Phase modifier bearing fault early warning system and method

Abstract

The embodiment of the invention provides a phase modifier bearing fault early warning system and method. The phase modifier bearing fault early warning system comprises a data acquisition module, a data preprocessing module, a model training module and an early warning module, the data acquisition module comprises a vibration sensor and a temperature sensor, and the vibration sensor acquires vibration amplitude and vibration frequency of a bearing in x / y / z directions to obtain vibration data; the temperature sensor collects the real-time temperature of the bearing to obtain temperature data. According to the fault early warning system provided by the embodiment of the invention, the marked temperature of the time sequence mark and the multi-dimensional data structure of the vibration amplitude and the vibration frequency in the x / y / z direction are designed as the pre-processing data, and the early warning model is utilized to carry out fault prediction based on the pre-processing data, so that the early warning advance can be increased to 3-5 hours, the fault window period is completely covered, and the fault early warning efficiency is improved. And enough maintenance time is provided for operation and maintenance personnel.

Description

Technical Field

[0001] This application relates to the field of early warning technology for synchronous condenser bearings, and in particular to a system and method for early warning of synchronous condenser bearing failures. Background Technology

[0002] Synchronous condensers, also known as power grid synchronous condensers, are key devices for maintaining grid voltage stability and compensating for reactive power. As a core rotating component, the condenser bearing operates under heavy loads and at high speeds of 1500-3000 rpm for extended periods, making it prone to failure due to wear, lubrication failure, and fatigue spalling. According to grid operation and maintenance data, 35%-40% of synchronous condenser failures originate from bearing problems, and the window from initial signs of failure to complete shutdown is only 8-12 hours, making traditional early warning methods ineffective.

[0003] Currently, the mainstream bearing fault detection methods mainly include the threshold method and the manual diagnostic method. The threshold method judges faults by setting fixed thresholds for vibration amplitude and temperature, such as the "effective value of vibration velocity ≤ 6.3 mm / s" specified in GB / T11348.3-2009. However, based on this specification, changes in the operating conditions of the synchronous condenser (such as load fluctuations and changes in ambient temperature) can cause the normal threshold to drift, thus leading to the possibility of false alarms. In past statistics, the false alarm rate has been as high as 15%-20%, and the threshold method cannot identify early and weak faults. The manual diagnostic method relies on maintenance personnel to conduct regular inspections and judge the condition of the synchronous condenser bearings through methods such as auscultation and infrared thermography. However, manual diagnostics is greatly affected by the experience of maintenance personnel, and the inspection interval (usually 24-48 hours) far exceeds the fault development window, making it easy to miss the best time for treatment.

[0004] In addition, machine learning methods (such as SVM and BP neural networks) are also used for early warning of synchronous condenser bearing failures. However, although current machine learning methods can process multi-dimensional data, they have weak fitting ability for long-term dependencies in time series data. The early warning time for bearing failures is short, often less than 1 hour, which is difficult to meet the needs of power grid operation and maintenance. Summary of the Invention

[0005] The synchronous condenser bearing fault early warning system and method provided in this application can improve the early warning lead time and provide maintenance personnel with sufficient maintenance time.

[0006] In a first aspect, embodiments of this application provide a synchronous condenser bearing fault early warning system, wherein the direction parallel to the horizontal plane and perpendicular to the bearing axis is the x-direction, the direction perpendicular to the horizontal plane and perpendicular to the bearing axis is the y-direction, and the bearing axis is the z-direction. The synchronous condenser bearing fault early warning system includes: The data acquisition module includes a vibration sensor, a temperature sensor, and a data acquisition terminal. The vibration sensor is configured to acquire the vibration amplitude and vibration frequency of the bearing in the x, y, and z directions to obtain vibration data. The temperature sensor acquires the real-time temperature of the bearing to obtain temperature data. The data acquisition terminal acquires the vibration data and the temperature data as the acquired data and transmits the acquired data to the data preprocessing module. The data preprocessing module is configured to preprocess the collected data to obtain preprocessed data, which includes temperature marked according to a standardized time sequence and vibration amplitude and frequency in the x / y / z directions. The model training module constructs an early warning model for predicting the failure risk level of the bearing based on historical preprocessed data, and optimizes the early warning model based on failure data within a set time period. The failure data includes the temperature marked according to a standardized time sequence when the failure occurs, as well as the vibration amplitude and vibration frequency in the x / y / z directions. The early warning module is configured to receive the preprocessed data, call the early warning model based on the preprocessed data to perform fault prediction and output a risk level based on the prediction result, trigger an early warning based on the risk level, and store the early warning record and the processing result. The early warning record includes the risk level and the fault data corresponding to the risk level.

[0007] Optionally, the vibration sensor includes a first vibration sensor, a second vibration sensor, and a third vibration sensor. The first vibration sensor collects the vibration amplitude and vibration frequency of the bearing in the x-direction, the second vibration sensor collects the vibration amplitude and vibration frequency of the bearing in the y-direction, and the third vibration sensor collects the vibration amplitude and vibration frequency of the bearing in the z-direction. The vibration sensor is connected to the data acquisition terminal via a BNC shielded cable, and the temperature sensor is connected via a three-wire RS485 cable. The sampling frequency for the vibration amplitude is set to 5kHz, and the sampling frequencies for the temperature and vibration frequency are set to 1Hz. The data acquisition terminal packages the vibration data and temperature data every 100ms as the acquired data, and transmits the acquired data to the preprocessing module via TCP / IP protocol or Modbus RTU communication protocol.

[0008] Optionally, the data preprocessing module is specifically configured to remove outliers from the collected data, complete the missing values ​​in the collected data by using a completion model, normalize the collected data after the missing values ​​are completed to eliminate dimensional differences, and divide and label the normalized collected data according to a preset ratio to obtain the preprocessed data. The preprocessed data is labeled with temperature, vibration amplitude and vibration frequency in the x / y / z directions according to a standardized time sequence of 60 time steps per hour.

[0009] Optionally, the early warning model is an LSTM early warning model, a GRU early warning model, a TCN early warning model, or a Transformer early warning model, and the model training module includes a double hidden layer, which includes a first hidden layer and a second hidden layer. The first hidden layer comprises 32 neurons, and the second hidden layer comprises 16 neurons; or The first hidden layer comprises 24 neurons, and the second hidden layer comprises 12 neurons; or The first hidden layer comprises 40 neurons, and the second hidden layer comprises 20 neurons.

