Fault identification analysis method based on multi-source signal analysis of hydraulic generator

Abstract

The invention belongs to the technical field of equipment fault identification, and particularly discloses a fault identification analysis method based on multi-source signal analysis of a hydro-generator, and the method comprises the steps: firstly carrying out the collection and synchronization of multi-part signals, and screening suspected abnormal signals; based on a fault propagation chain, double correlation characterization analysis of fault time and trend is carried out by analyzing the position of the signal in the chain so as to screen a real fault signal; when a plurality of chains exist, comprehensively evaluating fault dominant strength, signal density and abnormal degree of each chain, and determining a dominant fault propagation chain; and finally, forming an evidence body by the leading chain and the abnormal degree of the fault signal, performing similarity matching with a historical fault template, and outputting a final fault type. According to the invention, accurate traceability and diagnosis of faults are realized, and isolated false alarms are effectively eliminated, so that the accuracy of fault diagnosis is improved, and a clear direction is provided for subsequent fault traceability and fault processing.

Description

Technical Field

[0001] This invention belongs to the field of equipment fault identification technology, and more specifically, relates to a fault identification and analysis method based on multi-source signal analysis of hydro-generators. Background Technology

[0002] As the core equipment of a hydropower station, the turbine generator is susceptible to failures due to the combined effects of hydraulic, mechanical, and electrical factors during operation, such as bearing wear, rotor imbalance, and stator winding short circuits. These failures can not only cause downtime and huge economic losses, but also lead to safety accidents, highlighting the importance and necessity of fault identification.

[0003] The prior art, such as the Chinese invention patent application No. 202510015766.5, discloses a method for detecting and diagnosing anomalies in water turbines based on multi-sensor data fusion. It integrates the results of three unsupervised algorithms—clustering, local outlier factor, and autoencoder—to generate a comprehensive anomaly score through weighted summation, and finally diagnoses the anomaly by comparing it with historical fault templates.

[0004] The prior art 2, such as the multi-strategy fusion hydro-generator fault diagnosis method disclosed in Chinese invention patent application No. 202411326852.X, constructs a comprehensive model by combining a GRU-based neural network time series model and a random forest decision model based on industrial internet platform data. It makes full use of the advantages of different algorithms, avoids the limitations of single algorithm for fault judgment, improves the accuracy of fault diagnosis, and achieves efficient fault classification.

[0005] Since equipment failures have multiple manifestations, when a failure occurs, alarm signals from multiple sensors are often received. These signals are often correlated in terms of time and development trend. Existing technologies 1 and 2 have not deeply explored the correlation between different sensor signals in the time dimension and the consistency in development trend, resulting in diagnostic results that remain at the level of anomaly. They lack the ability to distinguish whether the abnormal signal is transient interference, and they cannot trace the source of the anomaly, which is not convenient for subsequent anomaly handling.

[0006] Secondly, the comprehensive model constructed by existing technology 2 relies on a large amount of complete and accurately labeled historical fault data for its performance, which limits the universality of the method. Furthermore, this approach assumes that the input data is reliable and synchronous. However, in the field environment, problems such as sensor drift, communication delays, and data loss are frequent, making it difficult to guarantee the authenticity of the data and the faults, thus affecting the accuracy of the final judgment.

[0007] Furthermore, existing technologies lack a comprehensive analysis and comparison mechanism for multiple potential fault propagation paths when faced with simultaneous alarms from multiple sensors. When a complex coupled fault occurs in a hydro-generator, abnormal signals are often distributed across multiple propagation chains with different correlations. Existing technologies struggle to identify the dominant fault chain from these concurrent abnormal paths, leading to unclear fault tracing directions and affecting the efficiency and accuracy of maintenance decisions. Summary of the Invention

[0008] In view of this, in order to solve the above problems, a fault identification and analysis method based on multi-source signal analysis of hydro-generators is proposed.

[0009] The objective of this invention can be achieved through the following technical solution: This invention provides a fault identification and analysis method based on multi-source signal analysis of a hydro-generator. The method includes: collecting signal data from multiple parts of the hydro-generator, unifying the timestamps of each signal, and filtering signals that continuously deviate from the normal range as suspected abnormal signals.

[0010] For all suspected abnormal signals on the same fault propagation chain, fault time correlation characterization and fault trend correlation characterization are determined according to their position in the chain, and the fault signal is determined based on the correlation characterization.

[0011] Extract all fault signals. If they are distributed across multiple fault propagation chains, determine the dominant fault propagation chain based on the chain's fault dominance strength, fault signal density, and the degree of fault signal anomaly. Otherwise, the chain containing the fault signal is designated as the dominant fault propagation chain.

[0012] The abnormality of the dominant fault propagation chain and each fault signal within it is used to form an evidence body. The cosine similarity between the evidence body and the feature templates of each historical fault category is calculated. The historical fault category with the highest cosine similarity is taken as the final detected fault type.

[0013] Compared with the prior art, the beneficial effects of the present invention are as follows: (1) By establishing a fault propagation chain, the present invention solves the problem that the prior art cannot achieve abnormal source tracing, effectively reduces the dependence on the quantity and quality of historical fault data samples, realizes the transformation from discrete alarm to continuous fault path, and thus can clearly depict the potential development path of the abnormality in the equipment, which is convenient for subsequent abnormality handling.

