Electrical equipment multi-mode fault diagnosis method based on dynamic self-adaption

Abstract

The invention relates to the technical field of electrical equipment fault diagnosis, and discloses an electrical equipment multi-modal fault diagnosis method based on dynamic self-adaption, and the method comprises the steps: obtaining an original multi-modal signal flow containing vibration, temperature and current signals, inputting the original multi-modal signal flow into a dynamic self-adaption diagnosis network, obtaining a fault feature matching result set, and determining each modal result. The network is trained by historical fault data, and the data comprises time domain, frequency domain and fusion feature parameters extracted from continuous multi-mode signals, and corresponding fault type labels and confidence coefficients. The network comprises a cross-modal feature fusion module, a spatial-temporal feature coding module and the like, and when the confidence coefficient of at least two modal results exceeds a dynamic threshold value, three-level early warning is triggered. The method improves the comprehensiveness, accuracy and real-time performance of diagnosis, and is suitable for fault diagnosis of electrical equipment.

Description

Technical Field

[0001] The present invention relates to the technical field of electrical equipment fault diagnosis, and specifically to a multimodal fault diagnosis method for electrical equipment based on dynamic adaptability. Background Art

[0002] In modern power systems, the safe and stable operation of electrical equipment is crucial for ensuring power supply reliability and the normal operation of the power system. With the development of power equipment towards high voltage, large capacity, and intelligence, its operating conditions are becoming increasingly complex, and the types of faults also show the characteristics of diversification and complexity. Traditional electrical equipment fault diagnosis methods usually only analyze single-modal signals (such as vibration signals, temperature signals, or current signals, etc.), which are difficult to comprehensively and accurately reflect the actual operating state of the equipment, and there are problems such as low diagnostic accuracy, high missed diagnosis or misdiagnosis rates.

[0003] On the one hand, the fault information provided by single-modal signals is limited. For example, vibration signals mainly reflect the operating state of mechanical components of the equipment, and are more sensitive to mechanical faults (such as bearing wear, gearbox faults, etc.), but the reflection of electrical faults (such as winding overheating, partial discharge, etc.) is relatively lagged or not obvious; temperature signals can directly reflect thermal faults such as local overheating of the equipment, but for some early mechanical or electrical faults, the subtle changes in temperature may not be captured in time; current signals are mainly related to the electrical parameters and load conditions of the equipment, and can be used to analyze faults in electrical circuits, but the fault diagnosis ability for mechanical structure faults is relatively weak. Therefore, relying solely on single-modal signals for fault diagnosis is likely to lead to misjudgment or missed judgment of complex faults.

[0004] On the other hand, most traditional diagnosis methods use fixed diagnosis models and parameters, lacking the ability to adapt to the dynamic changes in the operating state of the equipment. The signal characteristics of electrical equipment will change at different operating stages and under different load conditions. Traditional methods are difficult to automatically adjust the diagnosis model according to real-time signal characteristics, resulting in a decline in diagnostic performance as the operating state of the equipment changes. For example, at the transition stage between normal and abnormal operation of the equipment, the subtle changes in signal characteristics may not be effectively identified by traditional fixed models, thus delaying the fault diagnosis time.

[0005] In addition, although existing multimodal fault diagnosis methods fuse multiple signals, most of them use simple feature splicing or fixed-weight fusion methods, and do not fully consider the importance differences of different modal signals under different fault types. For example, in some fault types, vibration signals may carry more critical fault information, while in other fault types, temperature signals or current signals may be more important. Simple fusion methods cannot dynamically adjust the weights of each modal signal, resulting in the fused features not fully leveraging the advantages of multimodal signals and affecting diagnostic accuracy.

[0006] Meanwhile, with the development of the intelligence and digitization of the power system, a large amount of multi-modal data will be generated during the operation of electrical equipment. Traditional diagnostic methods are inefficient in data processing and feature extraction, and it is difficult to meet the needs of real-time fault diagnosis. How to quickly and accurately extract effective fault features from the large amount of multi-modal data and build a diagnostic model with dynamic adaptive ability has become a key problem to be solved urgently in the field of electrical equipment fault diagnosis. Summary of the Invention

[0007] The purpose of the present invention is to provide a multi-modal fault diagnosis method for electrical equipment based on dynamic adaptability to solve the problems raised in the above background technology.

[0008] To achieve the above purpose, the present invention provides the following technical solution: A multi-modal fault diagnosis method for electrical equipment based on dynamic adaptability, the method includes:

[0009] The method is applied to an electrical equipment fault diagnosis system, and the method includes:

[0010] Obtain the original multi-modal signal stream to be fault diagnosed, and the original multi-modal signal stream includes vibration signals, temperature signals and current signals;

[0011] Input the original multi-modal signal stream into a pre-trained dynamic adaptive diagnosis network to obtain a fault feature matching result set;

[0012] Based on the fault feature matching result set, determine the fault feature matching result corresponding to each modal signal in the original multi-modal signal stream;

[0013] The dynamic adaptive diagnosis network is a parameterized network obtained by training an initial diagnosis network to be trained with historical fault data. The historical fault data is generated by a batch of historical feature parameters respectively extracted from a batch of historical multi-modal signals, a fault type label and a confidence coefficient corresponding to each historical feature parameter. The fault type label reflects the abnormal state identification of a historical feature parameter, and the confidence coefficient reflects the correction factor corresponding to the abnormal state identification. The batch of historical multi-modal signals is a batch of continuously collected electrical equipment operation signals. Each historical feature parameter includes a feature parameter determined by a historical multi-modal signal and its adjacent historical multi-modal signals;

[0014] The fault feature matching result is used to represent the fault type identification of the multi-modal signal stream.

[0015] Preferably, the method further includes:

[0016] Extract time-domain characteristic parameters from the batch of historical multi-modal signals to obtain a first set of characteristic parameters, where one characteristic parameter in the first set of characteristic parameters corresponds to one historical multi-modal signal in the batch of historical multi-modal signals;

[0017] Extract frequency-domain characteristic parameters from the batch of historical multi-modal signals to obtain a second set of characteristic parameters, where the first characteristic parameter in the second set of characteristic parameters corresponds to the first historical multi-modal signal in the historical multi-modal signals, and the adjacent historical multi-modal signal of the first historical multi-modal signal is the second historical multi-modal signal, and the first characteristic parameter represents the frequency-domain correlation characteristic between the first historical multi-modal signal and the second historical multi-modal signal;

[0018] Fuse the first set of characteristic parameters and the second set of characteristic parameters respectively to obtain the batch of historical characteristic parameters.

