Hydro-generator carbon brush ignition diagnosis method based on multi-parameter fusion

Abstract

The invention provides a hydro-generator carbon brush ignition diagnosis method based on multi-parameter fusion. According to the hydro-generator carbon brush ignition diagnosis method based on multi-parameter fusion, early warning of hydro-generator carbon brush ignition and accurate fault root cause recognition can be achieved, the hysteresis quality and one-sidedness of a traditional monitoring means are effectively overcome, and the operation and maintenance efficiency and the equipment operation reliability are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of fault diagnosis and predictive maintenance of a carbon brush-slip ring system of a hydroelectric generator, and in particular to a carbon brush sparking diagnosis method for a hydroelectric generator based on multi-parameter fusion. BACKGROUND

[0002] As core equipment of a hydropower station, the stability of the carbon brush-slip ring system of a hydroelectric generator is directly related to the safety and economy of the operation of the unit. With the continuous advancement of the construction of smart power plants, the traditional carbon brush sparking monitoring method has been difficult to meet the high reliability operation requirements. In related technologies, the observation of the carbon brush state is usually realized through artificial inspection, online temperature monitoring, and video monitoring, which covers the whole process from data acquisition, state judgment to manual intervention, including key links such as infrared temperature measurement, spark classification, and fixed threshold alarm. Specifically, this system relies on the experience of the operation and maintenance personnel and the threshold response of a single parameter, which to some extent guarantees the operation of the equipment, but lacks the ability of systematic fusion and intelligent analysis of multi-source heterogeneous data, making it difficult to realize early warning and root cause diagnosis.

[0003] However, in the existing carbon brush sparking monitoring method, single-point temperature or visible sparks are directly used as the basis for judgment, and key features such as positive and negative temperature difference, current distribution uniformity, and early discharge signal are not effectively fused and analyzed, which may lead to problems such as delayed early warning, fuzzy diagnosis, and passive treatment, or high misjudgment rate under complex working conditions, thereby affecting the operation efficiency and maintenance cost of the equipment. In addition, the traditional method has obvious shortcomings in data acquisition density, feature extraction accuracy, and intelligent decision-making ability, and it is difficult to form a closed-loop control system from perception to execution, which limits the development of intelligent operation and maintenance of the slip ring system of the hydroelectric generator. SUMMARY

[0004] The present application aims to at least partially solve one of the technical problems in the related art.

[0005] To this end, a first object of the present application is to propose a carbon brush sparking diagnosis method for a hydroelectric generator based on multi-parameter fusion.

[0006] A second object of the present application is to propose a carbon brush sparking diagnosis device for a hydroelectric generator based on multi-parameter fusion.

[0007] A third object of the present application is to propose an electronic device.

[0008] A fourth object of the present application is to propose a computer-readable storage medium.

[0009] A fifth object of the present application is to propose a computer program product.

[0010] To achieve the above purpose, the first aspect of the embodiment of the present application provides a water turbine generator carbon brush sparking diagnosis method based on multi-parameter fusion, comprising: S1, collecting temperature data, current data, ultrasonic signal data, ultra-high frequency electromagnetic wave signal data and environmental parameter data of the water turbine generator carbon brush to form a multi-dimensional original data set; S2, based on the multi-dimensional original data set, extracting key features such as positive and negative carbon brush temperature difference, same-pole carbon brush current distribution uniformity, ultrasonic signal effective value and ultra-high frequency signal pulse count, and constructing a multi-parameter fusion feature vector; S3, inputting the multi-parameter fusion feature vector into an AI fault diagnosis model deployed in the cloud, the model is trained based on historical data and is used for identifying the fault root cause of the carbon brush sparking and outputting a diagnosis result; S4, generating a structured operation and maintenance instruction containing a specific fault position, a fault type and a processing suggestion according to the diagnosis result, and pushing it to a man-machine interaction platform for operation and maintenance personnel to execute.

[0011] In an embodiment of the present application, the collecting temperature data, current data, ultrasonic signal data, ultra-high frequency electromagnetic wave signal data and environmental parameter data of the water turbine generator carbon brush to form a multi-dimensional original data set further comprises: S11, using a DS18B20 digital temperature sensor installed on the brush holder of each carbon brush to realize millisecond-level synchronous temperature collection; S12, using a miniature open-type Hall current sensor to be sleeved on the flexible lead of each carbon brush to realize real-time acquisition of the shunt value of each carbon brush.

[0012] In an embodiment of the present application, the based on the multi-dimensional original data set, extracting key features such as positive and negative carbon brush temperature difference, same-pole carbon brush current distribution uniformity, ultrasonic signal effective value and ultra-high frequency signal pulse count, and constructing a multi-parameter fusion feature vector further comprises: S21, calculating the average temperature difference ΔT of the positive carbon brush group and the negative carbon brush group, and combining the standard deviation of the same-pole carbon brush current as the basis for judging poor contact or pressure abnormality; S22, performing frequency spectrum analysis on the ultrasonic signal, extracting the effective value and pulse count in the 20-100 kHz frequency band, and using them to identify early weak discharge phenomenon.

[0013] In an embodiment of the present application, the inputting the multi-parameter fusion feature vector into an AI fault diagnosis model deployed in the cloud, the model is trained based on historical data and is used for identifying the fault root cause of the carbon brush sparking and outputting a diagnosis result further comprises: S31, the AI fault diagnosis model uses a gradient boosting decision tree (GBDT) algorithm to classify the feature vector and identify more than 12 typical fault modes; S32, the AI fault diagnosis model further combines a convolutional neural network (CNN) to model the time series features to improve the recognition accuracy of progressive faults.

[0014] In an embodiment of the present application, further comprising: S5, based on the multi-parameter fusion feature vector and historical operation data, calculating the remaining useful life (RUL) of the carbon brush by using a predictive maintenance model, and generating a recommended suggestion for the next maintenance time according to the RUL.

