A method of coordinated temperature control of a rotor bearing

Abstract

The application discloses a rotor bearing cooperative temperature control method, which comprises the following steps: obtaining rotor bearing end face hot spot area information, millimeter level spatial resolution local temperature field distribution information and vibration frequency spectrum and local temperature information, and then fusing to generate rotor bearing global temperature field model information; according to the rotor bearing global temperature field model information, running a double closed loop control system or triggering a phase change microcapsule coolant; when the double closed loop control system is running, the viscosity and flow of lubricating oil are adjusted and / or the flow of cooling water is dynamically distributed through a variable nozzle array, and the running data of the double closed loop control system is output; when the phase change microcapsule coolant is triggered, the rotor bearing is spot cooled; and the running data of the double closed loop control system is fed back to the double closed loop control system after deep reinforcement learning dynamic optimization and fault mode knowledge graph prediction. The application can significantly improve the operation reliability and service life of the rotor bearing.

Description

Technical Field

[0001] This invention belongs to the technical field of hydropower generation systems, and particularly relates to a rotor bearing coordinated temperature control method. Background Technology

[0002] Large hydroelectric turbines are key components of hydroelectric power systems, and their rotor bearings play a crucial role. The rotor bearings support the turbine rotor, ensuring its stable operation at high speeds. During turbine operation, the rotor bearings inevitably generate heat due to the immense pressure, friction, and centrifugal force generated by high-speed rotation. Excessive rotor bearing temperature can lead to a series of serious problems, such as thinning or even rupture of the lubricating oil film, thus accelerating bearing wear. Furthermore, excessively high temperatures can degrade the mechanical properties of the bearing materials, reducing their lifespan. In severe cases, it can even cause rotor bearing seizure and rotor imbalance, leading to turbine shutdown and disrupting the normal operation of the entire hydroelectric power system.

[0003] Currently, traditional rotor bearing temperature control methods mainly include passive cooling, single-point temperature monitoring, PID-based linear feedback control, independently designed lubrication and cooling systems, and fixed threshold alarm mechanisms. Passive cooling relies on preset parameters and cannot dynamically adjust according to load or environmental changes, potentially leading to energy waste due to overcooling at low loads and localized overheating due to insufficient heat dissipation at high loads. Single-point temperature monitoring is limited by a single sensor deployment, making it difficult to capture temperature gradients at the rotor bearing edges or end faces, easily creating monitoring blind spots and causing false alarms or delayed alarms due to temperature differences. PID-based linear feedback control, due to its reliance on historical data and linearized algorithm, exhibits lag in response under nonlinear scenarios such as sudden load increases or deterioration of lubrication performance, resulting in significant overshoot. Independently designed lubrication and cooling systems do not consider thermodynamic coupling effects; excessive cooling water can lead to excessively thick oil films and increased oil churning losses, and the difference in response delay between the lubrication and cooling systems exacerbates temperature fluctuations. Fixed threshold alarm mechanisms lack adaptability to changes in operating conditions, posing potential risks of frequent false triggers or missed alarms during variable speeds or loads. Therefore, these traditional rotor bearing temperature control methods generally suffer from problems such as low temperature control accuracy, high energy consumption, and passive maintenance, making it difficult to cope with complex and ever-changing operating environments. Summary of the Invention

[0004] To overcome the shortcomings of existing technologies, this invention provides a rotor bearing coordinated temperature control method, which can significantly improve the operational reliability and lifespan of rotor bearings.

[0005] The objective of this invention is achieved through the following technical solution:

[0006] A rotor bearing coordinated temperature control method includes the following steps:

[0007] Acquire three types of real-time data: hot spot area information on rotor bearing end face, local temperature field distribution information with millimeter-level spatial resolution, and vibration spectrum and local temperature information;

[0008] By integrating three types of real-time data, a global temperature field model of the rotor bearing is generated.

[0009] Based on the rotor bearing's global temperature field model information, the dual closed-loop control system is operated or the phase change microcapsule coolant is triggered. When the dual closed-loop control system is operated, the viscosity and flow rate of the lubricating oil are adjusted and / or the flow rate of the cooling water is dynamically distributed through a variable nozzle array, and the operating data of the dual closed-loop control system is output. When the phase change microcapsule coolant is triggered, the rotor bearing is cooled in a point-like manner.

