[0055] Meanwhile, temperature sensors are placed on the surface of the motor housing to collect the overall temperature data of the motor in real time. An abnormal rise in temperature often indicates potential faults such as overload or poor heat dissipation inside the motor.

[0056] Step S11 involves real-time filtering, amplification, and noise reduction of the multidimensional data signal, followed by feature extraction using a feature extraction algorithm to extract multiple key feature parameters from the multidimensional data signal.

[0057] In this process, after acquiring multidimensional data signals from the drive motor using multimodal sensors, the raw data signals often contain a large amount of noise due to complex interference factors such as electromagnetic interference and mechanical vibration in the actual acquisition environment. Furthermore, some useful signals may be difficult to identify due to their small amplitude. Therefore, preprocessing and feature extraction are necessary for these multidimensional data signals. First, real-time filtering is performed by designing appropriate filters, such as low-pass filters, band-pass filters, or adaptive filters, based on the frequency characteristics of the signal. This filters out high-frequency noise or low-frequency interference mixed in with the data signal, retaining the effective signals in the frequency bands relevant to fault diagnosis. For example, a band-pass filter can be used to extract vibration components related to bearing faults within a specific frequency range from the vibration signal. Next, amplification is performed to increase the amplitude of the effective signal when the signal amplitude is weak, making the signal characteristics more obvious and facilitating subsequent processing, preventing key information from being ignored due to weak signals. Finally, noise reduction further employs algorithms such as wavelet denoising and mean filtering to suppress random noise in the signal, optimize signal quality, and ensure the accuracy and reliability of the data. After real-time filtering, amplification, and noise reduction, the multidimensional data signal has achieved high purity and clarity. At this point, feature extraction algorithms such as Fourier transform, wavelet transform, and principal component analysis (PCA) are used to extract key information that can characterize the operating status and fault characteristics of the drive motor from the processed multidimensional data signal. For example, fault characteristic frequencies are extracted from vibration signals, harmonic components are extracted from current signals, and temperature rise rates are extracted from temperature signals. This information is then transformed into multiple key feature parameters. These parameters serve as "feature labels" for the operating status of the drive motor, which can more accurately and concisely reflect the motor's operating condition and potential faults, providing an effective basis for subsequent fault diagnosis.

[0058] Specifically, in this embodiment of the invention, after real-time filtering, amplification, and noise reduction of the multidimensional data signal, various feature extraction algorithms are used to extract key feature parameters from different dimensions in order to deeply explore the drive motor fault information contained in the signal. First, time-domain features are extracted based on the statistical characteristics of the multidimensional data signal. By calculating statistical quantities such as the mean, variance, peak value, kurtosis, and skewness of the signal, the trend and fluctuation of the signal on the time axis are intuitively reflected. For example, when the drive motor is running normally, the mean of the vibration signal is within a stable range. If the mean suddenly increases, it may indicate that the motor is malfunctioning. The variance reflects the dispersion of the signal. When the bearing wear intensifies, the variance of the vibration signal will increase significantly. These time-domain features can quickly capture the overall trend of the signal and provide preliminary clues for fault diagnosis.

[0059] Secondly, Fourier transform is used to extract frequency domain features from multidimensional data signals. Fourier transform converts time-domain signals into frequency-domain representations, revealing the distribution of different frequency components. By analyzing the amplitude and phase of each frequency component, information such as the natural frequency and fault characteristic frequency of the drive motor can be identified. For example, when a motor bearing fails, a specific frequency peak will appear in the spectrum of the vibration signal. By comparing this peak with the spectrum under normal conditions, it can be determined whether the bearing has a fault and the type of fault.

[0060] Furthermore, by extracting the time-frequency domain features of multidimensional data signals using wavelet transform, wavelet transform overcomes the limitation of Fourier transform in reflecting the local time-domain features of signals. It can analyze signals simultaneously in both time and frequency dimensions. By selecting appropriate wavelet basis functions, the signal is decomposed into wavelet coefficients of different scales and frequencies, clearly showing the frequency component changes of the signal at different times. For non-stationary signals, such as current signals during motor starting and braking, wavelet transform can accurately capture the dynamic changes in signal frequency over time, which helps in the detection of transient and sudden faults.