[0014] (2) By performing dual verification of fault time correlation characterization and fault trend correlation characterization, the present invention determines the fault signal, effectively eliminates isolated abnormal signals caused by accidental false alarms of sensors or external interference, and can screen out real fault sources with consistency in the spatiotemporal dimension, thereby improving the credibility and anti-interference ability of fault source identification.

[0015] (3) This invention can automatically identify abnormal signal misalignment caused by communication delays by verifying the lag time in the fault time correlation characterization analysis. By calculating the time-delay cross-correlation and maximum information coefficient in the fault trend correlation characterization analysis, it can effectively filter out false abnormal signals caused by sensor drift. Thus, it can ensure that even under conditions where historical fault data is scarce and field data has noise and asynchronous problems, it can still maintain high diagnostic accuracy and reliability, effectively improving the universality and robustness of fault identification.

[0016] (4) In scenarios with multiple potential fault paths, this invention determines the dominant fault propagation chain by integrating and calculating the fault dominance strength, fault signal density and fault anomaly degree of each chain, thereby clarifying the fault tracing direction, realizing the objective assessment and comparison of the influence of multiple chains, and also improving the diagnostic efficiency and handling accuracy of complex faults, further improving the efficiency and accuracy of subsequent maintenance decisions.

[0017] (5) This invention reduces the dependence on the accuracy of single sensor data and enhances the stability of diagnostic results by forming evidence body from the abnormality of the dominant fault propagation chain and its internal fault sources, and matching it with the similarity of historical fault templates. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a schematic diagram of the overall implementation process of the present invention.

[0020] Figure 2 This is a schematic diagram of the overall implementation process of the fault time correlation characterization statistics of the present invention.

[0021] Figure 3 This is a schematic diagram of a partial implementation process of the fault time correlation characterization statistics of the present invention. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] Please see Figure 1As shown, the present invention provides a fault identification and analysis method based on multi-source signal analysis of hydro-generators. The method includes: S1, collecting signal data from multiple parts of the hydro-generator, unifying the timestamps of each signal, and filtering signals that continuously deviate from the normal range as suspected abnormal signals.

[0024] Considering the typical spatiotemporal correlation characteristics of the occurrence and development of faults in hydro-generators, signal monitoring of a single part cannot fully reflect the equipment status, and multiple asynchronous signals can directly lead to disordered fault timing and misjudgment of causal relationships. Therefore, a unified timestamp is required when collecting signal data from multiple parts of the hydro-generator.

[0025] Specifically, in one particular embodiment, the multiple parts include, but are not limited to, bearings, blades, and generator windings, and the signals include temperature signals, pressure signals, vibration signals, voltage signals, and current signals.

[0026] It should be noted that, in the process of screening signals that continuously deviate from the normal range, the duration of the signal deviation from the normal range must exceed the normal fluctuation duration threshold specified for the signal in the factory settings of the hydro-generator, in order to distinguish between real fault precursor signals and instantaneous interference fluctuations.

[0027] For example, a vibration signal must continuously exceed a threshold for three sampling cycles, and a temperature signal must continuously deviate for five sampling cycles before it is considered a suspected abnormal signal. By introducing a persistence requirement, misjudgments caused by instantaneous disturbances can be effectively avoided, ensuring the reliability of the reference signal data for subsequent fault analysis.

[0028] It should also be noted that the normal range is primarily determined based on the equipment's factory settings. Specifically, the rated values, permissible deviation values, and alarm values ​​of each signal in the manufacturer's technical specifications are used as benchmarks. For signals such as temperature and pressure, where a unidirectional increase indicates a deterioration in condition or a risk of failure, the rated value is used as the benchmark, and the factory-set alarm value is used as the upper (or lower) safety limit to define the boundaries of the normal range. The benchmark and the upper (or lower) safety limit constitute the normal range. For signals such as voltage and vibration, where both increases and decreases indicate a deterioration in condition or a risk of failure, the rated value plus or minus the factory permissible deviation value is used as the upper and lower limits of the normal range.

[0029] This invention ensures that signal data from different acquisition units have comparable time bases through a unified timestamp, providing data consistency for subsequent fault time correlation characterization analysis and fault propagation timing inference. Furthermore, by filtering signals deviating from normal ranges, it can identify potential anomalies in the early stages of hydro-generator faults, providing a target signal set for subsequent fault tracing analysis. Moreover, multi-site synchronous monitoring overcomes the limitations of traditional single-point diagnosis, further improving the effectiveness of subsequent fault diagnosis.

[0030] S2. For all suspected abnormal signals on the same fault propagation chain, determine the fault time correlation characterization and fault trend correlation characterization according to their position in the chain, and determine the fault signal based on the correlation characterization.

[0031] It needs to be explained that a fault propagation chain is essentially a physical path representation of the transmission of fault energy between hydraulic, mechanical, and electrical systems. To more clearly illustrate the construction logic and transmission rules of the fault propagation chain, we will now use the typical bearing fault propagation chain and generator stator winding fault propagation chain in a hydro-generator as examples for specific explanation.

[0032] For example, the construction and implementation process of a typical bearing failure propagation chain in the hydro-generator is as follows: When an early wear failure occurs in the thrust bearing of the hydro-generator unit, signal data from the following locations are collected using a unified timestamp: Signal A: Radial vibration of the thrust bearing housing.