[0010] The operating data of the dual closed-loop control system is dynamically optimized through deep reinforcement learning and predicted by fault mode knowledge graph, and then fed back to the dual closed-loop control system.

[0011] Furthermore, the acquisition of three types of real-time data includes: information on hot spot areas on the rotor bearing end face, millimeter-level spatial resolution information on local temperature field distribution, and vibration spectrum and local temperature information.

[0012] Real-time information on hotspot areas on the rotor bearing end face is obtained using an infrared thermal imager.

[0013] Real-time acquisition of local temperature field distribution information with millimeter-level spatial resolution is achieved using millimeter-wave radar temperature sensors;

[0014] Vibration spectrum and local temperature information are acquired in real time and synchronously using an embedded vibration-temperature composite sensor.

[0015] Furthermore, by integrating three types of real-time data, the generated rotor bearing global temperature field model information includes:

[0016] The data fusion unit receives three types of real-time data, fuses the three types of real-time data into multi-source perception data, and outputs the multi-source perception data to the CNN-GRU-Attention fusion network.

[0017] The CNN-GRU-Attention fusion network extracts spatial and temporal features from multi-source sensing data and focuses on key information to generate a global temperature field model for rotor bearings.

[0018] Furthermore, based on the rotor bearing global temperature field model information, the operation of the dual closed-loop control system includes:

[0019] The inner loop of the dual closed-loop control system adjusts the viscosity and flow rate of the lubricating oil based on the real-time temperature field model information of the rotor bearing.

[0020] The outer loop of the dual closed-loop control system dynamically distributes the flow rate of cooling water through a variable nozzle array based on the temperature field gradient data in the real-time temperature field model information of the rotor bearing.

[0021] Furthermore, based on the rotor bearing global temperature field model information, the phase change microcapsule coolant that triggers the phase change includes:

[0022] The temperature monitoring unit compares the information of the rotor bearing's global temperature field model with the preset normal temperature threshold in real time to trigger the phase change microcapsule coolant to perform point cooling on the rotor bearing.

[0023] Furthermore, the dynamic optimization of the operating data of the dual closed-loop control system through deep reinforcement learning includes:

[0024] Deep reinforcement learning dynamically optimizes the operating data of a dual closed-loop control system based on temperature stability, energy consumption, and mechanical life indicators.

[0025] Furthermore, before deep reinforcement learning dynamic optimization and fault mode knowledge graph prediction, three types of real-time data are cleaned, filtered, and format converted to drive a lightweight digital twin model.

[0026] Furthermore, the deep reinforcement learning algorithm includes a state space, an action space, a reward function, and a training mechanism; the state space includes the standard deviation of the temperature field, the amplitude of the dominant vibration frequency, the load rate, and the inlet temperature of the cooling water; the action space includes the opening degree of the lubricating oil flow valve, the control signal of the variable nozzle array, and the adjustment amount of the PID parameters; the training mechanism uses a digital twin model to generate 100,000 sets of virtual data covering extreme working conditions to pre-train the deep reinforcement learning agent, and then fine-tunes the strategy through online learning;

[0027] The reward function is:

[0028]

[0029] in, For instant reward value, For the standard deviation of the temperature field, For the real-time power of the cooling system, This is the maximum design power of the cooling system. This represents the change in the fatigue damage coefficient. This is the threshold for the incremental fatigue damage.

[0030] Furthermore, the operational data of the dual closed-loop control system are used to predict the following using a fault mode knowledge graph:

[0031] The fault mode knowledge graph predicts thermal runaway paths based on historical data and physical laws, and uses early intervention commands to adjust the operating data of the dual closed-loop control system.

[0032] Furthermore, the control method adopts an edge-cloud computing collaborative architecture, which includes edge nodes and the cloud. The edge nodes are used for real-time control and uploading data to the cloud. After simulation iteration, the cloud sends back the optimization parameters to the edge nodes.

