General generator electrical monitoring system with fault self-diagnosis function

Abstract

The invention relates to the technical field of electrical monitoring, provides a general generator electrical monitoring system with a fault self-diagnosis function, and aims to deeply excavate potential correlation between electrical and mechanical parameters by judging a coherence coefficient and a phase difference between a current harmonic component and a bearing vibration frequency band and marking a fault coupling identifier by using a coupling mode library. Electrical and mechanical coupling faults can be accurately identified, and the identification capability of complex faults can be greatly improved; meanwhile, the diagnosis threshold is dynamically adjusted in combination with the load rate and the winding temperature, so that the system can better adapt to different operation conditions of the generator, and the diagnosis accuracy and reliability are improved; the fault causal chain is analyzed through the causal inference algorithm, and the fault source is positioned, so that compared with the existing fault tracing mode lacking systematicness, the fault generation reason and process can be analyzed more comprehensively and deeply, the fault source can be positioned quickly and accurately, operation and maintenance personnel can take targeted measures in time, and the fault tracing efficiency is improved. And the operation safety and reliability of the generator are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of electrical monitoring, and particularly relates to a general generator electrical monitoring system with a fault self-diagnosis function. BACKGROUND

[0002] In modern power systems, as a core device, the operation stability of the general generator is crucial to the safety of the power grid. With the rapid development of the power industry, the existing generator electrical monitoring technology is also constantly evolving. Early monitoring systems mostly adopt a single parameter monitoring method, such as monitoring only electrical parameters such as current and voltage, or mechanical parameters such as vibration and temperature. This method cannot comprehensively reflect the operating state of the generator and is difficult to discover potential correlations between parameters, resulting in low accuracy of fault diagnosis.

[0003] With the advancement of sensor technology and data processing technology, multi-parameter monitoring systems have gradually become mainstream. These systems can simultaneously obtain multiple operating parameters of the generator and analyze and judge the data through simple threshold comparison or empirical formulas. However, the existing technology still has many shortcomings. On the one hand, traditional systems lack in-depth exploration of the coupling relationship between electrical and mechanical parameters when analyzing them, and cannot accurately identify complex faults caused by electrical and mechanical coupling. For example, when current harmonic abnormalities and bearing vibration abnormalities occur simultaneously, it is difficult to determine the correlation and fault root cause of the two. On the other hand, the threshold values are mostly fixed or adjusted only according to a single factor, which cannot adapt to changes in the generator under different load rates and winding temperatures, and is prone to misjudgment or omission. The existing technology does not solve the problem of how to analyze the coupling between electrical and mechanical parameters of the motor to achieve fault separation and fault tracing. SUMMARY

[0004] In view of the shortcomings of the prior art, the present application provides a general generator electrical monitoring system with a fault self-diagnosis function, which comprises a coupling perception module, a fault diagnosis module and a fault tracing module.

[0005] The coupling perception module is used to obtain current harmonic amplitude, bearing vibration speed, load rate and winding temperature in real time, synchronously extract current harmonic components and bearing vibration frequency bands, and respond to the coherence coefficient and phase difference of the current harmonic components and the bearing vibration frequency bands to mark a fault coupling identifier through a coupling mode library.

[0006] The fault diagnosis module is used to extract a diagnosis threshold, combine the load rate and the winding temperature after receiving the fault coupling identifier, dynamically adjust the diagnosis threshold, filter the current harmonic amplitude and the bearing vibration speed greater than the diagnosis threshold in the adjustment process, separate electrical faults and mechanical faults based on a double-channel network, simultaneously reconstruct target current harmonics and target bearing vibrations through the double-channel network, and respond to the comparison with the current harmonic amplitude and the bearing vibration speed to generate a fault contribution matrix.

[0007] The fault tracing module is configured to receive a fault contribution matrix, analyze a fault causal chain and locate a fault source through a causal inference algorithm, and respond to updating a diagnosis threshold and a coupling mode library.

[0008] As an optional implementation, the fault coupling identification marking logic comprises:

[0009] A coherence coefficient threshold is configured, associated frequency points with a coherence coefficient greater than the coherence coefficient threshold are screened, and a phase difference between a current harmonic component and a bearing vibration frequency band at the associated frequency points is calculated. A clustering algorithm is used to divide multiple phase difference intervals according to the size of the phase difference, and each phase difference interval represents a fault coupling mode;

[0010] A coupling mode library is established, the phase difference interval is corresponded to the fault coupling mode, and a fault coupling identification is assigned to each fault coupling mode to form an identification mapping table;

[0011] According to the phase difference interval to which the current phase difference belongs, the fault coupling identification is found in the identification mapping table, and the associated frequency points are marked.

[0012] As an optional implementation, the coherence coefficient judgment sub-logic comprises:

[0013] Real-time acquisition of current harmonic amplitude and bearing vibration speed, synchronous extraction of current harmonic component and bearing vibration frequency band;

[0014] The current harmonic component and the bearing vibration frequency band are divided into data segments according to time intervals, the dimension features of the current harmonic component and the bearing vibration frequency band in each data segment are extracted by fast Fourier transform, and the dimension features are processed by principal component analysis for dimension reduction;

[0015] Based on the dimension-reduced dimension features, the coherence coefficient of the current harmonic component and the bearing vibration frequency band at each frequency point is calculated, a coherence coefficient sequence is formed, and the coherence coefficient sequence is detected for abnormal values by a clustering algorithm.

[0016] As an optional implementation, the fault contribution matrix generation logic comprises:

[0017] Based on the attention mechanism, the contribution weights of the electrical fault feature and the mechanical fault feature to fault diagnosis are determined;

[0018] The difference between the target current harmonic and the current harmonic amplitude and the difference between the target bearing vibration and the bearing vibration speed are combined with the contribution weights to determine the fault contribution degrees of the electrical fault feature and the mechanical fault feature to fault diagnosis;

[0019] The fault contribution degrees of the electrical fault feature and the mechanical fault feature are arranged in order to generate a fault contribution matrix.

[0020] As an optional implementation, the adjustment sub-logic of the diagnosis threshold comprises:

[0021] Obtain historical operation data of the generator, and respectively count data distribution of the current harmonic amplitude and the bearing vibration speed under normal operation state as initial diagnosis threshold;

[0022] After receiving the fault coupling identifier, obtain the fault coupling mode corresponding to the fault coupling identifier from the coupling mode library, and count the change trend of the current harmonic amplitude and the bearing vibration speed under the fault coupling mode to determine the adjustment weight of the initial diagnosis threshold;

[0023] Real-time obtain the load rate and winding temperature, and analyze the influence factors of the load rate and winding temperature on the current harmonic amplitude and the bearing vibration speed through the support vector regression algorithm;

[0024] Obtain the adjustment coefficient by weighted summation of the adjustment weight of the initial diagnosis threshold and the influence factors, and adjust the diagnosis threshold in combination with the initial diagnosis threshold.

[0025] As an optional implementation, the separation sub-logic of the electrical fault and the mechanical fault comprises:

[0026] Screen out data points with current harmonic amplitude greater than the diagnosis threshold, and mark them as electrical candidate fault points, and screen out data points with bearing vibration speed greater than the diagnosis threshold, and mark them as mechanical candidate fault points;

[0027] Build a dual-channel network including a physical channel and a data channel, the physical channel performs feature enhancement on the electrical candidate fault points and the mechanical candidate fault points to obtain electrical fault features and mechanical fault features;

[0028] The data channel extracts the associated features of the current harmonic amplitude and the bearing vibration speed under the fault coupling mode from the coupling mode library based on the fault coupling identifier, and combines the load rate and the winding temperature to real-time correct the associated features to generate a fault feature template;

[0029] Calculate the cosine similarity of the electrical fault features, the mechanical fault features and the fault feature template respectively to determine the electrical fault probability and the mechanical fault probability, and configure a decision threshold to separate the electrical fault and the mechanical fault.

