Health monitoring platform for intelligently predicting and maintaining magnetic suspension flywheel energy storage system

Abstract

The invention discloses a health monitoring platform for intelligent prediction and maintenance of a magnetic suspension flywheel energy storage system, and the platform comprises a multi-source sensor module which is provided with a plurality of sensors in a distributed manner for collecting multi-modal signals, and the multi-modal signals are transmitted through a redundant bus; the adaptive data acquisition module performs amplitude normalization and noise suppression on the signals; the embedded data processing module carries an FPGA chip, processes vibration, temperature and current signals by using methods such as empirical mode decomposition and extracts features; the health state evaluation module integrates a deep belief network model to generate a comprehensive health index HI and predicts the decline trend of the comprehensive health index HI; and the decision-making module performs early warning based on a fuzzy logic decision-making tree, and generates a maintenance scheme including bearing clearance adjustment and the like by adopting an improved genetic algorithm when HI is less than or equal to 0.7 or the predicted residual life is less than a threshold value. According to the platform, omnibearing monitoring, accurate evaluation and scientific decision making are realized, and the reliability and safety of the magnetic suspension flywheel energy storage system are effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of magnetic levitation flywheel energy storage systems, and in particular to a health monitoring platform for intelligent predictive maintenance of magnetic levitation flywheel energy storage systems. Background Art

[0002] Magnetic levitation flywheel energy storage systems are widely used in smart grids, uninterruptible power supplies, rail transit, and other fields. These systems operate in a complex environment, and their core components, such as magnetic bearings and motor / generator sets, are susceptible to various factors during long-term operation, leading to performance degradation or even failure. This can cause a drop in overall system performance and, in severe cases, lead to safety incidents. While some existing monitoring systems achieve automated data collection, they lack significant deficiencies in data processing and health status assessment. When processing vibration, temperature, and current signals, they are unable to effectively extract key features reflecting the system's health status, resulting in delayed early warning and resolution of system failures, causing significant economic losses and potential safety risks. Summary of the Invention

[0003] In order to fill the market gap, the present invention provides a health monitoring platform for intelligent predictive maintenance of magnetic levitation flywheel energy storage systems.

[0004] The purpose of the present invention is to provide a health monitoring platform for intelligent predictive maintenance of magnetic levitation flywheel energy storage systems to solve the problems raised in the above background technology.

[0005] To achieve the above objectives, the present invention provides the following technical solutions: a health monitoring platform for intelligent predictive maintenance of a magnetic levitation flywheel energy storage system, comprising a multi-source sensor module, an adaptive data acquisition module, an embedded data processing module, a health status assessment module, and a decision module;

[0006] The multi-source sensor module consists of vibration sensors, temperature sensors, speed sensors, and current / voltage sensors distributed on the axial magnetic bearings, radial magnetic bearings, permanent magnetic bias bearings, and motor / generator units of the magnetically suspended flywheel rotor. The sensors are connected via a redundant bus topology.

[0007] The adaptive data acquisition module, equipped with a programmable gain amplifier and a dynamic anti-aliasing filter, collects multimodal physical signals output by the multi-source sensor module in real time and performs amplitude normalization and noise suppression on the collected signals through the signal conditioning circuit;

[0008] The embedded data processing module, equipped with an FPGA chip to implement a parallel computing architecture, performs the following processing steps: C1. Performs empirical mode decomposition on the vibration signal in the adaptive data acquisition module to extract the kurtosis index and energy entropy characteristics of each IMF component; C2. Establishes a time series autoregressive integrated moving average model for the temperature signal in the adaptive data acquisition module to calculate the 3σ confidence interval of the prediction residual; C3. Performs wavelet packet transform on the current signal in the adaptive data acquisition module to extract the characteristic quantity of energy proportion in specific frequency bands;

[0009] The health status assessment module integrates a pre-trained deep belief network model built on a historical fault database. The model inputs the feature vectors extracted in steps C1-C3 into the network input layer through feature-level fusion. The output layer generates a comprehensive health index HI∈[0,1] and predicts the decline trend of the HI index based on a long short-term memory network.