[0033] The beneficial effects of this invention are as follows:

[0034] 1. By collaboratively acquiring information on hot spot areas on the rotor bearing end face, local temperature field distribution information with millimeter-level spatial resolution, and vibration spectrum and local temperature information, the full-domain coverage of the three-dimensional temperature field of the rotor bearing and dynamic heat source positioning are achieved, eliminating the blind spot problem of traditional single-point monitoring and improving monitoring resolution.

[0035] 2. Based on multi-source data fusion and digital twin model, combined with deep reinforcement learning algorithm, the operating data parameter of the dual closed-loop control system is dynamically optimized, which helps to improve temperature control accuracy and shorten response time.

[0036] 3. It adopts a dual closed-loop design. The inner ring adjusts the viscosity and flow of the lubricating oil, while the outer ring precisely distributes the cooling water to different zones. At the same time, it combines phase change microcapsule coolant to quickly cool the rotor bearings in a point-like manner, so as to achieve global optimization of temperature control strategy and local emergency intervention, thereby reducing energy consumption.

[0037] 4. Integrate fault mode knowledge graphs and improve early warning of thermal runaway risk through multi-parameter trend analysis. Combine digital twin simulation to formulate intervention strategies to reduce unplanned downtime.

[0038] 5. The edge-cloud computing collaborative architecture enables efficient collaboration. Specifically, edge nodes can perform real-time control, while the cloud is responsible for high-precision simulation and model iteration updates. It balances real-time performance with computing resource optimization, thereby improving system adaptability and robustness. Attached Figure Description

[0039] The invention will now be described in more detail with reference to embodiments and the accompanying drawings.

[0040] Figure 1 An overall block diagram of the present invention is shown;

[0041] Figure 2 A schematic diagram of the lightweight digital twin model in this invention is shown;

[0042] Figure 3 The flowchart of the deep reinforcement learning algorithm in this invention is shown;

[0043] Figure 4 A block diagram of the dual closed-loop control system in this invention is shown;

[0044] Figure 5 The structure diagram of the fault mode knowledge graph in this invention is shown. Detailed Implementation

[0045] The invention will now be further described with reference to the accompanying drawings.

[0046] This invention provides a rotor bearing coordinated temperature control method, implemented through a rotor bearing coordinated temperature control system, such as... Figure 1 As shown, the rotor bearing coordinated temperature control system includes a multi-dimensional temperature field sensing module, a dynamic lubrication-cooling coordinated regulation module, and an intelligent regulation and fault classification module. The rotor bearing coordinated temperature control method includes the following steps:

[0047] In the multi-dimensional temperature field sensing module, three types of real-time data are acquired respectively through a thermal infrared imager, a millimeter-wave radar temperature sensor, and an embedded vibration-temperature composite sensor: information on hot spots on the rotor bearing end face, millimeter-level spatial resolution local temperature field distribution information, and synchronous vibration spectrum and local temperature information. These are used to construct a multi-dimensional temperature field sensing network, and then the three types of real-time data from the multi-dimensional temperature field sensing network are transmitted to the data fusion unit. The data fusion unit receives and fuses the three types of real-time data from the multi-dimensional temperature field sensing network to obtain multi-source sensing data, which is then transmitted to the CNN-GRU-Attention fusion network. The CNN-GRU-Attention fusion network extracts the spatial and temporal features of the multi-source sensing data and focuses on key information to generate a global temperature field model of the rotor bearing, which is then transmitted as input to the dynamic lubrication-cooling coordinated control module.

[0048] In the dynamic lubrication-cooling coordinated control module, based on the rotor bearing's global temperature field model information, a dual closed-loop control system is run or a phase change microcapsule coolant is triggered. When the dual closed-loop control system is run, the viscosity and flow rate of the lubricating oil are adjusted and / or the flow rate of the cooling water is dynamically distributed through a variable nozzle array, and the operating data of the dual closed-loop control system is output to the intelligent control and fault classification module. When the phase change microcapsule coolant is triggered, the rotor bearing is rapidly cooled in a point-like manner.

[0049] In the intelligent control and fault classification module, a multi-physics coupled digital twin model of the bearing, lubrication and cooling system is established to simulate thermodynamic behavior in real time. The deep reinforcement learning (DRL) algorithm is used to dynamically optimize parameters based on temperature stability, energy consumption and mechanical life indicators. The fault mode knowledge graph is embedded and combined with historical data and physical laws to predict thermal runaway path and send back intervention instructions in advance to adjust the operating data of the dual closed-loop control system.