[0019] Preferably, the extracting frequency-domain characteristic parameters from the batch of historical multi-modal signals to obtain a second set of characteristic parameters includes:

[0020] Extract frequency-domain characteristic parameters from the batch of historical multi-modal signals through the following steps to obtain a second set of characteristic parameters, where each historical multi-modal signal for which the frequency-domain characteristic parameters are extracted is used as the current historical multi-modal signal, and the obtained characteristic parameters are used as the current characteristic parameters, and the second set of characteristic parameters includes the current characteristic parameters;

[0021] Perform a spectrum decomposition operation on the current historical multi-modal signal to determine a first set of frequency-domain characteristic quantities, where the first set of frequency-domain characteristic quantities is used to describe the main frequency band energy distribution in the current historical multi-modal signal;

[0022] Perform a spectrum decomposition operation on the adjacent historical multi-modal signal of the current historical multi-modal signal to determine a second set of frequency-domain characteristic quantities, where the second set of frequency-domain characteristic quantities is used to describe the main frequency band energy distribution in the adjacent historical multi-modal signal;

[0023] Determine the current characteristic parameter based on the first set of frequency-domain characteristic quantities and the second set of frequency-domain characteristic quantities, where the current characteristic parameter is used to represent the energy deviation amount of the corresponding frequency bands in the first set of frequency-domain characteristic quantities and the second set of frequency-domain characteristic quantities.

[0024] Preferably, the method further includes:

[0025] Determine the fault type label and confidence coefficient corresponding to each historical feature parameter in the batch of historical feature parameters respectively through the following steps. Herein, each historical feature parameter for which the fault type label and confidence coefficient are determined each time is used as the current historical feature parameter, and the fault type label corresponding to the current historical feature parameter is used as the current fault type label and the current confidence coefficient:

[0026] Input the current historical feature parameter into the initial diagnosis network to obtain the current fault type label, wherein the initial diagnosis network is a parameterized network obtained by performing initialization processing in advance;

[0027] Determine the current confidence coefficient based on the current fault type label and the reference verification identifier corresponding to the current historical feature parameter. Herein, the reference verification identifier carries the calibration label in the historical multimodal signal corresponding to the current historical feature parameter, the calibration label includes the target fault type identifier of the multimodal signal stream, and the confidence coefficient is used to indicate whether the current fault type label is consistent with the target fault type identifier.

[0028] Preferably, the step of inputting the current historical feature parameter into the initial diagnosis network to obtain the current fault type label includes:

[0029] Input each fault type identifier in the pre-stored fault type identifier set into the first classification algorithm branch in sequence together with the current historical feature parameter to obtain a first set of classification confidence indicators, wherein the first set of classification confidence indicators includes the confidence indicators corresponding to each fault type identifier;

[0030] Determine the candidate fault type identifier corresponding to the highest first confidence indicator value in the first set of classification confidence indicators as the current fault type identifier corresponding to the current fault type label.

[0031] Preferably, before inputting each fault type identifier in the pre-stored fault type identifier set into the first classification algorithm branch in sequence together with the current historical feature parameter to obtain a first set of classification confidence indicators, the method further includes:

[0032] Determine the signal stability indicator through a preset mode;

[0033] Determine whether the signal stability indicator meets the set stability threshold;

[0034] On the basis that the signal stability indicator meets the set stability threshold, dynamically select the candidate fault type identifier in the pre-stored fault type identifier set as the current fault type label;

[0035] Based on the signal stability index not meeting the set stability threshold, input the current historical feature parameters into the initial diagnosis network to obtain the current fault type label.

[0036] Preferably, determining the current fault type identifier corresponding to the current fault type label by the first confidence index with the highest value among the first group of classification confidence indexes includes:

[0037] Based on the candidate fault type identifier corresponding to the first confidence index being the same as the target fault type identifier, determining the current confidence coefficient as the first confidence coefficient, where the first confidence coefficient indicates that the initial diagnosis network successfully generates a matching identifier;

[0038] Based on the candidate fault type identifier corresponding to the first confidence index being different from the target fault type identifier, determining the current confidence coefficient as the second confidence coefficient, where the second confidence coefficient indicates that the initial diagnosis network fails to generate a matching identifier.

[0039] Preferably, the method further includes:

[0040] Screen a number of groups of historical combined fault records from the historical fault data, where the k-th group of historical combined fault records in the number of groups of historical combined fault records includes the k-th historical feature parameter, the k-th fault type label corresponding to the k-th historical feature parameter, the k-th confidence coefficient, and the (k + 1)-th historical feature parameter, and k is a positive integer;

[0041] Train the initial diagnosis network to be trained based on the number of groups of historical combined fault records to obtain the dynamic adaptive diagnosis network, where, based on the number of iterations of training the initial diagnosis network reaching the set training threshold, determining the initial diagnosis network as the dynamic adaptive diagnosis network, and based on the number of iterations of training the initial diagnosis network not reaching the set training threshold, updating the network parameters of the initial diagnosis network according to a preset error function, and the input of each round of training process is a group of historical combined fault records in the number of groups of historical combined fault records.

[0042] Preferably, the dynamic adaptive diagnosis network includes:

[0043] A cross-modal feature fusion module that dynamically calculates the modal weight coefficients of vibration signals, temperature signals, and current signals using an attention mechanism;

[0044] A spatio-temporal feature encoder that encodes time series features using a bidirectional LSTM structure;

[0045] A multi-granularity classifier, including a parallel convolutional neural network branch and a fully connected classification branch, where the outputs of each branch are fused through an adaptive weighting strategy to generate the final fault type identification;

[0046] An online update module that automatically triggers incremental learning training of network parameters when the confidence coefficients of N consecutive signal samples are detected to be lower than a preset threshold.

[0047] Preferably, the method further includes:

[0048] When the confidence coefficients of the fault feature matching results of at least two modalities among the vibration signal, temperature signal, and current signal simultaneously exceed the dynamic threshold, a three-level early warning mechanism is triggered. Among them, the first-level early warning displays a yellow warning sign through the device operation interface and generates a diagnostic log, the second-level early warning superimposes sound and light alarms and automatically locks the device operation parameters, and the third-level early warning synchronously pushes an encrypted alarm message to the remote monitoring platform;

[0049] The dynamic threshold is dynamically adjusted according to the confidence distribution law of the same type of faults in the device historical operation data, and the threshold adjustment step size of different modality signals is negatively correlated with their signal sampling frequencies.

[0050] Compared with the prior art, the beneficial effects of the present invention are:

[0051] In terms of multi-modal signal processing, by extracting and fusing time-domain and frequency-domain features from historical multi-modal signals, the spatio-temporal features in the signals can be comprehensively captured. Time-domain feature extraction can directly reflect the instantaneous and statistical characteristics of the signal, such as mean, variance, peak value, etc., and effectively identify abnormal fluctuations in the signal; frequency-domain feature extraction analyzes the frequency components and energy distribution of the signal through spectral decomposition operations, and can detect changes in frequency components caused by faults and frequency-domain correlation features between adjacent signals. For example, by calculating the main frequency band energy deviation between the current signal and adjacent signals, the early frequency feature change trend of the fault can be keenly captured, providing a basis for early warning of the fault. By fusing time-domain and frequency-domain features, the complementary nature of different domain features can be fully utilized to form a more comprehensive and rich description of fault features, overcome the limitations of single-domain features, and improve the characterization ability of features for fault types.