[0061] Finally, based on the Hilbert-Huang transform, the transform domain features of the multidimensional data signal are extracted. This method decomposes the complex signal into multiple intrinsic mode functions (IMFs) through empirical mode decomposition (EMD), and then performs a Hilbert transform on each IMF to obtain the instantaneous frequency and amplitude information of the signal, forming a Hilbert spectrum, thereby more accurately describing the nonlinear and non-stationary characteristics of the signal. When dealing with complex signals generated by motor faults, especially those caused by compound faults, the Hilbert-Huang transform can effectively separate the signal components generated by different fault sources and accurately extract fault features, providing a more comprehensive and detailed basis for the diagnosis of complex faults. Through the above feature extraction of multidimensional data signals from multiple perspectives including the time domain, frequency domain, time-frequency domain, and transform domain, several key feature parameters that can comprehensively reflect the operating status and fault information of the drive motor are obtained, laying a solid foundation for subsequent fault diagnosis and analysis.

[0062] Step S12: Based on the preset dynamic weighted fusion algorithm, according to the signal quality and current operating condition information collected by the multimodal sensors, different weights are assigned to the key feature parameters acquired by different sensor channels in the multimodal sensors, so as to fuse multiple key feature parameters to obtain fused key feature parameters.

[0063] After acquiring and extracting multiple key feature parameters from the multi-modal sensors of the drive motor, the reliability and sensitivity of different sensors vary under different operating conditions, and the signal quality acquired by each sensor will also change due to environmental interference, its own performance, and other factors. Therefore, a preset dynamic weighted fusion algorithm is needed to process these key feature parameters.

[0064] The core logic of this algorithm lies in comprehensively considering the signal quality acquired by multimodal sensors and the current operating condition information, assigning different weights to the key feature parameters acquired by different sensor channels. Signal quality assessment primarily relies on indicators such as the signal-to-noise ratio (SNR) and stability of the sensor data. For example, when a vibration sensor experiences excessive signal noise due to environmental electromagnetic interference, its SNR decreases, and the algorithm correspondingly reduces the weight of the key feature parameters for that sensor channel; conversely, sensors with stable signals and low noise are assigned higher weights. Simultaneously, the current operating condition information encompasses factors such as the drive motor's speed, load, and running time. Because the sensitivity of different sensors to fault characteristics varies under different operating conditions—for instance, under high-speed operation, the vibration sensor is more sensitive to bearing faults, and the algorithm increases its weight—and under heavy-load conditions, the current sensor is more effective at detecting faults such as winding overheating and short circuits, thus increasing the weight of the key feature parameters for the current sensor channel. Through this dynamic weight adjustment, the key feature parameters from different channels of the multimodal sensors are weighted and fused to obtain the final fused key feature parameters. This fusion of key characteristic parameters integrates the effective information from each sensor under the current operating conditions. It avoids misjudgments caused by poor signal quality or insensitivity to specific operating conditions of a single sensor, and fully leverages the synergistic advantages of multi-modal sensors. This allows the fused parameters to more accurately and comprehensively reflect the actual operating status and potential fault characteristics of the drive motor.

[0065] Step S13: Establish a hybrid deep network architecture consisting of a long short-term memory network and a convolutional neural network, and input the fused key feature parameters into the pre-trained hybrid deep network architecture to obtain the corresponding fault information of the drive motor.