[0033] Signal B: Temperature of thrust bearing pad.

[0034] Signal C: Cooling water outlet temperature of the lower guide bearing.

[0035] Signal D: Vibration of the lower guide bearing housing.

[0036] After data acquisition, signal A (vibration amplitude consistently exceeding the normal range) and signal B (temperature consistently exceeding the normal range) were first identified as suspected abnormal signal sources. Since abnormal vibration and temperature rise in the thrust bearing can be transmitted through the shaft system to the adjacent lower guide bearing, leading to deterioration of the adjacent lower guide bearing's operating condition, this deterioration further manifests as increased cooling water temperature and intensified vibration in the lower guide bearing. Therefore, a fault propagation chain can be constructed: a thrust bearing failure (A, B) will affect the lower guide bearing (C, D). This chain accurately reflects the complete path of fault transmission along the shaft system from the thrust bearing to the lower guide bearing.

[0037] As another example, the process of constructing the generator stator winding fault propagation chain includes: when the generator stator winding experiences local overheating or inter-turn short circuit, signal data from the following locations are acquired through a unified timestamp acquisition system, specifically including signals D, E, and F.

[0038] Signal D: Stator current harmonic content.

[0039] Signal E: Stator core temperature.

[0040] Signal F: Unit output power signal.

[0041] After data acquisition, signals D (harmonic content consistently exceeding the normal range), E (temperature consistently exceeding the normal range), and F (power fluctuation consistently exceeding the normal range) were identified as suspected abnormal signal sources. Since the inter-turn short-circuit fault in the stator winding first causes magnetic field distribution distortion, this distortion manifests as an increase in the content of specific harmonics. This increase in harmonic content subsequently leads to increased electromagnetic losses, which in turn causes a localized rise in stator core temperature. Ultimately, this affects the stability of the electromagnetic torque, resulting in continuous power fluctuations exceeding the normal range. Therefore, a fault propagation chain can be constructed: the stator winding fault first manifests as abnormal current harmonic content in signal D; this electrical abnormality, through increased electromagnetic losses, leads to abnormal stator core temperature in signal E, ultimately reflected in abnormal unit output power in signal F.

[0042] As can be seen from the above examples of fault propagation chain construction, the fault propagation chain described in this invention can accurately reflect the transmission pattern of faults between different systems, providing a reliable information foundation for subsequent fault tracing and diagnostic analysis.

[0043] This invention solves the problem of existing technologies being unable to trace the source of anomalies by establishing a fault propagation chain. It effectively reduces the dependence on the quantity and quality of historical fault data samples and realizes the transformation from discrete alarms to continuous fault paths. This allows for a clear depiction of the potential development path of anomalies in the equipment, facilitating subsequent anomaly handling.

[0044] It is important to note that the signal data in the fault propagation chain are sorted according to the structural connection relationship, physical connection relationship, or energy / material flow transmission path of the part of the equipment from which the data was collected.

[0045] Because signal anomalies at different locations along the same fault propagation chain during the operation of a hydro-generator may exhibit temporal sequence and trend changes during operation, this invention analyzes the relative positional relationships of signals within the fault propagation chain to perform subsequent fault time correlation characterization and fault trend correlation characterization analysis.

[0046] In one specific embodiment, please refer to Figure 2 As shown, the determination of the fault time correlation characterization based on its position in the chain includes the following statistical implementation process: First, for any two suspected abnormal signals on the same fault propagation chain, determine their relative positions in the chain.

[0047] If the relative positions are adjacent, perform fault time correlation characterization statistics under adjacent positions. If the relative positions are not adjacent, perform fault time correlation characterization statistics under non-adjacent positions. Finally, based on the fault time correlation characterization of two suspected abnormal signals in the two cases of adjacent and non-adjacent positions, output the fault time correlation characterization between each suspected abnormal signal.

[0048] Preferably, please refer to Figure 3 As shown in (a), the specific implementation process of performing fault time correlation characterization statistics at adjacent locations includes: First, calculating the interval between the abnormal start points of the two suspected abnormal signals, denoted as the lag time. Then, determining whether the lag time is within a preset delay range. If the lag time is within the preset delay range, the fault time correlation characterization of the two suspected abnormal signals is output as a strong correlation; if the lag time exceeds the preset delay range, the fault time correlation characterization of the two suspected abnormal signals is output as a weak correlation.

[0049] It should be noted that when vibration signals propagate through metal structures, the damping characteristics within the material lead to energy dissipation and velocity changes. Reflection and refraction at structural interfaces alter the propagation path, thus changing the propagation time. Furthermore, other signals, such as temperature, change with operating conditions during transmission, resulting in fluctuations in propagation time. Therefore, it is necessary to determine the lag time.

[0050] In one specific embodiment, the preset delay range is calculated based on the equipment dimensions of a specific hydro-generator unit.

[0051] For example, taking vibration signal and temperature signal as examples, a specific implementation example of constructing a preset reference lag time range is as follows: 1) Vibration signal: Obtain the shaft section length between the thrust bearing and the lower guide bearing from the equipment dimensions, and determine the longitudinal wave propagation speed of the steel based on the material manual. Use the ratio of the shaft section length to the longitudinal wave propagation speed as the theoretical delay time of the vibration signal.