[0050] Understandably, multi-dimensional temperature field sensing and control overcomes the limitations of traditional single-point monitoring, facilitating the full-domain temperature gradient control of rotor bearings. Dual closed-loop regulation of the lubrication and cooling processes helps reduce cooling system energy consumption and stabilizes lubricating oil viscosity within the optimal range, thus reducing oil churning losses. Adaptive control driven by DRL shortens temperature stabilization time and reduces overshoot under sudden load conditions. Early fault warning based on a digital twin model can issue an alarm 24 hours before the Babbitt alloy layer detaches from the bearing, with higher accuracy.

[0051] It should be noted that before the deep reinforcement learning dynamic optimization and fault mode knowledge graph prediction, the three types of real-time data are cleaned, filtered and format converted to drive the lightweight digital twin model.

[0052] Specifically, such as Figure 2 As shown, after the rotor bearing collaborative temperature control system is initialized, a multi-dimensional temperature field sensing network is first constructed using a thermal infrared imager, a millimeter-wave radar temperature sensor, and an embedded vibration-temperature composite sensor to monitor the rotor bearing temperature distribution in real time.

[0053] Three types of real-time data are preprocessed through data cleaning, filtering, and format conversion. The preprocessed real-time data is then used to drive a lightweight digital twin model to quickly respond to and simulate the thermodynamic behavior of rotor bearings.

[0054] Real-time simulation is performed in the digital twin model to simulate the temperature changes and mechanical response of the rotor bearing under current operating conditions;

[0055] Based on simulation results, the future condition of the rotor bearing is predicted and its performance is analyzed to identify potential overheating risks or other failure modes.

[0056] Based on the performance analysis results, a control strategy (adjusting the viscosity of the lubricating oil, the flow rate of the lubricating oil, and the distribution of cooling water) is specified, and the control strategy is translated into specific control commands to be implemented by actuators (Peltier semiconductor devices, servo motors, and piezoelectric ceramics).

[0057] The results of the implementation are fed back to the physical entity of the rotor bearing to achieve closed-loop control;

[0058] In addition, the status and control effect of the rotor bearing can be displayed through a visual interface, allowing users to interact and manually adjust the model parameters. The parameters of the digital twin model can be updated based on user interaction and feedback from the rotor bearing-coordinated temperature control system to improve the accuracy and applicability of the model.

[0059] It should be noted that the rotor bearing collaborative temperature control system adopts an edge-cloud collaborative architecture, which includes edge nodes and the cloud. The edge nodes perform control operations with a delay of 10ms based on real-time data and upload the data to the cloud. The cloud then performs high-precision simulation and model iteration updates based on the data uploaded by the edge nodes, and sends the optimized model parameters back to the edge nodes to continuously improve the control effect.

[0060] Among them, the edge-cloud computing collaborative architecture, in a distributed computing environment, achieves efficient collaboration between edge devices and cloud computing platforms by rationally allocating tasks and resources, so as to meet performance requirements in terms of real-time performance, bandwidth utilization, energy consumption, and computing power.

[0061] It should also be noted that, such as Figure 3 As shown, in the deep reinforcement learning algorithm, the DRL agent and its training environment are first initialized, which is a simulation environment built based on multi-source sensing data and a digital twin model. Then, in each training round, the environment is reset to obtain an initial state composed of fused features such as temperature field and vibration spectrum. The agent outputs control actions (such as parameter adjustments for lubricating oil flow valve opening and variable nozzle array control signals) according to the current state, executes the actions in the simulation environment, and observes the reward signal (comprehensive temperature stability, energy consumption, and mechanical life indicators) and the new state. The rotor bearing collaborative temperature control system determines whether the termination condition has been met. If the termination condition has not been met, the DRL agent network is updated using this interactive experience and the decision-making and learning process continues in a loop. If the termination condition has been met, the performance of this round is evaluated and the current model is saved. Through multi-round iterative training, the DRL agent eventually learns to dynamically optimize control parameters under various operating conditions, thereby providing adaptive and high-precision decision support for the actual rotor bearing dual closed-loop temperature control system. In addition, the trained lightweight DRL agent model can be deployed to edge nodes to achieve real-time and intelligent collaborative temperature control.