[0066] To accurately extract drive motor fault information from fused key feature parameters, this invention constructs a hybrid deep network architecture composed of a Long Short-Term Memory (LSTM) network and a Convolutional Neural Network (CNN). In practical applications, the hybrid deep network architecture is first pre-trained using a large amount of drive motor data containing normal states and different fault types. The parameters in the network are continuously adjusted through a backpropagation algorithm, enabling the network to accurately identify feature patterns corresponding to different fault types. After pre-training is complete, the fused key feature parameters obtained through a dynamic weighted fusion algorithm are input into the hybrid deep network architecture. Finally, the corresponding drive motor fault information is output through the network's output layer, including fault type (such as bearing wear, winding short circuit, etc.), fault severity, and the probability of fault occurrence, thus achieving accurate diagnosis of drive motor faults.

[0067] Specifically, Long Short-Term Memory (LSTM) networks, with their unique gating mechanisms including input gates, forget gates, and output gates, can efficiently process time-series data fused with key feature parameters. During motor operation, the trends in parameters such as temperature and vibration over time contain important clues about the occurrence and development of faults. LSTM's gating mechanism can selectively retain and forget historical information, thereby learning long-term dependencies in this time-series data. For example, motor bearing wear is a gradual process; LSTM can capture the trend of vibration signals gradually increasing over a longer period, accurately remembering historical information related to fault development, avoiding the problem of traditional recurrent neural networks struggling to learn long-term dependencies due to gradient vanishing, and thus extracting global features reflecting the overall changing patterns of the time-series data. Convolutional Neural Networks (CNNs), on the other hand, automatically extract local structured features from the data by sliding convolutional kernels across the fused key feature parameters. For multidimensional data of drive motors, such as the distortion characteristics of current waveforms and specific frequency components of vibration signals, the convolutional kernels of CNNs can keenly capture the correlations and patterns between these local data. Through convolution and pooling operations, features are continuously compressed and abstracted, transforming complex raw data into more representative local structured features, such as extracting abnormal local patterns of harmonic components from current signals. After separately acquiring the temporal global features output by LSTM and the local structured features output by CNN, the two are fused to fully combine the advantages of temporal change trends and local data features, enabling the network to more comprehensively understand fault information in the data. The fused features are processed through fully connected layers, which map the fused features to classification or regression targets. If used for fault type diagnosis, the classification target outputs the probability of different fault types to determine what kind of fault has occurred in the drive motor; if used for fault severity assessment, the regression target outputs specific values ​​to quantify the severity of the fault, ultimately achieving accurate acquisition of drive motor fault information.

[0068] In addition, in some optional embodiments of the present invention, the method further includes:

[0069] The corresponding abnormal thresholds for different operating conditions are determined based on the failure probability distribution of normal samples under different operating conditions.

[0070] During the inference phase of the hybrid deep network architecture, the corresponding abnormal threshold is loaded according to the identified operating conditions, and the abnormal probability output by the hybrid deep network architecture is compared with the abnormal threshold to determine whether the drive motor has a fault.

[0071] In the actual operation of the drive motor, the normal operating status and data performance of the motor vary significantly under different operating conditions (such as no-load, light-load, full-load, and different speeds). To more accurately determine whether the motor has a fault, an anomaly threshold determination mechanism based on operating conditions is further introduced. First, a large amount of normal operating sample data of the drive motor under different operating conditions is collected. This data covers the fused key feature parameters of the multi-modal sensor signals of the motor under various typical operating conditions after processing and feature extraction. Through in-depth analysis of these normal sample data, statistical methods, such as probability density estimation and confidence interval calculation, are used to determine the fault probability distribution of normal samples under each operating condition. For example, under no-load conditions, a large number of fused feature parameters such as current, vibration, and temperature during normal operation are analyzed to construct a corresponding fault probability distribution model. Based on this distribution model, a reasonable anomaly threshold is set, which is used to define the boundary from normal state to abnormal state under no-load conditions. Similarly, for other operating conditions such as light-load and full-load, corresponding fault probability distributions are established and their respective anomaly thresholds are determined. When the hybrid deep network architecture enters the inference phase—that is, when diagnosing faults in the actual operating drive motor—it first identifies the current operating condition by collecting the motor's operating parameters (such as speed and load) through sensors. Then, a pre-determined corresponding abnormal threshold is applied based on the identified operating condition. At this point, the fused key feature parameters of the drive motor are input into the pre-trained hybrid deep network architecture. The network calculates and outputs a probability value indicating that the current motor state is abnormal. Finally, this abnormal probability is compared with the applied abnormal threshold for the corresponding operating condition. If the abnormal probability is greater than the abnormal threshold, it indicates that the current motor operating state deviates from the normal range under that condition, and the drive motor is determined to have a fault. By setting abnormal thresholds according to different operating conditions and comparing them with the operating conditions during diagnosis, misjudgments caused by differences in operating conditions are effectively avoided, significantly improving the accuracy and reliability of drive motor fault diagnosis, and ensuring timely and accurate identification of motor faults under various complex operating conditions.