[0015] To achieve the above object, the second aspect of the present application provides a water turbine generator carbon brush sparking diagnosis device based on multi-parameter fusion, comprising: a data acquisition module for acquiring temperature data, current data, ultrasonic signal data, very high frequency electromagnetic wave signal data and environmental parameter data of the water turbine generator carbon brush to form a multi-dimensional original data set; a feature extraction module for extracting key features such as positive and negative carbon brush temperature difference, same-pole carbon brush current distribution uniformity, ultrasonic signal effective value and very high frequency signal pulse count based on the multi-dimensional original data set, and constructing a multi-parameter fusion feature vector; an AI diagnosis module for inputting the multi-parameter fusion feature vector into an AI fault diagnosis model deployed in the cloud, the model being trained based on historical data and used for identifying the fault root cause of the carbon brush sparking and outputting a diagnosis result; an instruction generation and pushing module for generating a structured operation and maintenance instruction containing a specific fault position, fault type and processing suggestion according to the diagnosis result, and pushing to a man-machine interaction platform for operation and maintenance personnel to execute.

[0016] To achieve the above object, the third aspect of the present application provides an electronic device, comprising: a processor, and a memory in communication connection with the processor;

[0017] The memory stores computer execution instructions;

[0018] The processor executes the computer execution instructions stored in the memory to realize the method according to any one of the first aspect.

[0019] To achieve the above object, the fourth aspect of the present application provides a computer readable storage medium, the computer readable storage medium stores computer execution instructions, and the computer execution instructions are executed by the processor to realize the method according to any one of the first aspect.

[0020] To achieve the above object, the fifth aspect of the present application provides a computer program product, the computer program is executed by the processor to realize the method according to any one of the first aspect.

[0021] The technical scheme provided by the embodiments of the present application at least brings the following beneficial effects:

[0022] Additional aspects and advantages of the present application will be in part apparent and in part pointed out hereinafter. BRIEF DESCRIPTION OF DRAWINGS

[0023] The above and / or additional aspects and advantages of the present application will become apparent and more readily appreciated from the following description, taken in conjunction with the following drawings of which:

[0024] Figure 1 is a flow chart of a water turbine generator carbon brush sparking diagnosis method based on multi-parameter fusion according to an embodiment of the present application;

[0025] Figure 2 is a structural schematic diagram of a water turbine generator carbon brush sparking diagnosis device based on multi-parameter fusion according to an embodiment of the present application. DETAILED DESCRIPTION

[0026] Embodiments of the present application are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals represent the same or similar elements or elements having the same or similar functions throughout. The embodiments described below by reference to the accompanying drawings are exemplary and are intended to explain the present application, and cannot be understood as a limitation of the present application.

[0027] Figure 1 is a flow chart of a water turbine generator carbon brush sparking diagnosis method based on multi-parameter fusion according to an embodiment of the present application.

[0028] As shown in Figure 1 , the water turbine generator carbon brush sparking diagnosis method based on multi-parameter fusion includes the following steps:

[0029] S1, collecting temperature data, current data, ultrasonic signal data, ultra-high frequency electromagnetic wave signal data and environmental parameter data of the water turbine generator carbon brush to form a multi-dimensional original data set.

[0030] Specifically, this step is the key data acquisition link in the water turbine generator carbon brush sparking intelligent diagnosis system, and its technical implementation is based on the collaborative deployment and synchronous acquisition mechanism of multi-source heterogeneous sensors, aiming to build a high-dimensional and high-precision original data set, and provide a solid data foundation for subsequent edge computing and cloud platform AI fusion analysis.

[0031] In some implementations, this step achieves comprehensive perception of the carbon brush operating state by deploying a sensor array at multiple key locations of the hydroelectric generator carbon brush system. Specifically, a digital temperature sensor (such as a DS18B20) is embedded on the brush holder of each carbon brush, with a sampling frequency of up to 100 ms per time, an accuracy of ±0.5°C, support for multi-point parallel collection, and ensuring the synchronization and integrity of all carbon brush temperature data. At the same time, a miniature open-type Hall current sensor (such as a LEM HALL effect sensor) is fitted on the flexible lead of each carbon brush, with a measurement range of 0-200A, a sampling frequency of 1kHz, high linearity, and low phase delay, used to obtain real-time carbon brush shunt values and calculate the uniformity index (such as standard deviation, coefficient of variation) of the same-pole carbon brush current distribution.

[0032] Further, to capture early sparking phenomena, the system arranges high-frequency ultrasonic sensors (frequency range 20-100kHz) and ultra-high frequency (UHF) electromagnetic wave sensors (frequency range 300-1000MHz) near the brush holder, respectively for detecting weak mechanical vibration signals and electromagnetic wave signals generated by partial discharge. The ultrasonic sensor identifies abnormal friction or discharge behavior between the carbon brush and the slip ring through acoustic feature extraction (such as effective value, peak value, pulse count), while the UHF sensor captures early features of discharge events through electromagnetic wave spectrum analysis.

[0033] In addition, the system also integrates an environmental parameter monitoring module, including a temperature and humidity sensor (such as SHT31) and a particulate matter sensor (such as PMS5003), for collecting data on the environmental humidity (accuracy ±2%RH), temperature (±0.3°C), and carbon powder concentration (particle size 0.3-10μm) in the machine room, providing background data for environmental cause analysis of carbon brush sparking.

[0034] This step plays a core role in the "perception layer" of the entire system, and its multi-dimensional data collection capability provides rich, real-time, and synchronized data input for subsequent feature extraction, fault classification, and predictive maintenance model training. By achieving high-precision synchronous collection of temperature, current, ultrasonic, UHF, and environmental parameters, the system can effectively identify early signs of carbon brush sparking, significantly improving the timeliness of fault warning and the accuracy of diagnosis, thereby realizing a shift from "passive response" to "active prevention" in the operation and maintenance mode.

[0035] Further, S1 includes:

[0036] S11, a DS18B20 digital temperature sensor is installed on the brush holder of each carbon brush to achieve millisecond-level synchronous temperature collection.

[0037] Specifically, in the present application, a DS18B20 digital temperature sensor is installed on the brush holder of each carbon brush to achieve millisecond-level synchronous temperature acquisition, which is one of the key steps to build the "all-factor perception layer". DS18B20 is a digital temperature sensor based on 1-Wire bus protocol, with high precision (±0.5℃ in the range of -10℃ to +85℃), low power consumption, no external power supply, and multiple point series access, etc. It is suitable for distributed temperature monitoring in industrial environment.