[0010] The decision-making module includes a decision tree based on fuzzy logic. When the comprehensive health index (HI) is ≤0.7, a three-level early warning mechanism is triggered. When the predicted remaining service life is lower than the set threshold, a maintenance plan set including bearing clearance adjustment, active balancing compensation, and cooling system optimization is generated through a maintenance strategy optimization algorithm. The optimization algorithm uses an improved genetic algorithm to screen Pareto optimal solutions based on a multi-objective function of maintenance cost, downtime, and reliability improvement rate.

[0011] Furthermore, the installation positions of the vibration sensors in the multi-source sensor module at the axial magnetic bearing are four monitoring points radially symmetrically distributed on the stator base flange surface, and the measurement range covers the 0-10kHz frequency band; the temperature sensors are arranged at the permanent magnet bias bearing as embedded thermistors, with a temperature measurement accuracy of ±0.5°C and a response time of less than 200ms.

[0012] Furthermore, the redundant bus topology adopts a CAN bus and Ethernet dual-channel redundant architecture, the CAN bus communication rate is 1Mbps, and the number of nodes is not less than 32; the Ethernet channel supports the IEEE1588 time synchronization protocol, and the synchronization accuracy is better than 1μs.

[0013] Furthermore, the FPGA chip adopts the XC7K325T model chip, and its parallel computing architecture includes: a dedicated logic unit for vibration signal processing, configured with a 16-way parallel FIR filter group; a dedicated logic unit for temperature signal processing, integrated with a 32-bit floating-point operator; and a storage unit including 2MBBRAM and 1GBDDR3 cache.

[0014] Furthermore, the input layer structure of the deep belief network model is: vibration feature vector dimension, temperature feature vector dimension: and current feature vector dimension: All input feature vectors are input into the network after Z-score normalization.

[0015] Furthermore, the bearing clearance adjustment solutions in the maintenance solution set include: S1, applying an axial preload of 50-80 N·m through an electromagnetic actuator; S2, using a laser displacement sensor to monitor the clearance change in real time with a control accuracy of ±2 μm; S3, locking the adjustment mechanism when the radial runout drops below 10 μm.

[0016] Furthermore, the cooling system optimization specifically includes: adding a micro vortex tube cooler at the stator winding with a cooling power ≥ 200W; adjusting the coolant flow according to the PID control law based on the feedback from the temperature sensor, with a proportional coefficient Kp = 0.8 and an integral time Ti = 120s; and starting the dual-circulation cooling mode when the permanent magnet temperature exceeds 80°C.

[0017] Furthermore, the bearing clearance adjustment solution in the maintenance solution set includes: applying an axial preload through an electromagnetic actuator, with the preload force range set to 50-80 N·m; using a laser displacement sensor to monitor the clearance change in real time, with a control accuracy of ±2 μm; and locking the adjustment mechanism to fix the bearing clearance when the radial runout drops below 10 μm.

[0018] Furthermore, the input variables of the decision tree based on fuzzy logic include the comprehensive health index HI, the frequency N of the predicted residual exceeding the 3σ confidence interval, and the σ , the rate of change of the energy proportion of the specific frequency band of the current signal ΔE; the output variable is the three-level warning level (level I, level II, level III);

[0019] The fuzzy rule is defined as: when HI≤0.7 and N σ ≥3 times / minute, and ΔE≥15% triggers Level I warning, corresponding to the shutdown maintenance strategy; when 0.7<HI≤0.8, and N σ ≥2 times / minute, and 10%≤ΔE<15% triggers Level II warning, corresponding to load reduction monitoring strategy; when 0.8<HI≤0.9 and N σ When the rate is ≥1 time / minute and 5%≤ΔE<10%, a Level III warning is triggered, corresponding to the regular inspection strategy.