[0072] In summary, the drive motor fault diagnosis method based on artificial intelligence technology in the above embodiments of the present invention collects multi-dimensional data signals by deploying multi-modal sensors at preset locations on the drive motor. After real-time filtering, amplification, and noise reduction, key feature parameters are extracted. A dynamic weighted fusion algorithm is used to assign weights to the parameters of different sensor channels based on signal quality and operating condition information and then fuse them. Finally, the fused key feature parameters are input into a hybrid deep network architecture composed of a long short-term memory network and a convolutional neural network for fault diagnosis. Multi-modal sensors overcome the limitations of single-sensor information dimensions and can cover multiple types of fault features. Dynamic weight optimization improves data reliability and enhances operating condition adaptability. Deep feature learning enhances the distinguishability of complex fault modes, ultimately effectively improving the accuracy of fault diagnosis. This solves the problem of low accuracy in drive motor fault diagnosis in the prior art.

[0073] Example 2

[0074] This embodiment also proposes a drive motor fault diagnosis method based on artificial intelligence technology. The difference between the drive motor fault diagnosis method based on artificial intelligence technology in this embodiment and the drive motor fault diagnosis method based on artificial intelligence technology in Embodiment 1 is as follows:

[0075] The steps of the preset dynamic weighted fusion algorithm, which assigns different weights to key feature parameters acquired by different sensor channels in the multimodal sensors according to the signal quality and current operating condition information of the multimodal sensors, and fuses multiple key feature parameters to obtain fused key feature parameters, include:

[0076] The signal-to-noise ratio (SNR) and signal variance of the data signals collected by the multimodal sensors are obtained respectively. The SNR and signal variance are linearly mapped to a preset interval and then weighted and fused to obtain the corresponding sensor single-channel signal quality score.

[0077] Based on the current operating condition information, find the basic weight coefficient of each sensor channel corresponding to the current operating condition information in the preset basic weight mapping table;

[0078] The initial weights of different sensor channels are calculated by combining the single-channel signal quality score of the sensor with the basic weight coefficients under the corresponding operating conditions.

[0079] The fused key feature parameters are obtained by weighting and summing the corresponding key feature parameters according to the initial weights of different sensor channels.