[0038] In some implementations, a DS18B20 sensor is embedded on the brush holder of each carbon brush, which is closely attached to the brush holder through its packaging structure to ensure that the sensor can accurately perceive the temperature change of the contact surface between the carbon brush and the slip ring. The sensor communicates with the industrial intelligent gateway of the edge computing layer through the 1-Wire bus, and multiple sensors can be connected in series to the same bus to realize centralized data acquisition. Further, to ensure the synchronicity and real-time of temperature acquisition, the system uses a timing polling mechanism with a 100ms acquisition cycle to periodically read all sensors, ensuring that the temperature data of each carbon brush remains consistent in the time dimension with an error of no more than ±5ms.

[0039] In terms of parameter index, the temperature measurement range of DS18B20 is -55℃ to +125℃, which is suitable for monitoring the carbon brush of the hydroelectric generator in normal operation (about 60℃ to 100℃) and abnormal temperature rise (up to 120℃ or more). Its resolution is 9-12 bits adjustable, which is set to 12 bits in the present application to achieve a temperature accuracy of 0.0625℃. The collected data is checked by CRC to ensure transmission integrity, and filtering processing (such as moving average or median filtering) is performed in the edge computing layer to eliminate transient interference and noise.

[0040] This step is deployed in the excitation system of the hydroelectric generator in practical application, especially at the position of each carbon brush holder in the slip ring area, forming a global temperature monitoring network. By collecting temperature data in real time, the system can calculate the average temperature (T+, T-) of the positive and negative carbon brush groups and their temperature difference (ΔT), providing key input features for subsequent AI fusion diagnosis.

[0041] In terms of technical effect, this step realizes high-density, high-precision, and millisecond-level synchronous acquisition of carbon brush temperature, overcoming the defects of monitoring blind area and response lag in traditional methods. Through the real-time and synchronicity of temperature data, the system can capture the local temperature rise caused by increased contact resistance, sticking, insufficient spring pressure, etc. in advance, thereby realizing early warning before sparking occurs, significantly improving the safety and stability of equipment operation.

[0042] S12, a miniature open-type Hall current sensor is fitted on the flexible lead of each carbon brush to obtain the shunt value of each carbon brush in real time.

[0043] Specifically, in the present application, the step of "obtaining the shunt value of each carbon brush in real time by sleeving a miniature open-type Hall current sensor on the flexible lead of each carbon brush" is based on the technical principle of non-contact current detection technology based on the Hall effect. The sensor adopts an open structure, which is convenient for quick installation outside the existing flexible lead without the need to disconnect the circuit, thereby realizing online, real-time and high-precision measurement of the shunt current of each carbon brush. In some implementations, the selected miniature open-type Hall sensor has an IP67 protection level, is suitable for the high-humidity and high-dust operating environment inside the hydroelectric generator, and has a measurement accuracy of within 0.5% and a 100A range, which can meet the monitoring needs of typical current values in the carbon brush system.

[0044] In terms of specific operation, the sensor senses the magnetic field generated by the current flowing in the flexible lead through a magnetic core closed structure, and the internal Hall element converts the magnetic field strength into a voltage signal. After processing by a signal conditioning circuit (including amplification, filtering, analog-to-digital conversion, etc.), a digital signal is output to the edge computing terminal. Optionally, the sensor supports RS-485 or Modbus RTU communication protocols to ensure the stability and real-time performance of data transmission. Further, the system can calculate the standard deviation or coefficient of variation of the current between the same-pole carbon brushes by synchronously collecting the shunt values of all carbon brushes, which serves as a key indicator for judging the balance of current distribution.

[0045] This step plays a core role in the entire technical solution, and the shunt data collected is one of the basic inputs for building a multi-parameter fusion diagnosis model, especially in identifying current abnormalities caused by uneven spring pressure, carbon brush sticking or slip ring surface oxidation. By monitoring the shunt value of each carbon brush in real time, the system can detect current distribution imbalance in advance, providing a reliable basis for early warning and root cause diagnosis of sparking faults.

[0046] S2, based on the multi-dimensional original data set, extracting key features such as positive and negative carbon brush temperature difference, same-pole carbon brush current distribution uniformity, ultrasonic signal effective value, and ultra-high frequency signal pulse count, and constructing a multi-parameter fusion feature vector.

[0047] Specifically, in the "multi-parameter fusion feature vector construction" step of the present application, the technical implementation core lies in extracting key features with physical meaning and diagnostic value from the multi-dimensional original data set of the hydroelectric generator carbon brush system, and integrating them into a structured multi-parameter fusion feature vector as the input of the subsequent AI diagnosis model. This step is a key link in data processing between the edge computing layer and the cloud platform intelligent analysis layer in the "perception-decision-execution" system, and plays a decisive role in improving diagnosis accuracy and achieving early warning.

[0048] In some implementations, the multi-dimensional original data set includes temperature data, current data, ultrasonic signal data, ultra-high frequency (UHF) signal data, and environmental parameters (such as humidity, carbon powder concentration) of each carbon brush. Among them, the positive and negative carbon brush temperature difference (ΔT) is collected synchronously at the millisecond level by a digital temperature sensor (such as DS18B20) installed on each brush holder, and the difference between the average temperature of the positive carbon brush group (T+) and the average temperature of the negative carbon brush group (T-) is calculated as an important indicator reflecting the electrochemical state of the slip ring and the contact quality of the carbon brush. The uniformity of the current distribution of the carbon brush of the same polarity is calculated by a miniature open Hall current sensor to collect the shunt value of each carbon brush, and the standard deviation or coefficient of variation is calculated to quantify the dispersion degree of the current distribution, so as to determine whether there is an abnormal contact resistance or uneven spring pressure problem.

[0049] Further, the ultrasonic signal effective value (RMS) and the ultra-high frequency signal pulse count (Pulse Count) are used as early discharge characteristics, which are collected by a high-frequency ultrasonic sensor and a UHF partial discharge sensor respectively. The ultrasonic signal is used to capture the mechanical vibration caused by the sparking of the carbon brush, and the effective value thereof can reflect the discharge intensity; the UHF signal is used to detect electromagnetic pulses, and the pulse count thereof can represent the discharge frequency, and the combination of the two can significantly improve the recognition ability of weak discharge events.