[0030] As an optional implementation, the reconstruction sub-logic of the target current harmonic and the target bearing vibration comprises:

[0031] Real-time obtain the current harmonic amplitude and the bearing vibration speed of the generator in the physical channel;

[0032] The historical current harmonic amplitude and the historical bearing vibration speed of the generator in the normal operation state are called in the data channel, and are classified and stored according to the combination of the load rate and the winding temperature to form a feature mapping relationship;

[0033] The current harmonic amplitude and the bearing vibration speed obtained in the physical channel are matched in the feature mapping relationship in combination with the load rate and the winding temperature to screen normal working condition features, and the normal working condition features are fed back to the physical channel to reconstruct target current harmonics and target bearing vibrations;

[0034] The reconstructed target current harmonics and target bearing vibrations are transmitted to the physical channel for error verification to determine whether to re-screen normal working condition features and reconstruct target current harmonics and target bearing vibrations.

[0035] As an optional implementation, the update logic of the diagnostic threshold value includes:

[0036] The change trend of the current harmonic amplitude and the bearing vibration speed at the fault source is analyzed, and the threshold offset of the fault source causing the change of the current harmonic amplitude and the bearing vibration speed is determined;

[0037] Through a reinforcement learning algorithm, the false positive rate and the false negative rate are taken as a reward function, and the diagnostic threshold value is updated in combination with the threshold offset;

[0038] The effectiveness of the updated diagnostic threshold value is evaluated by a cross-validation method to determine whether to update the diagnostic threshold value again.

[0039] As an optional implementation, the positioning sub-logic of the fault source includes:

[0040] A fault contribution matrix is received, and a causal inference algorithm is used to automatically search for the causal relationship among the current harmonic amplitude, the bearing vibration speed, the load rate and the winding temperature to generate a causal network;

[0041] The causal network is converted into a Bayesian network, each node represents a fault feature, and each edge represents a causal relationship, and the prior probability distribution of each node is set according to the fault contribution matrix;

[0042] The posterior probability of each fault feature as a fault source is determined by Bayesian inference to screen out potential fault sources, and the trust degree of each potential fault source is determined to locate the fault source.

[0043] As an optional implementation, the update logic of the coupling mode library includes:

[0044] The abnormal data of the unmatched fault coupling mode is clustered by a DBSCAN algorithm, and the clustering area of the new fault coupling mode is identified according to the data density;

[0045] The mapping relationship between the phase difference interval division and the fault coupling identifier in the new fault coupling mode is dynamically adjusted by a gradient descent algorithm to update the coupling mode library.

[0046] Compared with the prior art, the beneficial effects of the application are: by judging the coherence coefficient and phase difference of the current harmonic component and the bearing vibration frequency band, and using the coupling mode library to mark the fault coupling identifier, the potential correlation between electrical and mechanical parameters can be deeply mined, and electrical and mechanical coupling faults can be accurately identified, compared with the traditional single parameter or simple correlation analysis method, the identification ability of complex faults is greatly improved, and fault omission is avoided; at the same time, the diagnosis threshold is dynamically adjusted combined with the load rate and the winding temperature, the limitations of traditional fixed threshold or single factor adjustment are overcome, the system can better adapt to different operating conditions of the generator, effectively reduces the misjudgment rate, improves the accuracy and reliability of diagnosis; and the fault cause chain is analyzed and the fault source is located through the causal inference algorithm, compared with the existing fault tracing method which lacks systematicness, the causes and processes of the fault can be more comprehensively and deeply analyzed, the fault source can be quickly and accurately located, the operation and maintenance personnel can take targeted measures in time, the fault processing time is shortened, the downtime loss is reduced, and the safety and reliability of the generator operation are improved. BRIEF DESCRIPTION OF DRAWINGS

[0047] In order to more clearly illustrate the technical solutions of the embodiments of the application, the drawings needed to be used in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the application, and other drawings can be obtained by those skilled in the art without creative labor. Among them:

[0048] Figure 1 The system flowchart of the general generator electrical monitoring system with fault self-diagnosis function provided by the embodiments of the application;

[0049] Figure 2 The marking logic diagram of the fault coupling identifier of the general generator electrical monitoring system with fault self-diagnosis function provided by the embodiments of the application;

[0050] Figure 3 The separation sub-logic diagram of electrical faults and mechanical faults of the general generator electrical monitoring system with fault self-diagnosis function provided by the embodiments of the application;

[0051] Figure 4 The positioning sub-logic diagram of the fault source of the general generator electrical monitoring system with fault self-diagnosis function provided by the embodiments of the application. DETAILED DESCRIPTION

[0052] In order to make the objectives, technical solutions and advantages of the embodiments of the present application more obvious and easy to understand, the technical solutions in the embodiments of the present application are clearly and completely described below in conjunction with the drawings in the specification. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments.

[0053] Example

[0054] like Figure 1 As shown, a system flow chart of a universal generator electrical monitoring system with fault self-diagnosis function is provided for an embodiment of the present application. The system includes a coupling perception module, a fault diagnosis module and a fault tracing module.

[0055] The coupling perception module is used to obtain the current harmonic amplitude, bearing vibration speed, load rate and winding temperature in real time, and synchronously extract the current harmonic components and bearing vibration frequency bands. In response, the coherence coefficient and phase difference between the current harmonic components and the bearing vibration frequency band are judged to mark the fault coupling identifier through the coupling mode library.

[0056] Furthermore, the judgment sub-logic of the coherence coefficient includes:

[0057] Acquire the current harmonic amplitude and bearing vibration velocity in real time, and simultaneously extract the current harmonic components and bearing vibration frequency bands;

[0058] The current harmonic components and bearing vibration frequency bands are divided into data segments according to time intervals. The dimensional features of the current harmonic components and bearing vibration frequency bands in each data segment are extracted by fast Fourier transform, and the dimensional features are reduced by principal component analysis.

[0059] Based on the dimensional features after dimensionality reduction, the coherence coefficient between the current harmonic components and the bearing vibration frequency band at each frequency point is calculated to form a coherence coefficient sequence, and the outlier detection of the coherence coefficient sequence is performed using a clustering algorithm.

[0060] When the generator is running, the faults of the electrical system and the mechanical system can be shown through the abnormal coupling of current harmonics and bearing vibration, and the two types of signals need to be acquired in real time and synchronously to capture the time-frequency domain correlation characteristics in the early stage of the fault; the electrical signal and the vibration signal are acquired through the current sensor and the vibration sensor to monitor the current harmonic amplitude and the bearing vibration speed, and the sampling time of the current sensor and the vibration sensor is uniformly calibrated to ensure that there is no timestamp error between the two types of signals, then the current signal is processed through band-pass filtering to retain the harmonic frequency band related to the fault, including the integer multiple frequency of the fundamental frequency, and the vibration signal is processed through wavelet packet decomposition to extract the sensitive frequency band of the bearing fault characteristics, and the frequency domain separation of the signals is realized through the filter bank to extract the current harmonic component and the bearing vibration frequency band; to ensure the strict synchronization of the electrical signal and the vibration signal in the time domain and the highlighting of the fault characteristics in the frequency domain, and to provide the original data with consistent timing and clear characteristics for subsequent coherence analysis, the signals processed synchronously provide standardized input for data segmentation and feature dimension reduction, and ensure that the subsequent dimension feature extraction can accurately reflect the coupling relationship between the signals.

