A method and system for evaluating maintenance time of a T1 class temperature sensor of a nuclear power plant

Abstract

The application provides a nuclear power plant T1 level temperature sensor maintenance time evaluation method and an evaluation system, and belongs to the field of temperature sensor fault prediction. The method solves the problem that the existing nuclear power plant T1 level temperature sensor maintenance technology cannot accurately predict the optimal maintenance time in advance. The method comprises the following steps: reading the target and redundant sensor, and identifying the fault sensor; according to the identified fault sensor, the historical operation data and maintenance record of the fault sensor are called; a deep learning model is established, the existing operation data and the historical operation data of the fault sensor are analyzed, and the future operation data prediction value is obtained; a Monte Carlo mathematical model is constructed, the prediction result of the deep learning model is subjected to uncertainty analysis, and the maintenance time evaluation result is obtained. The method is mainly used for nuclear power plant T1 level temperature sensor maintenance time evaluation.

Description

Technical Field

[0001] This invention belongs to the field of temperature sensor fault prediction, and in particular relates to a method and system for assessing the maintenance time of a T1-level temperature sensor in a nuclear power plant. Background Technology

[0002] As a complex system composed of numerous systems, equipment, and components, the operational safety of a nuclear power plant highly depends on the reliability of its various equipment. With the increasing years of operation, equipment performance degradation and aging gradually become apparent. Temperature sensors, in particular, which operate under harsh conditions such as high temperature, high pressure, and radiation for extended periods, are prone to drift, decreased accuracy, and even malfunction, posing a potential threat to the safe and stable operation of the nuclear power plant. While equipment aging typically does not show obvious effects in the early stages of a power plant's lifespan, its cumulative effect gradually increases as the plant's lifespan progresses, becoming a significant factor affecting the safe operation of the nuclear power plant.

[0003] To promptly detect equipment anomalies, the industry widely employs online monitoring technology to monitor the operational status of critical equipment. Existing online monitoring systems primarily identify faults by collecting equipment operating parameters and combining them with threshold alarms or simple data analysis methods. However, these methods mostly only issue alarms after a fault occurs or when fault characteristics are obvious, failing to predict the remaining lifespan of the equipment and optimal maintenance time in advance. This often results in a reactive approach to maintenance, increasing the risk of sudden failures and potentially leading to unnecessary cost waste due to over-maintenance.

[0004] In recent years, with the development of artificial intelligence technology, deep learning-based fault prediction methods have been gradually applied to the field of nuclear power plant equipment monitoring. However, most existing deep learning-based prediction methods only focus on the prediction accuracy of the model and lack quantitative analysis of the uncertainty of the prediction results, making it difficult to assess the reliability of the prediction results and failing to provide a reliable basis for maintenance decisions. At the same time, in the fault identification process, existing technologies often fail to make full use of the redundant sensor data that are commonly found in nuclear power plants, which can easily lead to misjudgment or missed judgment due to abnormal data from a single sensor, affecting the reliability of the monitoring system. Therefore, developing a predictive maintenance system that can accurately predict the failure time of temperature sensors and quantify the uncertainty of prediction is of great significance for improving the operational safety and maintenance efficiency of nuclear power plants. Summary of the Invention

[0005] In view of this, the present invention aims to propose a method and system for evaluating the maintenance time of T1-level temperature sensors in nuclear power plants, in order to solve the problems that the existing maintenance technology for T1-level temperature sensors in nuclear power plants can only perform post-event repairs or simple threshold alarms, cannot accurately predict the optimal maintenance time in advance, and does not make full use of redundant sensor data for fault identification and lacks the quantification of the uncertainty of the prediction results, resulting in safety hazards and high maintenance costs.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a method for assessing the maintenance time of a T1-level temperature sensor in a nuclear power plant, comprising the following steps: S1. Take readings from the target and redundant sensors to identify faulty sensors; S2. Based on the identified faulty sensor, retrieve the historical operating data and maintenance records of the faulty sensor; S3. Establish a deep learning model, analyze the existing and historical operating data of the fault sensor, and obtain the predicted value of future operating data. S4. Construct a Monte Carlo mathematical model, perform uncertainty analysis on the prediction results of the deep learning model, and obtain the maintenance time evaluation results.

[0007] Furthermore, step S1 includes the following steps: S11. Install temperature sensors within the nuclear power plant; S12. Set up the sensor network and determine the main sensor and redundant sensors for the same monitored object; S13. Monitor sensor data in real time and identify faulty sensors with abnormal readings.

[0008] Furthermore, step S2 includes the following steps: S21. Obtain the location, type, and real-time monitoring data of the faulty sensor; S22. Retrieve historical operating data and maintenance records of the faulty sensor, as well as steady-state and transient operating data of the redundant sensor; S23. Organize, preprocess, and summarize the retrieved data.