[0062] It should also be noted that the rotor bearing collaborative temperature control system incorporates a fault mode knowledge graph, such as... Figure 5 As shown, the fault mode knowledge graph describes various fault modes, causes, effects, detection methods, preventive measures, and response strategies. When abnormal temperature or vibration data is detected, the rotor bearing cooperative temperature control system quickly identifies possible fault causes through the fault mode knowledge graph and automatically triggers corresponding preventive or repair measures. In addition, the rotor bearing cooperative temperature control system also has user interaction functions. Operators can monitor the rotor bearing status through a visual interface and manually adjust model parameters or intervene in control strategies. Through this cooperative temperature control method, the rotor bearings of large hydro turbines can maintain optimal operating conditions under various operating conditions, effectively preventing and reducing the occurrence of faults, thereby significantly improving the operating efficiency and reliability of the hydro turbine.

[0063] It should also be noted that vibration spectrum and local temperature information are collected simultaneously to identify mechanical friction heat sources; the CNN-GRU-Attention fusion network utilizes both spatial and temporal features to adapt to the dynamic changes in rotor bearing operating conditions; among them, CNN is typically used to process data from devices such as infrared thermal imagers to extract useful spatial features; GRU is used to process time series data such as temperature change data to identify and predict time-related patterns; the Attention mechanism focuses on key information in the time series to improve the accuracy of fault diagnosis.

[0064] In one embodiment, the operation of the dual closed-loop control system, based on the rotor bearing global temperature field model information, includes:

[0065] The inner loop of the dual closed-loop control system adjusts the viscosity and flow rate of the lubricating oil based on the real-time temperature field model information of the rotor bearing.

[0066] The outer loop of the dual closed-loop control system dynamically distributes the flow rate of cooling water through a variable nozzle array based on the temperature field gradient data in the real-time temperature field model information of the rotor bearing.

[0067] It should be noted that both the inner and outer loops of the dual closed-loop control system operate under the condition of "keeping the rotor bearings within the normal operating range".

[0068] It should be noted that, as Figure 4 As shown, by adjusting the viscosity and flow rate of the lubricating oil and dynamically distributing the flow rate of cooling water through a variable nozzle array, the temperature changes of the rotor bearing can be adapted. Specifically, the inner ring adjusts the viscosity of the lubricating oil through a Peltier semiconductor device. This device achieves heating and cooling by changing the current, thereby controlling the viscosity changes of the lubricating oil. The servo motor in the inner ring adjusts the speed of the gear pump according to the set lubricating oil flow rate requirements to achieve precise flow control. The outer ring is responsible for distributing the cooling water flow rate according to the temperature gradient. This can be achieved by using a piezoelectric ceramic-driven variable micro-nozzle array to dynamically adjust the cooling water flow rate in 64 independently controlled areas, thus achieving precise heat dissipation with "strong cooling in hot spots and energy saving in low-temperature zones." Therefore, this dual-closed-loop control system can not only quickly respond to temperature changes in the rotor bearing but also self-adjust and optimize based on real-time data, ensuring that the rotor bearing remains within the optimal temperature range under various operating conditions, thereby improving the operating efficiency and reliability of the turbine rotor bearing.

[0069] In one embodiment, the phase change microcapsule coolant is triggered based on rotor bearing global temperature field model information, including:

[0070] The temperature monitoring unit compares the information of the rotor bearing's global temperature field model with the preset normal temperature threshold in real time to trigger the phase change microcapsule coolant for rapid point cooling of the rotor bearing.

[0071] It should be noted that the preset normal temperature threshold is usually the optimal operating temperature range for the rotor bearing material and lubricating oil. If it exceeds this operating temperature range by 5°C or more, or if the temperature change rate exceeds 2°C / min and lasts for more than 10 seconds, it is defined as a temperature abnormality. At this time, the phase change microcapsule coolant is triggered to perform rapid point cooling.