[0050] Optionally, environmental parameters (such as humidity, carbon powder concentration) can also be introduced in this step as auxiliary features to exclude environmental interference factors and improve the robustness of the diagnostic model. After preliminary extraction of all feature values in the edge computing layer, they are packaged as a multi-parameter fusion feature vector, which usually includes 5-8 key parameters such as ΔT, current standard deviation, ultrasonic RMS, UHF pulse number, and the specific parameter combination can be adjusted according to actual working conditions and model training requirements.

[0051] The technical value of this step lies in the fusion processing of multi-source heterogeneous data, which constructs a feature vector with physical interpretability and diagnostic sensitivity, provides high-quality input for the AI model, and thus realizes early identification and root cause diagnosis of carbon brush sparking. Compared with the monitoring method relying on only a single parameter (such as temperature or current) in the prior art, the present application significantly improves the accuracy and comprehensiveness of diagnosis by multi-parameter fusion, and provides a solid data foundation for the intelligent operation and maintenance of the hydroelectric generator.

[0052] Further, S2 includes:

[0053] S21, calculating the average temperature difference ΔT of the positive carbon brush group and the negative carbon brush group, and combining the standard deviation of the current of the carbon brush of the same polarity as the basis for judging the poor contact or pressure abnormality

[0054] Specifically, the step aims to achieve early identification and diagnosis of poor contact or pressure abnormalities of the hydroelectric generator carbon brush by calculating the average temperature difference ΔT of the positive and negative carbon brush groups and combining the standard deviation of the carbon brush current. This method is based on the principles of thermodynamics and current conduction, and uses the statistical characteristics of temperature and current to build a multi-parameter fusion fault identification model.

[0055] In some implementations, the temperature data of the positive and negative carbon brush groups is collected in real time by digital temperature sensors (such as DS18B20) embedded in each brush holder, with a sampling frequency of 100 ms per time, ensuring dynamic response to temperature changes. The system first performs sliding window averaging on the temperature data of the positive and negative carbon brush groups respectively (window length of 10-30 seconds), calculates the positive average temperature T+ and the negative average temperature T-, and then calculates ΔT = |T+ - T-|. Under normal operating conditions, ΔT should be maintained within ±2℃, and if ΔT continuously exceeds the set threshold (such as 5℃), it may indicate that there is uneven contact resistance or abnormal surface state of the slip ring.

[0056] Further, the system synchronously collects current data of each carbon brush, and realizes high-precision measurement of shunt values through miniature open-type Hall current sensors (such as LEM HRS series), with a sampling frequency of 100 Hz-1 kHz. In the same-pole carbon brush group, the standard deviation σI of the current is calculated to evaluate the uniformity of the current distribution. When σI exceeds the set threshold (such as 5%-10% of the rated current), it indicates that there is uneven pressure or poor contact between the carbon brushes.

[0057] This step is implemented in the edge computing layer, and ΔT and σI are input into the multi-parameter fusion diagnosis model of the cloud platform as key feature parameters through the pre-set statistical analysis algorithm and AI model input interface. The technical value lies in that by quantifying the statistical characteristics of temperature difference and current distribution, it can effectively distinguish the sparking risk caused by different reasons such as mechanical pressure abnormality, slip ring oxidation, and carbon brush sticking, and provide reliable basis for subsequent root cause diagnosis and treatment suggestions, thereby realizing the transition from "after-treatment" to "early-warning".

[0058] S22, performing frequency spectrum analysis on the ultrasonic signal to extract the effective value and pulse count in the 20-100 kHz frequency band for identifying early weak discharge phenomena.

[0059] Specifically, in the present application, performing frequency spectrum analysis on the ultrasonic signal and extracting the effective value and pulse count in the 20-100 kHz frequency band is one of the key steps to realize early identification of hydroelectric generator carbon brush sparking. This step captures the high-frequency mechanical vibration signal generated by the contact surface of the carbon brush and the slip ring when weak discharge occurs, and combines frequency domain feature analysis, thereby realizing early warning of potential faults before visible sparks or temperature abnormalities occur.

[0060] In some implementations, the ultrasonic sensor is installed near the carbon brush holder, with a working frequency range of 20-100 kHz to cover the high-frequency vibration signals generated in the early stage of carbon brush sparking. The analog signal output by the sensor is gain-adjusted by a low-noise preamplifier (usually set to 20-40 dB), and then digitized by an anti-aliasing filter (cutoff frequency of 120 kHz) and a 16-bit ADC (sampling rate not less than 200 kHz). The collected raw ultrasonic signals are then sent to the intelligent gateway of the edge computing layer for real-time spectral analysis.

[0061] The spectral analysis uses the Fast Fourier Transform (FFT) algorithm to convert the time-domain signal to the frequency-domain signal and analyze the energy distribution in the 20-100 kHz frequency band. The root mean square (RMS) value of the signal in this frequency band reflects the persistence of the discharge intensity, while the pulse count is used to characterize the frequency and burstiness of the discharge event. In specific parameter settings, the system can set the RMS threshold to 0.5-2.0 mV (adjusted according to the device background noise level) and the pulse count threshold to 10-50 times per second to distinguish between normal friction noise and abnormal discharge signals.

[0062] This step plays an important role in the system, and its output is one of the input features of the multi-parameter fusion diagnosis model, which, together with temperature, current, UHF signals, etc., forms a fault feature vector. By capturing the ultrasonic features of early discharge, the system can identify fault modes such as carbon brush sticking, poor contact, slip ring surface oxidation, etc., thereby realizing the transition from “post-processing” to “pre-warning” and significantly improving the operation reliability and maintenance efficiency of the carbon brush system of the hydroelectric generator.

[0063] S3, inputting the multi-parameter fusion feature vector into an AI fault diagnosis model deployed in the cloud, which is trained based on historical data to identify the root cause of carbon brush sparking and output a diagnosis result

[0064] Specifically, in some implementations, inputting the multi-parameter fusion feature vector into the AI fault diagnosis model deployed in the cloud is the core decision-making link in the “perception-decision-execution” integrated system of the present application. This model is trained offline based on historical operation data and fault samples, using a supervised learning framework, through feature fusion and pattern recognition, to realize root cause diagnosis and intelligent warning of carbon brush sparking faults in hydroelectric generators.