[0061] The original signal has time continuity, resulting in a large amount of data, and there is correlation between the multi-dimensional features, which needs to be processed through segmentation and dimensionality reduction to eliminate redundant information and improve computational efficiency; the signal is segmented through the sliding window mechanism to obtain multiple data segments, the window length is set according to the time-varying characteristics of the fault characteristics, and the window overlap rate is set to 50% to ensure data continuity, the time-frequency transformation of each data segment is performed through fast Fourier transform to extract the amplitude characteristics of the current harmonic component at each frequency point and the energy distribution characteristics of the bearing vibration frequency band to extract the dimensional features, then the covariance matrix of the dimensional features is calculated through principal component analysis, and the principal components are selected according to the cumulative variance contribution rate to compress the high-dimensional features to a low-dimensional space and eliminate the linear correlation between the dimensional features; thereby the continuous signal is converted into discrete data segments, the computational complexity is reduced, the key fault sensitive features are retained through dimensionality reduction, and the interference of the dimension disaster on the subsequent analysis is avoided.

[0062] The coherence coefficient can quantify the coupling strength of the current signal and the vibration signal at a specific frequency point, and the abnormal coherence coefficient indicates the coupling state of early failure, and the atypical coupling mode is identified through the clustering algorithm; based on the dimension-reduced feature vector, the cross power spectral density of each frequency point is calculated through the Welch method, the coherence coefficient is obtained through the ratio relationship between the cross power spectrum and the self-power spectrum, and a sequence reflecting the coupling strength of each frequency point is formed to form a coherence coefficient sequence, the coherence coefficient sequence is analyzed through the DBSCAN clustering algorithm, and the data points are divided into core clusters and abnormal points by setting the neighborhood radius and the minimum sample number parameters, wherein the data points in the core cluster satisfy the high-density distribution condition and represent the coupling mode of normal operation or typical failure, and the data points deviating from the core cluster are determined as abnormal values, corresponding to the coupling characteristics of non-steady state or early failure; thereby realizing the quantitative analysis of the coupling strength of the current and vibration signals, identifying the abnormal coupling characteristics of potential failure through a data-driven manner, providing a basis for fault mode classification, screening the abnormal coherence coefficient frequency points and corresponding characteristics to provide a candidate set for the marking of the fault coupling identifier, and narrowing the range of subsequent phase difference analysis.

[0063] Specifically, as shown in Figure 2 the marking logic of the fault coupling identifier includes:

[0064] The coherence coefficient threshold is configured, the associated frequency points with the coherence coefficient greater than the coherence coefficient threshold are screened, and the phase difference between the current harmonic component and the bearing vibration frequency band at the associated frequency point is calculated, a plurality of phase difference intervals are divided according to the size of the phase difference through the clustering algorithm, and each phase difference interval represents a fault coupling mode;

[0065] A coupling mode library is established, the phase difference interval is corresponded to the fault coupling mode, and the fault coupling identifier is assigned to each fault coupling mode to form an identifier mapping table;

[0066] According to the phase difference interval to which the current phase difference belongs, the fault coupling identifier is found in the identifier mapping table, and the associated frequency point is marked.

[0067] The coherence coefficient represents the signal coupling strength, and the phase difference reflects the phase relationship between signals, and the combination of the two can uniquely determine the fault coupling mode, which needs to be screened by threshold and divided into typical phase difference intervals; The coherence coefficient threshold is determined by statistical analysis of historical fault data, and the frequency points with coherence coefficient greater than the coherence coefficient threshold are selected as the correlation frequency points to ensure that the analysis object is the strong coupling characteristic frequency, and then the phase difference between the current signal and the vibration signal is calculated at the correlation frequency point, the instantaneous phase is obtained by the phase correlation algorithm, the random noise is eliminated by time average processing, then the K-means clustering algorithm is used to classify the phase difference of the correlation frequency point, the number of clusters is determined according to the prior knowledge of the fault coupling mode, and the continuous phase difference space is divided into several discrete phase difference intervals according to the size of the phase difference, each phase difference interval corresponds to a typical fault coupling mode; The threshold screening focuses on strong coupling characteristics, and the phase difference clustering discretizes the continuous characteristic space, reduces the complexity of fault coupling mode recognition, lays the foundation for standardized description of fault coupling mode, and provides classification units for the establishment of coupling mode library, ensuring one-to-one correspondence between fault coupling mode and characteristic parameters.

[0068] The abstract phase difference interval and the fault coupling mode need to be standardized to establish a mapping relationship, and the mode is quickly searched and called through a digital identifier; A database structure is constructed to store the description of the phase difference interval, the fault coupling mode and the corresponding fault coupling identifier, a key-value pair form is used to establish a mapping relationship, and a unique identifier is assigned to each fault coupling mode, the identifier uses a coding rule, i.e. a letter prefix plus a number sequence, to ensure the uniqueness and scalability of the fault coupling identifier, and the correspondence between the fault coupling identifier and the fault coupling mode is maintained through a data dictionary, supporting dynamic updating; Thus, a standardized coupling mode library is formed, realizing the fast mapping from signal characteristics to fault coupling mode, providing prior knowledge support for the fault diagnosis module, improving the diagnosis efficiency, and the identifier mapping table provides a query basis for real-time phase difference matching, which is a key data interface connecting signal characteristics and fault diagnosis.

[0069] During the operation of the generator, the current signal characteristics need to be converted into fault coupling identifiers in real time to provide direct input for subsequent fault diagnosis, realize real-time warning of faults, and determine the phase difference interval to which the real-time calculated phase difference belongs through binary search, query the corresponding fault coupling identifier in the identifier mapping table, bind the fault coupling identifier with the correlation frequency point, and transmit the marked results to the fault diagnosis module in real time through a data bus or a message queue to ensure the timeliness of identifier transmission; Thus, the real-time digital marking of fault coupling characteristics is realized, the fault diagnosis module is provided with an identifier-bearing characteristic input, and the marked fault coupling identifier directly drives the threshold adjustment and fault separation of the fault diagnosis module, which is a key interaction node between the perception layer and the diagnosis layer.

[0070] The fault diagnosis module is used to extract a diagnosis threshold, combine a load rate and a winding temperature after receiving a fault coupling identifier, dynamically adjust the diagnosis threshold, screen current harmonic amplitudes and bearing vibration speeds greater than the diagnosis threshold in the adjustment process, and separate electrical faults and mechanical faults based on a dual-channel network, while respectively reconstructing target current harmonics and target bearing vibrations through the dual-channel network to generate a fault contribution matrix in response to comparison with the current harmonic amplitudes and the bearing vibration speeds.

[0071] Further, the adjustment sub-logic of the diagnosis threshold comprises:

[0072] Historical operation data of the generator are acquired, and data distributions of the current harmonic amplitudes and the bearing vibration speeds in the normal operation state are respectively counted to serve as initial diagnosis thresholds;

[0073] After receiving the fault coupling identifier, a fault coupling mode corresponding to the fault coupling identifier is acquired from a coupling mode library, and variation trends of the current harmonic amplitudes and the bearing vibration speeds under the fault coupling mode are counted to determine adjustment weights of the initial diagnosis thresholds;

[0074] The load rate and the winding temperature are acquired in real time, and support vector regression algorithms are used to analyze influence factors of the load rate and the winding temperature on the current harmonic amplitudes and the bearing vibration speeds;

[0075] The adjustment weights of the initial diagnosis thresholds and the influence factors are summed by weighting to obtain adjustment coefficients, and the initial diagnosis thresholds are combined to adjust the diagnosis threshold.