[0020] Compared with the prior art, the present invention has the following beneficial effects:

[0021] The multi-source sensor module, through the distributed deployment of multiple sensors in key components, can collect multimodal physical signals such as vibration, temperature, speed, and current / voltage in real time, enabling comprehensive system monitoring. A redundant bus topology ensures reliable data transmission, ensuring that even if one channel fails, the other can still function normally. The adaptive data acquisition module performs amplitude normalization and noise suppression on the collected signals, improving data quality and providing a reliable basis for subsequent analysis. The embedded data processing module, equipped with an FPGA chip, implements a parallel computing architecture, significantly increasing data processing speed. Empirical mode decomposition, time series modeling, and wavelet packet transform are used to extract features from vibration, temperature, and current signals, respectively. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 It is a schematic diagram of the process of the present invention. DETAILED DESCRIPTION

[0023] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0024] See also Figure 1 : A health monitoring platform for intelligent predictive maintenance of magnetic levitation flywheel energy storage system, including multi-source sensor module, adaptive data acquisition module, embedded data processing module, health status assessment module and decision module;

[0025] The multi-source sensor module consists of a variety of sensors distributed in key locations of the magnetic levitation flywheel rotor. At the axial magnetic bearing, vibration sensors are installed at four monitoring points that are radially symmetrically distributed on the flange surface of the stator base. This layout can capture the vibration of the axial magnetic bearing in all directions. Its measurement range covers the 0-10kHz frequency band and can effectively monitor various vibration signals from low frequency to high frequency, providing rich data for analyzing the operating status of the bearing. At the permanent magnet biased bearing, an embedded thermistor is used as a temperature sensor with a temperature measurement accuracy of ±0.5°C and a response time of less than 200ms. It can quickly and accurately measure the bearing temperature and promptly detect potential faults caused by abnormal temperature. In addition, speed sensors and current / voltage sensors are arranged at the radial magnetic bearing and the electric / generator set to monitor system parameters such as speed, current and voltage. Each sensor is connected to the Ethernet dual-channel redundant architecture via the CAN bus. The CAN bus communication rate is 1Mbps, and the number of nodes is no less than 32, which can ensure the high-speed transmission of large amounts of sensor data. The Ethernet channel supports the IEEE1588 time synchronization protocol with a synchronization accuracy better than 1μs, ensuring the temporal consistency of the sensor data and facilitating subsequent analysis and processing.

[0026] The adaptive data acquisition module is equipped with a programmable gain amplifier and a dynamic anti-aliasing filter. The programmable gain amplifier can automatically adjust the amplification factor according to the amplitude of the sensor output signal to ensure that the amplitude of the collected signal is within the appropriate range and improve the accuracy of data acquisition. The dynamic anti-aliasing filter can effectively suppress high-frequency noise, avoid its interference with useful signals, and ensure the quality of the collected signals. This module collects multi-modal physical signals output by multi-source sensor modules in real time, and performs amplitude normalization and noise suppression processing on the collected signals through signal conditioning circuits. Amplitude normalization unifies signals of different amplitude ranges into a specific interval, which is convenient for subsequent data processing and comparison; noise suppression processing further removes noise from the signal, improves the signal-to-noise ratio of the signal, and provides high-quality data for the embedded data processing module;

[0027] The embedded data processing module, equipped with an FPGA chip to implement a parallel computing architecture, performs the following processing steps: C1. Performs empirical mode decomposition on the vibration signal to extract the kurtosis index and energy entropy characteristics of each IMF component; C2. Establishes a time series autoregressive integrated moving average model for the temperature signal and calculates the 3σ confidence interval of the prediction residual; C3. Performs wavelet packet transform on the current signal to extract the characteristic quantity of energy proportion in specific frequency bands;