[0010] Optionally, the model training module is specifically configured to perform weight training based on at least 3000 historical preprocessed data points, using a batch size of 32, forward propagation to calculate predicted values, backpropagation to calculate gradients using the cross-entropy loss function, and batch training to obtain a batch-trained model by updating parameters using the Adam optimizer or RMSprop. The batch-trained model is then validated after each round of training based on a preset validation set. If the validation set loss does not decrease for n consecutive rounds, an early stopping mechanism is triggered, and the batch-trained model obtained from the last batch training is used as the base module. Otherwise, batch training continues until the set number of training iterations is reached to obtain the base model. The base model is then evaluated for performance. If the warning lead time is ≥3 hours, the accuracy is ≥95%, and the false alarm rate is ≤5%, the base model meets the standards and is used as the warning model. Otherwise, the number of neurons is adjusted, and batch training is repeated until the performance evaluation of the base model meets the standards, where n is 3, 4, or 5.

[0011] Optionally, the early warning module is specifically configured to call the early warning model once per minute, and perform fault prediction based on the preprocessed data of the most recent hour, and output a risk level based on the prediction result. The risk level is 0-5, where level 0 indicates that the bearing is in a normal state, levels 1-3 indicate that the bearing is in an early fault state, and the triggered early warning is to send early fault information to the maintenance personnel, and levels 4-5 indicate that the bearing is in a mid-term fault state, and the triggered early warning is to send mid-term fault information to the maintenance personnel, trigger an audible and visual alarm, and send a load reduction command to the control system of the synchronous condenser.

[0012] Secondly, embodiments of this application provide a method for early warning of bearing failure in a camera condenser. The direction parallel to the horizontal plane and perpendicular to the bearing axis is the x-direction, the direction perpendicular to the horizontal plane and perpendicular to the bearing axis is the y-direction, and the bearing's axial direction is the z-direction. The early warning method includes: Vibration data is obtained by collecting vibration amplitude and vibration frequency in the x, y, and z directions of the bearing, and real-time temperature data is obtained by collecting the bearing. The vibration data and the temperature data are used as the collected data. The collected data is preprocessed to obtain preprocessed data, which includes temperature marked according to a standardized time sequence, as well as vibration amplitude and vibration frequency in the x / y / z directions; An early warning model for predicting the failure risk level of the bearing is constructed based on historical preprocessed data, and the early warning model is optimized based on the failure data within a set time period. The failure data includes the temperature marked according to a standardized time sequence when the failure occurs, as well as the vibration amplitude and vibration frequency in the x / y / z directions. Based on the preprocessed data, the early warning model is invoked to predict faults and output a risk level based on the prediction results. An early warning is triggered based on the risk level, and the early warning record and processing results are stored. The early warning record includes the risk level and the fault data corresponding to the risk level.

[0013] Optionally, the collected data may be preprocessed to obtain preprocessed data, including: Remove outliers from the collected data; The missing values ​​in the collected data are filled by using a completion model to remove the outliers; The collected data, after being filled with missing values, is normalized to eliminate dimensional differences. The normalized collected data is divided and labeled according to a preset ratio to obtain the preprocessed data. The preprocessed data is labeled with temperature, vibration amplitude and vibration frequency in the x / y / z directions according to a standardized time sequence of 60 time steps per hour.

[0014] Optionally, an early warning model for predicting the failure risk level of the bearing is constructed based on historical preprocessed data, including: Weight training is performed based on at least 3,000 historical preprocessed data entries. A batch training model is obtained by using a batch size of 32, forward propagation to calculate the predicted value, back propagation to calculate the gradient through the cross-entropy loss function, and updating the parameters through Adam optimizer or RMSprop. The batch training model is validated based on a preset validation set after each round of training. If the loss of the validation set does not decrease for n consecutive rounds, an early stopping mechanism is triggered and the batch training model obtained in the last batch training is used as the base module. Otherwise, batch training continues until the number of training times reaches a set number to obtain the base model, where n is 3, 4 or 5. The basic model is evaluated for performance. If it meets the requirements of ≥3 hours advance warning, ≥95% accuracy, and ≤5% false alarm rate, the basic model is qualified and is used as the early warning model. Otherwise, the parameters are adjusted and batch training is performed again until the performance evaluation of the basic model is qualified.

[0015] Optionally, the preprocessed data is used to call the early warning model to predict faults and output a risk level based on the prediction results, and an early warning is triggered based on the risk level, etc. The early warning model is invoked once per minute, and fault prediction is performed based on the preprocessed data of the most recent hour. The risk level is output based on the prediction result, and the risk level is 0-5. Level 0 indicates that the bearing is in a normal state; Levels 1-3 indicate that the bearing is in an early failure state, and the triggered warning is to send early failure information to the maintenance personnel; Levels 4-5 indicate that the bearing is in a mid-term failure state, and the triggered warning is to send mid-term failure information to the maintenance personnel, trigger an audible and visual alarm, and send a load reduction command to the control system of the synchronous condenser.

[0016] The technical solution provided in this application, namely the synchronous condenser bearing fault early warning system and method, uses a multi-dimensional data structure of temperature marked with time-series markers and vibration amplitude and frequency in the x / y / z directions as preprocessed data. The early warning model uses this preprocessed data to predict faults, increasing the early warning lead time to 3-5 hours, fully covering the fault window period and providing sufficient repair time for maintenance personnel. Furthermore, the provided synchronous condenser bearing fault early warning system and method, through "tiered linkage + automatic iteration," reduces ineffective maintenance responses by 60%, shortens emergency fault handling time to 30 minutes, and maintains a long-term stable false alarm rate of ≤5%. Specifically, it designs a 0-5 level tiered early warning system, with two levels (4-5) pushing inspection reminders (triggering audible and visual alarms), and five levels (1-5) linking to the power grid to reduce load. This achieves the goal of "early detection of hidden dangers and automatic risk control in emergency faults," reducing ineffective maintenance responses by 60% and shortening emergency fault handling time from 2 hours to 30 minutes, avoiding production stoppage losses for industrial users due to untimely responses. Attached Figure Description

[0017] Figure 1 This is a structural schematic diagram of the camera bearing from a first-person perspective. Figure 2 This is a structural schematic diagram of the camera bearing from a first-person perspective. Figure 3 This is a schematic diagram of a frame for early warning of synchronous condenser bearing failure provided in an embodiment of this application; Figure 4 This is a schematic diagram of the data acquisition module in the early warning system for synchronous condenser bearing failure provided in an embodiment of this application; Figure 5 This is a flowchart illustrating a method for early warning of synchronous condenser bearing failure provided in an embodiment of this application; Figure 6 yes Figure 5 The flowchart of step S2 in the synchronous condenser bearing fault early warning method is shown. Figure 7 yes Figure 5 The flowchart of step S3 in the synchronous condenser bearing fault early warning method is shown. Figure 8 yes Figure 5 The flowchart shown is a schematic diagram of step S4 in the synchronous condenser bearing fault early warning method.