[0052] The construction and training mechanism of the dynamic adaptive diagnosis network has significant advantages. The network adopts a cross-modal feature fusion module, which uses an attention mechanism to dynamically calculate the weight coefficients of each modal signal, and can automatically adjust the importance of vibration, temperature, and current signals according to different fault types. For example, in mechanical fault diagnosis, the weight coefficient of the vibration signal automatically increases, making it dominant in feature fusion, thus more accurately identifying mechanical faults; in electrical fault diagnosis, the weight coefficient of the current signal increases, highlighting the key role of the current signal and improving the diagnostic accuracy of electrical faults. The spatio-temporal feature encoder adopts a bidirectional LSTM structure, which can effectively capture the forward and backward dependencies in time series signals and deeply encode the temporal features of the signals, and is suitable for analyzing the operation signals of electrical equipment with dynamic change characteristics. The multi-granularity classifier includes parallel convolutional neural network branches and fully connected classification branches. Each branch can classify features from different granularities, and the outputs of the branches are fused through an adaptive weighting strategy to improve the reliability and robustness of the classification results. The online update module can automatically trigger the incremental learning training of network parameters when the confidence coefficient of N consecutive signal samples is detected to be lower than the preset threshold, enabling the network to update model parameters in real time according to new fault data, adapt to changes in the equipment operation state, and avoid the decline in diagnostic performance caused by factors such as equipment aging and environmental changes.

[0053] The confidence coefficient and early warning mechanism in the fault diagnosis process further enhance the reliability and practicality of the diagnosis. By comparing the current fault type label with the reference verification identifier (carrying a calibration label) to determine the confidence coefficient, the credibility of the diagnostic result can be quantitatively evaluated. When the confidence coefficients of the fault feature matching results of at least two modalities simultaneously exceed the dynamic threshold, a three-level early warning mechanism is triggered, and different countermeasures can be taken according to the severity of the fault. The first-level early warning displays a yellow warning sign on the device operation interface and generates a diagnostic log to achieve timely monitoring and recording of early faults; the second-level early warning superimposes sound and light alarms and automatically locks the device operation parameters to facilitate maintenance personnel to detect faults in time and perform on-site processing; the third-level early warning synchronously pushes encrypted alarm messages to the remote monitoring platform to ensure remote fast response and collaborative processing in case of serious faults. The dynamic threshold is dynamically adjusted according to the confidence distribution law of the same type of faults in the historical operation data of the device, and the threshold adjustment step size of different modal signals is negatively correlated with their signal sampling frequencies, enabling the early warning mechanism to adapt to the fault feature changes of different devices and different operation stages and improving the accuracy and timeliness of the early warning. Description of the Drawings

[0054] Figure 1 It is the working principle diagram of the multi-modal fault diagnosis method for electrical equipment based on dynamic adaptability according to the present invention;

[0055] Figure 2 It is the flow chart for extracting historical feature parameters;

[0056] Figure 3 It is a flowchart for extracting frequency-domain characteristic parameters;

[0057] Figure 4 It is a flowchart for determining fault type labels and confidence coefficients. Specific implementation manners

[0058] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0059] Please refer to Figures 1 - 4 , the multi-modal fault diagnosis method for electrical equipment based on dynamic adaptability involved in the present invention is specifically implemented as follows:

[0060] Obtain the original multi-modal signal stream to be fault diagnosed, and the original multi-modal signal stream includes vibration signals, temperature signals and current signals. Through the sensor array arranged at the key parts of the electrical equipment, multi-dimensional real-time signals during the operation of the equipment are continuously collected. Among them, the vibration signal is obtained by an acceleration sensor, the temperature signal is collected through a thermocouple or an infrared temperature measurement module, and the current signal is extracted from the electrical circuit through a current transformer. All signals are synchronously collected at a preset sampling frequency (such as 10 kHz) and transmitted to the signal preprocessing unit of the diagnosis system.

[0061] Input the original multi-modal signal stream into the pre-trained dynamic adaptive diagnosis network to obtain a fault feature matching result set. After the signal preprocessing unit performs preliminary processing such as denoising and normalization on the original signal, a continuous signal stream is formed according to the time series and input into the dynamic adaptive diagnosis network. Through the collaborative operation of components such as the cross-modal feature fusion module, spatio-temporal feature encoder, and multi-granularity classifier inside the network, the fault features in the multi-modal signal are extracted, fused and matched, and finally a result set containing the possibility of fault feature matching of each modal signal is output.

[0062] Based on the fault feature matching result set, determine the fault feature matching result corresponding to each modal signal in the original multi-modal signal stream. The diagnosis system separates the fault feature matching results corresponding to the vibration, temperature, and current signals respectively according to the distribution of the characteristic parameters of each modal signal in the fault feature matching result set. Each result includes a fault type identifier and a corresponding confidence coefficient, providing a basis for subsequent fault location and early warning.

[0063] The dynamic adaptive diagnosis network is a parameterized network obtained by inputting historical fault data into the initial diagnosis network to be trained. The construction process of historical fault data is as follows: first, a batch of continuous electrical equipment operation signals are collected as historical multimodal signals, and for each historical multimodal signal, the corresponding characteristic parameters are extracted by combining its adjacent historical multimodal signals to form a batch of historical characteristic parameters; at the same time, the corresponding fault type label and confidence coefficient are marked for each historical characteristic parameter, wherein the fault type label reflects the abnormal state identification of the characteristic parameter, and the confidence coefficient reflects the correction factor corresponding to the abnormal state identification.

[0064] The fault feature matching result is used to represent the fault type identification of the multi-modal signal stream, and accurate identification of the input signal fault type is achieved through network classification and matching operations.

[0065] The technical solution of the present invention is further described in detail below in conjunction with specific embodiments.

[0066] Embodiment 1:

[0067] Based on the overall implementation scheme, this embodiment describes in detail the extraction process of historical feature parameters. When extracting time domain feature parameters for a batch of historical multimodal signals, for each historical multimodal signal (such as a combination of vibration, temperature, and current signals collected at a certain moment), multiple statistical features are calculated in the time domain to form a first set of feature parameters. Taking the vibration signal as an example, its mean value needs to be calculated. The mean value reflects the average energy level of the signal in the time domain, which can be obtained by summing the values of all sampling points of the vibration signal and dividing it by the number of sampling points; the variance is calculated to reflect the degree of fluctuation of the signal. The larger the variance, the more dispersed the signal energy distribution and the more violent the fluctuation. The calculation method is the average value of the square of the difference between the values of each sampling point and the mean; the peak parameter can indicate whether there is impact vibration, which is determined by identifying the maximum instantaneous value in the signal; the kurtosis is used to measure the sharpness of the signal amplitude distribution, and the skewness is used to describe the symmetry of the signal amplitude distribution. These two features can be calculated by the corresponding probability statistics formula. For temperature signals, time domain characteristic parameters such as mean can reflect the current average temperature level of the device, and variance can analyze the stability of temperature changes. If the variance is small, it means that the temperature fluctuation is small and the device operating temperature is stable. Time domain characteristic parameters of current signals such as mean can determine the average size of the current, and variance can be used to analyze the distortion of the current waveform. If the variance increases abnormally, it may mean that the current waveform changes irregularly, indicating that the device may have an electrical fault. Each historical multimodal signal corresponds to a characteristic parameter composed of the time domain statistical features of each mode. Such characteristic parameters of all historical multimodal signals together constitute the first set of characteristic parameters.