[0080] In the fault diagnosis of drive motors, to achieve effective fusion of multimodal sensor data, this technical solution adopts a dynamic weighted fusion algorithm, which differentiates the key characteristic parameters of each sensor channel by combining signal quality and operating condition information. First, for the data signals collected by each sensor, the signal-to-noise ratio (SNR) and signal variance are calculated separately. The SNR reflects the ratio of effective information to noise in the signal; a higher SNR indicates better signal quality. The signal variance measures the degree of signal fluctuation; excessive variance may indicate signal instability or abnormal interference. These two indicators are mapped to a preset interval (e.g., [0,1]) through a linear mapping function, making indicators of different dimensions comparable. The mapped results are then weighted and fused to obtain a signal quality score for each sensor channel. This score directly reflects the reliability of the data collected by each sensor at the current moment. Simultaneously, based on the current operating condition information of the drive motor (e.g., speed, load), the corresponding basic weight coefficients for each sensor channel are looked up in a pre-established basic weight mapping table. This mapping table, built upon extensive historical data and expert knowledge, clarifies the varying importance of different sensors for fault diagnosis under different operating conditions. For instance, vibration sensors have a relatively higher base weight during high-speed operation, while current sensors have a higher base weight under heavy-load conditions. Subsequently, by combining the single-channel signal quality score with the base weight coefficient for the corresponding operating condition, the initial weight of each sensor channel is calculated. This can be achieved through multiplication or a more complex nonlinear combination, thus considering both the impact of operating conditions on sensor importance and dynamically adjusting for reliability differences caused by signal quality variations. Finally, the calculated initial weights are used to weight and sum the key characteristic parameters of each sensor channel, fusing the multi-source heterogeneous characteristic parameters into a comprehensive key characteristic parameter. This parameter fully integrates the advantageous information of each sensor under the current operating condition, effectively suppressing interference from low-quality signals and providing a more accurate and reliable basis for subsequent fault diagnosis.

[0081] In addition, before the step of obtaining the fused key feature parameters by weighted summation of the corresponding key feature parameters according to the initial weights of different sensor channels, the method further includes:

[0082] A penalty factor is introduced into the initial weights of sensor channels with transmission lag, and an exponential moving average filter is used for smoothing.

[0083] The expression for the initial weighting process is as follows:

[0084] ;

[0085] ;

[0086] ;

[0087] in, No. The initial weights of each sensor channel are adjusted by introducing a penalty factor. For the first The initial weights of each sensor channel, As a penalty factor, It is the natural logarithm. The attenuation coefficient controls the strength of the effect of delay on the weights. For the first Transmission delay time of each sensor channel For the first The final smoothed weights of each sensor channel. For smoothing coefficients, For the previous moment The final smoothing weights for each sensor channel.

[0088] In the process of multimodal sensor data fusion for drive motors, due to factors such as the hardware performance and transmission lines of each sensor, some sensor channels may experience data transmission lag. This can lead to asynchronous data acquisition from these channels with other sensor data. If fusion is performed directly according to the initial weights, errors may be introduced, potentially even resulting in diagnostic errors. To address this issue, before obtaining the fused key feature parameters by weighted summation based on the initial weights, the initial weights of sensor channels with transmission lag are specially processed. First, a penalty factor is introduced to adjust the initial weights of sensor channels with transmission lag. However, simple penalty adjustment may cause excessively drastic weight changes. Therefore, an exponential moving average filter is further used to smooth the adjusted weights, making the weight changes more stable and continuous, avoiding the impact of sudden weight changes on the stability of data fusion. This effectively eliminates the negative impact of transmission lag, resulting in more accurate and reliable fused key feature parameters, thus improving the accuracy and stability of drive motor fault diagnosis.

[0089] In summary, the drive motor fault diagnosis method based on artificial intelligence technology in the above embodiments of the present invention collects multi-dimensional data signals by deploying multi-modal sensors at preset locations on the drive motor. After real-time filtering, amplification, and noise reduction, key feature parameters are extracted. A dynamic weighted fusion algorithm is used to assign weights to the parameters of different sensor channels based on signal quality and operating condition information and then fuse them. Finally, the fused key feature parameters are input into a hybrid deep network architecture composed of a long short-term memory network and a convolutional neural network for fault diagnosis. Multi-modal sensors overcome the limitations of single-sensor information dimensions and can cover multiple types of fault features. Dynamic weight optimization improves data reliability and enhances operating condition adaptability. Deep feature learning enhances the distinguishability of complex fault modes, ultimately effectively improving the accuracy of fault diagnosis. This solves the problem of low accuracy in drive motor fault diagnosis in the prior art.

[0090] Example 3

[0091] Please see Figure 2 The image shows a drive motor fault diagnosis device based on artificial intelligence technology proposed in the third embodiment of the present invention. The device includes:

[0092] The acquisition module 100 is used to deploy multi-modal sensors at preset locations on the drive motor and acquire multi-dimensional data signals related to drive motor fault diagnosis collected by the multi-modal sensors.