[0018] Figure label: 1-Fault early warning system; 11-Data acquisition module; 111-Vibration sensor; 1111-First vibration sensor; 1112-Second vibration sensor; 1113-Third vibration sensor; 123-Temperature sensor; 113-Data acquisition terminal; 2-Power grid server; 3-Control system for synchronous condensers; 4-Maintenance terminal. Detailed Implementation

[0019] The existing power synchronous condenser bearing fault early warning technology has three major technical problems: poor early warning timeliness, low accuracy, and weak adaptability. The specific problems and their causes are as follows.

[0020] Poor early warning timeliness leads to the inability to detect early faults. The main problem is that traditional methods can only issue alarms when bearing fault characteristics are significant (such as vibration amplitude far exceeding the threshold or a sharp rise in temperature), leaving maintenance personnel less than 2 hours, or even less than 1 hour, to handle the fault. This can easily lead to synchronous condenser shutdowns, grid voltage fluctuations, or even large-scale power outages. Different fault prediction methods cause this problem for different reasons. Threshold methods rely on fixed standards (such as the vibration threshold specified in GB / T11348.3-2009), while the characteristic signals of early bearing faults (such as slight wear or minor lubrication failure) are weak and do not meet the threshold standard, making them unidentifiable. Manual diagnosis relies on inspection intervals of 24-48 hours, which far exceeds the 8-12 hour fault window period of "early hidden danger - shutdown failure" for bearings, easily missing the best time for handling. Traditional machine learning methods (SVM, BP neural network) have weak fitting ability for long-term dependencies in time series data and can only capture short-term fault characteristics, with an early warning lead time of less than 1 hour.

[0021] Low accuracy in early warning systems, i.e., a high false alarm rate, easily leads to a waste of operational resources. The main problem is that single-parameter monitoring (such as monitoring only vibration or temperature) cannot distinguish between "normal operating condition fluctuations" and "fault signals," with a false alarm rate as high as 15%-20%. This results in frequent and ineffective inspections by maintenance personnel, wasting human and material resources. The main reason for this problem is the complex operating conditions of synchronous condensers (load fluctuation range ±20%, ambient temperature change 5-15℃). The normal range of a single parameter drifts with the operating conditions, and fixed thresholds or single-feature models are difficult to adapt, easily misjudging normal fluctuations as faults. Furthermore, traditional machine learning methods do not fully integrate multi-dimensional features (such as X / Y / Z three-dimensional data of vibration, temperature data), and cannot filter interference signals through data complementarity, resulting in low recognition accuracy.

[0022] The problem stems from the poor adaptability of traditional fault prediction methods. Traditional methods cannot adapt to varying operating conditions and require frequent manual adjustments of thresholds or model parameters, leading to high complexity and cost. The root causes of this problem differ between fault prediction methods. Threshold methods rely on fixed standards that cannot dynamically adjust to changing operating conditions. For example, under high loads, the normal vibration amplitude of a bearing is higher than under low loads; using the same threshold can easily result in false alarms under high loads and missed alarms under low loads. Traditional machine learning methods use fixed model parameters that do not consider the impact of different operating conditions on fault characteristics. For instance, the temperature rise rate of a bearing differs between low-temperature and high-temperature environments, making it impossible for the model to adaptively identify faults, requiring manual retraining or parameter adjustment.

[0023] To address the aforementioned issues, this application provides a synchronous condenser bearing fault early warning system and method. By designing a multi-dimensional data structure of temperature of time-series markers and vibration amplitude and frequency in the x / y / z directions as preprocessed data, and using an early warning model to predict faults based on the preprocessed data, the early warning lead time can be increased to 3-5 hours, fully covering the fault window period, and providing sufficient maintenance time for operation and maintenance personnel, effectively solving the above-mentioned technical problems.

[0024] The present application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0025] This embodiment provides a phase shifter bearing fault early warning system. Figure 1 This is a structural schematic diagram of the camera bearing from a first-person perspective. Figure 2 This is a structural diagram of the first-view camera bearing adjustment mechanism. (For example...) Figure 1 and Figure 2 As shown, the direction parallel to the horizontal plane and perpendicular to the bearing axis is the x-direction, the direction perpendicular to the horizontal plane and perpendicular to the bearing axis is the y-direction, and the bearing axis is the z-direction.

[0026] Figure 3 This is a schematic diagram of a frame for early warning of synchronous condenser bearing failure provided in an embodiment of this application. Figure 3 As shown, the fault early warning system includes a data acquisition module, a data preprocessing module, a model training module, and an early warning module.

[0027] The synchronous condenser bearing fault early warning system provided in this embodiment is implemented through a power grid server. The power grid server used can be a Dell PowerEdge R750, which has high performance and supports model training and real-time inference. The maintenance personnel use industrial tablet computers, such as Advantech industrial tablet computers, which are adaptable to the harsh environment of substations.

[0028] Figure 4This is a schematic diagram of the data acquisition module in the synchronous condenser bearing fault early warning system provided in this application embodiment. Figure 3 and Figure 4 As shown, in the fault early warning system provided in this embodiment, the data acquisition module includes a vibration sensor, a temperature sensor, and a data acquisition terminal. The vibration sensor is configured to acquire the vibration amplitude and vibration frequency of the bearing in the x, y, and z directions to obtain vibration data. The temperature sensor acquires the real-time temperature of the bearing to obtain temperature data. The data acquisition terminal acquires the vibration data and temperature data as acquisition data and transmits the acquisition data to the data preprocessing module.