[0009] Furthermore, in step S22, the retrieved historical data of the faulty sensor includes its historical drift data, inspection records, and maintenance records.

[0010] Furthermore, step S3 includes the following steps: S31. Establish an LSTM deep learning model, with the input being the historical operating data of the fault sensor, and the output being the known true value, the predicted value, and the predicted value of the future operating time. S32. Set a fault threshold, compare the predicted value with the threshold, and predict the time when the predicted value will reach the threshold. S33. Record the predicted data and trigger an alarm according to the set conditions.

[0011] Furthermore, in step S31, before inputting the data into the LSTM deep learning model, the data is preprocessed to extract valid data, including running time, maintenance records, and real-time collected values, and then normalized.

[0012] Furthermore, in step S32, a 1% deviation is set as a warning threshold and a 3% deviation is set as an alarm threshold; the prediction step size for future deviations is 300 steps.

[0013] Furthermore, step S4 includes the following steps: S41. Perform Monte Carlo model calculations on the fault sensor error prediction results obtained from the deep learning model. S42. Based on the Monte Carlo calculation results, give the 95% confidence interval of the predicted value; S43. The confidence interval and predicted value are compiled and fed back to form the final maintenance time assessment decision.

[0014] Furthermore, in step S43, a hierarchical decision is made based on the predicted probability and the confidence interval width.

[0015] The present invention also provides a predictive maintenance time assessment system for T1-level temperature sensors in nuclear power plants, which is used to implement a predictive maintenance time assessment method for T1-level temperature sensors in nuclear power plants, including a data acquisition module, a deep learning module and a fault prediction module connected in sequence; The data acquisition module is used to acquire real-time data measurement values ​​from sensors deployed at different locations in the nuclear power plant; The deep learning module is used to learn and process historical data from fault sensors to derive predicted values. The fault feedback module performs Monte Carlo processing on the predicted values ​​to obtain the confidence interval and the final decision.

[0016] Compared with existing technologies, the beneficial effects of this invention are as follows: This invention uses a combination of LSTM deep learning models and Monte Carlo algorithms to provide early warning of identified faulty sensors, enabling effective predictive maintenance of T1-level temperature sensors in nuclear power plants. On the one hand, it can achieve real-time monitoring of equipment performance status, promptly detect abnormal operating conditions, and take necessary measures to prevent serious consequences caused by unexpected equipment failure. On the other hand, based on quantitative assessment of equipment status and accurate prediction of future equipment status development, targeted aging mitigation measures can be formulated, transforming traditional reactive maintenance into proactive preventive maintenance, thereby reducing the probability of reactor failure and post-failure maintenance costs, while avoiding unnecessary waste caused by over-maintenance.

[0017] This invention makes full use of the redundant sensor data that are common in nuclear power plants. By comparing the readings of the main sensor and redundant sensors under the same monitoring object, faulty sensors can be identified. This effectively solves the problem of misjudgment or missed judgment caused by abnormal data from a single sensor, and significantly improves the accuracy of fault identification and the reliability of the monitoring system.

[0018] This invention, building upon deep learning prediction, further constructs a Monte Carlo mathematical model to analyze the uncertainty of the prediction results. It can provide a 95% confidence interval for the predicted value, quantifying the reliability of the prediction results and solving the problem of difficulty in assessing the reliability of existing deep learning prediction methods. Simultaneously, this invention integrates the confidence interval with the predicted value and the expected time to reach the threshold, and performs tiered decision-making based on the predicted probability and the width of the confidence interval, providing targeted maintenance recommendations. This offers nuclear power plant maintenance personnel a reliable basis for decision-making, significantly improving the scientific rigor and effectiveness of maintenance decisions. Attached Figure Description

[0019] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings: Figure 1 This is a flowchart of a method for assessing the maintenance time of a T1-level temperature sensor in a nuclear power plant, as described in this invention. Figure 2 This is a flowchart of the method for step S1 described in this invention; Figure 3 This is a flowchart of the method for step S2 of the present invention; Figure 4 This is a flowchart of the method for step S3 of the present invention; Figure 5 This is a flowchart of the method for step S4 of the present invention; Figure 6This is a block diagram of a predictive maintenance time assessment system for T1-level temperature sensors in nuclear power plants, as described in this invention. Figure 7 This is a diagram illustrating the parameter change process of the fault sensor described in this invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. It should be noted that, unless otherwise specified, the embodiments and features in the embodiments of the present invention can be combined with each other, and the described embodiments are only some embodiments of the present invention, not all embodiments.

