Nuclear control power supply device TTF distribution calculation method and system based on fusion model

Abstract

The invention discloses a nuclear control power supply device TTF distribution calculation method and system based on a fusion model. The method comprises the following steps: acquiring first data and second data; the first data are data parameters before and after failure of the nuclear control power supply device, and the second data are historical use data of the nuclear control power supply device; based on the first data and the second data, constructing a failure physical sub-model and a data driving sub-model; fusing the failure physical sub-model and the data driving sub-model to obtain a data driving model with an embedded failure mechanism, namely a fusion model; based on the fusion model, constructing and training a TTF distribution calculation model; and based on the TTF distribution calculation model, performing fitting calculation to obtain TTF probability distribution of the nuclear control power supply device. Decision support is provided for an active operation and maintenance system, the reactor operation risk and maintenance cost are reduced, and the safe operation level and guarantee capacity are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of nuclear control rod drive power supply and the field of electronic equipment fault prediction, and in particular to a TTF distribution calculation method and system for a nuclear control power supply device based on a fusion model. Background Art

[0002] Nuclear energy is a low-carbon, clean and efficient energy source. The development of nuclear energy has become an important means for my country to ensure energy security and optimize its energy structure. Nuclear safety is the lifeline of nuclear energy development. On the other hand, the technical and economic feasibility of nuclear energy is also receiving more and more attention. The nuclear control power supply device is one of the important equipment for reactor control. It performs normal reactor startup and shutdown functions according to the operator's instructions, and manually operates the control rods to change the core reactivity under various effective working conditions; and performs safe shutdown and power reduction operation functions according to the operator or protection system instructions; when the reactor power is automatically adjusted, the regulating pound group receives the automatic rod speed signal and direction signal sent by the reactor control system through the rod control system, so that the reactor power can automatically track the load changes and maintain it at the required power level. The external interfaces of the nuclear control power supply device and the reactor control, protection and other devices are as follows: Figure 1 shown.

[0003] Although the nuclear control power supply device has been strictly inspected during the manufacturing stage, as the operation time increases, some faults often occur, and the device needs to be inspected and repaired. The quality of the maintenance work of the nuclear control power supply device directly reflects the guarantee capability of the entire reactor instrumentation and control system, and is also related to the safety and economy of the reactor. At present, operation and maintenance personnel lack effective special tools or technical means to monitor the working status of the nuclear control power supply device, determine the potential failure of the equipment, and evaluate the performance status of the equipment, which limits the improvement of the pertinence, timeliness and effectiveness of maintenance activities.

[0004] At present, the reliability prediction of reactor instrumentation and control system equipment is still at a preliminary level. It mainly uses prototype tests or general failure physical models of key components to calculate the time to failure (TTF) of a single sample, and integrates multiple sample data to calculate the mean time to failure (MTTF). This statistical method uses the MTTF obtained to estimate the individual TTF, ignoring the impact of uneven TTF distribution and unable to reflect the true distribution of equipment TTF. Its prediction error is too large to provide accurate decision-making information for active operation and maintenance, and is insufficient to meet users' urgent needs for equipment operation and maintenance assurance.

[0005] In view of this, this application is hereby filed. Summary of the invention

[0006] The purpose of the present invention is to provide a TTF distribution calculation method and system for a nuclear control power supply device based on a fusion model. By collecting and organizing the failure data of the nuclear control power supply device and analyzing the failure mechanism of the equipment under different working conditions, a reliability prediction method suitable for the reactor instrumentation and control system is formed, and the probability distribution of the time to failure (TTF) of the nuclear control power supply device is calculated. The prediction error is small, and decision support is provided for the active operation and maintenance system, thereby reducing the operating risk and maintenance cost of the reactor and improving the safe operation level and guarantee capability.

[0007] The present invention is achieved through the following technical solutions:

[0008] In a first aspect, the present invention provides a TTF distribution calculation method for a nuclear-controlled power supply device based on a fusion model, the method comprising:

[0009] Acquire first data and second data; the first data is data parameters of the nuclear-controlled power supply device before and after failure, and the second data is historical usage data of the nuclear-controlled power supply device;

[0010] Based on the first data and the second data, a failure physics sub-model and a data-driven sub-model are constructed; and the failure physics sub-model and the data-driven sub-model are fused to obtain a data-driven model with an embedded fault mechanism, that is, a fusion model;

[0011] Based on the fusion model, a TTF distribution calculation model is constructed and trained; based on the TTF distribution calculation model, the TTF probability distribution of the nuclear control power supply device is obtained by fitting calculation.

[0012] Furthermore, the method also includes: according to the TTF probability distribution, setting an alarm to provide a repair or replacement time for the nuclear control power supply device.

[0013] Further, the data parameters include temperature data, voltage and current data, and control signal data;

[0014] The temperature data is the temperature data measured for the full-bridge IGBT device;

[0015] The voltage and current data include the voltage and current data measured on the bus, the phase voltage and current data measured on the output three-phase AC power supply, and the current and voltage data measured on the optocoupler and electrolytic capacitor;

[0016] The control signal data is the IGBT drive signal phase difference data measured for the full-bridge IGBT control signal.