[0068] When extracting frequency-domain characteristic parameters from a batch of historical multi-modal signals to obtain a second set of characteristic parameters, the historical multi-modal signals for each frequency-domain characteristic parameter extraction are defined as the current historical multi-modal signals, and their corresponding characteristic parameters are the current characteristic parameters. In specific operations, first, perform a spectrum decomposition operation on the current historical multi-modal signals. Convert the time-domain signals into frequency-domain signals through the Fast Fourier Transform (FFT), and then determine the first set of frequency-domain characteristic quantities, which are used to describe the energy distribution in the main frequency band of the current historical multi-modal signals. Taking vibration signals as an example, through spectrum decomposition, the main frequency components where the vibration energy is concentrated and their energy proportions can be identified. If there is a significant increase in energy at a certain specific frequency (such as the bearing fault characteristic frequency), it may indicate the existence of corresponding mechanical faults. For temperature signals and current signals, the frequency-domain characteristic quantities can be used to analyze whether there are abnormalities in the energy distribution of their signals at different frequencies. For example, if there is an increase in the energy of high-order harmonics other than the power frequency in the current signal, it may indicate the existence of harmonic interference or faults in the internal components of the equipment. Next, perform the same spectrum decomposition operation on the adjacent historical multi-modal signals (such as the signals at the previous or next moment) of the current historical multi-modal signals to determine the second set of frequency-domain characteristic quantities, which are used to describe the energy distribution in the main frequency band of the adjacent signals. Then, based on the first set of frequency-domain characteristic quantities and the second set of frequency-domain characteristic quantities, calculate the energy deviation of the corresponding frequency bands. This deviation is the current characteristic parameter. For example, if the energy of the main frequency band (such as 100Hz - 200Hz) of the current vibration signal is E1, and the energy of the same frequency band of the vibration signal at the adjacent moment is E2, the energy deviation can be expressed as |E1 - E2| / E1, which is used to reflect the change trend of the adjacent signals in the frequency domain. If this deviation is large, it may mean that the fault has developed significantly in a short period of time and needs to be focused on. All current characteristic parameters constitute the second set of characteristic parameters.

[0069] After the extraction of time-domain feature parameters and frequency-domain feature parameters is completed, the first set of feature parameters and the second set of feature parameters need to be fused separately to obtain a batch of historical feature parameters. The fusion process adopts a feature-level fusion strategy, that is, the time-domain features and frequency-domain features of the same modality are combined according to the modality (vibration, temperature, current) to form a feature vector containing multi-dimensional information. Taking the vibration modality as an example, the features such as the mean, variance, peak value, kurtosis, and skewness in its time domain are merged with the main frequency band energy distribution feature and the energy deviation amount feature between adjacent moments in the frequency domain to form the complete feature parameters of this modality. These feature parameters simultaneously contain the statistical characteristics of the vibration signal in the time domain and the energy distribution and change trend information in the frequency domain. Similarly, the same feature fusion operation is performed on the temperature modality and the current modality, and their respective time-domain features and frequency-domain features are combined into complete modal feature parameters. Finally, after all historical multi-modal signals are processed as described above, a batch of historical feature parameters containing the spatio-temporal features of each modality are obtained. These feature parameters can comprehensively and deeply reflect the state information during the operation of electrical equipment, providing rich and effective input data for the training of the subsequent dynamic adaptive diagnosis network, enabling the network to learn the feature manifestations and evolution laws of different fault types in multi-modal signals.

[0070] Embodiment 2:

[0071] On the basis of the completion of the extraction of historical feature parameters in Embodiment 1, this embodiment elaborates in detail the determination process of the fault type labels and confidence coefficients corresponding to the historical feature parameters. This process realizes the accurate labeling of the fault type and the confidence evaluation by inputting and calculating the initial diagnosis network and combining the benchmark verification identifier.

[0072] The historical feature parameters for determining the fault type labels and confidence coefficients each time are defined as the current historical feature parameters, and the corresponding fault type labels and confidence coefficients are respectively called the current fault type labels and the current confidence coefficients. The initial diagnosis network is a parameterized network that has been pre-initialized. Its structure includes an input layer, a hidden layer, and an output layer. The neurons in the hidden layer perform signal transmission and calculation through preset weights and biases. When determining the current fault type label, a multi-step classification algorithm input strategy is adopted.

[0073] In the first step, each fault type identifier in the pre-stored fault type identifier set is input into the first classification algorithm branch in sequence together with the current historical feature parameters. The pre-stored fault type identifier set covers various types of faults that may occur in electrical equipment, such as bearing wear, winding overheating, load abnormality, insulation aging, poor contact, etc. The first classification algorithm branch can use classification algorithms such as support vector machine (SVM) and random forest to calculate the combination of input feature parameters and fault type identifiers. Through the internal operation of the algorithm, the first set of classification confidence indicators is obtained, which contains the confidence indicators corresponding to each fault type identifier. These indicators reflect the possibility that the current historical feature parameters belong to the fault type. For example, for the bearing wear fault type identifier, the algorithm may output a confidence indicator of 0.75, indicating that the current historical feature parameters have a 75% probability of corresponding to the bearing wear fault.

[0074] In the second step, the candidate fault type identification corresponding to the first confidence index with the highest value is selected from the first group of classification confidence indicators. Assuming that among the confidence indicators corresponding to all fault type identifications, the confidence index 0.85 corresponding to the bearing wear fault is the highest, the bearing wear is determined as the candidate fault type identification, and further determined as the current fault type identification corresponding to the current fault type label. This process ensures that the current fault type label is determined based on the maximum matching possibility between the historical feature parameters and the fault type.

[0075] Next, the current confidence coefficient is determined based on the current fault type label and the benchmark verification identifier corresponding to the current historical feature parameter. The benchmark verification identifier is a calibration label pre-set for the historical multimodal signal corresponding to the current historical feature parameter. The label is determined by manual diagnosis or other authoritative detection means and contains the target fault type identifier of the multimodal signal stream, that is, the actual fault type.