[0072] It should also be noted that phase change microcapsule coolant is a new type of coolant that combines phase change material (PCM) and microcapsule technology. Its core is to encapsulate the phase change material in microcapsules, enabling it to absorb or release a large amount of latent heat through phase change within a specific temperature range, thereby achieving efficient thermal energy storage and release. Its advantages include high heat storage capacity, enhanced stability, multifunctionality, and ease of operation.

[0073] In one embodiment, the deep reinforcement learning algorithm includes a state space, an action space, a reward function, and a training mechanism. The state space includes 12-dimensional features such as the standard deviation of the temperature field, the amplitude of the dominant vibration frequency, the load rate, and the inlet temperature of the cooling water. The action space includes 9-dimensional features such as the opening degree of the lubricating oil flow valve, the control signal of the variable nozzle array, and the adjustment amount of the PID parameters. The training mechanism uses a digital twin model to generate 100,000 sets of virtual data covering extreme working conditions and pre-trains the DRL agent using the TD3 algorithm, and then fine-tunes the strategy through online learning.

[0074] The reward function is:

[0075]

[0076] in, For instant reward value, For the standard deviation of the temperature field, For the real-time power of the cooling system, This is the maximum design power of the cooling system. This represents the change in the fatigue damage coefficient. This is the threshold for the incremental fatigue damage.

[0077] It should be noted that the standard deviation of the temperature field is obtained by analyzing the dispersion of the full-domain temperature data collected by the millimeter-wave radar temperature sensor and the thermal infrared imager of the multi-dimensional temperature field sensing module through the CNN-GRU-Attention fusion network; the amplitude of the vibration dominant frequency can be obtained by extracting the amplitude value corresponding to the dominant frequency from the vibration spectrum data collected by the embedded vibration-temperature composite sensor; the load rate is calculated from the operating parameters such as rotor speed and output power in the turbine control system; the cooling water inlet temperature is directly collected by the temperature sensor in the cooling water circuit; and the other dimensions of features, such as lubricating oil viscosity, are obtained by collecting data from the corresponding sensors or by deducing data from the system operation data, and are finally integrated into 12-dimensional features in the state space.

[0078] The opening degree of the lubricating oil flow valve is output to the servo motor in the inner loop to adjust the speed of the gear pump, thereby controlling the actual flow rate of the lubricating oil. The control signal of the variable nozzle array is output to the piezoelectric ceramic drive unit in the outer loop to independently adjust the cooling water flow rate of the 64 micro-nozzle areas, thereby achieving precise heat dissipation. The PID parameter adjustment is output to the control unit of the dual closed-loop control system to dynamically optimize the PID control logic of the inner and outer loops, thereby improving the temperature control response speed and stability. Other characteristics, such as the phase change microcapsule coolant trigger threshold, are output to the corresponding actuators to jointly achieve multi-dimensional coordinated regulation of the rotor bearing temperature.

[0079] in addition, The real-time power of the cooling system is calculated in real time using the operating parameters of equipment such as water pumps and piezoelectric ceramic drive units in the cooling system (such as water pump operating current, voltage, and piezoelectric ceramic drive power) and combined with the equipment power formula, in order to reflect the current energy consumption level of the cooling system.

[0080] The maximum design power or rated power of the cooling system is the upper limit of the total energy consumption of all actuators when running at full power. This value is obtained by arithmetic summation of the system hardware specifications and on-site measurement, so as to convert the real-time cooling power into a relative occupancy rate between 0 and 1, so that the DRL algorithm can fairly weigh the energy-saving target in the reward function and learn to achieve the cooling effect with the minimum energy consumption.

[0081] The change in fatigue damage coefficient is calculated based on the rotor bearing stress data and running time simulated by the digital twin model, combined with the material fatigue damage model (such as Miner's linear cumulative damage rule), to calculate the change in rotor bearing fatigue damage per unit time. At the same time, it is corrected by state parameters such as the amplitude of the dominant vibration frequency, and finally the change in fatigue damage coefficient is obtained to reflect the mechanical wear trend of the rotor bearing.