[0029] like Figure 4 As shown, the vibration sensor includes a first vibration sensor, a second vibration sensor, and a third vibration sensor. The first vibration sensor collects the vibration amplitude and frequency of the bearing in the x-direction, the second vibration sensor collects the vibration amplitude and frequency of the bearing in the y-direction, and the third vibration sensor collects the vibration amplitude and frequency of the bearing in the z-direction.

[0030] Specifically, the vibration sensor is connected to the data acquisition terminal via a BNC shielded cable, and the temperature sensor is connected via a three-wire RS485 cable. This connection method provides good anti-interference characteristics. The sampling frequency for vibration amplitude is set to 5kHz, and the sampling frequencies for temperature and vibration frequency are set to 1Hz. The data acquisition terminal packages vibration and temperature data every 100ms as acquired data and transmits the acquired data to the preprocessing module via TCP / IP or Modbus RTU communication protocol.

[0031] Specifically, the vibration sensor is mounted on the bearing housing of the camera condenser. The vibration sensor is a piezoelectric vibration sensor. In an optional embodiment, the vibration sensor is a PCB356A15 type, which has the advantages of high accuracy, anti-interference, and the ability to capture minute vibrations.

[0032] Specifically, the temperature sensor is installed on the end cover of the camera condenser bearing. The temperature sensor is a platinum resistance temperature sensor. In an optional embodiment, the temperature sensor is a PT100 temperature sensor. This type of temperature sensor is a Class A temperature sensor with an accuracy of ±0.1℃ and the advantage of fast response.

[0033] Specifically, the data acquisition terminal can be the NICDAQ-9178, which has the advantages of multi-channel compatibility and high sampling rate.

[0034] like Figure 3As shown, in the synchronous condenser bearing fault early warning system provided in this embodiment, the data preprocessing module is configured to preprocess the collected data to obtain preprocessed data. The preprocessed data includes temperature marked according to a standardized time sequence, as well as vibration amplitude and frequency in the x / y / z directions. Specifically, the preprocessing of the collected data adopts a "time step × 5-dimensional feature" structure, where the time step is the standardized time sequence, for example, marked once per minute (60 time steps per hour), and the 5-dimensional features refer to temperature, vibration frequency, and vibration amplitude in the x / y / z directions.

[0035] like Figure 3 As shown, in the synchronous condenser bearing fault early warning system provided in this embodiment, the model training module constructs an early warning model for predicting the fault risk level of the bearing based on historical preprocessed data, and optimizes the early warning model based on fault data within a set time period. The fault data includes the temperature marked according to a standardized time sequence, as well as the vibration amplitude and vibration frequency in the x / y / z directions when the fault occurred. Specifically, the set time period can be several months, such as 1-3 months. In a specific embodiment, the early warning model is optimized using fault data from the most recent 3 months.

[0036] like Figure 3 As shown, in the synchronous condenser bearing fault early warning system provided in this embodiment, the early warning module is configured to receive preprocessed data, call the early warning model based on the preprocessed data to perform fault prediction and output the risk level based on the prediction result, trigger an early warning based on the risk level, and store the early warning record and processing result. The early warning record includes the risk level and the fault data corresponding to the risk level.

[0037] The synchronous condenser bearing fault early warning system provided in this embodiment uses a multi-dimensional data structure of temperature marked by a time sequence marker, as well as vibration amplitude and frequency in the x / y / z directions, as preprocessed data. The early warning model uses the preprocessed data to predict faults, which can increase the early warning lead time to 3-5 hours, fully covering the fault window period and providing maintenance personnel with sufficient maintenance time.

[0038] like Figure 3 As shown, in the synchronous condenser bearing fault early warning system provided in this embodiment, the data preprocessing module is specifically configured to remove outliers from the collected data. It uses a completion model to fill in missing values ​​in the collected data after removing outliers. The collected data after missing value completion is then normalized to eliminate dimensional differences. The normalized collected data is then divided and labeled according to a preset ratio to obtain preprocessed data. The preprocessed data is labeled with temperature and vibration amplitude and frequency in the x / y / z directions according to a standardized time sequence of 60 time steps per hour. Specifically, the data preprocessing module can use Python software.

[0039] In one specific embodiment, the data preprocessing module is specifically configured to remove outliers from the collected data based on the Python Pandas library. The specific process is to first remove extreme values ​​caused by electromagnetic interference (such as a sudden surge in vibration amplitude to 10 mm / s) by using the 3σ principle, and then remove "slow drift anomalies" (such as a temperature exceeding the normal range by 5% for 10 consecutive minutes) by judging the time series trend, so as to ensure the authenticity of the data.

[0040] In one specific embodiment, the data preprocessing module is specifically configured to fill in missing values ​​in the collected data by removing outliers based on a lightweight TensorFlow LSTM model. Specifically, for short-term missing values ​​(≤5 sampling points), the trained lightweight LSTM model is used to predict and fill in the missing values ​​(based on 100 normal time-series points before and after), with an error of ≤3%. For long-term missing values ​​(>5 points), linear interpolation is performed first, and then a sliding window is used to correct the missing values ​​after subsequent data collection to avoid data breakage.

[0041] In a specific embodiment, the data preprocessing module is specifically configured to use the Python NumPy library to normalize the collected data after missing values ​​are filled in in order to eliminate dimensional differences. Specifically, the Min-Max normalization is used to map the vibration amplitude (unit: mm / s), temperature (unit: ℃), and vibration frequency (unit: kHz) to the [0,1] interval to eliminate dimensional differences. In the formula, x_min / x_max is obtained by statistically analyzing 6 months of historical normal data to ensure the stability of the normalization standard.

[0042] In one specific embodiment, the data preprocessing module is specifically configured to divide and label the normalized collected data according to a preset ratio using the Python Scikit-learn library to obtain preprocessed data. The preprocessed data is labeled with temperature, vibration amplitude, and vibration frequency in the x / y / z directions according to a standardized time sequence of 60 time steps per hour. The preset ratio is 7:2:1, 8:1:1, or 6:3:1. Combined with the synchronous condenser maintenance records, three levels of labels are added (e.g., 12-6 hours before bearing replacement is labeled as level 1-3, and 6-0 hours before is labeled as level 4-5).

[0043] like Figure 3As shown, in the synchronous condenser bearing fault early warning system provided in this embodiment, the early warning model is an LSTM early warning model, a GRU early warning model, a TCN early warning model, or a Transformer early warning model. The model training module includes a double hidden layer, which includes a first hidden layer and a second hidden layer. The first hidden layer includes 32 neurons, and the second hidden layer includes 16 neurons; or the first hidden layer includes 24 neurons, and the second hidden layer includes 12 neurons; or the first hidden layer includes 40 neurons, and the second hidden layer includes 20 neurons.