[0076] When determining the current confidence coefficient, the following judgment is made: If the candidate fault type identifier corresponding to the first confidence index is the same as the target fault type identifier in the benchmark verification identifier, it means that the initial diagnosis network has successfully generated an identifier that matches the actual fault type. At this time, the current confidence coefficient is determined as the first confidence coefficient, which can be set to 1.0, indicating that the network's diagnostic results are highly reliable. If the two are different, it means that the initial diagnosis network has not successfully generated a matching identifier. At this time, the current confidence coefficient is determined as the second confidence coefficient, which can be set to 0.0, indicating that the network's diagnostic results are unreliable. In this way, a confidence coefficient is labeled for each historical feature parameter, which is used to measure the accuracy of the initial diagnosis network's judgment of the fault type of the historical feature parameter, providing an important reference for the subsequent training of the dynamic adaptive diagnosis network.

[0077] Before entering the first classification algorithm branch, it is also necessary to judge the signal stability to ensure the reliability of the input data and the accuracy of the diagnostic results. The signal stability index is determined through a preset mode, which can be designed based on the time-domain statistical characteristics or frequency-domain energy distribution of the signal. For example, calculate the change amplitude of the time-domain characteristic parameters (such as mean and variance) of the current historical multimodal signal, or the fluctuation range of the energy in the main frequency band of the frequency domain, and use this as an index to measure the signal stability.

[0078] Then, compare the determined signal stability index with the set stability threshold. The stability threshold is preset according to the statistical law of the signal characteristics during the normal operation of the electrical equipment. For example, it is set that the change amplitude of the time-domain variance does not exceed ±10% of the normal range, and the energy fluctuation in the main frequency band of the frequency domain does not exceed ±5%.

[0079] If the signal stability index meets the set stability threshold, it indicates that the signal is in a stable state. In this case, dynamically select the candidate fault type identifier in the pre-stored set of fault type identifiers as the current fault type label. The dynamic selection strategy can give priority to selecting the fault type most likely to occur under this operating condition as a candidate according to factors such as the operating conditions of the equipment and the historical fault occurrence frequency, so as to improve the diagnostic efficiency. For example, when the equipment is operating at high load, give priority to considering the fault types related to the load.

[0080] If the signal stability index does not meet the set stability threshold, it indicates that the signal may have abnormal fluctuations. At this time, input the current historical characteristic parameters into the initial diagnostic network, and obtain the current fault type label through the complex operations of the network to ensure accurate fault diagnosis even when the signal is unstable. This process realizes the adaptive judgment of signal stability, adopts different diagnostic strategies according to different situations, and improves the flexibility and reliability of the diagnostic system.

[0081] In the whole process of determining the fault type label and the confidence coefficient, the collaborative work of multiple key steps and algorithms is involved. First, the parameter settings of the initial diagnostic network have an important impact on the classification results, and the initial values of its weights and biases determine the initial response mode of the network to the input features. Second, the selection of the first classification algorithm branch is also crucial. Different classification algorithms (such as SVM, random forest) have different characteristics and application scenarios, and need to be reasonably selected according to the characteristics of the historical characteristic parameters and the distribution of fault types. In addition, the accuracy of the reference verification identifier is directly related to the reliability of the confidence coefficient. Therefore, when setting the reference verification identifier, rigorous and accurate detection means need to be adopted to ensure that it can truly reflect the fault state of the equipment.

[0082] Through the method of this embodiment, corresponding fault type labels and confidence coefficients are determined for each historical feature parameter, forming a complete set of historical fault data annotation systems. These annotated data will be used as training samples to train the dynamic adaptive diagnosis network, enabling the network to learn the mapping relationship between fault features and fault types, as well as how to adjust the diagnosis strategy according to the confidence of feature parameters, thereby improving the accuracy and reliability of multi-modal fault diagnosis of electrical equipment.

[0083] Embodiment 3:

[0084] Based on Embodiment 2, this embodiment further illustrates the judgment of signal stability and the corresponding processing flow before inputting into the first classification algorithm branch. This process determines the signal stability index through a preset mode, compares it with the set stability threshold, and adopts different fault type label determination strategies according to the comparison results, thereby realizing the adaptive processing of signal stability and improving the flexibility and reliability of the diagnosis system.

[0085] When determining the signal stability index through a preset mode, this preset mode is designed based on the time-domain statistical characteristics or frequency-domain energy distribution of the signal. For time-domain statistical characteristics, the change amplitudes of parameters such as the mean, variance, and peak value of the current historical multi-modal signal can be calculated. For example, for vibration signals, calculate the mean change rate within consecutive multiple sampling periods. If the mean changes significantly within a short time, it may indicate that there are unstable factors in the signal. For temperature signals, calculate the fluctuation of its variance. A sudden increase in variance may mean that the temperature has abnormal fluctuations and the signal stability decreases. For frequency-domain energy distribution, the change of energy distribution in different frequency bands of the signal can be analyzed. For example, after converting the signal to the frequency domain through fast Fourier transform, calculate the change in the energy ratio of a specific frequency band (such as the fault feature frequency band). If the energy ratio of a certain frequency band suddenly increases or decreases, it may indicate that the signal stability is affected. These parameter change situations based on the time domain and frequency domain together constitute the signal stability index, which is used to measure the stability degree of the current historical multi-modal signal.

[0086] After determining the signal stability index, compare it with the set stability threshold. The stability threshold is preset according to the statistical law of signal characteristics during the normal operation of the electrical equipment. For the threshold of the change amplitude of time-domain parameters, for example, set that the mean change rate of vibration signals does not exceed ±5% of the normal range, and the variance fluctuation of temperature signals does not exceed ±8% of the normal range. For the threshold of frequency-domain energy distribution, set that the change in the energy ratio of a specific frequency band does not exceed ±10% of the normal range. The setting of these thresholds is based on a large amount of historical data statistical analysis to ensure that during the normal operation of the equipment, the signal stability index can be maintained within the threshold range, while when the equipment has abnormalities or faults, the signal stability index will exceed the threshold.

[0087] On the basis that the signal stability index meets the set stability threshold, it indicates that the signal is in a stable state. At this time, a candidate fault type identifier in the pre-stored fault type identifier set is dynamically selected as the current fault type label. The dynamic selection strategy comprehensively considers factors such as the operating conditions of the equipment and the historical fault occurrence frequency. For example, under the high-load operating conditions of the equipment, fault type identifiers related to the load (such as overload, winding overheating, etc.) are preferentially selected as candidates because the likelihood of these fault types occurring is relatively high under such operating conditions. Another example is that according to the historical fault records of the equipment, if a certain type of fault (such as bearing wear) has occurred frequently in the past, then when the signal is stable, the probability of selecting this fault type identifier as a candidate is appropriately increased. Through this dynamic selection strategy, when the signal is stable and the fault characteristics are relatively obvious, the most likely fault type label can be quickly determined, improving the diagnostic efficiency.