[0076] To distinguish between the normal operation state and the fault state, initial diagnosis thresholds need to be established according to historical operation data of the generator, the initial diagnosis thresholds serve as a basis for subsequent dynamic adjustment, and the initial diagnosis thresholds ensure reliability of initial judgment in a statistical sense; operation data of the generator in the normal operation state are called from a database, including the current harmonic amplitudes and the bearing vibration speeds under different load rates and winding temperatures, the current harmonic amplitudes and the bearing vibration speeds are cleaned, and outliers generated by sensor abnormalities or short-time interference are removed, and the cleaned data are analyzed by a probability statistical method, including fitting a probability distribution curve of the data, determining the initial diagnosis thresholds based on a confidence principle, and making most data points in the normal operation fall within a range of the initial diagnosis thresholds; thus, the initial diagnosis thresholds conform to actual operation characteristics of the generator, provide reasonable references for subsequent dynamic adjustment, reduce a misjudgment probability in the initial state, and provide baseline data for calculation of subsequent adjustment weights and analysis of working condition influence factors, and ensure that the dynamic adjustment process has a clear starting point and a reference standard.

[0077] The influence degree of different fault coupling modes on current harmonic amplitude and bearing vibration speed is different, and the adjustment weight of the initial diagnosis threshold value needs to be determined according to the historical fault data statistical results of the fault coupling mode, so that the initial diagnosis threshold value can be adaptively offset according to the characteristic change of the specific fault coupling mode; when the system receives the fault coupling identifier, the historical fault data corresponding to the fault coupling identifier is retrieved from the coupling mode library, including the change records of current harmonic amplitude and bearing vibration speed when the same type of fault occurs, the time sequence analysis is performed on these historical fault data, the characteristic parameters including amplitude change rate and trend slope are extracted, and the average change amplitude of current harmonic amplitude and bearing vibration speed under the fault coupling mode is calculated through statistical method, which is compared with the initial diagnosis threshold value to determine the corresponding adjustment proportion, which is used as the adjustment weight; so that the initial diagnosis threshold value can be adjusted according to the characteristic change of the specific fault coupling mode, avoiding the false alarm or false alarm problem caused by the unified threshold, improving the sensitivity of the initial diagnosis threshold value to the target fault, and the adjustment weight provides the fault coupling mode dimension parameter for the threshold value calculation of the subsequent combined influence factor, ensuring that the diagnosis threshold value adjustment considers the comprehensive action of fault type and real-time working condition.

[0078] The load rate and winding temperature of the generator will directly affect the normal fluctuation range of the electrical signal and the vibration signal, and the influence degree of the working condition parameters on the signal characteristics needs to be quantified to realize the adaptive adjustment of the diagnosis threshold value with the change of working condition, avoiding the misdiagnosis caused by the fluctuation of working condition; the working condition parameters of load rate and winding temperature are obtained in real time through sensors, and the current harmonic amplitude and bearing vibration speed at the corresponding time are recorded, a historical data set containing working condition parameters and signal characteristics is constructed, the historical data set is trained through support vector machine, and the mapping relationship between working condition parameters and current harmonic amplitude and bearing vibration speed is constructed, and the influence factor of each working condition parameter on current harmonic amplitude and bearing vibration speed is obtained through the mapping relationship, which reflects the quantitative influence degree of working condition change on current harmonic amplitude and bearing vibration speed; so as to clarify the influence law of running working condition on current harmonic amplitude and bearing vibration speed, so that the diagnosis threshold value can be dynamically corrected with the change of working condition, that is, the threshold value is appropriately increased according to the influence factor under high load, avoiding the misjudgment of signal fluctuation under normal working condition as fault.

[0079] The adjustment weight and the influence factor of the diagnosis threshold are combined, a final adjustment coefficient is obtained by weighted calculation, the real-time dynamic update of the diagnosis threshold is realized, and the diagnosis threshold can adapt to the real-time change of the generator operating state. The adjustment weight and the influence factor of the initial diagnosis threshold are combined, the adjustment coefficient is obtained by weighted summation, and the diagnosis threshold is adjusted in combination with the initial diagnosis threshold. The diagnosis threshold is adjusted through a corresponding transformation method, and the transformation method includes linear transformation or nonlinear transformation based on historical data. Therefore, the diagnosis threshold can respond to the characteristics of the fault coupling mode and the operating condition change at the same time, the dynamic adaptability of the diagnosis threshold is improved, the accuracy of fault detection can be maintained under different loads and temperature conditions, and the updated diagnosis threshold is used to screen abnormal features in the current harmonic amplitude and the bearing vibration speed, so as to provide accurate judgment criteria for subsequent separation of electrical faults and mechanical faults.

[0080] Further, as shown in Figure 3 The separation of electrical faults and mechanical faults includes:

[0081] The data points with current harmonic amplitudes greater than the diagnosis threshold are screened out and marked as electrical candidate fault points, and the data points with bearing vibration speeds greater than the diagnosis threshold are screened out and marked as mechanical candidate fault points.

[0082] A dual-channel network including a physical channel and a data channel is constructed, the physical channel performs feature enhancement on the electrical candidate fault points and the mechanical candidate fault points to obtain electrical fault features and mechanical fault features;

[0083] The data channel extracts the correlation features of the current harmonic amplitude and the bearing vibration speed under the fault coupling mode from the coupling mode library based on the fault coupling identification, and combines the load rate and the winding temperature to correct the correlation features in real time, and generates a fault feature template;

[0084] The cosine similarity of the electrical fault features and the mechanical fault features and the fault feature template is calculated to determine the electrical fault probability and the mechanical fault probability, and a decision threshold is configured to separate the electrical fault and the mechanical fault.

[0085] Based on the dynamically updated diagnostic threshold, abnormal points in current harmonic amplitude and bearing vibration speed are identified to provide candidate objects for subsequent fault separation, thereby narrowing the range of fault feature analysis and improving processing efficiency. The real-time acquired current harmonic amplitude and bearing vibration speed are compared with the current diagnostic threshold to screen out data points greater than the diagnostic threshold. The data points of current harmonic amplitude greater than the diagnostic threshold are marked as electrical candidate fault points, and the data points of bearing vibration speed greater than the diagnostic threshold are marked as mechanical candidate fault points. Context information such as time stamp, load rate and winding temperature is attached to each candidate point to form a complete fault feature data record. Through threshold screening, abnormal features in the signal are quickly located to reduce the calculation amount of subsequent feature analysis, while retaining the working condition information at the time of fault occurrence to provide auxiliary judgment basis for fault classification.

[0086] The original features of the candidate fault points contain noise or redundant information, which need to be enhanced through a physical channel to extract key features that can effectively distinguish electrical faults from mechanical faults and improve the accuracy of fault separation. A physical channel containing a deep learning network is constructed, and different feature extraction paths are designed for electrical candidate fault points and mechanical candidate fault points. For electrical candidate fault points, harmonic distortion rate and characteristic frequency amplitude ratio are extracted through frequency domain analysis method, and transient harmonic abnormalities are captured through time-frequency analysis method. For mechanical candidate fault points, the energy distribution of bearing fault characteristic frequency is extracted through time-frequency domain decomposition, and early fault features are enhanced through envelope analysis to obtain electrical fault features and mechanical fault features. Thus, high-discriminability fault features are extracted from candidate fault points, noise and irrelevant features are suppressed, and electrical fault features and mechanical fault features have stronger separability in feature space to provide high-quality feature input for subsequent classification.

[0087] The prior knowledge of fault coupling mode and real-time working condition are combined to generate a fault feature template reflecting the expected fault features as a reference standard for fault separation to ensure that fault classification has a clear comparison benchmark. The data channel retrieves the historical associated features of the corresponding fault coupling mode from the coupling mode library according to the received fault coupling identification, constructs a fault feature template, and simultaneously acquires the load rate and winding temperature in real time. The fault feature template is corrected through interpolation method to compensate for the influence of current working condition changes on fault features, so that the fault feature template can reflect the expected fault features under the current working condition. Thus, a fault feature template matching the current fault coupling mode and operating condition is generated to provide a dynamic reference benchmark for fault separation, avoid classification deviation caused by working condition differences, and improve the accuracy and adaptability of fault separation.