[0044] like Figure 3 As shown, in the early warning system for camera condenser bearing faults provided in this embodiment, the model training module is specifically configured to perform weight training based on at least 3000 historical preprocessed data, using a batch size of 32, forward propagation to calculate predicted values, backpropagation to calculate gradients using the cross-entropy loss function, and batch training to obtain a batch-trained model by updating parameters using the Adam optimizer or RMSprop. The batch-trained model is then validated based on a preset validation set after each round of training. If the validation set loss does not decrease for n consecutive rounds, an early stopping mechanism is triggered, and the batch-trained model obtained from the last batch training is used as the base module. Otherwise, batch training continues until the number of training iterations reaches the set number to obtain the base model. The performance of the base model is then evaluated. If the early warning lead time is ≥3 hours, the accuracy is ≥95%, and the false alarm rate is ≤5%, the base model meets the standards and is used as the early warning model. Otherwise, the number of neurons is adjusted, and batch training is repeated until the performance evaluation of the base model meets the standards, where n is 3, 4, or 5.

[0045] Specifically, the validation set is a pre-set dataset with data in the same format as the preprocessed data. The data in the validation set includes data on bearings in different states, so as to more comprehensively verify the performance of the early warning model.

[0046] Specifically, taking the LSTM early warning model as an example, the model training module is specifically configured to train weights based on at least 5000 historical preprocessed data points, using a batch size of 32. Forward propagation calculates the predicted value, backpropagation calculates the gradient using the cross-entropy loss function, and parameter updates are performed through the Adam optimizer to obtain a batch-trained model. The batch-trained model is then validated after each round of training based on a preset validation set. If the validation set loss does not decrease for five consecutive rounds, an early stopping mechanism is triggered, and the batch-trained model obtained from the last batch training is used as the base module. Otherwise, batch training continues until the set number of training iterations is reached to obtain the base model. The performance of the base model is then evaluated. If the early warning lead time is ≥3 hours, the accuracy is ≥95%, and the false alarm rate is ≤5%, the base model meets the standards and is used as the early warning model. Otherwise, the number of neurons is adjusted, and batch training is repeated until the performance evaluation of the base model meets the standards.

[0047] Existing technologies use BP neural networks (single hidden layer, static point input, no temporal dimension), which lack early stopping mechanisms during training (prone to overfitting) and have slow convergence. The fault warning system provided in this embodiment improves the warning lead time from 0.8-1.2 hours to 3-5 hours through a "double hidden layer LSTM + temporal input + training optimization" approach, with a warning model accuracy of ≥95% and a 15% reduction in overfitting rate. Furthermore, the model training module obtains weights through pre-training, which helps reduce the number of training iterations and improve training efficiency.

[0048] like Figure 3 As shown, in the synchronous condenser bearing fault early warning system provided in this embodiment, the early warning module is specifically configured to call the early warning model once per minute, and perform fault prediction based on the preprocessed data of the most recent hour and output the risk level based on the prediction result. The risk level is 0-5, where level 0 indicates that the bearing is in a normal state, levels 1-3 indicate that the bearing is in an early fault state, and the triggered early warning is to send early fault information to the operation and maintenance personnel. Levels 4-5 indicate that the bearing is in a mid-term fault state, and the triggered early warning is to send mid-term fault information to the operation and maintenance personnel, trigger an audible and visual alarm, and send a load reduction command to the synchronous condenser control system.

[0049] Specifically, level 0 indicates that the bearing is in a normal state, and only logs are recorded to the MySQL database. It should be noted that levels 1-5 represent fault states, and fault data also needs to be stored in the MySQL database.

[0050] Specifically, the audible and visual alarm can be implemented using a red indicator light and a buzzer.

[0051] Specifically, sending a load reduction command to the synchronous condenser's control system can be done through the IEC 61850 protocol, which is linked to the power grid's SCADA system. In one optional embodiment, the IEC 61850 protocol is used to link the power grid's SCADA system to output a "reduce load by 30%" command to the synchronous condenser's control system. Furthermore, in addition to sending load reduction commands to the synchronous condenser's control system when the risk level is 4-5, load reduction commands can also be sent when the risk level is 1-3. That is, regardless of whether it's an early fault condition or a mid-term warning condition, the power grid is linked to reduce load.

[0052] Specifically, by integrating SMS API to send early and mid-term fault information to maintenance personnel, early warnings can be pushed to their mobile phones within 10 seconds, enabling them to promptly detect the fault status of the synchronous condenser bearing and carry out timely repairs.

[0053] Existing technologies only offer "binary early warning (normal / fault)" without grid linkage (emergency faults cannot be controlled), and model parameters are fixed (false alarm rate rises to 12%+ after 6 months). The synchronous condenser bearing fault early warning system provided in this embodiment reduces ineffective maintenance responses by 60% and shortens emergency fault handling time to 30 minutes through "tiered linkage + automatic iteration." The model's false alarm rate remains stable at ≤5% over the long term. Specifically, it is designed with 0-5 levels of tiered early warning, with 2 levels (levels 4-5) pushing inspection reminders (triggering audible and visual alarms), and 5 levels (levels 1-5) linking with the grid to reduce load. This achieves the goal of "early detection of hidden dangers and automatic control of emergency faults," reducing ineffective maintenance responses by 60% and shortening emergency fault handling time from 2 hours to 30 minutes, thus avoiding production stoppage losses for industrial users due to untimely responses.

[0054] Based on the same inventive concept, this application also provides a method for early warning of camera bearing failure, wherein the direction parallel to the horizontal plane and perpendicular to the bearing axis is the x-direction, the direction perpendicular to the horizontal plane and perpendicular to the bearing axis is the y-direction, and the bearing axis is the z-direction. Figure 5 This is a flowchart illustrating a method for early warning of synchronous condenser bearing failure provided in an embodiment of this application. Figure 5 As shown, the synchronous condenser bearing fault early warning method provided in this embodiment includes: S1: Collect vibration amplitude and frequency data of the bearing in the x, y, and z directions to obtain vibration data, and collect real-time temperature data of the bearing to obtain temperature data. Use the vibration data and temperature data as the collected data.