[0088] On the basis that the signal stability index does not meet the set stability threshold, it indicates that the signal may have abnormal fluctuations. At this time, the current historical feature parameters are input into the initial diagnostic network, and the current fault type label is obtained through the complex operations of the network. The initial diagnostic network has strong non-linear mapping ability and feature extraction ability, and can handle the abnormal fluctuations and complex features in the signal. When the signal is unstable, the network can, through its multi-layer structure and the connections between neurons, deeply analyze and process the input feature parameters, and dig out the fault features hidden in the unstable signal, so as to accurately judge the fault type. For example, when the vibration signal has short-term abnormal fluctuations due to external interference, the initial diagnostic network can comprehensively analyze the time-domain and frequency-domain features of the signal to distinguish whether it is a true fault feature or an interference signal, and thus give a reliable fault type label.

[0089] In the entire signal stability judgment and processing process, there are multiple key links. The design of the preset mode directly affects the accuracy of the signal stability index and needs to be designed specifically according to different types of signals (vibration, temperature, current) and the characteristics of the equipment. The setting of the stability threshold is also crucial. An overly high threshold may lead to insensitivity to signal abnormalities, while an overly low threshold may lead to frequent misjudgments. The rationality of the dynamic selection strategy determines whether the fault type label can be quickly and accurately determined when the signal is stable, and the actual operating conditions of the equipment and historical data need to be fully considered. The performance of the initial diagnostic network directly affects the fault diagnosis ability when the signal is unstable, and its accuracy and reliability need to be ensured through reasonable network structure design and parameter training.

[0090] Through the method of this embodiment, the adaptive judgment and processing of signal stability are realized. When the signal is stable, an efficient dynamic selection strategy is adopted to quickly determine the fault type label; when the signal is unstable, the powerful analysis ability of the initial diagnosis network is used for accurate diagnosis. This way of flexibly adjusting the diagnosis strategy according to the signal state improves the adaptability and reliability of the entire fault diagnosis system, can better handle various complex signal conditions during the operation of electrical equipment, and provides a strong guarantee for the safe and stable operation of the equipment.

[0091] Embodiment 4:

[0092] Based on Embodiment 3, the training process and network structure of the dynamic adaptive diagnosis network are described in detail in this embodiment. Through learning historical fault data, this network can adaptively process the multi-modal fault diagnosis of electrical equipment, improving the diagnosis accuracy and flexibility.

[0093] Select several groups of historical combined fault records from the historical fault data. Each group of records includes the kth historical feature parameter, the corresponding kth fault type label, the kth confidence coefficient, and the (k + 1)th historical feature parameter (k is a positive integer). These records not only contain the feature parameters and labels at a single moment, but also consider the continuity of feature parameters at adjacent moments, which helps the network learn the evolution law of fault features over time. For example, for the bearing wear fault, the vibration signal features at adjacent moments may show a gradually changing trend. By including this time continuity information, the network can better capture the feature change patterns during the fault development process.

[0094] Based on the selected historical combined fault records, train the initial diagnosis network to be trained. The input for each round of training is a group of historical combined fault records, and the network calculates the output result through forward propagation. In this process, the input historical feature parameters are processed and transformed through each layer of the network, and finally the prediction of the fault type is output. Then, according to a preset error function (such as the cross-entropy loss function), calculate the error between the prediction result and the true label (i.e., the fault type label). The error function measures the deviation degree between the network prediction result and the actual situation. The smaller the error, the more accurate the network prediction.

[0095] Update the network parameters of the initial diagnosis network, such as weights and biases, through the backpropagation algorithm. The backpropagation algorithm adjusts the weights of each connection in the network and the biases of the neurons according to the gradient information of the error function, so that the value of the error function gradually decreases. This process is continuously iterated until the number of iterations for training the initial diagnosis network reaches the set training threshold. At this time, it is considered that the network has fully learned the features and laws in the historical fault data, and the initial diagnosis network is determined as the dynamic adaptive diagnosis network. If the number of iterations does not reach the set training threshold, continue training until the condition is met.

[0096] The dynamic adaptive diagnosis network includes multiple functional modules that work together to achieve accurate diagnosis of multi-modal fault signals.

[0097] The cross-modal feature fusion module uses the attention mechanism to dynamically calculate the modal weight coefficients of vibration signals, temperature signals, and current signals. The attention mechanism can automatically adjust the weights of each modality according to the importance of different modal signals in the current fault diagnosis. For example, in the case of bearing wear faults, vibration signals usually contain richer fault feature information. Therefore, the module will assign a higher weight to the vibration signal, thus highlighting the features of this modality and improving the diagnosis accuracy. In the case of winding overheating faults, the temperature signal may be more critical, and the module will correspondingly increase the weight of the temperature signal. This way of dynamically adjusting the weights enables the network to adaptively focus on the most relevant modal signals according to the specific fault type.

[0098] The spatio-temporal feature encoder uses a bidirectional LSTM structure to encode time series features. The bidirectional LSTM can capture both the past and future information of the signal, effectively extracting the long-term dependencies in the time series. For the fault diagnosis of electrical equipment, many fault features need to be accurately identified by analyzing the signal changes over a period of time. For example, by encoding the current signals at consecutive moments, the bidirectional LSTM can analyze the trend of current changes and determine whether there are gradually developing faults, such as a slow increase in current caused by insulation aging. Through the processing of the bidirectional LSTM, the network can better understand the evolution law of the signal in the time dimension and improve the ability to capture fault features.

[0099] The multi-granularity classifier includes parallel convolutional neural network branches and fully connected classification branches, where the outputs of each branch are fused through an adaptive weighting strategy to generate the final fault type identification. The convolutional neural network branch is good at extracting local features of the signal, such as specific frequency components in vibration signals. It slides the convolutional kernel over the signal to extract representative local feature patterns. The fully connected classification branch can comprehensively judge based on global features, integrating and classifying all the extracted features. The adaptive weighting strategy dynamically adjusts the output weights according to the performance of different branches in different situations. For example, in some fault types, local features are more critical, and the weight of the convolutional neural network branch will increase accordingly; in other fault types, global features play a dominant role, and the weight of the fully connected classification branch will be larger. Through this fusion method, the multi-granularity classifier can make full use of the advantages of different types of features to generate more reliable fault type identifications.

[0100] The online update module is an important part of the dynamic adaptive diagnosis network. When the confidence coefficients of N consecutive signal samples are detected to be lower than the preset threshold, it indicates that the current diagnostic performance of the network may decline and parameter update is required. At this time, the online update module automatically triggers the incremental learning training of the network parameters. Incremental learning allows the network to update parameters in real time according to newly emerging fault data without forgetting the knowledge learned before. For example, when a new fault mode appears in the device, through incremental learning, the network can incorporate this new fault feature into its knowledge system, thus adapting to the changes in the device operating state and maintaining the stability and advancement of diagnostic performance.