[0028] The FPGA uses the XC7K325T model. Its parallel computing architecture includes a dedicated logic unit for vibration signal processing, equipped with a 16-way parallel FIR filter bank; and a dedicated logic unit for temperature signal processing, integrating a 32-bit floating-point arithmetic unit (FPU) for temperature signal processing. A time series autoregressive integrated moving average model is established to predict future temperature values using historical temperature data and calculate a 3σ confidence interval for the prediction residual. Exceeding this confidence interval indicates an abnormal temperature change, potentially indicating a potential system failure. The storage unit includes 2MB of RAM and 1GB of DDR3 cache.

[0029] The health status assessment module integrates a pre-trained deep belief network model built based on a historical fault database. The vibration, temperature, and current feature vectors extracted by the embedded data processing module are fused at the feature level and input into the deep belief network model's input layer. The input layer structure is designed based on the dimensions of the vibration, temperature, and current feature vectors. All input feature vectors are normalized using the Z-score before being input into the network to eliminate the influence of different feature dimensions and improve the accuracy of model training and evaluation. The output layer of the deep belief network model generates a comprehensive health index (HI) with a value range of [0, 1], where 0 indicates a complete system failure and 1 indicates an ideal system health state. Furthermore, the long short-term memory network (LSTM) is used to predict the decline trend of the HI indicator. LSTM can effectively process time series data, capture the changes in system health status over time, and predict potential system failures in advance, providing a basis for formulating maintenance strategies.

[0030] The decision-making module includes a decision tree based on fuzzy logic. When the comprehensive health index HI ≤ 0.7, the three-level early warning mechanism is triggered. When the predicted remaining service life is lower than the set threshold, a maintenance plan set including bearing clearance adjustment, active balancing compensation and cooling system optimization is generated through the maintenance strategy optimization algorithm. The optimization algorithm uses an improved genetic algorithm to screen the Pareto optimal solution based on the multi-objective function of maintenance cost, downtime and reliability improvement rate.

[0031] The input variables of the decision tree based on fuzzy logic include the comprehensive health index HI, the frequency N of the predicted residual exceeding the 3σ confidence interval, and the σ , the rate of change of the energy proportion of the specific frequency band of the current signal ΔE; the output variable is the three-level warning level (level I, level II, level III);

[0032] The fuzzy rule is defined as: when HI≤0.7 and N σ ≥3 times / minute, and ΔE≥15% triggers Level I warning, corresponding to the shutdown maintenance strategy; when 0.7<HI≤0.8, and N σ≥2 times / minute, and 10%≤ΔE<15% triggers Level II warning, corresponding to load reduction monitoring strategy; when 0.8<HI≤0.9 and N σ When the rate is ≥1 time / minute and 5%≤ΔE<10%, a Level III warning is triggered, corresponding to the regular inspection strategy.

[0033] The bearing clearance adjustment solution in the maintenance solution suite includes: An electromagnetic actuator applies axial preload within a set range of 50-80 N·m, and a laser displacement sensor monitors clearance changes in real time, achieving a control accuracy of ±2 μm. When radial runout drops below 10 μm, the adjustment mechanism locks to maintain the bearing clearance. Accurate clearance adjustment helps improve bearing operating accuracy and stability, reduce wear and vibration, and extend bearing life.

[0034] Cooling system optimization specifically includes the addition of micro-vortex tube coolers with a cooling power of ≥200W at the stator windings to enhance the system's heat dissipation capacity. Coolant flow is regulated using PID control principles based on temperature sensor feedback, with a proportional coefficient Kp = 0.8 and an integral time Ti = 120s, ensuring that coolant flow can be adjusted promptly to system temperature fluctuations. When the permanent magnet temperature exceeds 80°C, a dual-circulation cooling mode is activated to further enhance cooling efficiency, prevent permanent magnet demagnetization due to high temperatures, and ensure normal system operation.