[0052] Subsequently, the historical vibration propagation time series of the hydro-generator described in this invention under normal operating conditions was retrieved. For each measured propagation time in the series, the deviation from the theoretical delay time was calculated. All these deviation values ​​were divided into two groups, positive and negative, and the average value of the positive deviation group was calculated for each group. ) and the mean of the negative deviation group ( ), combining the theoretical delay time with The sum of these values ​​serves as the lower limit of the preset range, combining the theoretical delay time with... The sum of the values ​​is used as the upper limit of the preset range, thereby outputting the preset reference lag time range of the vibration signal. When the measured lag time is within this range, it is judged to be strongly correlated.

[0053] 2) Temperature signal: Obtain the oil passage length from the thrust bearing to the lower guide bearing from the equipment dimensions. Simultaneously, extract the oil reference flow rate and oil density from the equipment instruction manual, and determine the convective heat transfer coefficient and specific heat capacity of the oil based on the material handbook. Use the ratio of oil passage length to oil reference flow rate as the flow delay. Subsequently, calculate the heat transfer delay according to the heat balance equation. , ,in For the density of the oil, This is the specific heat capacity of the oil. The reference flow rate for the oil is... The convective heat transfer coefficient of the oil is given. Given the heat exchange area, the combined flow delay and heat transfer delay are used as the theoretical delay time of the temperature signal.

[0054] Retrieve historical temperature transmission time series under normal operating conditions and calculate the deviation between each measured value and the theoretical value. Similarly, divide all these deviation values ​​into two groups according to positive and negative, calculate the average value of each group, and similarly determine the preset reference lag time range of the temperature signal in the same way as the determination method of the preset reference lag time range of the vibration signal. When the measured lag time is within this range, it is judged as a strong correlation.

[0055] Alternatively, please refer to Figure 3 As shown in (b), the specific implementation process of performing fault time correlation characterization statistics under non-adjacent locations includes: identifying all adjacent signal pairs between the two suspected abnormal signals; for each pair of adjacent signals, statistically analyzing the adjacent signal fault time correlation characterization based on the lag time of the adjacent signals in the same way as the statistical method of fault time correlation characterization under adjacent locations.

[0056] If the fault time correlation characteristics of all adjacent signal pairs are strong, then the final output will show a strong correlation between the fault time correlation characteristics of these two non-adjacent suspected abnormal signals. Otherwise, the final output will show a weak correlation.

[0057] It should be noted that for long fault propagation chains, parallel computing can be used to improve real-time performance, and the same approach should be taken in the subsequent fault trend correlation characterization process.

[0058] To better understand the fault time correlation characterization of the present invention, the present invention takes the bearing fault propagation chain as an example to carry out a statistical example of fault time correlation characterization: First, based on the above-collected signals A, B, C and D, it is assumed that signals A and D and signals B and C are two sets of suspected abnormal signals, signals A and D are taken as the first set of signal pairs, signals B and C are taken as the second set of signal pairs, and the preset delay range is assumed to be 5 to 10 seconds.

[0059] Next, based on the above-mentioned construction process of the bearing failure propagation chain, it can be seen that the thrust bearing failure (A, B) will affect the lower guide bearing (C, D). That is, in this failure propagation chain, signal A and signal D are not adjacent positions, and signal B and signal C are adjacent positions. That is, the first pair of signal pairs are not adjacent positions, and the second pair of signal pairs are adjacent positions.

[0060] For the second set of signal pairs, the starting time of signal B's first deviation from the normal range is T1=14:20:15.000, and the starting time of signal C's first deviation from the normal range is T2=14:20:22.000. The lag time of both is 7 seconds. Since 7 seconds is within the preset range, the fault time correlation of these two suspected abnormal signals (B, C) in the output set is characterized as a strong correlation.

[0061] For the first group of signal pairs, since they are non-adjacent positions, identify all adjacent signal pairs (AB, BC, CD) in the middle. According to the determination method of the first group, characterize the fault time correlation of all adjacent signals in the second group. After determination, the fault time characterization of AB, BC, and CD are all strongly correlated. Then, output the fault time correlation characterization of the two suspected abnormal signals (A, D) in the first group as strongly correlated.

[0062] In one specific embodiment, the specific implementation process for determining the fault trend correlation characterization includes: selecting any two signals from each suspected abnormal signal in the same fault propagation chain to form a target signal pair.

[0063] If two signals are adjacent in position on the fault propagation chain, the data sequence of these two signals within a preset time window is extracted. If they are not adjacent, the data sequence of all adjacent signal pairs on the shortest fault propagation path connecting these two signals within a preset time window is extracted.

[0064] Based on the extracted data sequence, the time-delay cross-correlation index between two signals and the maximum information coefficient between the two signal pairs data sequences are calculated. The time-delay cross-correlation index includes the time-delay correlation coefficient and the target time-delay value.

[0065] If two signals are located adjacent to each other and simultaneously meet all of the following conditions, their fault trend correlation is considered strong. Otherwise, it is considered weak correlation: the time-delay correlation coefficient is higher than the first preset threshold.

[0066] The target time delay value is within the preset delay range.

[0067] The maximum information coefficient is higher than the second preset threshold.

[0068] For non-adjacent signals, if the fault trend correlation characteristics of all adjacent signal pairs on the path connecting them are strong, then the fault trend correlation characteristics between the two non-adjacent signals are determined to be strong. If any adjacent signal pair has a weak correlation, then it is determined to be a weak correlation.