[0101] During the training and operation of the entire dynamic adaptive diagnosis network, each module cooperates with each other to form a complete fault diagnosis system. The cross-modal feature fusion module ensures that signals of different modalities can be reasonably integrated and utilized, the spatio-temporal feature encoder captures the time-dependent relationships of signals, the multi-granularity classifier analyzes features from different perspectives and generates diagnostic results, and the online update module ensures that the network can adapt to the continuously changing device state. This design enables the dynamic adaptive diagnosis network to efficiently and accurately handle the multi-modal fault diagnosis problems of electrical equipment, providing a strong guarantee for the safe operation of the equipment.

[0102] Embodiment 5:

[0103] Based on the dynamic adaptive diagnosis network constructed in Embodiment 4, this embodiment further designs a three-level early warning mechanism to achieve hierarchical response to electrical equipment faults through comprehensive analysis of multi-modal signals. This mechanism triggers different levels of early warnings based on the comparison results of the confidence coefficients and dynamic thresholds, ensuring that corresponding treatment measures are taken at different stages of fault development.

[0104] When the confidence coefficients of the matching results of fault features in at least two modalities among vibration signals, temperature signals, and current signals simultaneously exceed the dynamic threshold, the three-level early warning mechanism is triggered. The dynamic threshold is dynamically adjusted according to the confidence distribution law of the same type of faults in the historical operation data of the device. Specifically, by statistically analyzing the distribution of confidence coefficients of each modality when a certain type of fault occurred in history, the dynamic threshold is set as:

[0105]

[0106] where, is the dynamic threshold, is the average value of the confidence coefficients of historical same-type faults, is the standard deviation, is the adjustment coefficient (usually taking 1.5 - 2.5). This formula ensures that the threshold can adapt to the changes in device characteristics and operating environments, and the thresholds for different devices or the same device under different working conditions will be automatically adjusted according to actual historical data.

[0107] The threshold adjustment step size of different modal signals is negatively correlated with their signal sampling frequencies. For vibration signals with a relatively high sampling frequency (such as 10 kHz), the threshold adjustment step size is small (such as 0.02) to ensure high sensitivity to rapidly changing signal features; for temperature signals with a relatively low sampling frequency (such as 1 Hz), the threshold adjustment step size is large (such as 0.05) to avoid frequent triggering of warnings due to slow signal changes. This design enables the system to optimize the warning trigger logic according to the characteristics of each modal signal.

[0108] The specific implementation method of the three-level warning mechanism is as follows:

[0109] First-level warning: When the triggering conditions are met, the system first displays a yellow warning sign through the device operation interface. This sign uses a prominent color and a flashing effect to ensure that the operator can notice the abnormal state of the device in a timely manner. At the same time, the system automatically generates a detailed diagnostic log, recording information such as the fault feature matching results, the confidence coefficient of each mode, the occurrence time, and the device operation parameters. The diagnostic log is stored in a structured format for easy subsequent traceability and analysis.

[0110] Second-level warning: On the basis of the first-level warning, an audible and visual alarm system is superimposed. The audible and visual alarm is installed at the device site and in the control room, emitting an alarm sound with a specific frequency and a red flashing light to attract the operator's attention. At the same time, the system automatically locks the current device operation parameters to prevent further parameter fluctuations from affecting device safety and providing stable reference data for troubleshooting. The locked parameters include key indicators such as current, voltage, temperature, and vibration amplitude, and are highlighted in a special color on the operation interface.

[0111] Third-level warning: On the basis of the second-level warning, the system synchronously pushes an encrypted alarm message to the remote monitoring platform. The alarm message is transmitted using a secure encryption protocol and contains key information such as the device unique identifier, the fault type, the severity, and the real-time operation parameters. After receiving the message, the remote monitoring platform immediately triggers SMS, email, or APP push notifications to relevant responsible persons to achieve real-time fault notification across regions.

[0112] The trigger logic of the entire warning mechanism also takes into account the factor of signal duration. To avoid false alarms caused by instantaneous interference, only when the state with a confidence coefficient exceeding the threshold lasts for more than a set time (such as 30 seconds for vibration signals and 5 minutes for temperature signals), will the corresponding level of warning be officially triggered. This design effectively improves the reliability of the warning system.

[0113] During the implementation of the early warning mechanism, the system also integrates the function of predicting the development trend of faults. By analyzing the change trend of the confidence coefficient at multiple consecutive time points, it predicts the possible development direction and severity of the faults. For example, if the confidence coefficient of a certain mode shows a continuous upward trend, it indicates that the fault may be intensifying and the processing speed needs to be accelerated. The prediction results are displayed in the form of charts on the operation interface to provide decision-making support for the operators.

[0114] The design of the three-level early warning mechanism fully considers the diversity and complexity of electrical equipment faults. Through the coordinated action of functions such as dynamic threshold adjustment, comprehensive judgment of multi-modal signals, hierarchical response, and trend prediction, this mechanism can provide timely and accurate early warning information under different fault scenarios, provide strong guarantee for equipment maintenance and fault handling, and minimize the downtime and maintenance costs of the equipment to the greatest extent.

[0115] It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprising", "including" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements not only includes those elements, but also includes other elements not expressly listed, or further includes elements inherent to such process, method, article or device.

[0116] Although the embodiments of the present invention have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention, and the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A multi-modal fault diagnosis method for electrical equipment based on dynamic adaptability, characterized in that The method is applied to an electrical equipment fault diagnosis system, and the method includes: Obtain an original multi-modal signal stream to be fault diagnosed, where the original multi-modal signal stream includes vibration signals, temperature signals, and current signals; Input the original multi-modal signal stream into a pre-trained dynamic adaptive diagnosis network to obtain a fault feature matching result set; Based on the fault feature matching result set, determine the fault feature matching result corresponding to each modal signal in the original multi-modal signal stream; The dynamic adaptive diagnosis network is a parameterized network obtained by training an initial diagnosis network to be trained with historical fault data. The historical fault data is generated from a batch of historical feature parameters respectively extracted from a batch of historical multi-modal signals, as well as fault type labels and confidence coefficients corresponding to each historical feature parameter. The fault type label reflects the abnormal state identification of a historical feature parameter, and the confidence coefficient reflects the correction factor corresponding to the abnormal state identification. The batch of historical multi-modal signals is a batch of continuously collected electrical equipment operation signals. Each historical feature parameter includes feature parameters determined by a historical multi-modal signal and its adjacent historical multi-modal signals; The fault feature matching result is used to represent the fault type identification of the multi-modal signal stream.

2. The method for multi-modal fault diagnosis of electrical equipment based on dynamic adaptability according to claim 1, wherein The method further includes: Extract time-domain feature parameters from the batch of historical multi-modal signals to obtain a first set of feature parameters, where one feature parameter in the first set of feature parameters corresponds to one historical multi-modal signal in the batch of historical multi-modal signals; Extract frequency-domain feature parameters from the batch of historical multi-modal signals to obtain a second set of feature parameters, where the first feature parameter in the second set of feature parameters corresponds to the first historical multi-modal signal in the historical multi-modal signals. The adjacent historical multi-modal signal of the first historical multi-modal signal is the second historical multi-modal signal, and the first feature parameter represents the frequency-domain correlation feature between the first historical multi-modal signal and the second historical multi-modal signal; Fuse the first set of feature parameters and the second set of feature parameters respectively to obtain the batch of historical feature parameters.