[0021] The predictive maintenance time assessment method for T1-level temperature sensors in nuclear power plants provided in this embodiment has the following overall process: Figure 1 As shown, it includes the following steps: S1. Take readings from the target and redundant sensors to identify faulty sensors; S2. Based on the identified faulty sensor, retrieve the historical operating data and maintenance records of the faulty sensor; S3. Establish a deep learning model, analyze the existing and historical operating data of the fault sensor, and obtain the predicted value of future operating data. S4. Construct a Monte Carlo mathematical model, perform uncertainty analysis on the prediction results of the deep learning model, and obtain the maintenance time evaluation results.

[0022] like Figure 2 As shown, step S1 specifically includes the following steps: S11. Install temperature sensors within the nuclear power plant; S12. Set up the sensor network and determine the main sensor and redundant sensors for the same monitored object; S13. Monitor sensor data in real time and identify faulty sensors with abnormal readings.

[0023] In this embodiment, the monitored object is the reactor coolant outlet temperature. A main sensor and two independent redundant sensors are configured for this object to ensure that data from each sensor can be compared and verified. The system monitors the data from all sensors in real time. By comparing the readings of different sensors for the same monitored object, faulty sensors with significantly different readings are identified, effectively avoiding misjudgments or missed diagnoses caused by abnormal data from a single sensor.

[0024] Once the faulty sensor is identified, the system enters the data retrieval and processing phase, such as... Figure 3 As shown, step S2 specifically includes the following steps: S21. Obtain the location, type, and real-time monitoring data of the faulty sensor; S22. Retrieve historical operating data and maintenance records of the faulty sensor, as well as steady-state and transient operating data of the redundant sensor; S23. Organize, preprocess, and summarize the retrieved data to form a standardized dataset, providing data support for subsequent model analysis.

[0025] In step S22, the retrieved historical data of the faulty sensor includes its historical drift data, inspection records, and maintenance records.

[0026] After data processing is complete, the system enters the deep learning prediction stage, such as... Figure 4 As shown, step S3 includes the following steps: S31. Establish an LSTM deep learning model, with the input being the historical operating data of the fault sensor, and the output being the known true value, the predicted value, and the predicted value of the future operating time. S32. Set a fault threshold, compare the predicted value with the threshold, and predict the time when the predicted value will reach the threshold. S33. Record the predicted data and trigger an alarm according to the set conditions.

[0027] In step S31, before inputting the data into the LSTM deep learning model, the preprocessed dataset is further processed to extract valid data such as runtime, maintenance records, and real-time acquisition values ​​and perform normalization processing. The processed data is then matched with the model's input and output formats and the corresponding parameters are set.

[0028] In step S32, a 1% deviation is set as the warning threshold and a 3% deviation is set as the alarm threshold; the prediction step size for future deviations is 300 steps.

[0029] To quantify the reliability of the prediction results, the system further performs uncertainty analysis, such as... Figure 5 As shown, step S4 includes the following steps: S41. Perform Monte Carlo model calculations on the fault sensor error prediction results obtained from the deep learning model. S42. Based on the Monte Carlo calculation results, give the 95% confidence interval of the predicted value; S43. The confidence interval and predicted value are compiled and fed back to form the final maintenance time assessment decision.

[0030] In step S43, a tiered decision is made based on the predicted probability and the confidence interval width. The specific tiering rules are as follows: When the predicted probability is high and the confidence interval is narrow, it indicates that the model is very certain about the prediction result; for example, P(warning) = 0.9, confidence interval [0.85, 0.95], indicating that the sensor is about to fail. When the predicted probability is high but the confidence interval is wide, it indicates that the model tends towards a fault state but the confidence level is not high; for example, P(fault) = 0.7, confidence interval [0.4, 0.9], requiring the integration of other information or an increase in monitoring frequency. When the probability distribution is flat and the confidence interval is wide, it indicates that the model is very uncertain about the current state, possibly indicating insufficient data or a state transition critical point; this should be considered a high-risk signal, and immediate inspection is recommended.

[0031] This embodiment also provides a predictive maintenance time assessment system for T1-level temperature sensors in nuclear power plants that implements the above method, and its structural block diagram is shown below. Figure 6 As shown, the system includes a data acquisition module, a deep learning module, and a fault prediction module connected in sequence. The data acquisition module is used to acquire real-time data measurements from sensors located at different locations in the nuclear power plant. Specifically, it is required to acquire distributed sensor data, including redundant data from multiple sensors, data that can reflect steady-state and transient operation, as well as historical drift data, inspection records, and maintenance records of the sensors.

[0032] The deep learning module is used to learn and process historical data from fault sensors to obtain predicted values. Its working process involves organizing the sensor data, extracting effective data such as running time, maintenance records, and real-time acquisition values, and performing normalization processing. The processed data is then matched with the input and output of the model, corresponding parameters are set, and an appropriate LSTM deep learning algorithm is selected for calculation and output of the prediction results.

[0033] The fault feedback module performs Monte Carlo processing on the predicted values ​​output by the deep learning module to obtain a 95% confidence interval for the predicted values. It then integrates the prediction results with the expected time to reach the threshold and provides targeted maintenance suggestions.