[0069] It is understood that the first preset threshold and the second preset threshold can be determined based on historical data or experiments of a specific hydro-generator unit.

[0070] The process of setting the first preset threshold is as follows: statistically analyze the historical time-delay correlation coefficient distribution of the two signals under normal operating conditions of the hydro-generator unit, and take the 95th percentile of the historical time-delay correlation coefficient distribution as the first preset threshold.

[0071] The process of setting the second preset threshold is as follows: extract the historical maximum information distribution of these two signals under normal operating conditions of the hydro-generator, and take the 99th percentile of the historical maximum information distribution as the second preset threshold.

[0072] Given the uncertain time delay in the actual propagation of fault signals, directly calculating signal correlation may underestimate the true degree of correlation due to time delay misalignment. Therefore, time delay search and calculation of time delay cross-correlation indices are used to compensate for the error in determining fault trend correlation caused by time delay misalignment.

[0073] In a preferred embodiment of the present invention, the calculation process of the time-delay cross-correlation index includes: First, setting the time-delay search range to -50ms to +50ms, with a sampling interval of 10ms as the step size; then, within the preset time-delay search range, changing the time offset between the corresponding data sequences of the two signals and realigning them with the data sampling interval as the step size. For example, starting from -50ms, increasing the time offset by 10ms each time, and realigning the corresponding data sequences of the two signals. Each time the time offset is increased, the current time-delay position is recorded, and the Pearson correlation coefficient of the aligned data sequences of the two signals at each time-delay position is calculated.

[0074] The maximum Pearson correlation coefficient is selected from the Pearson correlation coefficients calculated at each time delay position. The maximum Pearson correlation coefficient is used as the time delay correlation coefficient between the two signals. Then, the time offset corresponding to the time delay correlation coefficient is taken as the target time delay value between the two signals.

[0075] It should be noted that the above-mentioned time delay search range and sampling interval are only a preferred example of the present invention, and can be set by the implementers themselves in specific practical scenarios. Furthermore, the Pearson correlation coefficient is an existing function, and its formula will not be elaborated further.

[0076] This invention, through time delay search and correlation coefficient calculation, can accurately capture the optimal correlation between signals and their corresponding time delay, thereby providing a quantitative basis for judging the trend of fault propagation.

[0077] Furthermore, given the complex nonlinear relationships often existing between actual fault signals, traditional correlation coefficients primarily reflect linear relationships and are insufficient to comprehensively characterize the degree of dependence between signals. Therefore, supplementary analysis is needed using statistics capable of capturing nonlinearity. This invention preferably selects the maximum information coefficient as the nonlinear statistic.

[0078] In a preferred embodiment of the present invention, the calculation process of the maximum information coefficient includes: constructing a scatter plot of the data sequence, dividing it into grids of different densities in the X-axis and Y-axis directions, for example, dividing it into 2 rows and 2 columns, 2 rows and 3 columns, 3 rows and 2 columns, etc., until a maximum number of grids is reached according to the data volume, and the density range of the grid division can preferably be from 10×10 to 30×30.

[0079] For each grid division method, data points are assigned to the corresponding grid cells. Based on the distribution of data points in each cell, the mutual information value of the two signals under the grid division is calculated, and the mutual information value is subjected to minimum-maximum normalization.

[0080] Iterate through all grid partitioning methods, record the normalized mutual information value obtained at each step, and select the maximum value from all recorded normalized mutual information values ​​as the maximum information coefficient.

[0081] It should be noted that the mutual information value calculation involved is a prior art method, and its specific calculation process will not be described in detail. By comparing the calculated maximum information coefficient with the second preset threshold, data services are provided for determining the correlation of fault trends along the fault propagation chain.

[0082] By statistically analyzing the maximum information coefficient, this invention can effectively capture various dependencies between signals, including nonlinear relationships, complementing the aforementioned linear time-delay correlation coefficient, and together ensuring the comprehensiveness and accuracy of fault trend correlation trend judgment.

[0083] Secondly, this invention, through lag duration verification in fault time correlation characterization analysis, can automatically identify abnormal signal misalignment problems caused by communication delays. By calculating time-delay cross-correlation and maximum information coefficient in fault trend correlation characterization analysis, it can effectively filter out false abnormal signals caused by sensor drift. This ensures high diagnostic accuracy and reliability even under conditions where historical fault data is scarce and field data contains noise and asynchronous issues, effectively improving the universality and robustness of fault identification.

[0084] In another specific embodiment, determining the fault signal includes: for each of the suspected abnormal signals, extracting its fault time correlation characterization and fault trend correlation characterization with at least one other suspected abnormal signal in the same fault propagation chain.

[0085] If a suspected abnormal signal is strongly correlated with both the fault time correlation characterization and the fault trend correlation characterization of at least one other suspected abnormal signal, then the suspected abnormal signal is determined to be a fault signal.

[0086] S3. Extract all fault signals. If they are distributed across multiple fault propagation chains, determine the dominant fault propagation chain based on the chain's fault dominance strength, fault signal density, and the degree of fault signal anomaly. Otherwise, take the chain containing the fault signal as the dominant fault propagation chain.

[0087] In actual operation, multiple faults in a hydro-generator may occur simultaneously or have multiple potential propagation paths. Based on this, this invention considers whether the fault is distributed across multiple fault propagation chains, and when it is distributed across multiple fault propagation chains, determines the dominant fault propagation chain from the perspectives of fault dominance intensity, fault signal density, and fault anomaly degree.