[0055] In the synchronous condenser bearing fault early warning method provided in this embodiment, step S1 is implemented by the data acquisition module of the above-mentioned fault early warning system. The data acquisition module includes a vibration sensor, a temperature sensor and a data acquisition terminal. The vibration sensor is configured to acquire the vibration amplitude and vibration frequency of the bearing in the x, y and z directions to obtain vibration data. The temperature sensor acquires the real-time temperature of the bearing to obtain temperature data. The data acquisition terminal acquires the vibration data and temperature data as acquisition data and transmits the acquisition data to the data preprocessing module.

[0056] Specifically, the sampling frequency for vibration amplitude is set to 5kHz, and the sampling frequency for temperature and vibration frequency is set to 1Hz. The data acquisition terminal packages vibration data and temperature data every 100ms as the acquired data, and transmits the acquired data to the preprocessing module via TCP / IP protocol or Modbus RTU communication protocol.

[0057] S2: Preprocess the collected data to obtain preprocessed data, which includes temperature tagged according to a standardized time sequence, as well as vibration amplitude and frequency in the x / y / z directions. Specifically, the preprocessing of the collected data adopts a "time step × 5-dimensional feature" structure, where the time step is the standardized time sequence, for example, tagged once per minute (60 time steps per hour), and the 5-dimensional features refer to temperature, vibration frequency, and vibration amplitude in the x / y / z directions.

[0058] S3: Construct an early warning model for predicting bearing failure risk levels based on historical preprocessed data, and optimize the early warning model based on failure data within a set time period. The failure data includes the temperature marked according to a standardized time sequence, as well as the vibration amplitude and frequency in the x / y / z directions when the failure occurred. Specifically, the set time period can be several months, such as 1-5 months. In a specific embodiment, the early warning model is optimized using failure data from the most recent 3 months.

[0059] S4: Based on the preprocessed data, call the early warning model to predict the fault and output the risk level based on the prediction result. Trigger the early warning based on the risk level and store the early warning record and processing result. The early warning record includes the risk level and the fault data corresponding to the risk level.

[0060] The synchronous condenser bearing fault early warning method provided in this embodiment uses a multi-dimensional data structure of temperature of the time-series markers and vibration amplitude and frequency in the x / y / z directions as preprocessed data. The early warning model uses the preprocessed data to predict the fault, which can increase the early warning lead time to 3-5 hours, fully cover the fault window period, and provide maintenance personnel with sufficient maintenance time.

[0061] Figure 6yes Figure 5 The flowchart illustrates step S2 in the synchronous condenser bearing fault early warning method shown. Figure 6 As shown, in the synchronous condenser bearing fault early warning method provided in this embodiment, step S2 includes: S21: Remove outliers from the collected data.

[0062] Optionally, step S21 specifically includes: removing outliers from the collected data based on the Python Pandas library. The specific process is to first remove extreme values ​​caused by electromagnetic interference (such as a sudden surge in vibration amplitude to 10 mm / s) using the 3σ principle, and then remove "slow drift anomalies" (such as a temperature exceeding the normal range by 5% for 10 consecutive minutes) by judging the time series trend, so as to ensure the authenticity of the data.

[0063] Optionally, step S21 may also specifically include: "3σ + time series trend" can be replaced by DBSCAN clustering (to identify time series outliers) or moving average + 3 times the deviation threshold (three consecutive points exceeding the threshold are judged as abnormal) to remove outliers.

[0064] S22: The missing values ​​in the collected data are filled in by using the completion model to remove outliers.

[0065] Optionally, step S22 specifically includes: using a lightweight TensorFlow LSTM model to fill in missing values ​​in the collected data when removing outliers. Specifically, for short-term missing values ​​(≤5 sampling points), the trained lightweight LSTM model is used to predict and fill in the missing values ​​(based on 100 normal time-series points before and after), with an error of ≤3%; for long-term missing values ​​(>5 points), linear interpolation is performed first, and then a sliding window is used to correct the missing values ​​after subsequent data collection to avoid data breakage.

[0066] Optionally, step S22 may also specifically include: replacing the lightweight LSTM model with a GRU model (fewer gates, faster training) or an ARIMA model (statistical time series prediction, adapted to stationary data) to complete the missing values.

[0067] S23: Normalize the collected data after filling in missing values ​​to eliminate dimensional differences.

[0068] Optionally, step S23 specifically includes: using the Python NumPy library to normalize the collected data after missing values ​​are filled in in order to eliminate dimensional differences. Specifically, the process is to use Min-Max normalization to map the vibration amplitude (unit: mm / s), temperature (unit: ℃), and vibration frequency (unit: kHz) to the [0,1] interval to eliminate dimensional differences. In the formula, x_min / x_max is obtained by statistically analyzing 6 months of historical normal data to ensure that the normalization standard is stable.

[0069] S24: Divide and label the normalized collected data according to a preset ratio to obtain preprocessed data. The preprocessed data is labeled with temperature, vibration amplitude and vibration frequency in the x / y / z directions according to a standardized time sequence of 60 time steps per hour.

[0070] Optionally, the normalized collected data is divided and labeled according to a preset ratio using the Python Scikit-learn library to obtain preprocessed data. The preprocessed data is labeled with temperature, vibration amplitude and frequency in the x / y / z directions according to a standardized time sequence of 60 time steps per hour. The preset ratio is 7:2:1, 8:1:1 or 6:3:1. Combined with the synchronous condenser maintenance records, three levels of labels are added (e.g., 12-6 hours before bearing replacement is labeled as level 1-3, and 6-0 hours before is labeled as level 4-5).

[0071] Figure 7 yes Figure 5 The flowchart shown is a schematic diagram of step S3 in the synchronous condenser bearing fault early warning method. Figure 7 As shown, in the synchronous condenser bearing fault early warning method provided in this embodiment, step S3 includes: S31: Weight training is performed based on at least 3000 historical preprocessed data.

[0072] Specifically, taking the LSTM early warning model as an example, step S31 specifically includes: weight training based on at least 5000 historical preprocessed data.

[0073] Specifically, taking the RMSprop early warning model as an example, step S31 specifically includes: weight training based on 3000 historical preprocessed data.

[0074] S32: A batch training model is obtained by using a batch size of 32, forward propagation to calculate the predicted value, back propagation to calculate the gradient through the cross-entropy loss function, and updating the parameters through the Adam optimizer or RMSprop.