[0088] By calculating the similarity of the fault feature and the fault feature template, the occurrence probability of the electrical fault and the mechanical fault is quantified, and the automatic separation of the fault is realized by combining the preset decision threshold, which provides clear fault type information for subsequent fault tracing; by using the cosine similarity calculation method, the similarity of the electrical fault feature and the mechanical fault feature with the fault feature template is calculated respectively, the similarity results are converted into fault probability values, and the preset decision threshold is set; when the electrical fault probability is greater than the mechanical fault probability and greater than the decision threshold, it is determined as an electrical fault, otherwise it is determined as a mechanical fault, if the fault probability of both is less than the decision threshold, it is determined as an unknown fault mode; thereby realizing the quantitative separation of the electrical fault and the mechanical fault, reducing the manual intervention, improving the automation degree and classification accuracy of the fault diagnosis, providing clear fault type input for the fault tracing module, narrowing the tracing range, and the separated fault type result is used to drive the generation of the fault contribution matrix and provide key fault type clues for the fault tracing, guiding the causal analysis direction in the tracing process.

[0089] Further, the reconstruction sub-logic of the target current harmonic and the target bearing vibration includes:

[0090] In the physical channel, the current harmonic amplitude and the bearing vibration speed of the generator are acquired in real time;

[0091] In the data channel, the historical current harmonic amplitude and the historical bearing vibration speed of the generator in the normal operation state are called, and are classified and stored according to the combination of the load rate and the winding temperature to form a feature mapping relationship;

[0092] The current harmonic amplitude and the bearing vibration speed acquired in the physical channel are matched in the feature mapping relationship combined with the load rate and the winding temperature to screen normal working condition characteristics, and the normal working condition characteristics are fed back to the physical channel to reconstruct the target current harmonic and the target bearing vibration;

[0093] The reconstructed target current harmonic and target bearing vibration are transmitted to the physical channel for error verification to determine whether to re-screen the normal working condition characteristics and reconstruct the target current harmonic and the target bearing vibration.

[0094] To ensure that the reconstruction process is closely combined with the actual operating state of the generator, the real current harmonic amplitude and bearing vibration speed at the current time need to be obtained as the reconstruction basis. Only based on real-time data, the operating characteristics of the generator under specific working conditions can be accurately reflected, and reliable basis can be provided for subsequent target data reconstruction. The current harmonic amplitude and bearing vibration speed are obtained through current sensors and vibration sensors, respectively. The obtained current harmonic amplitude and bearing vibration speed are preprocessed, including low-pass filtering to remove high-frequency noise and eliminating random interference through the sliding average method. For the preprocessed current harmonic amplitude and bearing vibration speed, time series data streams including multiple frequency points are formed. Thus, high-quality real-time data are obtained, the authenticity and integrity of the data are ensured, reliable data sources are provided for the subsequent reconstruction process, and the reconstruction error caused by data quality problems is reduced.

[0095] To accurately identify the normal operating state under complex working conditions, the feature mapping relationship between different combined working conditions and historical normal characteristics needs to be established as the reference standard for real-time data matching. The normal operating data of the generator under various typical working conditions, including historical current harmonic amplitude and historical bearing vibration speed under different load rates and winding temperature combinations, are called from the historical database. Then, the historical data are feature extracted, the statistical characteristics of the current harmonic amplitude and bearing vibration speed at each frequency point, including mean, standard deviation and skewness, are calculated, a multi-dimensional feature vector is formed, the multi-dimensional feature vector is organized based on the KD tree index structure, the load rate and winding temperature are taken as the index key, the feature mapping relationship between the combined working conditions and the historical normal characteristics is constructed, and fast working condition matching query is realized. Thus, a structured feature mapping relationship is formed, which facilitates fast retrieval and matching of the normal characteristics under the current combined working conditions, improves the reconstruction efficiency and accuracy, and provides a reliable comparison benchmark for real-time data.

[0096] By matching real-time data with feature mapping relationship, normal working condition features are screened out, and these normal working condition features are used to reconstruct target current harmonics and target bearing vibration to separate abnormal components caused by faults; real-time current harmonic amplitudes and bearing vibration speeds obtained by physical channels, combined with current load rate and winding temperature, are subjected to nearest neighbor search in the KD tree index of the data channel to find one or more groups of normal working condition features most similar to the current working condition, similarity scores of the current working condition and each group of normal working condition features are calculated, weights are distributed according to the similarity scores through a weighted fusion algorithm, the screened normal working condition features are linearly combined, inverse Fourier transform is performed on the fused normal working condition features, and target current harmonics and target bearing vibration are reconstructed, so that the target current harmonics and the target bearing vibration can reflect the normal running state under the current working condition in the frequency domain and the time domain; thus, normal running features are separated from complex running conditions, and target current harmonics and target bearing vibration reflecting the current working condition are reconstructed to provide a clear contrast benchmark for subsequent fault contribution analysis, facilitating accurate identification of abnormalities caused by faults.

[0097] The accuracy of the reconstructed target current harmonics and target bearing vibration needs to be verified to ensure that it can truly reflect the normal running state, and if the error exceeds the acceptable range, the normal working condition features need to be reselected for iterative reconstruction; error indicators between the reconstructed target current harmonics and target bearing vibration and real-time measurement values are calculated, including time domain error and frequency domain error, wherein the time domain error includes mean square error, and the frequency domain error includes frequency spectrum similarity; the error indicators are compared with the preset error threshold value, if the error is greater than the error threshold value, the screening conditions are adjusted, including expanding the working condition matching range, increasing the number of candidate normal working conditions and changing the feature fusion weight distribution, then the normal working condition features are reselected based on the adjusted screening conditions and the target current harmonics and target bearing vibration are reconstructed, the error verification process is repeated until the error meets the requirements or the maximum number of iterations is reached; the closed-loop verification mechanism ensures the reliability of the reconstructed waveform, improves the accuracy of fault feature extraction, avoids fault misjudgment caused by reconstruction error, and ensures that the reconstruction result can accurately reflect the normal running state; the verified reconstructed target current harmonics and target bearing vibration provide stable and reliable input for fault contribution matrix generation, ensuring the accuracy of subsequent fault tracing, and if the reconstructed waveform error is large, it will cause distortion of the fault contribution matrix, affecting the fault source positioning result.

[0098] Specifically, the generation logic of the fault contribution matrix includes:

[0099] Determine the contribution weights of electrical fault features and mechanical fault features to fault diagnosis based on attention mechanism;

[0100] The difference between the target current harmonic and the current harmonic amplitude and the difference between the target bearing vibration and the bearing vibration speed are combined with the contribution weight to determine the fault contribution degree of the electrical fault feature and the mechanical fault feature to the fault diagnosis.

[0101] The fault contribution degrees of the electrical fault feature and the mechanical fault feature are arranged in order to generate a fault contribution matrix.

[0102] The contribution degrees of different fault features to the final diagnosis result are different, and the importance of each fault feature needs to be quantified through the attention mechanism to provide a weight distribution basis for accurately calculating the fault contribution degree and avoid the diagnosis deviation caused by treating all fault features equally. A neural network based on the attention mechanism is constructed, the electrical fault feature and the mechanical fault feature are input, the contribution weight of the electrical fault feature and the mechanical fault feature to the fault diagnosis is calculated through the self-attention mechanism, in the training stage of the neural network, the historical fault data is used as a sample, the known fault type label is used as a supervision signal, the back propagation algorithm is used to optimize the contribution weight allocated by the attention mechanism, so that the neural network can automatically identify the fault feature with the most discriminative power to the fault diagnosis, and a regularization term is introduced to constrain the contribution weight to prevent overfitting and improve the generalization ability of the neural network, ensuring that the contribution weight of the fault feature can be accurately allocated in different fault scenarios. Thus, the adaptive weighting of the fault feature is realized, the key fault feature is highlighted, the noise feature is suppressed, the accuracy and robustness of the fault diagnosis are improved, and the diagnosis result is more focused on the real fault feature.