[0088] In one specific embodiment, the specific implementation process of determining the dominant fault propagation chain includes: for each fault propagation chain, traversing the remaining fault propagation chains and checking whether they share fault signals or preset strong correlation signals with the current chain.

[0089] If it exists, increment the association mark count of the current chain by 1, and calculate the fault dominance strength of each fault propagation chain based on the association mark count. If it does not exist, traverse the next fault propagation chain.

[0090] The fault signal density of a chain is obtained by dividing the number of identified fault signals on all fault propagation chains by the total number of signal sources in that chain.

[0091] For each fault signal in the fault propagation chain, calculate the absolute value of the relative deviation between its real-time data value and the median corresponding to the normal range, and use this as the anomaly degree. Based on the anomaly degree, calculate the degree of fault anomaly in the fault propagation chain.

[0092] By comparing the fault dominance strength of each fault propagation chain, the coefficient of variation of the fault dominance strength of the fault propagation chain is calculated.

[0093] If the coefficient of variation is less than or equal to a preset threshold, the linear weighted sum of the fault dominance strength, fault signal density, and fault anomaly degree is used as the final anomaly dominance score, and the fault propagation chain with the highest score is used as the dominant fault propagation chain.

[0094] If the coefficient of variation is greater than a preset threshold, then the fault propagation chain with the strongest fault dominance is taken as the dominant fault propagation chain.

[0095] Furthermore, since a single association marker count cannot distinguish between widespread influence and repeated associations, this invention introduces repeated association ratio and marker ratio to perform fault dominance strength statistics. The specific statistical process for calculating the fault dominance strength of each fault propagation chain is as follows: For each fault propagation chain, count its association marker count and record all fault propagation chains that generate the marker to form a list of associated chains for that chain.

[0096] For each chain in the list of associated chains, count the number of times it is repeatedly associated with the current chain. The number of times it is repeatedly associated refers to the number of times it is associated with a certain chain during the traversal and inspection process due to the existence of multiple shared fault sources or multiple strong association signals.

[0097] For each fault propagation chain, count the number of chains in its associated chain list where the number of repeated association markers is not zero, and record this number as the number of valid associated chains.

[0098] Divide the number of valid associated chains corresponding to each fault propagation chain by the difference between the total number of fault propagation chains and 1 to obtain the duplication association ratio. The duplication association ratio indicates how many other chains in the network are deeply associated with the current chain. The larger the duplication association ratio, the more widespread the influence of the chain.

[0099] From the number of associated markers for all fault propagation chains, select the maximum value as the highest number of markers.

[0100] The ratio of the number of association tags for each fault propagation chain to the highest number of tags is denoted as the tagging ratio. The tagging ratio indicates the total number of associations between the current chain and the entire network. The larger the tagging ratio, the stronger the overall activity and influence of the chain.

[0101] For each fault propagation chain, the fault dominance strength is obtained by multiplying its repetition correlation ratio by its labeling ratio.

[0102] It's important to note that two common interference scenarios exist in fault propagation networks. One is where a chain has multiple repeated associations with only a very few other chains, exhibiting high association depth but insufficient breadth. The other is where a chain has only a single association with numerous other chains, exhibiting large association breadth but insufficient depth. If a simple linear weighted summation is used, both of these interference scenarios may lead to an inflated final result due to a high score on one particular indicator. By introducing multiplication, the dominance strength score only increases when both the repetition association ratio and the labeling ratio reach high values. This allows us to filter out fault propagation chains that possess both broad influence and deep coupling within the fault propagation network, thereby improving the accuracy and reliability of fault tracing analysis.

[0103] In scenarios with multiple potential fault paths, this invention determines the dominant fault propagation chain by integrating and calculating the fault dominance strength, fault signal density, and fault anomaly degree of each chain, thereby clarifying the fault tracing direction, achieving an objective assessment and comparison of the influence of multiple chains, and improving the diagnostic efficiency and handling accuracy of complex faults, further enhancing the efficiency and accuracy of subsequent maintenance decisions.

[0104] Furthermore, considering that the field data is filled with various instantaneous interferences and sensor false alarms, if only a few signals are abnormal, it is likely to be an occasional fault or a false alarm. That is, the degree of abnormality of a single fault signal cannot represent the level of abnormality of the entire chain. Therefore, this invention determines the final degree of fault abnormality representing the entire chain based on the degree of abnormality of the fault signal. The calculation process of the degree of fault abnormality of the fault propagation chain includes: comparing the degree of abnormality of all fault signals and taking the maximum value as the benchmark degree of fault abnormality of the corresponding fault propagation chain.

[0105] For the same fault propagation chain, the average anomaly degree of all fault signals is calculated, and the proportion of fault signals with anomalies exceeding the average value is also calculated.

[0106] For each fault propagation chain, the fault signal ratio is multiplied by the baseline fault anomaly degree to obtain the final fault anomaly degree.

[0107] This invention combines the baseline anomaly degree with the proportion of fault signals exceeding the average value, taking into account both the impact of the most severe fault point and the breadth of anomaly distribution within the chain, thus comprehensively reflecting the overall severity of anomalies in the fault propagation chain.