[0065] At the technical implementation level, the multi-parameter fusion feature vector is generated in real time by the edge computing layer, including but not limited to the average temperature difference of positive and negative carbon brushes (ΔT), the standard deviation of the current of the same pole carbon brush (σI), the effective value of the ultrasonic signal (RMS_US), the pulse count of the ultra-high frequency (UHF) partial discharge signal (Pulse_UHF), the environmental humidity (RH), and the carbon powder concentration (PM), and other key parameters. These features are standardized (such as normalization, Z-score standardization), packaged as structured data packets, and uploaded to the cloud AI platform via industrial Ethernet or 5G communication links. In the cloud platform, the AI model uses a multi-layer perception (MLP) or convolutional neural network (CNN) deep learning architecture to perform non-linear mapping and classification on the input features, and outputs specific fault root cause labels.

[0066] At the parameter index level, the training data set of the AI model needs to cover multiple typical fault modes, such as carbon brush sticking, insufficient spring pressure, and slip ring surface oxidation, and each type of fault sample should contain at least 300 labeled multi-dimensional feature data. Model evaluation indicators include accuracy (Accuracy), recall (Recall), F1 score, and AUC value, and the recognition accuracy of key fault modes should be no less than 92%. In addition, the model needs to have online incremental learning ability to support continuous injection of new samples and dynamic updating of model parameters.

[0067] At the application scenario level, this step is suitable for remote intelligent operation and maintenance systems of large hydroelectric generators, especially in hydropower stations with multiple units operating in parallel, which can realize centralized monitoring and intelligent diagnosis of the carbon brush systems of each unit. Through cloud deployment, the system can support cross-regional data aggregation and expert knowledge sharing, and improve the overall operation and maintenance response efficiency.

[0068] At the technical effect level, this step realizes the transition from "abnormality detection" to "root cause identification", significantly improves the accuracy and timeliness of fault diagnosis, provides a reliable basis for subsequent precise execution, effectively reduces the risk of unplanned downtime, and improves the reliability and intelligence level of equipment operation.

[0069] Further, S3 includes:

[0070] S31, the AI fault diagnosis model uses the gradient boosting decision tree (GBDT) algorithm to classify the feature vector and identify more than 12 typical fault modes.

[0071] Specifically, in the intelligent analysis layer of the cloud platform of the present application, the AI fault diagnosis model uses the Gradient Boosting Decision Tree (GBDT) algorithm to classify the multi-dimensional feature vectors extracted from the edge computing layer, thereby identifying more than 12 typical fault modes. This step is the core link to realize fault root cause diagnosis and intelligent decision-making, and has high technical integration and engineering practicality.

[0072] In some implementations, the GBDT model gradually optimizes the prediction error by integrating multiple weak decision trees to model complex nonlinear relationships. Specifically, the model input is a feature vector composed of temperature, current, ultrasonic wave, UHF signal, humidity, and other parameters. Each feature dimension is standardized to eliminate dimensional differences. For example, temperature data is in Celsius, current is in amperes, ultrasonic wave signal is in decibels (dB), UHF signal is in pulses per second, and humidity is in relative humidity (%RH). The model output is the classification result of more than 12 fault modes, such as "carbon brush jam", "spring pressure deficiency", "slip ring surface oxidation", "carbon brush wear", etc.

[0073] Further, the training process of the GBDT model uses a supervised learning method to construct a training set based on historical operation data and manually labeled fault samples. Model parameters include learning rate, tree depth, number of iterations, etc. The learning rate is usually set to between 0.01 and 0.1 to balance the model convergence speed and generalization ability; the tree depth is controlled between 3 and 6 layers to avoid overfitting; the number of iterations is set between 100 and 500 times according to the data volume and model performance requirements.

[0074] This step is deployed in the cloud platform in actual application, and performs online inference on the feature data uploaded in real time by the edge computing layer, realizing dynamic diagnosis of the carbon brush system of the hydroelectric generator. Its technical value lies in that through multi-parameter fusion and the high-precision classification ability of GBDT, the accuracy and interpretability of fault identification can be significantly improved, providing clear fault types and processing suggestions for maintenance personnel, thereby realizing the intelligent transition from "abnormal alarm" to "root cause diagnosis".

[0075] S32, the AI fault diagnosis model further combines a convolutional neural network (CNN) to model time series features to improve the identification accuracy of progressive faults.

[0076] Specifically, in the intelligent analysis layer of the cloud platform of the present application, the AI fault diagnosis model further combines a convolutional neural network (CNN) to model the time series features, in order to improve the recognition accuracy of the progressive fault in the brush sparking of the hydro-generator. The core of this step is to use the powerful local feature extraction capability of the CNN to deeply model the multi-dimensional time series data (such as carbon brush temperature, current distribution, ultrasonic signal, UHF pulse number, etc.) from the edge computing layer, so as to capture the dynamic evolution process of the fault development.

[0077] In some implementations, the CNN model adopts a one-dimensional convolution structure (1D-CNN) to adapt to the linear features of the time series data. The input layer receives the pre-processed time series data, each channel corresponds to a physical parameter (such as temperature, current, ultrasonic intensity, etc.), the data sampling frequency is 100 Hz to 1 kHz, and the time window length is usually set to 10 seconds to 1 minute to balance the real-time performance and feature integrity. The convolution layer uses multiple filters (kernel size = 3-5, stride = 1) to extract local time series features such as the rising trend of the temperature curve, the mutation point of the current distribution, and the periodic fluctuation of the ultrasonic signal through a sliding window. The pooling layer (such as max pooling) is used to reduce the data dimension and enhance the robustness of the model while preserving the key features.

[0078] Further, the model realizes step-by-step abstraction of features through multi-layer stacking, and finally connects a fully connected layer for classification output. In the training process, a supervised learning method is adopted, the input is the labeled historical fault data, and the output is the corresponding fault type (such as "sticking", "uneven oxidation", "spring fatigue", etc.). The model evaluation indicators include accuracy (Accuracy), F1-score, AUC-ROC curve, etc. to ensure high sensitivity and specificity in identifying progressive faults.

[0079] This step is deployed in the cloud platform in actual application, forming a collaborative analysis architecture with the edge computing layer, and is suitable for remote intelligent operation and maintenance scenarios of large hydropower stations. Through the modeling of the CNN on the time series, the system can identify early fault features that are difficult to detect by traditional methods, such as the slow rising trend of temperature caused by the oxidation of the slip ring surface, thereby realizing the transition from "after-treatment" to "pre-warning", and significantly improving the accuracy and foresight of the diagnosis.