[0103] By comparing the differences between the reconstructed target current harmonic and the target bearing vibration and the real-time measurement values, and combining the contribution weight of the fault feature, the contribution degree of each fault feature to the overall fault is quantified to provide a quantitative basis for fault tracing, so as to accurately identify the main fault cause. The difference between the target current harmonic and the current harmonic amplitude and the difference between the target bearing vibration and the bearing vibration speed are calculated to obtain the residual vector of the original fault, the residual vector of the original fault is dot multiplied with the contribution weight determined based on the attention mechanism to obtain the fault contribution degree of each fault feature, the influence of the important fault feature is highlighted, then the fault contribution degree is normalized so that the sum is 1, which is convenient for intuitive comparison of the relative importance of different fault features, and a standardized fault contribution degree vector is formed. Thus, the difference of the fault feature is quantified as a specific contribution degree value, the influence degree of each fault feature on the overall fault is clearly reflected, a quantifiable basis is provided for fault source positioning, and the fault tracing process is more scientific and objective.

[0104] The fault contribution degrees of each fault feature are arranged according to a specific logic to form a structured matrix representation, and the correlation relationship and relative importance between the fault features are intuitively displayed, which provides systematic input for fault tracing and facilitates analysis and decision-making; the contribution degrees of the electrical fault features and the mechanical fault features are arranged according to a preset rule to form a two-dimensional matrix structure, the rows of the matrix represent different types of fault features, the columns represent different fault dimensions, including time dimension, frequency dimension and space dimension, and the matrix element values are the fault contribution degrees of the corresponding fault features in the corresponding dimensions; the correlation and transmission relationship between different fault features can be analyzed through matrix operation, the fault contribution matrix is visualized, including generating a heat map, using color depth to represent the contribution degree, enhancing the intuitive expression of the fault feature relationship, and facilitating manual analysis and understanding; thereby forming a structured and visualized fault contribution matrix, which clearly displays the relative importance and mutual relationship of each fault feature, provides systematic and hierarchical input information for the fault tracing module, and helps the diagnostic personnel to quickly locate the fault source; the fault contribution matrix as the input of the fault tracing, its structure and content directly affect the efficiency and accuracy of the causal inference algorithm, and determine the effect of fault source positioning, if the fault contribution matrix is designed unreasonably, it will lead to complex causal inference process and inaccurate results.

[0105] The fault tracing module is configured to receive the fault contribution matrix to analyze the fault causal chain and locate the fault source by a causal inference algorithm, and respond to updating the diagnostic threshold and the coupling mode library.

[0106] Further, as shown in Figure 4 The positioning of the fault source includes:

[0107] The fault contribution matrix is received, and the causal relationship of the current harmonic amplitude, the bearing vibration speed, the load rate and the winding temperature is automatically searched by a causal inference algorithm to generate a causal network;

[0108] The causal network is converted into a Bayesian network, each node represents a fault feature, and each edge represents a causal relationship, and the prior probability distribution of each node is set according to the fault contribution matrix;

[0109] The posterior probability of each fault feature as a fault source is determined by Bayesian inference to screen out potential fault sources, and the trust degree of each potential fault source is determined to locate the fault source.

[0110] Generator failure is usually caused by the interaction of multiple factors, and the causal relationship between variables such as current harmonic amplitude, bearing vibration speed, load rate and winding temperature needs to be revealed by a systematic method. The causal network can intuitively present the transmission path between the fault characteristics, provide a logical framework for locating the fault source, and avoid blind troubleshooting. The fault contribution matrix output by the fault diagnosis module is received, and the electrical and mechanical fault characteristics in the fault contribution matrix and the load rate and winding temperature are taken as nodes. The correlation strength between nodes is set based on the initial basis of fault contribution degree. The higher the fault contribution degree, the higher the priority of the connection edge. Then the PC algorithm is used to deduce the causal relationship between the nodes. The PC algorithm removes the edges with no causal relationship by conditional independence test. First, the unconditional independence between all node pairs is calculated, and the connection of independent nodes is removed. Then, an intermediate node is introduced as a conditional variable, and the independence is tested again. The process is repeated until no more edges can be removed, forming a causal network with a directed acyclic graph structure. Finally, the generated causal network is manually checked in combination with the operation principle of the generator and expert experience. If the network shows that the winding temperature change causes the load rate to be abnormal, which is not consistent with the actual physical logic, the incorrect connection is corrected to ensure that the causal network conforms to the operation rules of the generator. Thus, a causal network that conforms to the physical characteristics of the generator is constructed, clearly presenting the logical chain between the fault characteristics, reducing the uncertainty of fault troubleshooting, and improving the positioning efficiency.

[0111] The causal network only describes the logical relationship between variables and cannot quantify the possibility of fault occurrence. The Bayesian network incorporates the uncertainty of fault characteristics into the analysis through a probability model, providing data support for evaluating the possibility of each node as a fault source. The causal network is directly converted into a Bayesian network, each node corresponds to a random variable in the Bayesian network, and the directed edge represents the dependency relationship of the conditional probability. According to the fault contribution matrix, the nodes are assigned prior probabilities, where fault characteristics with high fault contribution degree are given higher prior probabilities, and the load rate and winding temperature are set based on the probability distribution of historical normal data. The current harmonic amplitude and bearing vibration speed are discretized, where the bearing vibration speed is divided into "normal", "slightly abnormal" and "severe abnormal" intervals. By statistical analysis of the frequency of each interval in historical data, a conditional probability table is constructed. Then the Bayesian network is trained through historical fault data. Through the maximum likelihood estimation method or the Bayesian estimation method, the probability parameters in the conditional probability table are adjusted to make the Bayesian network more consistent with the actual fault scenario. Thus, the uncertainty of multiple factors can be considered comprehensively, providing a more scientific evaluation of the possibility of the fault source and reducing subjective judgment errors.

[0112] Based on Bayesian inference combined with the fact data of the current fault, the posterior probability of each node as the fault source is calculated, thereby screening out the fault source and providing a clear direction for maintenance decision; the real-time data of the current fault is input as observation evidence, wherein the real-time data includes abnormal current harmonic amplitude and bearing vibration speed, the posterior probability of each node is calculated based on Bayesian formula, all nodes are sorted according to the posterior probability, the first several nodes with the highest posterior probability are selected as potential fault sources, then a trustworthiness evaluation mechanism is introduced to improve the positioning result, the historical occurrence frequency of fault features, the reliability of detection equipment and the mutual confirmation relationship between fault features are comprehensively considered to weight and correct the posterior probability, if a fault feature has a high posterior probability but the detection sensor has an abnormality in recent period, the trustworthiness of the fault feature is reduced, and a trustworthiness threshold is set, only the nodes with the trustworthiness greater than the trustworthiness threshold are determined as the final fault source, if all nodes do not meet the standard, the step of generating a causal network is returned to recheck the causal relationship or supplement data; thereby the accurate positioning of the fault source is realized, the search range of maintenance personnel is reduced, the fault handling time is shortened, and the maintenance cost is reduced, the positioned fault source will be used as the basis for updating the diagnosis threshold and optimizing the coupling mode library, if the fault source judgment is wrong, it will lead to the failure of subsequent system parameter adjustment.

[0113] Specifically, the updating logic of the diagnosis threshold includes:

[0114] analyzing the change trend of the current harmonic amplitude and the bearing vibration speed at the fault source, and determining the threshold offset of the change of the current harmonic amplitude and the bearing vibration speed caused by the fault source;

[0115] updating the diagnosis threshold by reinforcement learning algorithm with false positive rate and false negative rate as reward function, and combining the threshold offset;

[0116] evaluating the effectiveness of the updated diagnosis threshold by cross-validation method to determine whether to update the diagnosis threshold again.