[0108] S4. The abnormality of the dominant fault propagation chain and each fault signal within it is used to form an evidence body. The cosine similarity between the evidence body and the feature templates of each historical fault category is calculated. The historical fault category with the highest cosine similarity is taken as the final detected fault type.

[0109] Specifically, the unique identifier of the identified dominant fault propagation chain and the anomaly degree of all fault signals within it are combined in a preset order to form a multi-dimensional vector, denoted as evidence body E. , This indicates the identifier of the dominant fault propagation chain, which can be represented by a specific number for subsequent analysis. to This indicates the degree of abnormality of each fault signal in the dominant fault propagation chain. This indicates the total number of fault signals.

[0110] Based on the historical fault case database, a corresponding feature template is constructed for each known fault category, such as bearing wear or winding overheating. Each template is a vector with the same dimension as the evidence body, and the components of the vector consist of the dominant chain identifier and the average value of the fault signal anomaly degree corresponding to that historical fault case.

[0111] The cosine similarity of evidence body E with each historical fault feature template is calculated, and the results are compared. The historical fault category with the highest similarity value is selected and determined as the final fault type for this diagnosis.

[0112] It should be noted that the data in the historical fault case database described in this invention comes from real fault events recorded and ultimately confirmed during the past operation of the hydro-generator unit or units of the same model. Each fault event should include a complete multi-source signal data sequence and a fault type label confirmed by post-analysis, such as thrust bearing wear and stator winding interlayer short circuit. For each historical fault event, the same analysis process as the fault identification in this invention is adopted, namely, constructing a fault propagation chain, identifying the dominant fault propagation chain, and calculating the anomaly degree of each fault signal on the chain.

[0113] In practice, the identifier of the dominant fault propagation chain and the average anomaly of each fault signal on that chain can be combined in a preset order to form a feature vector. This feature vector constitutes a feature template for this type of fault. For the same fault category, multiple historical events can be aggregated to generate multiple templates, or the average anomaly of each fault signal in each historical event can be taken as the final anomaly of each fault signal, and the feature template can be formed.

[0114] This invention reduces the reliance on the accuracy of single sensor data and enhances the stability of diagnostic results by forming an evidence body composed of the anomalies of the dominant fault propagation chain and its internal fault sources, and matching it with historical fault templates based on similarity.

[0115] The above content is merely an example and illustration of the concept of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the concept of the invention or exceed the scope defined by the present invention, and all such modifications and additions should fall within the protection scope of the present invention.

Claims

1. A method for fault identification analysis based on multi-source signal analysis of a hydroelectric generator, characterized in that, The method comprises: Collecting multi-site signal data of the hydroelectric generator, and unifying timestamps of all signals, and screening signals continuously deviating from normal range as suspected abnormal signals; For all suspected abnormal signals on the same fault propagation chain, determining fault time correlation representation and fault trend correlation representation according to the positions of the signals in the chain, and determining fault signals based on the correlation representations; Extracting all fault signals, if distributed in multiple fault propagation chains, determining a dominant fault propagation chain based on fault dominant strength of the chains, fault signal density and abnormal degree of the fault signals; otherwise, taking the chain where the fault signals are located as the dominant fault propagation chain; Assembling the dominant fault propagation chain and abnormal degrees of all fault signals in the chain into an evidence body, calculating cosine similarity of the evidence body and each historical fault category feature template, and taking the historical fault category with the highest cosine similarity as the final detected fault type.

2. The method for fault identification analysis based on multi-source signal analysis of hydroelectric generator as claimed in claim 1, characterized in that: The signal data in the fault propagation chain is sorted according to the structural connection relationship, physical connection relationship or energy flow / material flow transmission path of the collected sites in the equipment.

3. The method for fault identification analysis based on multi-source signal analysis of hydroelectric generator as claimed in claim 1, characterized in that: The determination process of the fault time correlation representation is as follows: Determining the relative positions of any two suspected abnormal signals in the chain; If the relative positions are adjacent positions, calculating the interval time length between the abnormal starting points of the two suspected abnormal signals, and recording the interval time length as a lag time length; When the lag time length is within a preset delay range, outputting the fault time correlation representation of the two suspected abnormal signals as strong correlation; When the lag time length exceeds the preset delay range, outputting the fault time correlation representation of the two suspected abnormal signals as weak correlation; If the relative positions are not adjacent positions, identifying all adjacent signal pairs between the two suspected abnormal signals, and determining the fault time correlation representation of each adjacent signal pair based on the lag time length of the adjacent signal; If the fault time correlation representations of all adjacent signal pairs are strong correlation, outputting the fault time correlation representation of the two suspected abnormal signals as strong correlation, otherwise, outputting the fault time correlation representation as weak correlation, thereby obtaining the fault time correlation representation between each suspected abnormal signal.