[0035] The bearing clearance adjustment solution in the maintenance solution set includes: applying an axial preload through an electromagnetic actuator with a preload range of 50-80 N·m; using a laser displacement sensor to monitor the clearance change in real time with a control accuracy of ±2 μm; and locking the adjustment mechanism to fix the bearing clearance when the radial runout drops below 10 μm.

[0036] The multi-source sensor module collects multimodal physical signals such as vibration, temperature, speed, current / voltage, etc. of key parts of the magnetic levitation flywheel energy storage system in real time, and transmits the data to the adaptive data acquisition module through a redundant bus topology.

[0037] The adaptive data acquisition module performs amplitude normalization and noise suppression on the collected signals and then transmits them to the embedded data processing module.

[0038] The embedded data processing module uses the parallel computing architecture of the FPGA chip to extract features of vibration, temperature, and current signals respectively, and transmits the extracted feature vectors to the health status assessment module.

[0039] The deep belief network model of the health status assessment module generates a comprehensive health index HI based on the input feature vector, predicts the decline trend of the HI indicator through the long short-term memory network, and transmits the results to the decision module.

[0040] The decision module is based on the comprehensive health index HI and the frequency N of the predicted residual exceeding the 3σ confidence interval. σ The rate of change (ΔE) of energy in a specific frequency band of the current signal is used to trigger the corresponding warning level through a fuzzy logic-based decision tree. If the predicted remaining useful life falls below a set threshold, a maintenance strategy optimization algorithm is used to generate a set of maintenance plans to guide system maintenance work.

Claims

1. A health monitoring platform for intelligent predictive maintenance of magnetic levitation flywheel energy storage systems, characterized by: It includes multi-source sensor module, adaptive data acquisition module, embedded data processing module, health status assessment module and decision-making module; The multi-source sensor module consists of vibration sensors, temperature sensors, speed sensors, and current / voltage sensors distributed on the axial magnetic bearings, radial magnetic bearings, permanent magnetic bias bearings, and motor / generator units of the magnetically suspended flywheel rotor. The sensors are connected via a redundant bus topology. The adaptive data acquisition module, equipped with a programmable gain amplifier and a dynamic anti-aliasing filter, collects multimodal physical signals output by the multi-source sensor module in real time and performs amplitude normalization and noise suppression on the collected signals through the signal conditioning circuit; The embedded data processing module, equipped with an FPGA chip to implement a parallel computing architecture, performs the following processing steps: C1. Performs empirical mode decomposition on the vibration signal in the adaptive data acquisition module to extract the kurtosis index and energy entropy characteristics of each IMF component; C2. Establish a time series autoregressive integrated moving average model for the temperature signal in the adaptive data acquisition module and calculate the 3σ confidence interval of the prediction residual; C3. Perform wavelet packet transform on the current signal in the adaptive data acquisition module to extract the characteristic quantity of energy proportion in a specific frequency band; The health status assessment module integrates a pre-trained deep belief network model built on a historical fault database. The model inputs the feature vectors extracted in steps C1-C3 into the network input layer through feature-level fusion. The output layer generates a comprehensive health index HI∈[0,1] and predicts the decline trend of the HI index based on a long short-term memory network. The decision-making module includes a decision tree based on fuzzy logic. When the comprehensive health index (HI) is ≤0.7, a three-level early warning mechanism is triggered. When the predicted remaining service life is lower than the set threshold, a maintenance plan set including bearing clearance adjustment, active balancing compensation, and cooling system optimization is generated through a maintenance strategy optimization algorithm. The optimization algorithm uses an improved genetic algorithm to screen Pareto optimal solutions based on a multi-objective function of maintenance cost, downtime, and reliability improvement rate.