[0093] The extraction module 200 is used to perform real-time filtering, amplification and noise reduction on the multidimensional data signal, and then use the feature extraction algorithm to extract features from the multidimensional data signal to obtain multiple key feature parameters.

[0094] The fusion module 300 is used to assign different weights to the key feature parameters acquired by different sensor channels in the multimodal sensors based on the signal quality and current operating condition information of the multimodal sensors according to the preset dynamic weighted fusion algorithm, so as to fuse multiple key feature parameters to obtain fused key feature parameters.

[0095] The diagnostic module 400 is used to establish a hybrid deep network architecture composed of a long short-term memory network and a convolutional neural network, and input the fused key feature parameters into the pre-trained hybrid deep network architecture to obtain the corresponding fault information of the drive motor.

[0096] The functions or operation steps implemented by the above modules are largely the same as those in the above method embodiments, and will not be repeated here.

[0097] Example 4

[0098] In another aspect, the present invention provides a readable storage medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the steps of the method described in any one of Embodiments 1 to 2 above.

[0099] Example 5

[0100] In another aspect, the present invention provides an electronic device, the electronic device including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the program to implement the steps of any one of the methods described in Embodiments 1 to 2 above.

[0101] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0102] Those skilled in the art will understand that the logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequential list of executable instructions for implementing logical functions, and can be embodied in any computer-readable storage medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable storage medium" can mean any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0103] More specific examples (a non-exhaustive list) of computer-readable storage media include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable storage media can even be paper or other suitable media on which the program can be printed, since the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0104] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0105] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0106] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A drive motor fault diagnosis method based on artificial intelligence technology, characterized in that, The method comprises: respectively, the preset part of the driving motor is arranged with multi-modal sensors, wherein ultrasonic probes are arranged on the stator, rotor and bearing of the driving motor, vibration and temperature sensors are installed outside and inside the driving motor, and multi-dimensional data signals related to driving motor fault diagnosis collected by the multi-modal sensors are acquired; respectively, the multi-dimensional data signals are filtered, amplified and denoised in real time, and then the multi-dimensional data signals are extracted by a feature extraction algorithm to obtain a plurality of key feature parameters; based on the preset dynamic weighted fusion algorithm, according to the signal quality and current working condition information collected by the multi-modal sensors, different weights are given to the key feature parameters obtained by different sensor channels in the multi-modal sensors to fuse the plurality of key feature parameters to obtain fused key feature parameters; a hybrid deep network architecture composed of long short-term memory network and convolutional neural network is established, and the fused key feature parameters are input into the pre-trained hybrid deep network architecture to obtain corresponding driving motor fault information; wherein, the step of based on the preset dynamic weighted fusion algorithm, according to the signal quality and current working condition information collected by the multi-modal sensors, different weights are given to the key feature parameters obtained by different sensor channels in the multi-modal sensors to fuse the plurality of key feature parameters to obtain fused key feature parameters comprises: respectively, the signal-to-noise ratio and signal variance of the data signals collected by the multi-modal sensors are obtained, and the signal-to-noise ratio and signal variance are linearly mapped to a preset interval and then weighted fused to obtain corresponding sensor single-channel signal quality scores; according to the current working condition information, the basic weight coefficients of each sensor channel corresponding to the current working condition information are found in the preset basic weight mapping table; the initial weights of different sensor channels are calculated by combining the sensor single-channel signal quality scores and the basic weight coefficients under the corresponding working conditions; a penalty factor is introduced to the initial weight of the sensor channel with transmission lag, and exponential moving average filtering is used for smoothing processing; wherein, the expression of the initial weight processing process is: ； ； ； in, No. The initial weights of each sensor channel are adjusted by introducing a penalty factor. For the first The initial weights of each sensor channel, As a penalty factor, It is the natural logarithm. The attenuation coefficient controls the strength of the effect of delay on the weights. For the first Transmission delay time of each sensor channel For the first The final smoothed weights of each sensor channel. For smoothing coefficients, For the previous moment The final smoothing weights for each sensor channel; the fused key feature parameters are obtained by weighting and summing the corresponding key feature parameters according to the initial weights of different sensor channels. 2.The drive motor fault diagnosis method based on artificial intelligence technology according to claim 1, characterized in that, The step of using a feature extraction algorithm to extract a plurality of key feature parameters from the multi-dimensional data signals comprises: time domain features are extracted from the multi-dimensional data signals based on the statistics of the multi-dimensional data signals; frequency domain features are extracted from the multi-dimensional data signals based on Fourier transform of the multi-dimensional data signals; time-frequency domain features are extracted from the multi-dimensional data signals based on wavelet transform of the multi-dimensional data signals; transform domain features are extracted from the multi-dimensional data signals based on Hilbert-Huang transform of the multi-dimensional data signals. 3.The drive motor fault diagnosis method based on artificial intelligence technology according to claim 1, characterized in that, The step of establishing a hybrid deep network architecture composed of long short-term memory network and convolutional neural network, and inputting the fused key feature parameters into the pre-trained hybrid deep network architecture to obtain corresponding driving motor fault information comprises: The long short-term memory network learns the long-term and short-term dependencies in the time series data in the key feature parameters through a gating mechanism, and the convolution kernel in the convolutional neural network slides to extract local structured features. The long short-term memory network outputs the time series global features, and the convolutional neural network outputs the local structured features, which are fused through a fully connected layer to map to a classification / regression target, thereby obtaining the fault information of the drive motor. 4.The method of claim 3, wherein, The method further comprises: determining an abnormal threshold for different working conditions according to the fault probability distribution of normal samples under different working conditions; In the inference stage of the hybrid deep network architecture, the corresponding abnormal threshold is loaded according to the identified working condition, and it is determined whether the drive motor has a fault according to the comparison between the abnormal probability output by the hybrid deep network architecture and the abnormal threshold. 5.The drive motor fault diagnosis method based on artificial intelligence technology according to claim 1, characterized in that, The step of arranging multiple modal sensors at the preset positions of the drive motor comprises: ultrasonic probes are arranged on the stator, rotor and bearing of the drive motor, and vibration and temperature sensors are installed outside and inside the drive motor.