[0080] S4, according to the diagnosis result, generating a structured operation and maintenance instruction containing the specific fault location, fault type and treatment suggestion, and pushing it to the human-computer interaction platform for operation and maintenance personnel to execute.

[0081] Specifically, in the present application, the step of generating structured operation and maintenance instructions according to the diagnosis results and pushing them to the human-computer interaction platform is the key link to realize the intelligent operation and maintenance closed loop. Based on the fault root cause diagnosis results output by the cloud platform intelligent analysis layer, combined with the pre-set expert knowledge base and standardized processing flow, the structured instructions containing the specific fault location, fault type and processing suggestions are automatically generated, and are pushed to the terminal equipment of the operation and maintenance personnel through a safe and efficient communication mechanism.

[0082] In some implementations, the system first receives the output results from the AI diagnosis model, which are presented in a structured data format (such as JSON or XML), containing the specific pole number, carbon brush number, fault type (such as jamming, excessive contact resistance, insufficient spring pressure, etc.) and preliminary fault characteristic description. The system then calls the built-in expert knowledge base, which is based on standards such as IEC 60034-21-1 "Rotating electrical machines - Carbon brush and slip ring systems", combined with historical maintenance records and typical fault handling schemes, to match the corresponding processing suggestions for each fault type. For example, when the diagnosis is "excessive contact resistance", the system will automatically associate the standard operation flow of "check the wear state of the carbon brush, clean the brush holder, measure and adjust the spring pressure to 0.15-0.25 MPa".

[0083] Further, the generated operation and maintenance instructions need to meet the structured requirements, usually including fault location (such as "No. 1 unit, negative pole, No. 2 carbon brush"), fault type (such as "abnormal contact resistance"), processing priority (such as "urgent" or "general"), operation steps, required tools, safety precautions, etc. The system supports multi-language output of instructions and can perform hierarchical pushing according to user permissions, ensuring the executability and safety of the instructions.

[0084] Optionally, the system pushes the instructions to the Web monitoring platform or mobile APP through industrial communication protocols such as OPC UA, MQTT or HTTP, realizing cross-platform, real-time and visual operation and maintenance response. The pushing delay is controlled within 500 ms, ensuring that the operation and maintenance personnel can obtain and execute the processing measures in a timely manner.

[0085] The technical effect of this step is to convert the original fuzzy judgment relying on human experience into executable and traceable standardized instructions, significantly improving the efficiency and accuracy of fault handling, reducing the risk of misoperation, and realizing the transition from "abnormal alarm" to "intelligent decision-making", providing solid support for predictive maintenance and intelligent operation and maintenance of the carbon brush system of the hydroelectric generator.

[0086] The multi-parameter fusion-based carbon brush sparking diagnosis method for the hydroelectric generator of the embodiments of the present application realizes early warning and root cause accurate diagnosis of the carbon brush sparking of the hydroelectric generator, improves the timeliness and accuracy of fault identification, and reduces the risk of equipment damage and operation and maintenance dependence.

[0087] S5, based on the multi-parameter fusion feature vector and historical operation data, calculating the remaining useful life (RUL) of the carbon brush using a predictive maintenance model, and generating a recommended suggestion for the next maintenance time according to the RUL.

[0088] Specifically, in some implementations, the predictive maintenance model of the present application dynamically predicts the remaining useful life (RUL) of the carbon brush of the hydro-generator by fusing the multi-parameter feature vector with the historical operation data, using a time series-based machine learning algorithm such as LSTM, Prophet or Weibull regression model. Specifically, the model input includes the currently collected multi-dimensional features such as carbon brush temperature, current distribution uniformity, ultrasonic signal intensity, UHF partial discharge signal characteristic value, environmental humidity and carbon powder concentration, forming a high-dimensional feature vector. At the same time, the system splices the historical operation data (such as temperature trend, current fluctuation, maintenance record, etc. in the past 30 days) as time context information to construct input samples with time sequence dependence.

[0089] At the parameter index level, a sliding window mechanism is used in the model training process, with a window length of 7 to 15 days and a step size of 1 day to capture the gradual process of carbon brush performance degradation. The temperature threshold is set to trigger the model attention when AT > 10℃, and the current unevenness is set to be abnormal with a variation coefficient (CV) > 0.15. The RUL output by the model is usually in units of hours, combined with the carbon brush wear rate (such as 0.1mm / 1000h) and replacement standard (such as wear amount > 2mm), the system can automatically calculate the next maintenance time and generate a structured suggestion such as "it is recommended to replace or check the carbon brush after 120 hours".

[0090] Further, this step plays a key decision support role in the whole system, upgrading the traditional "threshold alarm + manual judgment" mode to an "intelligent prediction + automatic suggestion" mode, significantly improving the forward-looking and precision of maintenance. In actual application scenarios, the model can be deployed on a cloud platform or an edge computing node, suitable for carbon brush state management in large hydropower stations, especially in high-load, long-period operation environments, which can effectively reduce the risk of unplanned downtime and improve equipment operation reliability and maintenance efficiency.

[0091] To achieve the above-mentioned embodiments, the present application also proposes a water turbine generator carbon brush fire diagnosis device based on multi-parameter fusion. Figure 2 A structure diagram of a water turbine generator carbon brush fire diagnosis device based on multi-parameter fusion provided by an embodiment of the present application is shown in Figure 2 As shown, the device comprises:

[0092] The data acquisition module 100 is configured to acquire temperature data, current data, ultrasonic signal data, ultra-high frequency electromagnetic wave signal data and environmental parameter data of the carbon brush of the hydro-generator, and form a multi-dimensional original data set.

[0093] The feature extraction module 200 is configured to extract key features such as the temperature difference between the positive and negative carbon brushes, the current distribution uniformity of the carbon brushes of the same polarity, the effective value of the ultrasonic signal and the pulse count of the ultra-high frequency signal based on the multi-dimensional original data set, and construct a multi-parameter fusion feature vector.

[0094] The AI diagnosis module 300 is configured to input the multi-parameter fusion feature vector into an AI fault diagnosis model deployed in the cloud, the model is trained based on historical data and is used to identify the root cause of the carbon brush sparking fault and output a diagnosis result.