[0082] The fatigue damage increment threshold is set based on fatigue cumulative damage theory (such as the Mainner criterion) and the design life of the rotor bearing and is allowed within a specific evaluation time window. Its value comes from the SN curve of the rotor bearing material, the design load spectrum and industry safety standards. It converts the change in fatigue damage coefficient calculated by the digital twin model into a ratio relative to the safety limit, thereby guiding the DRL algorithm to actively find control strategies that can significantly suppress thermal stress and extend the service life of the rotor bearing.

[0083] In addition, by introducing and These two engineering parameters, which respectively characterize the system hardware capability boundary and the equipment safe life scale, enable DRL to automatically and dynamically optimize the complex balance between temperature stability, operating energy efficiency and mechanical reliability within a unified quantitative framework, thereby achieving a fundamental leap from passive response to proactive prediction and multi-objective collaborative optimization.

[0084] Finally, it should be noted that edge nodes are equipped with edge servers or high-performance embedded computing units and other computing power modules to build lightweight digital twin models, while high-precision digital twin models are deployed synchronously on the cloud platform for simulation and iteration.

[0085] An embedded controller is set up at the edge node as the control unit of the dual closed-loop control system. The embedded controller receives real-time status data of temperature field, lubricating oil and cooling water, and performs parameter calculations and commands to control the Peltier semiconductor device, servo motor and gear pump in the inner loop, and also facilitates the control of the piezoelectric ceramic in the outer loop.

[0086] In summary, this invention achieves full coverage of the three-dimensional temperature field of the rotor bearing and dynamic heat source positioning by collaboratively acquiring information on the hot spot area of ​​the rotor bearing end face, local temperature field distribution information with millimeter-level spatial resolution, and vibration spectrum and local temperature information. This eliminates the blind spot problem of traditional single-point monitoring and improves the monitoring resolution.

[0087] This invention is based on multi-source data fusion and digital twin model, combined with deep reinforcement learning algorithm to dynamically optimize the operating data parameter of dual closed-loop control system, which is beneficial to improve temperature control accuracy and shorten response time;

[0088] This invention employs a dual closed-loop design. The inner loop adjusts the viscosity and flow rate of the lubricating oil, while the outer loop precisely distributes the cooling water to different zones. Simultaneously, it combines phase change microcapsule coolant to rapidly cool the rotor bearings in a point-like manner, thereby achieving global optimization of the temperature control strategy and local emergency intervention, thus reducing energy consumption.

[0089] This invention integrates a fault mode knowledge graph, improves early warning of thermal runaway risk through multi-parameter trend analysis, and combines digital twin simulation to formulate intervention strategies to reduce unplanned downtime.

[0090] In this invention, the edge-cloud computing collaborative architecture achieves high-efficiency collaboration. Specifically, edge nodes can perform real-time control, while the cloud is responsible for high-precision simulation and model iterative updates. It balances real-time performance with computing resource optimization, thereby improving system adaptability and robustness.

[0091] In the description of this invention, it should be understood that the terms "upper", "lower", "bottom", "top", "front", "rear", "inner", "outer", "left", "right", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0092] While the invention has been described herein with reference to specific embodiments, it should be understood that these embodiments are merely examples of the principles and applications of the invention. Therefore, it should be understood that many modifications can be made to the exemplary embodiments, and other arrangements can be designed without departing from the spirit and scope of the invention as defined by the appended claims. It should be understood that different dependent claims and features described herein can be combined in ways different from those described in the original claims. It is also understood that features described in conjunction with individual embodiments can be used in other described embodiments.

Claims

1. A method of rotor bearing cooperative temperature control, characterized by, The method comprises the following steps: acquiring rotor bearing end face hot spot area information, millimeter level spatial resolution local temperature field distribution information, and vibration spectrum and local temperature information of three types of real-time data; fusing the three types of real-time data to generate rotor bearing global temperature field model information; according to the rotor bearing global temperature field model information, running a double closed-loop control system or triggering phase change microcapsule coolant; when the double closed-loop control system is running, adjusting the viscosity and flow of the lubricating oil and / or dynamically distributing the flow of the cooling water through a variable nozzle array, and outputting the running data of the double closed-loop control system; when the phase change microcapsule coolant is triggered, point cooling is performed on the rotor bearing; the running data of the double closed-loop control system is fed back to the double closed-loop control system after deep reinforcement learning dynamic optimization and fault mode knowledge graph prediction.