[0034] like Figure 7 The graph showing the parameter changes of the faulty sensor intuitively demonstrates the evaluation effect of this invention. The time axis in the graph presents the historical operating error of the faulty sensor, the actual and predicted values ​​within the error occurrence period, and the predicted error values ​​and Monte Carlo 95% confidence interval distribution for future periods. In the graph, a green background indicates that the error is within the healthy range, a yellow background indicates that the error has reached the warning range, and a red background indicates that the error has reached the alarm threshold. This allows maintenance personnel to intuitively grasp the trend of sensor operating status changes and assists in making scientific maintenance decisions.

[0035] The specific embodiments of the present invention disclosed above are merely illustrative of the invention. These embodiments do not exhaustively describe all details, nor do they limit the invention to the specific embodiments described. Many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention.

Claims

1. A method for assessing the maintenance time of a T1-level temperature sensor in a nuclear power plant, characterized in that: Includes the following steps: S1. Take readings from the target and redundant sensors to identify faulty sensors; S2. Based on the identified faulty sensor, retrieve the historical operating data and maintenance records of the faulty sensor; S3. Establish a deep learning model, analyze the existing and historical operating data of the fault sensor, and obtain the predicted value of future operating data. S4. Construct a Monte Carlo mathematical model, perform uncertainty analysis on the prediction results of the deep learning model, and obtain the maintenance time evaluation results.

2. The method for assessing the maintenance time of a T1-level temperature sensor in a nuclear power plant according to claim 1, characterized in that: Step S1 includes the following steps: S11. Install temperature sensors within the nuclear power plant; S12. Set up the sensor network and determine the main sensor and redundant sensors for the same monitored object; S13. Monitor sensor data in real time and identify faulty sensors with abnormal readings.

3. The method for assessing the maintenance time of a T1-level temperature sensor in a nuclear power plant according to claim 1, characterized in that: Step S2 includes the following steps: S21. Obtain the location, type, and real-time monitoring data of the faulty sensor; S22. Retrieve historical operating data and maintenance records of the faulty sensor, as well as steady-state and transient operating data of the redundant sensor; S23. Organize, preprocess, and summarize the retrieved data.

4. The method for assessing the maintenance time of a T1-level temperature sensor in a nuclear power plant according to claim 3, characterized in that: In step S22, the retrieved historical data of the faulty sensor includes its historical drift data, inspection records, and maintenance records.

5. The method for assessing the maintenance time of a T1-level temperature sensor in a nuclear power plant according to claim 1, characterized in that: Step S3 includes the following steps: S31. Establish an LSTM deep learning model, with the input being the historical operating data of the fault sensor, and the output being the known true value, the predicted value, and the predicted value of the future operating time. S32. Set a fault threshold, compare the predicted value with the threshold, and predict the time when the predicted value will reach the threshold. S33. Record the predicted data and trigger an alarm according to the set conditions.

6. The method for assessing the maintenance time of a T1-level temperature sensor in a nuclear power plant according to claim 5, characterized in that: In step S31, before inputting the data into the LSTM deep learning model, the data is preprocessed to extract valid data, including running time, maintenance records, and real-time collected values, and then normalized.

7. The method for assessing the maintenance time of a T1-level temperature sensor in a nuclear power plant according to claim 5, characterized in that: In step S32, a 1% deviation is set as the warning threshold and a 3% deviation is set as the alarm threshold; the prediction step size for future deviations is 300 steps.

8. The method for assessing the maintenance time of a T1-level temperature sensor in a nuclear power plant according to claim 1, characterized in that: Step S4 includes the following steps: S41. Perform Monte Carlo model calculations on the fault sensor error prediction results obtained from the deep learning model. S42. Based on the Monte Carlo calculation results, give the 95% confidence interval of the predicted value; S43. The confidence interval and predicted value are compiled and fed back to form the final maintenance time assessment decision.

9. The method for assessing the maintenance time of a T1-level temperature sensor in a nuclear power plant according to claim 8, characterized in that: In step S43, a hierarchical decision is made based on the predicted probability and the confidence interval width.

10. A predictive maintenance time assessment system for T1-level temperature sensors in nuclear power plants, used to implement the predictive maintenance time assessment method for T1-level temperature sensors in nuclear power plants as described in any one of claims 1 to 10, characterized in that: It includes a data acquisition module, a deep learning module, and a fault prediction module connected in sequence; The data acquisition module is used to acquire real-time data measurement values ​​from sensors deployed at different locations in the nuclear power plant; The deep learning module is used to learn and process historical data from fault sensors to derive predicted values. The fault feedback module performs Monte Carlo processing on the predicted values ​​to obtain the confidence interval and the final decision.

Images