3. The multi-modal fault diagnosis method for electrical equipment based on dynamic adaptability according to claim 2, wherein The extracting frequency-domain feature parameters from the batch of historical multi-modal signals to obtain a second set of feature parameters includes: Extract frequency-domain feature parameters from the batch of historical multi-modal signals through the following steps to obtain a second set of feature parameters. Each time the historical multi-modal signal for frequency-domain feature parameter extraction is used as the current historical multi-modal signal, and the obtained feature parameter is used as the current feature parameter. The second set of feature parameters includes the current feature parameter; Perform a spectrum decomposition operation on the current historical multi-modal signal to determine a first set of frequency-domain feature quantities, where the first set of frequency-domain feature quantities is used to describe the main frequency band energy distribution in the current historical multi-modal signal; Perform a spectrum decomposition operation on the adjacent historical multimodal signals of the current historical multimodal signal to determine a second set of frequency-domain feature quantities, where the second set of frequency-domain feature quantities is used to describe the main frequency band energy distribution in the adjacent historical multimodal signals; Determine the current feature parameter based on the first set of frequency-domain feature quantities and the second set of frequency-domain feature quantities, where the current feature parameter is used to represent the energy deviation amount of the corresponding frequency bands in the first set of frequency-domain feature quantities and the second set of frequency-domain feature quantities.

4. The multimodal fault diagnosis method for electrical equipment based on dynamic adaptability according to claim 1, characterized in that The method further includes: Determine the fault type label and confidence coefficient corresponding to each historical feature parameter in the batch of historical feature parameters through the following steps. Each time, the historical feature parameter for determining the fault type label and confidence coefficient is used as the current historical feature parameter, and the fault type label corresponding to the current historical feature parameter is used as the current fault type label and current confidence coefficient: Input the current historical feature parameter into the initial diagnosis network to obtain the current fault type label, where the initial diagnosis network is a parameterized network obtained by performing initialization processing in advance; Determine the current confidence coefficient based on the current fault type label and the reference verification identifier corresponding to the current historical feature parameter. The reference verification identifier carries the calibration label in the historical multimodal signal corresponding to the current historical feature parameter, and the calibration label includes the target fault type identifier of the multimodal signal stream. The confidence coefficient is used to indicate whether the current fault type label is consistent with the target fault type identifier.

5. The multi-modal fault diagnosis method for electrical equipment based on dynamic adaptability according to claim 4, wherein The inputting the current historical feature parameter into the initial diagnosis network to obtain the current fault type label includes: Input each fault type identifier in the pre-stored fault type identifier set into the first classification algorithm branch in sequence together with the current historical feature parameter to obtain a first set of classification confidence indicators, where the first set of classification confidence indicators includes the confidence indicators corresponding to each fault type identifier; Determine the candidate fault type identifier corresponding to the highest first confidence indicator value in the first set of classification confidence indicators as the current fault type identifier corresponding to the current fault type label.

6. The method for multimodal fault diagnosis of electrical equipment based on dynamic adaption according to claim 5, wherein Before inputting each fault type identifier in the pre-stored fault type identifier set into the first classification algorithm branch in sequence together with the current historical feature parameter to obtain a first set of classification confidence indicators, the method further includes: Determine the signal stability index through a preset mode; Determine whether the signal stability index meets the set stability threshold; Based on the signal stability index meeting the set stability threshold, dynamically select the candidate fault type identifier in the pre-stored fault type identifier set as the current fault type label; Based on the signal stability index not meeting the set stability threshold, input the current historical feature parameter into the initial diagnosis network to obtain the current fault type label.

7. The method for multi-modal fault diagnosis of electrical equipment based on dynamic adaption according to claim 5, wherein Determining the current fault type identifier corresponding to the current fault type label by identifying the candidate fault type identifier corresponding to the first confidence index with the highest value among the first set of classification confidence indices includes: On the basis that the candidate fault type identifier corresponding to the first confidence index is the same as the target fault type identifier, determining the current confidence coefficient as the first confidence coefficient, where the first confidence coefficient indicates that the initial diagnostic network successfully generates a matching identifier; On the basis that the candidate fault type identifier corresponding to the first confidence index is different from the target fault type identifier, determining the current confidence coefficient as the second confidence coefficient, where the second confidence coefficient indicates that the initial diagnostic network fails to generate a matching identifier.

8. The multi-modal fault diagnosis method for electrical equipment based on dynamic adaptability according to claim 1, characterized in that, The method further includes: Screening several groups of historical combined fault records from the historical fault data, where the k-th historical combined fault record in the several groups of historical combined fault records includes the k-th historical feature parameter, the k-th fault type label corresponding to the k-th historical feature parameter, the k-th confidence coefficient, and the (k + 1)-th historical feature parameter, and k is a positive integer; Training the initial diagnostic network to be trained based on the several groups of historical combined fault records to obtain the dynamic adaptive diagnostic network, where when the number of training iterations for training the initial diagnostic network reaches the set training threshold, determining the initial diagnostic network as the dynamic adaptive diagnostic network, and when the number of training iterations for training the initial diagnostic network does not reach the set training threshold, updating the network parameters of the initial diagnostic network according to a preset error function, and the input of each round of training process is a group of historical combined fault records in the several groups of historical combined fault records.

9. The method for multi-modal fault diagnosis of electrical equipment based on dynamic adaptability according to claim 8, wherein, The dynamic adaptive diagnostic network includes: A cross-modal feature fusion module that dynamically calculates the modal weight coefficients of vibration signals, temperature signals, and current signals using an attention mechanism; A spatio-temporal feature encoder that encodes time series features using a bidirectional LSTM structure; A multi-granularity classifier that includes parallel convolutional neural network branches and fully connected classification branches, where the outputs of each branch are fused through an adaptive weighting strategy to generate the final fault type identifier; An online update module that automatically triggers incremental learning training of network parameters when it is detected that the confidence coefficients of N consecutive signal samples are lower than a preset threshold.

10. The method for multi-modal fault diagnosis of electrical equipment based on dynamic adaptability according to claim 1, wherein The method further includes: When the confidence coefficients of the fault feature matching results of at least two modalities among the vibration signal, temperature signal, and current signal simultaneously exceed the dynamic threshold, triggering a three-level early warning mechanism, where the first-level early warning displays a yellow warning identifier through the device operation interface and generates a diagnostic log, the second-level early warning superimposes an audible and visual alarm and automatically locks the device operation parameters, and the third-level early warning synchronously pushes an encrypted alarm message to the remote monitoring platform; The dynamic threshold is dynamically adjusted according to the confidence distribution law of similar faults in the device historical operation data, and the threshold adjustment step size of different modality signals is negatively correlated with their signal sampling frequencies.

Images