[0117] The signal change amplitude caused by different fault sources is different, and the load rate and winding temperature will affect the normal fluctuation range of the signal, therefore, the diagnosis threshold needs to be dynamically updated according to the fault source characteristics and the current working condition to avoid false positives or false negatives caused by fixed threshold; the change trend of the current harmonic amplitude and the bearing vibration speed at the fault source in the fault development process is analyzed, if the fault source is rotor eccentricity, the mutation amplitude and rising rate of the one frequency component amplitude in the vibration signal are extracted, the typical threshold offset of the signal change caused by the fault source is counted combined with historical fault data, if the historical fault data is insufficient, the initial range of the threshold offset is set according to expert experience, and the threshold offset is corrected combined with the load rate and winding temperature to avoid misjudgment of normal fluctuation as fault; thereby the diagnosis threshold can adapt to the fault source characteristics and the change of working condition, improve the sensitivity and specificity of fault detection, and reduce the diagnosis error caused by unreasonable diagnosis threshold.

[0118] It is difficult to adjust the threshold value only relying on experience or fixed rules to adapt to complex and variable operating scenarios. Reinforcement learning automatically searches for the optimal threshold value through a simulated trial-and-error process with false alarm rate and missed alarm rate as feedback, thereby improving the system's adaptive ability. The state space of reinforcement learning is defined to include the current diagnostic threshold value, load rate and winding temperature, historical fault detection results, and threshold offset, etc. The action space is set to the step range of diagnostic threshold value adjustment, each action corresponds to an increase or decrease in the diagnostic threshold value, and the reward function is designed to reduce the false alarm rate and missed alarm rate. A weighted reward method is designed and trained through the Q-learning algorithm. The system starts from the initial diagnostic threshold value, selects an action according to the current state, observes the fault detection results in the new state after execution, calculates the reward and updates the Q value table, and gradually learns the optimal threshold adjustment strategy. Thus, the autonomous update of the diagnostic threshold value is realized, enabling the system to automatically adjust to the optimal detection state under different operating conditions and fault modes, reducing manual intervention and improving diagnostic accuracy.

[0119] The threshold value obtained by reinforcement learning may be over-fitted or ineffective in certain scenarios, so it needs to be verified by multiple sets of data to ensure that the diagnostic threshold adjustment does not reduce the overall system performance. In specific applications, the historical fault data and normal operating data are divided into 5 subsets using k-fold cross-validation. Each time, 4 subsets are used to train the model and adjust the threshold value, and the remaining 1 subset is used for validation. The process is repeated 5 times to ensure that each subset is used as a validation set. During the validation process, the performance indicators of the updated diagnostic threshold value are calculated, including false alarm rate, missed alarm rate, and F1 score. The false alarm rate represents the proportion of normal data that is misjudged as fault, the missed alarm rate represents the proportion of fault data that is not detected, and the F1 score represents the comprehensive accuracy and recall rate. If the F1 score of the updated diagnostic threshold value is improved by more than a certain threshold value compared to the original diagnostic threshold value, and the false alarm rate and missed alarm rate are within an acceptable range, then the diagnostic threshold value update is confirmed to be effective. Otherwise, return to the reinforcement learning step to adjust the learning parameters and re-optimize the diagnostic threshold value. A strict validation mechanism is used to ensure the reliability of the threshold adjustment, avoid performance degradation due to blind updates, and ensure system stability. The validated diagnostic threshold value is applied to subsequent fault detection. If it fails the validation, it needs to be re-optimized to avoid affecting the system's response speed to new faults.

[0120] Specifically, the update logic of the coupling mode library includes:

[0121] Anomaly data of the un-matched fault coupling mode is clustered by the DBSCAN algorithm, and the clustering area of the new fault coupling mode is identified according to the data density.

[0122] The mapping relationship between the phase difference interval division and the fault coupling identifier in the new fault coupling mode is dynamically adjusted by the gradient descent algorithm to update the coupling mode library.

[0123] As the operating conditions of the generator change or new faults occur, the existing coupling mode library cannot cover all fault scenarios, so new fault coupling modes need to be mined from unmatched abnormal data to expand the fault recognition capability of the system; abnormal data that cannot be explained by existing modes during fault diagnosis is collected, including the coherence coefficient and phase difference of current harmonic components and bearing vibration frequency bands, as well as the corresponding combined working conditions, including load rate and winding temperature. The abnormal data is processed by DBSCAN density clustering algorithm, the neighborhood radius and minimum sample size are set, the data points with the same density are divided into the same cluster, and if there are multiple abnormal data points in a certain area, a new cluster is formed. Then, feature analysis is performed on each cluster, the center phase difference, coherence coefficient mean and working condition parameter distribution range of the data in the cluster are calculated, a preliminary description of the new fault coupling mode is formed, and the clustering area of the new fault coupling mode is identified. If the characteristics of a certain cluster show that the phase difference between the current harmonic amplitude and the bearing vibration speed is concentrated at 180°, and the probability of occurrence is high under high load rate, it is inferred that it is a new fault coupling mode related to load. Thus, unknown fault coupling modes are automatically discovered, the coverage blind spot of the coupling mode library is filled, and the diagnosis capability of the system for new faults is improved.

[0124] The newly identified fault coupling mode has problems such as fuzzy parameter boundary and conflict with existing modes, and needs to be optimized in parameters and adjusted in mapping relationship to ensure the accuracy and compatibility of the new fault coupling mode; an optimization objective function is constructed, considering the clustering compactness, class separation degree and matching degree with historical fault data, the clustering compactness includes the data dispersion in the cluster, the class separation degree includes the distance between clusters, the boundaries of phase difference intervals and the coherence coefficient thresholds are adjusted by gradient descent algorithm to minimize the objective function value, if the phase difference interval of a new fault coupling mode overlaps with that of an existing fault coupling mode, the boundary is fine-tuned to avoid confusion, and a unique fault coupling identifier is assigned to the new fault coupling mode, following the existing identifier coding rules, including the letter prefix to distinguish the mode type and the number suffix to represent the serial number, to ensure that the new fault coupling identifier does not repeat with the existing fault coupling mode, and update the identifier mapping table. The historical fault data is used to verify the new fault coupling mode, which is substituted into the judgment logic of the new fault coupling mode to calculate the classification accuracy, if the classification accuracy is less than the classification threshold, return to the parameter optimization step to adjust again. By optimizing the parameters and logic of the new fault coupling mode, it is seamlessly integrated with the existing coupling mode library, improving the consistency and accuracy of fault diagnosis.

[0125] The new fault coupling mode passed the verification is incorporated into the coupling mode library, so that the system can directly call in fault diagnosis, realize the continuous accumulation of knowledge and the iterative upgrade of system ability; a new mode record is created in the coupling mode library, including mode name, phase difference interval, coherence coefficient condition, fault coupling identification and applicable working condition range and other information, each time the coupling mode library is updated, the record updates the time, the update content and the operator and other information, and keeps the historical version, if the new version has a problem, it can quickly roll back to the stable version; so as to timely integrate new knowledge into the system, enhance the system's ability to diagnose complex faults, prolong the system life cycle, reduce the diagnostic blind area caused by the lack of mode, the updated coupling mode library will directly affect the fault coupling identification mark of the coupling perception module and the threshold adjustment of the fault diagnosis module, if the integration process has an error, it will cause a chain of diagnostic failure.