2. A method of coordinated temperature control of a rotor bearing according to claim 1, characterized in that The acquisition of the rotor bearing end face hot spot area information, millimeter level spatial resolution local temperature field distribution information, and vibration spectrum and local temperature information of the three types of real-time data comprises: acquiring the rotor bearing end face hot spot area information in real time through an infrared thermal imager; acquiring the millimeter level spatial resolution local temperature field distribution information in real time through a millimeter wave radar temperature sensor; acquiring the vibration spectrum and local temperature information in real time and synchronously through an embedded vibration-temperature composite sensor.

3. A method of coordinated temperature control of a rotor bearing according to claim 1, characterized in that The fusion of the three types of real-time data to generate the rotor bearing global temperature field model information comprises: a data fusion unit receives the three types of real-time data, fuses the three types of real-time data into multi-source perception data, and outputs the multi-source perception data to a CNN-GRU-Attention fusion network; the CNN-GRU-Attention fusion network extracts spatial features and time sequence features of the multi-source perception data and focuses on key information to generate the rotor bearing global temperature field model information.

4. A method of coordinated temperature control of a rotor bearing according to claim 1, characterized in that According to the rotor bearing global temperature field model information, running the double closed-loop control system comprises: the inner loop of the double closed-loop control system adjusts the viscosity and flow of the lubricating oil according to the rotor bearing real-time temperature field model information; the outer loop of the double closed-loop control system dynamically distributes the flow of the cooling water through a variable nozzle array according to the temperature field gradient data in the rotor bearing real-time temperature field model information.

5. A method of coordinated temperature control of a rotor bearing according to claim 1 or 4, characterized in that According to the rotor bearing global temperature field model information, triggering the phase change microcapsule coolant comprises: a temperature monitoring unit compares the rotor bearing global temperature field model information with a preset normal temperature threshold in real time to trigger the phase change microcapsule coolant to perform point cooling on the rotor bearing.

6. A method of coordinated temperature control of a rotor bearing as claimed in claim 1, wherein, The deep reinforcement learning dynamic optimization of the running data of the double closed-loop control system comprises: the deep reinforcement learning dynamically optimizes the running data of the double closed-loop control system according to temperature stability, energy consumption, and mechanical life indicators.

7. A method of coordinated temperature control of a rotor bearing according to claim 1 or 6, characterized in that Before the deep reinforcement learning dynamic optimization and fault mode knowledge graph prediction, the three types of real-time data are driven to a lightweight digital twin model after data cleaning, filtering, and format conversion.

8. A method of coordinated temperature control of a rotor bearing according to claim 7, characterized in that The algorithm of the deep reinforcement learning comprises a state space, an action space, a reward function, and a training mechanism; the state space comprises a temperature field standard deviation, a vibration main frequency amplitude, a load rate, and a cooling water inlet temperature; the action space comprises a lubricating oil flow valve opening degree, a variable nozzle array control signal, and a PID parameter adjustment amount. The training mechanism utilizes a digital twin model to generate 100,000 sets of virtual data covering extreme working conditions to pre-train a deep reinforcement learning agent, and then fine-tunes the strategy through online learning; The reward function is: wherein, is an instant reward value, is a temperature field standard deviation, is a cooling system real-time power, is a maximum design power of the cooling system, is a fatigue damage coefficient change amount, is a fatigue damage increment threshold value.

9. A method of coordinated temperature control of a rotor bearing according to claim 1 or 6, characterized in that The running data of the double closed-loop control system is pre-judged by the fault mode knowledge graph, including: The fault mode knowledge graph predicts the thermal runaway path based on historical data and physical laws to return intervention instructions in advance to adjust the running data of the double closed-loop control system.

10. A method of coordinated temperature control of a rotor bearing according to claim 1, characterized in that The control method adopts an edge-cloud computing collaborative architecture, which includes an edge node and a cloud end. The edge node is used for real-time control and uploading data to the cloud end, and the cloud end returns optimized parameters to the edge node after simulation iteration.

Images