6. An apparatus for diagnosing a fault of a driving motor based on an artificial intelligence technology, characterized by, The device for implementing the artificial intelligence technology-based drive motor fault diagnosis method of any one of claims 1 to 5 comprises: an acquisition module for arranging multiple modal sensors at the preset positions of the drive motor and acquiring multi-dimensional data signals related to drive motor fault diagnosis collected by the multiple modal sensors; an extraction module for performing real-time filtering, amplification and noise reduction processing on the multi-dimensional data signals, and then performing feature extraction on the multi-dimensional data signals by a feature extraction algorithm to obtain multiple key feature parameters; a fusion module for assigning different weights to the key feature parameters obtained by different sensor channels in the multiple modal sensors according to the signal quality and current working condition information collected by the multiple modal sensors based on a preset dynamic weighted fusion algorithm, so as to fuse the multiple key feature parameters to obtain fused key feature parameters; a diagnosis module for establishing a hybrid deep network architecture composed of a long short-term memory network and a convolutional neural network, and inputting the fused key feature parameters into the pre-trained hybrid deep network architecture to obtain corresponding drive motor fault information.

7. A readable storage medium, having stored thereon a computer program, characterized in that, The program is executed by the processor to implement the steps of the method of any one of claims 1 to 5.

8. An electronic device, comprising: The computer program is stored in the memory and executed on the processor, and the processor implements the steps of the method of any one of claims 1 to 5 when executing the program. The computer program is stored in the memory and executed on the processor, and the processor implements the steps of the method of any one of claims 1 to 5 when executing the program.

Images