[0075] S33: Validate the batch-trained model after each round of training based on the preset validation set. If the validation set loss does not decrease for n consecutive rounds, the early stopping mechanism is triggered and the batch-trained model obtained from the last batch training is used as the base module. Otherwise, continue batch training until the number of training times reaches the set number to obtain the base model. n is 3, 4 or 5.

[0076] Specifically, taking the LSTM early warning model as an example, step S31 specifically includes: verifying the batch training model after each round of training based on the preset validation set. If the validation set loss does not decrease for 5 consecutive rounds, the early stopping mechanism is triggered and the batch training model obtained in the last batch training is used as the basic module. Otherwise, batch training continues until the number of training times reaches the set number to obtain the basic model.

[0077] Specifically, taking the RMSprop early warning model as an example, step S33 specifically includes: the early stop mechanism can be replaced with "stop if the accuracy of the validation set does not increase for 3 consecutive rounds".

[0078] S34: Evaluate the performance of the base model. If it meets the requirements of ≥3 hours advance warning, ≥95% accuracy, and ≤5% false alarm rate, the base model is considered qualified and used as the early warning model. Otherwise, adjust the parameters and retrain the model in batches until the performance evaluation of the base model is qualified. Specifically, adjusting the parameters can involve increasing the number of neurons, for example, from 32 to 48.

[0079] Figure 8 yes Figure 5 The flowchart illustrates step S4 in the synchronous condenser bearing fault early warning method shown. Figure 8 As shown, in the synchronous condenser bearing fault early warning method provided in this embodiment, step S4 includes: S41: Call the early warning model once per minute, perform fault prediction based on preprocessed data from the most recent hour, and output the risk level based on the prediction results. The risk level is 0-5.

[0080] S42: Level 0 indicates the bearing is in normal condition; Levels 1-3 indicate the bearing is in an early failure state, and the triggered warning is to send early failure information to maintenance personnel; Levels 4-5 indicate the bearing is in a mid-term failure state, and the triggered warning is to send mid-term failure information to maintenance personnel, trigger an audible and visual alarm, and send a load reduction command to the synchronous condenser control system.

[0081] Specifically, the audible and visual alarm can be implemented using a red indicator light and a buzzer.

[0082] Specifically, sending a load reduction command to the synchronous condenser's control system can be done through the IEC 61850 protocol, which is linked to the power grid's SCADA system. In one optional embodiment, the IEC 61850 protocol is used to link the power grid's SCADA system to output a "reduce load by 30%" command to the synchronous condenser's control system. Furthermore, in addition to sending load reduction commands to the synchronous condenser's control system when the risk level is 4-5, load reduction commands can also be sent when the risk level is 1-3. That is, regardless of whether it's an early fault condition or a mid-term warning condition, the power grid is linked to reduce load.

[0083] Specifically, by integrating an SMS API, early and mid-stage fault information can be sent to maintenance personnel. This allows for alerts to be pushed to their mobile phones within 10 seconds, enabling them to promptly detect faults in the synchronous condenser bearing and perform timely repairs. It should be noted that the SMS API integration method can be replaced with an industrial-grade SMS modem (locally deployed, no platform dependency).

[0084] The synchronous condenser bearing fault early warning method provided in this embodiment is designed with a 0-5 level graded early warning system. Among them, two levels (levels 4-5) push inspection reminders (triggering audible and visual alarms), and five levels (levels 1-5) link the power grid to reduce load. This achieves the goal of "early detection of hidden dangers and automatic control of emergency faults", reducing ineffective operation and maintenance responses by 60% and shortening emergency fault handling time from 2 hours to 30 minutes, thus avoiding production stoppage losses for industrial users due to untimely responses.

[0085] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0086] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented using software plus a general-purpose hardware platform, or of course, using hardware. Based on this understanding, the above technical solutions, in essence or the parts that contribute to the related technology, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0087] The above embodiments are merely illustrative examples and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations. However, obvious variations or modifications derived therefrom are still within the scope of protection of this application.

Claims

1. A phase modulator bearing failure warning system, a direction parallel to a horizontal plane and perpendicular to a bearing axis direction being an x direction, a direction perpendicular to the horizontal plane and perpendicular to the bearing axis direction being a y direction, an axial direction of the bearing being a z direction, characterized by, The fault early warning system comprises: A data acquisition module comprising a vibration sensor, a temperature sensor and a data acquisition terminal, the vibration sensor is configured to collect the vibration amplitude and vibration frequency of the bearing in x, y and z directions to obtain vibration data, the temperature sensor collects the real-time temperature of the bearing to obtain temperature data, and the data acquisition terminal acquires the vibration data and the temperature data as the acquisition data and transmits the acquisition data to a data preprocessing module; The data preprocessing module is configured to preprocess the acquisition data to obtain preprocessing data, the preprocessing data comprising temperature and vibration amplitude and vibration frequency in x / y / z directions marked by standardized time sequence; A model training module, based on historical preprocessing data, constructs an early warning model for predicting the fault risk level of the bearing, and optimizes the early warning model based on fault data in a set time period, the fault data comprising temperature and vibration amplitude and vibration frequency in x / y / z directions marked by standardized time sequence; An early warning module configured to receive the preprocessing data, call the early warning model based on the preprocessing data to predict the fault and output the risk level based on the prediction result, trigger the early warning based on the risk level, and store the early warning record and the processing result, the early warning record comprising the risk level and the fault data corresponding to the risk level.

2. The pre-failure warning system for a phase-modulator bearing as claimed in claim 1, wherein, The vibration sensor comprises a first vibration sensor, a second vibration sensor and a third vibration sensor, the first vibration sensor collects the vibration amplitude and vibration frequency of the bearing in x direction, the second vibration sensor collects the vibration amplitude and vibration frequency of the bearing in y direction, and the third vibration sensor collects the vibration amplitude and vibration frequency of the bearing in z direction; The vibration sensor is connected to the data acquisition terminal through a BNC interface shielding cable, the temperature sensor is connected through a three-wire RS485 cable, the sampling frequency of the vibration amplitude is set to 5 kHz, the sampling frequency of the temperature and the vibration frequency is set to 1 Hz, the data acquisition terminal packs the vibration data and the temperature data once every 100 ms as the acquisition data, and transmits the acquisition data to the preprocessing module through TCP / IP protocol or ModbusRTU communication protocol.