2. The health monitoring platform for intelligent predictive maintenance of magnetic levitation flywheel energy storage systems according to claim 1 is characterized by: The vibration sensors in the multi-source sensor module are installed at four monitoring points on the axial magnetic bearing that are radially symmetrically distributed on the stator base flange surface, and the measurement range covers the 0-10kHz frequency band; the temperature sensors are arranged at the permanent magnet bias bearing as embedded thermal resistors, with a temperature measurement accuracy of ±0.5°C and a response time of less than 200ms.

3. The health monitoring platform for intelligent predictive maintenance of magnetic levitation flywheel energy storage systems according to claim 1 is characterized by: The redundant bus topology adopts a CAN bus and Ethernet dual-channel redundant architecture, the CAN bus communication rate is 1Mbps, and the number of nodes is not less than 32; the Ethernet channel supports the IEEE1588 time synchronization protocol, and the synchronization accuracy is better than 1μs.

4. The health monitoring platform for intelligent predictive maintenance of magnetic levitation flywheel energy storage systems according to claim 1 is characterized by: The FPGA chip uses the XC7K325T model chip, and its parallel computing architecture includes: a dedicated logic unit for vibration signal processing, configured with a 16-way parallel FIR filter group; a dedicated logic unit for temperature signal processing, integrated with a 32-bit floating-point arithmetic unit; and a storage unit including 2MB BRAM and 1GB DDR3 cache.

5. The health monitoring platform for intelligent predictive maintenance of magnetic levitation flywheel energy storage systems according to claim 1 is characterized by: The input layer structure of the deep belief network model is: vibration feature vector dimension, temperature feature vector dimension: and current feature vector dimension: All input feature vectors are input into the network after Z-score normalization.

6. The health monitoring platform for intelligent predictive maintenance of magnetic levitation flywheel energy storage systems according to claim 1 is characterized by: The bearing clearance adjustment solutions in the maintenance solution set include: S1, applying an axial preload of 50-80 N·m through an electromagnetic actuator; S2, using a laser displacement sensor to monitor the clearance change in real time with a control accuracy of ±2 μm; S3, locking the adjustment mechanism when the radial runout drops below 10 μm.

7. The health monitoring platform for intelligent predictive maintenance of magnetic levitation flywheel energy storage systems according to claim 1 is characterized by: The cooling system optimization specifically includes: adding a micro vortex tube cooler at the stator winding with a cooling power of ≥200W; adjusting the coolant flow according to the PID control law based on temperature sensor feedback, with a proportional coefficient Kp = 0.8 and an integration time Ti = 120s; and starting the dual-circulation cooling mode when the permanent magnet temperature exceeds 80°C.

8. The health monitoring platform for intelligent predictive maintenance of magnetic levitation flywheel energy storage systems according to claim 1 is characterized by: The bearing clearance adjustment solution in the maintenance solution set includes: applying an axial preload through an electromagnetic actuator, with the preload force range set to 50-80N·m; using a laser displacement sensor to monitor the clearance change in real time with a control accuracy of ±2μm; and locking the adjustment mechanism to fix the bearing clearance when the radial runout drops below 10μm.

9. The health monitoring platform for intelligent predictive maintenance of magnetic levitation flywheel energy storage systems according to claim 1 is characterized by: The input variables of the decision tree based on fuzzy logic include the comprehensive health index HI, the frequency N of the predicted residual exceeding the 3σ confidence interval σ , the rate of change of the energy proportion of the specific frequency band of the current signal ΔE; the output variable is the three-level warning level (level I, level II, level III); The fuzzy rule is defined as: when HI≤0.7 and N σ ≥3 times / minute, and ΔE≥15% triggers Level I warning, corresponding to the shutdown maintenance strategy; when 0.7<HI≤0.8, and N σ ≥2 times / minute, and 10%≤ΔE<15% triggers Level II warning, corresponding to load reduction monitoring strategy; when 0.8<HI≤0.9 and N σ When the rate is ≥1 time / minute and 5%≤ΔE<10%, a Level III warning is triggered, corresponding to the regular inspection strategy.

Images