[0095] The instruction generation and pushing module 400 is configured to generate a structured operation and maintenance instruction containing a specific fault location, a fault type and a processing suggestion according to the diagnosis result, and push the instruction to a man-machine interaction platform for an operation and maintenance personnel to execute.

[0096] As to the device in the above embodiment, the specific manner in which each module performs the operation has been described in detail in the embodiment related to the method, and will not be described in detail here.

[0097] In order to realize the above-mentioned embodiment, the present application also proposes an electronic device, comprising: a processor, and a memory connected with the processor in communication; the memory stores computer execution instructions; the processor executes the computer execution instructions stored in the memory to realize the method provided by the foregoing embodiments.

[0098] In order to realize the above-mentioned embodiment, the present application also proposes a computer readable storage medium, the computer readable storage medium stores computer execution instructions, and the computer execution instructions are executed by the processor to realize the method provided by the foregoing embodiments.

[0099] In order to realize the above-mentioned embodiment, the present application also proposes a computer program product, comprising a computer program, the computer program is executed by the processor to realize the method provided by the foregoing embodiments.

[0100] The collection, storage, use, processing, transmission, provision and disclosure of user personal information involved in the present application comply with relevant laws and regulations and do not violate public order and good customs.

[0101] It is important to note that user's personal information should be collected for legitimate and reasonable uses of the entity and not shared or sold outside of those legitimate uses. Further, such collection / sharing should occur after the user is informed of and individually agrees to the purposes for which it is gathered / shared, including for purposes specified in this document and the user's consent to same, and that the user is able to review, modify, and / or remove said consent and request the deletion of cookies, as further described herein. Additionally, steps should be taken to secure and safeguard this personal information data and that other persons with access to this data are bound to keep it confidential and not use it for any unauthorized purposes. Further, upon termination of the service to a user, user personal information should be deleted.

[0102] The present disclosure contemplates that the systems and methods described herein will be provided services by entities over whom the present disclosure has varying degrees of direct or indirect control. For instance, the present disclosure contemplates providing media guidance data from a service over the Internet, mobile network or other transport. In this example, the service might receive information from a user and / or a media guidance data source and / or provide information to a user, e.g., a client device, etc. Such a service might deliver content, e.g., text, graphics, audio, video, etc. information to individuals although the users' individual circumstances can differ and can change over time. Thus, the present disclosure contemplates the alternative embodiments in which a service might provide content to a plurality of users and / or receive information from a plurality of users.

[0103] In the foregoing description, reference has been made to specific embodiments of the application by way of example. It is to be understood that other embodiments can be utilized and that changes can be made without departing from the scope of the present application. For description purposes, the terminology "one embodiment", "some embodiments", "an example", "a specific example", or "some examples" has been used. It is to be understood that the specific features, structures, or characteristics described can be combined in any suitable manner in one or more embodiments or examples. It is to be understood that the terminology "some" and "some" are used in the context of describing specific features, structures, or characteristics, which are included in at least one embodiment or example of the present application. Thus, the terminology "some" and "some" does not necessarily refer to the same embodiment or example of the present application. Furthermore, the description herein of any particular feature, structure, or characteristic is not necessarily limited to that particular feature, structure, or characteristic. Thus, it is contemplated to further combine such features, structures, or characteristics and / or further modifications into other features, structures, or characteristics described herein, and further combinations and / or modifications into other features, structures, or characteristics described herein, as would be understood by those skilled in the art.

[0104] In addition, the terms "first", "second", etc. are used herein only to describe various features, and do not imply a relative importance or a specific order of precedence. Thus, features defined with "first", "second", etc. can include at least one of the features, explicitly or implicitly. In the description of the present application, the meaning of "a plurality" is at least two, such as two, three, etc., unless otherwise specifically limited.

[0105] Any process or method described in a flowchart or otherwise described herein can be understood as representing code modules, segments, or portions of code that include one or more executable instructions for performing specific logic functions (or steps) or portions of the application, and the various embodiments of the application can include additional or fewer functions (or steps) performed in the same order or in a different order, combined with other functions (or steps), performed at the same time or at different times, or even omitted. All of these permutations are intended to fall within the scope of the present application.

[0106] The logic and / or steps represented in flow diagrams or otherwise described herein, for example, can be considered as a sequence of instructions to implement logic functions, and can be embodied in any computer-readable medium for use by an instruction execution system, apparatus, or device, such as a computer-based system, processor- containing system, or other system that can fetch the instructions from the instruction execution system, apparatus, or device and execute the instructions. In the context of this specification, a "computer-readable medium" can be any means that can contain, store, communicate, propagate or transport the program for use by or in connection with the instruction execution system, apparatus, or device. The computer-readable medium can be a machine-readable storage device (e.g., magnetic, optical or other) a machine-readable storage diskette (e.g., floppy disk, optical disk, CD- ROM, etc.), a machine- readable storage card (e.g., PCMCIA card, etc.), a machine-readable storage tape (e.g., magnetic tape, optical tape, etc.), a machine-readable storage medium (e.g., RAM, ROM, etc.), a machine-readable signal (e.g., electrical, optical, etc.), a machine-readable medium (e.g., carrier wave, etc.) or any other suitable medium or means of embodying the program. More specific examples (a non-exhaustive list) of the computer-readable medium include the following: an electrical connection having one or more wires, a portable computer diskette, a hard disk, a RAM, a ROM, an EPROM, a FLASH memory card, an optical fiber, and a portable compact disc read-only memory (CD-ROM). Additionally, the computer-readable medium can be paper or another suitable medium upon which the program is printed, as the program can be electronically captured, for example, via optical scanning of the paper or other medium, then compiled, interpreted or otherwise processed in a suitable manner if necessary, and stored in a computer memory.

[0107] It should be understood that aspects of the application can be implemented in hardware, software, firmware or combinations thereof. In the above embodiments, various steps or methods can be implemented in software or firmware that is stored in memory and executed by a suitable instruction execution system. As such, in some embodiments, the steps or methods can be implemented in hardware such as a special purpose computer, a programmed microprocessor or microcontroller, a microprocessor-based or a microcontroller-based application-specific integrated circuit, a peripheral integrated circuit element, a digital signal processor, a highly-parallelized architecture or other similar or well-known computing devices. In other embodiments, the steps or methods can be implemented in software that is stored in a memory and executed on a suitable instruction execution system. In other embodiments, the steps or methods can be implemented in a combination of both software and hardware.