3. The pre-failure warning system for a phase-modulator bearing as claimed in claim 1, wherein, The data preprocessing module is specifically configured to eliminate outliers in the acquisition data, complete missing values in the acquisition data after eliminating the outliers through a completion model, eliminate dimension differences based on the acquisition data after completing the missing values, divide and label the acquisition data after normalization processing based on a preset ratio to obtain the preprocessing data, and the preprocessing data is temperature and vibration amplitude and vibration frequency in x / y / z directions marked by standardized time sequence with 60 time steps per hour.

4. The system for early warning of failures according to claim 3, characterized in that, The early warning model is an LSTM early warning model, a GRU early warning model, a TCN early warning model or a Transformer early warning model, the model training module comprises double hidden layers, and the double hidden layers comprise a first hidden layer and a second hidden layer; The first hidden layer comprises 32 neurons, and the second hidden layer comprises 16 neurons; or The first hidden layer comprises 24 neurons, and the second hidden layer comprises 12 neurons; or The first hidden layer comprises 40 neurons, and the second hidden layer comprises 20 neurons.

5. The pre-failure warning system for a phase-modulator bearing as claimed in claim 4, wherein, The model training module is specifically configured to perform weight training based on at least 3000 pieces of historical pretreatment data, adopt a batch size of 32, calculate a predicted value through forward propagation, calculate a gradient through a cross-entropy loss function through back propagation, and update parameters through an Adam optimizer or RMSprop to perform batch training to obtain a batch training model, verify the batch training model after each round of training based on a preset verification set, trigger an early stop mechanism if the verification set loss does not decrease for n consecutive rounds, and use the batch training model obtained through the last batch training as the base module, otherwise continue batch training until the number of training reaches a set number to obtain the base model, perform performance evaluation on the base model, and if the early warning lead time is greater than or equal to 3 hours, the accuracy is greater than or equal to 95%, and the false alarm rate is less than or equal to 5%, the base model meets the standard and the base model is used as the early warning model, otherwise the number of neurons is adjusted to re-perform batch training until the performance evaluation of the base model meets the standard, wherein n is 3, 4 or 5.

6. The pre-failure warning system for a phase-modulator bearing as claimed in claim 1, wherein, The early warning module is specifically configured to call the early warning model once every minute, perform fault prediction based on the pretreatment data in the last 1 hour, and output a risk level based on the prediction result, wherein the risk level is 0-5 levels, 0 level means that the bearing is in a normal state, 1-3 levels mean that the bearing is in an early fault state, the triggered early warning is to send early fault information to an operation and maintenance personnel, 4-5 levels mean that the bearing is in a medium-term fault, the triggered early warning is to send medium-term fault information to the operation and maintenance personnel, trigger a sound-light alarm, and send a load reduction instruction to a control system of the phase modifier.

7. A method for early warning of bearing failure in a phase shifter, wherein the direction parallel to the horizontal plane and perpendicular to the bearing axis is the x-direction, the direction perpendicular to the horizontal plane and perpendicular to the bearing axis is the y-direction, and the bearing axis is the z-direction, characterized in that... The fault early warning method comprises: Collecting vibration amplitudes and vibration frequencies of the bearing in x, y and z directions to obtain vibration data, collecting real-time temperature of the bearing to obtain temperature data, and using the vibration data and the temperature data as the collected data; Pretreating the collected data to obtain pretreatment data, wherein the pretreatment data comprises temperature and vibration amplitudes and vibration frequencies of x / y / z directions marked in a standardized time sequence; Constructing an early warning model for predicting a fault risk level of the bearing based on historical pretreatment data, and optimizing the early warning model based on fault data in a set time period, wherein the fault data comprises temperature and vibration amplitudes and vibration frequencies of x / y / z directions marked in a standardized time sequence; Based on the preprocessed data, the early warning model is called for fault prediction, and a risk level is output based on the prediction result, and early warning is triggered based on the risk level, and early warning records and processing results are stored, the early warning records including the risk level and the fault data corresponding to the risk level.

8. The method of claim 7, wherein the step of determining the bearing fault is performed by a phase modulator bearing fault early warning system. The collected data is preprocessed to obtain preprocessed data, including: Removing outliers in the collected data; Completing missing values in the collected data removed from the outliers by a completion model; Normalizing the collected data after missing value completion to eliminate dimensional differences; The normalized collected data is divided and labeled according to a preset ratio to obtain the preprocessed data, which is marked with temperature and vibration amplitude and frequency in x / y / z directions for every hour with 60 time steps.

9. The method of claim 8, wherein the step of determining the bearing fault is performed by a phase modulator bearing fault early warning system. Based on historical preprocessed data, an early warning model for predicting the risk level of the bearing is constructed, including: Based on at least 3000 pieces of historical preprocessed data, weight training is performed; Batch training is performed in the manner of batch size of 32, forward propagation calculation of prediction value, backward propagation calculation of gradient through cross-entropy loss function, and parameter update through Adam optimizer or RMSprop to obtain a batch training model; After each round of training, the batch training model is verified based on a preset validation set, if the validation set loss does not decrease for n consecutive rounds, an early stopping mechanism is triggered, and the batch training model obtained after the last batch training is taken as the base module, otherwise batch training is continued until the number of training reaches a set number, and the base model is obtained, n is 3, 4 or 5; The performance of the base model is evaluated, if the early warning lead time ≥ 3 hours, the accuracy rate ≥ 95%, and the false alarm rate ≤ 5%, the base model meets the standard, and the base model is taken as the early warning model, otherwise the parameters are adjusted and batch training is performed again until the performance evaluation of the base model meets the standard.

10. The method of claim 9, wherein the method further comprises: Based on the preprocessed data, the early warning model is called for fault prediction, and a risk level is output based on the prediction result, and early warning is triggered based on the risk level: The early warning model is called once a minute, and fault prediction is performed based on the preprocessed data in the last 1 hour, and a risk level is output based on the prediction result, the risk level being 0-5; Level 0 means that the bearing is in a normal state, levels 1-3 mean that the bearing is in an early fault state, the triggered early warning is to send early fault information to the operation and maintenance personnel, levels 4-5 mean that the bearing is in a medium-term fault, the triggered early warning is to send medium-term fault information to the operation and maintenance personnel, trigger sound and light alarm, and send a load reduction instruction to the control system of the phase modifier.

Images