Claims

1. Universal generator electrical monitoring system with fault self-diagnosis function, characterized by: include: Coupling perception module, fault diagnosis module and fault tracing module; The coupling sensing module is used to obtain the current harmonic amplitude, bearing vibration speed, load rate and winding temperature in real time, synchronously extract the current harmonic components and the bearing vibration frequency band, and in response to the coherence coefficient and phase difference between the current harmonic components and the bearing vibration frequency band, mark the fault coupling identifier through the coupling mode library; The fault diagnosis module is used to extract a diagnostic threshold, and after receiving a fault coupling identifier, dynamically adjust the diagnostic threshold by combining the load factor and the winding temperature. During the adjustment process, current harmonic amplitudes and bearing vibration velocities greater than the diagnostic threshold are screened, and electrical faults are separated from mechanical faults based on a dual-channel network. Target current harmonics and target bearing vibrations are reconstructed through the dual-channel network, and the response is compared with the current harmonic amplitudes and the bearing vibration velocities to generate a fault contribution matrix. The fault tracing module is used to receive the fault contribution matrix, analyze the fault causal chain and locate the fault source through the causal inference algorithm, and respond to update the diagnosis threshold and coupling mode library.

2. The universal generator electrical monitoring system with fault self-diagnosis function according to claim 1, characterized in that: The marking logic of the fault coupling identifier includes: Configure a coherence coefficient threshold, screen for associated frequency points with a coherence coefficient greater than the threshold, and calculate the phase difference between the current harmonic component and the bearing vibration frequency band at the associated frequency point. Use a clustering algorithm to divide the phase difference into multiple phase difference intervals based on the size of the phase difference. Each phase difference interval represents the fault coupling mode. Establish a coupling mode library, correspond the phase difference interval to the fault coupling mode, and assign a fault coupling identifier to each fault coupling mode to form an identifier mapping table; According to the phase difference interval to which the current phase difference belongs, the fault coupling identifier is searched in the identifier mapping table, and the associated frequency points are marked.

3. The universal generator electrical monitoring system with fault self-diagnosis function according to claim 2, characterized in that: The judgment sub-logic of the coherence coefficient includes: Acquire the current harmonic amplitude and bearing vibration velocity in real time, and simultaneously extract the current harmonic components and bearing vibration frequency bands; The current harmonic components and bearing vibration frequency bands are divided into data segments according to time intervals. The dimensional features of the current harmonic components and bearing vibration frequency bands in each data segment are extracted by fast Fourier transform, and the dimensional features are reduced by principal component analysis. Based on the dimensional features after dimensionality reduction, the coherence coefficient between the current harmonic components and the bearing vibration frequency band at each frequency point is calculated to form a coherence coefficient sequence, and the outlier detection of the coherence coefficient sequence is performed using a clustering algorithm.

4. The universal generator electrical monitoring system with fault self-diagnosis function according to claim 3, characterized in that: The generation logic of the fault contribution matrix includes: Determine the contribution weights of electrical fault features and mechanical fault features to fault diagnosis based on the attention mechanism; The difference between the target current harmonic and the current harmonic amplitude, as well as the difference between the target bearing vibration and the bearing vibration velocity, are combined with contribution weights to determine the fault contribution of the electrical fault feature and the mechanical fault feature to fault diagnosis; The fault contributions of the electrical fault features and the mechanical fault features are arranged in order to generate a fault contribution matrix.

5. The universal generator electrical monitoring system with fault self-diagnosis function according to claim 4, characterized in that: The diagnostic threshold adjustment sub-logic includes: Obtain historical operating data of the generator and calculate the data distribution of current harmonic amplitude and bearing vibration speed under normal operating conditions to serve as initial diagnostic thresholds; After receiving the fault coupling identifier, the fault coupling mode corresponding to the fault coupling identifier is obtained from the coupling mode library, and the change trends of the current harmonic amplitude and bearing vibration speed under the fault coupling mode are statistically analyzed to determine the adjustment weight of the initial diagnosis threshold; Obtain load factor and winding temperature in real time, and use support vector regression to analyze the factors affecting current harmonic amplitude and bearing vibration velocity. The adjustment weight of the initial diagnostic threshold and the influencing factor are weighted and summed to obtain the adjustment coefficient, which is then combined with the initial diagnostic threshold to adjust the diagnostic threshold.

6. The universal generator electrical monitoring system with fault self-diagnosis function according to claim 5, characterized in that: The separation sub-logic of electrical fault and mechanical fault includes: Filter out data points with current harmonic amplitudes greater than the diagnostic threshold and mark them as candidate electrical fault points. Filter out data points with bearing vibration speeds greater than the diagnostic threshold and mark them as candidate mechanical fault points. A dual-channel network consisting of a physical channel and a data channel is constructed. The physical channel enhances the features of the electrical and mechanical candidate fault points to obtain electrical and mechanical fault features. The data channel extracts the correlation features between the current harmonic amplitude and the bearing vibration speed under the fault coupling mode from the coupling mode library based on the fault coupling identifier, and performs real-time correction on the correlation features by combining the load factor and winding temperature to generate a fault feature template. The cosine similarity between the electrical fault signature, mechanical fault signature and fault signature template is calculated to determine the electrical fault probability and the mechanical fault probability, and a decision threshold is configured to separate the electrical fault and the mechanical fault.

7. The universal generator electrical monitoring system with fault self-diagnosis function according to claim 6, characterized in that: The reconstruction sub-logic of the target current harmonics and target bearing vibration includes: Obtain the generator's current harmonic amplitude and bearing vibration velocity in real time in the physical channel; The historical current harmonic amplitude and historical bearing vibration velocity of the generator under normal operating conditions are retrieved from the data channel and classified and stored according to the combined working conditions of load rate and winding temperature to form a feature mapping relationship; The current harmonic amplitude and bearing vibration velocity obtained from the physical channel are matched with the load factor and winding temperature in a feature mapping relationship to filter normal operating condition characteristics. The normal operating condition characteristics are then fed back to the physical channel to reconstruct the target current harmonics and target bearing vibration. The reconstructed target current harmonics and target bearing vibrations are transmitted to the physical channel for error verification to determine whether the normal operating condition characteristics should be re-screened and the target current harmonics and target bearing vibrations should be reconstructed.

8. The universal generator electrical monitoring system with fault self-diagnosis function according to claim 7, characterized in that: The updating logic of the diagnostic threshold includes: Analyze the changing trends of the current harmonic amplitude and bearing vibration velocity at the fault source, and determine the threshold offset of the change of the current harmonic amplitude and bearing vibration velocity caused by the fault source; Through the reinforcement learning algorithm, the false alarm rate and missed alarm rate are used as reward functions, and the diagnostic threshold is updated in combination with the threshold offset; The effectiveness of the updated diagnostic threshold was evaluated by cross-validation method to determine whether the diagnostic threshold should be updated again.

9. The universal generator electrical monitoring system with fault self-diagnosis function according to claim 8, characterized in that: The fault source location sub-logic includes: Receive the fault contribution matrix and automatically search for the causal relationship between current harmonic amplitude, bearing vibration speed, load factor and winding temperature through causal inference algorithm to generate a causal network; The causal network is converted into a Bayesian network, where each node represents the fault feature and each edge represents the causal relationship. The prior probability distribution of each node is set according to the fault contribution matrix. The posterior probability of each fault feature as a fault source is determined through Bayesian reasoning to screen out potential fault sources, and the confidence level of each potential fault source is determined to locate the fault source.

10. The universal generator electrical monitoring system with fault self-diagnosis function according to claim 9, characterized in that: The update logic of the coupled mode library includes: The DBSCAN algorithm is used to cluster abnormal data that do not match the fault coupling pattern, and the clustering area of ​​the new fault coupling pattern is identified based on the data density; The mapping relationship between the phase difference interval division and the fault coupling identification in the new fault coupling mode is dynamically adjusted through the gradient descent algorithm to update the coupling mode library.

Images