[0108] Those of skill in the art would understand that information and signals can be represented using any of a variety of technologies and techniques. For example, data, instructions, commands, information, signals, bits, symbols, and chips that can be referenced throughout the above description can be represented by voltages, currents, electromagnetic waves, magnetic fields or particles, optical fields or particles, or any combination thereof.

[0109] In addition, each function unit in each embodiment of the present application can be integrated in one processing module, or each unit can be physically present separately, or two or more units can be integrated in one module. The integrated module can be realized in the form of hardware or in the form of a software function module. When the integrated module is realized in the form of a software function module and sold or used as an independent product, it can also be stored in a computer readable storage medium.

[0110] The storage medium mentioned above can be a read-only memory, a magnetic disk or an optical disk, etc. Although the embodiments of the present application have been shown and described above, it should be understood that the above embodiments are exemplary and should not be construed as limiting the present application, and those skilled in the art can make changes, modifications, replacements and variations to the above embodiments within the scope of the present application.

[0111] It should be understood that various forms of flow shown above can be used to reorder, add or delete steps. For example, each step described in the present application can be executed in parallel, sequentially or in different order, as long as the desired results of the technical solutions of the present application can be achieved, which is not limited herein.

[0112] The above detailed description does not constitute a limitation on the scope of protection of the present application. Those skilled in the art should understand that various modifications, combinations, sub-combinations and replacements can be made according to design requirements and other factors. Any modifications, equivalent replacements and improvements within the spirit and principles of the present application should be included in the scope of protection of the present application.

Claims

1. A multi-parameter fusion-based carbon brush sparking diagnosis method for a hydroelectric generator, characterized in that, The method comprises the following steps: S1, collecting temperature data, current data, ultrasonic signal data, ultra-high frequency electromagnetic wave signal data and environmental parameter data of the carbon brush of the hydraulic generator to form a multi-dimensional original data set; S2, based on the multi-dimensional original data set, extracting the positive and negative carbon brush temperature difference, the same pole carbon brush current distribution uniformity, the ultrasonic signal effective value and the ultra-high frequency signal pulse count, and constructing a multi-parameter fusion feature vector; S3, inputting the multi-parameter fusion feature vector into an AI fault diagnosis model deployed in the cloud, the model is trained based on historical data and is used for identifying the fault root cause of carbon brush sparking and outputting a diagnosis result; S4, according to the diagnosis result, generating a structured operation and maintenance instruction containing the specific fault position, fault type and processing suggestion, and pushing it to a man-machine interaction platform for operation and maintenance personnel to execute.

2. The method of claim 1, wherein, The method for collecting temperature data, current data, ultrasonic signal data, ultra-high frequency electromagnetic wave signal data and environmental parameter data of the carbon brush of the hydraulic generator to form a multi-dimensional original data set further comprises: A DS18B20 digital temperature sensor is installed on each carbon brush holder to realize millisecond-level synchronous temperature collection; A miniature open-type Hall current sensor is sleeved on the flexible lead of each carbon brush to realize real-time acquisition of the shunt value of each carbon brush.

3. The method of claim 1, wherein, The method for extracting the positive and negative carbon brush temperature difference, the same pole carbon brush current distribution uniformity, the ultrasonic signal effective value and the ultra-high frequency signal pulse count based on the multi-dimensional original data set to construct a multi-parameter fusion feature vector further comprises: The average temperature difference ΔT of the positive carbon brush group and the negative carbon brush group is calculated, and the standard deviation of the same pole carbon brush current is calculated as the basis for judging poor contact or pressure abnormality; The ultrasonic signal is subjected to frequency spectrum analysis, the effective value and pulse count in the 20-100 kHz frequency band are extracted, and the early weak discharge phenomenon is identified.

4. The method of claim 1, wherein, The method for inputting the multi-parameter fusion feature vector into an AI fault diagnosis model deployed in the cloud, the model is trained based on historical data and is used for identifying the fault root cause of carbon brush sparking and outputting a diagnosis result further comprises: The AI fault diagnosis model uses a gradient boosting decision tree (GBDT) algorithm to classify the feature vector and identify more than 12 typical fault modes; The AI fault diagnosis model further combines a convolutional neural network (CNN) to model the time series features to improve the identification accuracy of progressive faults.

5. The method of claim 1, wherein, The method further comprises: Based on the multi-parameter fusion feature vector and historical operation data, a predictive maintenance model is used to calculate the remaining useful life (RUL) of the carbon brush, and a recommended suggestion for the next maintenance time is generated according to the RUL.

6. A multi-parameter fusion-based water turbine generator carbon brush sparking diagnosis device, characterized in that, The method comprises: A data acquisition module is used to collect temperature data, current data, ultrasonic signal data, ultra-high frequency electromagnetic wave signal data and environmental parameter data of the carbon brush of the hydraulic generator to form a multi-dimensional original data set; A feature extraction module is used to extract key features such as the positive and negative carbon brush temperature difference, the same pole carbon brush current distribution uniformity, the ultrasonic signal effective value and the ultra-high frequency signal pulse count based on the multi-dimensional original data set, and construct a multi-parameter fusion feature vector; An AI diagnosis module is configured to input the multi-parameter fusion feature vector into an AI fault diagnosis model deployed in the cloud, which is trained based on historical data to identify the root cause of carbon brush sparking and output a diagnosis result. An instruction generation and pushing module is configured to generate a structured operation and maintenance instruction including a specific fault location, a fault type and a processing suggestion according to the diagnosis result, and push the instruction to a human-computer interaction platform for an operation and maintenance personnel to execute.

7. An electronic device, comprising: The method comprises: a processor, and a memory connected to the processor in communication; the memory stores computer execution instructions; the processor executes the computer execution instructions stored in the memory to implement the method of any one of claims 1-5.

8. A computer-readable storage medium, characterized in that, The computer readable storage medium stores computer execution instructions, and the computer execution instructions are executed by the processor to implement the method of any one of claims 1-5.

9. A computer program product, characterised in that, The computer program is executed by the processor to implement the method of any one of claims 1-5.

Images