4. The method for fault identification analysis based on multi-source signal analysis of hydroelectric generator as claimed in claim 1, characterized in that: The determination process of the fault trend correlation representation is as follows: Selecting any two signals from the suspected abnormal signals in the same fault propagation chain to form a target signal pair; If the positions of the target signal pair in the fault propagation chain are adjacent, extracting data sequences of the two signals within a preset time window, if not adjacent, extracting data sequences of all adjacent signal pairs on the shortest fault propagation path connecting the two signals within a preset time window; Based on the extracted data sequences, calculating a time-lag cross-correlation index between the two signals and a maximum information coefficient between the data sequences of the two signal pairs, the time-lag cross-correlation index including a time-lag correlation coefficient and a target time-lag value; If the positions of the two signals are adjacent, and all the following conditions are met at the same time, the fault trend correlation representation is judged as strong correlation, otherwise, as weak correlation: The time-lag correlation coefficient is higher than a first preset threshold value; The target time-lag value is within a preset delay range; The maximum information coefficient is higher than a second preset threshold value; For non-adjacent signals, if the fault trend correlation characteristics of all adjacent signal pairs on the path connecting them are strong correlation, the fault trend correlation characteristics between the two non-adjacent signals are judged as strong correlation; if there is any adjacent signal pair that is weak correlation, it is judged as weak correlation.

5. The method for fault identification analysis based on multi-source signal analysis of hydroelectric generator as claimed in claim 4, characterized in that: The time-lag cross-correlation index calculation process comprises: Within a preset time-lag search range, the time offset between the data sequences corresponding to the two signals is changed at a data sampling interval as a step, and each time the time offset is increased, the data sequences corresponding to the two signals are realigned; The Pearson correlation coefficient of the aligned data sequences corresponding to the two signals at each time-lag position is calculated, and the maximum value is screened out as the time-lag correlation coefficient between the two signals; The time offset corresponding to the time-lag correlation coefficient is taken as the target time-lag value between the two signals.

6. The method for fault identification analysis based on multi-source signal analysis of hydroelectric generator as claimed in claim 4, characterized in that: The calculation process of the maximum information coefficient comprises: The data sequences are formed into a scatter plot, and a grid division of multiple different densities is performed in the X and Y axis directions; For each grid division method, the data points are assigned to the corresponding grid cells, and the mutual information value of the two signals under the grid division is calculated according to the distribution of the data points in each cell, and the mutual information value is normalized; All grid division methods are traversed, and the normalized mutual information value obtained at each step is recorded, and the maximum value is screened out from all recorded normalized mutual information values as the maximum information coefficient.

7. The method for fault identification analysis based on multi-source signal analysis of hydroelectric generator as claimed in claim 1, characterized in that: The determination of the fault signal comprises: For each suspected abnormal signal, the fault time correlation characteristic and the fault trend correlation characteristic of at least one other suspected abnormal signal in the same fault propagation chain are extracted; If both the fault time correlation characteristic and the fault trend correlation characteristic of the suspected abnormal signal and the at least one other suspected abnormal signal satisfy strong correlation, the suspected abnormal signal is determined as a fault signal.

8. The method for fault identification analysis based on multi-source signal analysis of hydroelectric generator as claimed in claim 1, characterized in that: The determination of the dominant fault propagation chain comprises: For each fault propagation chain, the rest of the fault propagation chains is traversed to check whether there is a shared fault signal or a preset strongly correlated signal between the current chain; If there is, the correlation mark number of the current chain is increased by 1, and the fault dominance strength of each fault propagation chain is counted based on the correlation mark number; The number of fault signals identified on all fault propagation chains is divided by the total number of signal sources of the corresponding chain to obtain the fault signal density of the chain; For each fault signal on the fault propagation chain, the relative deviation absolute value of the real-time data value and the median of the normal range is calculated as the abnormality degree, and the fault abnormality degree of the fault propagation chain is calculated based on the abnormality degree; The fault dominance strength variation coefficient of the fault propagation chain is calculated by comparing the fault dominance strength of each fault propagation chain; If the variation coefficient is less than or equal to a preset threshold, the linear weighted sum of the fault dominance strength, the fault signal density and the fault abnormality degree is taken as the final abnormality dominance score, and the fault propagation chain with the highest score is taken as the dominant fault propagation chain; If the variation coefficient is greater than the preset threshold, the fault propagation chain with the maximum fault dominance strength is taken as the dominant fault propagation chain.

9. The method for fault identification analysis based on multi-source signal analysis of hydroelectric generator as claimed in claim 8, characterized in that: The determination of the fault signal comprises: For each fault propagation chain, count the number of associated labels, and record all fault propagation chains that produce the label to form an associated chain list of the chain; For each chain in the associated chain list, count the number of repeated associated labels with the current chain; For each fault propagation chain, count the number of chains in the associated chain list whose repeated associated label number is not zero, and record the number as the number of valid associated chains; Divide the number of valid associated chains of each fault propagation chain by the difference between the total number of fault propagation chains and 1 to obtain the repeated associated ratio; From the associated label numbers of all fault propagation chains, filter out the maximum value as the highest label number; Calculate the ratio of the associated label number of each fault propagation chain to the highest label number, and record it as the label ratio; For each fault propagation chain, multiply its repeated associated ratio by the label ratio to obtain the fault dominant strength.

10. The method for fault identification analysis based on multi-source signal analysis of hydroelectric generators according to claim 8, characterized in that: The calculation process of the fault abnormality degree of the fault propagation chain includes: Compare the abnormality degrees of all fault signals, and take the maximum value as the reference fault abnormality degree of the corresponding fault propagation chain; For the same fault propagation chain, count the average value of the abnormality degrees of all fault signals, and count the proportion of fault signals whose abnormality degrees exceed the average value; For each fault propagation chain, multiply the fault signal proportion by the reference fault abnormality degree to obtain the final fault abnormality degree.

Images