[0017] Furthermore, the failure physics sub-model and data-driven sub-model are constructed, including:

[0018] The construction process of the failure physics sub-model is as follows:

[0019] Analyze the failure mechanism of the drive module and key components of the nuclear power supply device, study the aging failure mechanism of the drive module and key components under the optimal characteristic parameter failure mode, and determine the failure signs and characteristic parameters;

[0020] Based on the failure mechanism, a multi-failure mechanism physical evolution model of key components and drive modules is established and used as a failure physics sub-model;

[0021] The effectiveness of selecting characteristic parameters of the aging process of each device is verified by simulation, and the effectiveness of failure mechanism analysis of key components is verified by conducting accelerated aging experiments on IGBTs, capacitors, and optocouplers;

[0022] The construction process of the data-driven sub-model is as follows:

[0023] Clarify the aging and failure characteristic parameters of the drive module and its key components, build a database of aging and failure characteristic parameters of the drive module and key components, and analyze the coupling relationship between each key aging characteristic parameter and the aging characteristic parameter of the driver board to obtain a data-driven sub-model.

[0024] Furthermore, a TTF distribution calculation model is constructed and trained, including:

[0025] Integrate the fault mechanism model of IGBT devices and find the operating boundary and model initial value of the failed device; construct the boundary residual term, model residual term and key feature data residual term of the fusion model; the key feature data residual term includes characteristic parameters such as temperature, current and voltage of key nodes;

[0026] Construct a multi-input and multi-output neural network, and define the activation function of the neural network as tanh and normal Glorot initialization; after defining the optimization domain of each residual, construct an overall loss function including boundary residual term, model residual term and key feature data residual term; accordingly, the constraint function of the mathematical model of multiple failure mechanisms, the residual function of the target value and the actual estimated value, the constraint function of the initial condition and boundary condition, and other relevant physical system information are substituted into the loss function of the neural network through operators;

[0027] Use automatic differentiation technology to build Jacobi matrix and Hessian matrix modules, and realize high-order differential calculations through nesting; set the optimizer mode, learning rate, number of iterations and loss weight, and compile the fusion model; and train the neural network through the gradient descent method and minimize the loss function to find the optimal value until the loss drops below the set value to obtain a trained fusion model.

[0028] Furthermore, based on the TTF distribution calculation model, the TTF probability distribution of the nuclear control power supply device is obtained by fitting calculation, including:

[0029] Determine failure signs and characteristic parameters, and use failure threshold setting and TTF distribution calculation model to perform data trend analysis and Trend function estimation on important parameters and time;

[0030] The remaining life of the nuclear control power supply device is estimated by using the time when the parameter does not exceed the failure threshold; and the TTF distribution calculation model is repeated multiple times to fit the TTF probability distribution of the nuclear control power supply device.

[0031] In a second aspect, the present invention further provides a TTF distributed calculation system for a nuclear-controlled power supply device based on a fusion model, the system comprising:

[0032] An acquisition unit, used to acquire first data and second data; the first data is data parameters of the nuclear-controlled power supply device before and after failure, and the second data is historical usage data of the nuclear-controlled power supply device;

[0033] A model building unit is used to build a failure physics sub-model and a data-driven sub-model based on the first data and the second data; and fuse the failure physics sub-model and the data-driven sub-model to obtain a data-driven model with an embedded fault mechanism, that is, a fusion model; and build and train a TTF distribution calculation model based on the fusion model;

[0034] The TTF calculation unit is used to obtain the TTF probability distribution of the nuclear control power supply device by fitting calculation based on the TTF distribution calculation model.

[0035] Furthermore, the system also includes:

[0036] The alarm unit is used to set an alarm to provide the repair or replacement time of the nuclear control power supply device according to the TTF probability distribution.

[0037] In a third aspect, the present invention provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the TTF distribution calculation method of a nuclear-controlled power supply device based on a fusion model when executing the computer program.

[0038] In a fourth aspect, the present invention further provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the above-mentioned TTF distribution calculation method of a nuclear-controlled power supply device based on a fusion model is implemented.

[0039] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0040] 1. The present invention provides a method and system for calculating the TTF distribution of a nuclear control power supply device based on a fusion model. By collecting and collating the failure data of the nuclear control power supply device and analyzing the failure mechanism of the equipment under different working conditions, a reliability prediction method suitable for the reactor instrumentation and control system is formed. The probability distribution of the TTF of the nuclear control power supply device before failure is calculated, and the prediction error is small, which provides decision support for the active operation and maintenance system, reduces the operating risk and maintenance cost of the reactor, and improves the level of safe operation and guarantee capability.

[0041] 2. The TTF distribution calculation method and system of the nuclear power supply device based on the fusion model of the present invention has a high degree of fit between the physical test data and the historical operation and maintenance data in the equipment operation and maintenance guarantee activities, and the prediction effect is better, so that the TTF reliability prediction index estimation of this type of equipment is no longer a fuzzy fixed value estimation. According to certain data and model inputs, the degradation curves of multiple groups of characteristic parameters and the TTF life distribution of the equipment can be obtained, which is conducive to users to grasp the failure information of the equipment throughout its life, understand and predict when the equipment failure will occur, and carry out active operation and maintenance of the equipment. When an unexpected failure occurs, a simple maintenance activity is triggered, thereby realizing autonomous guarantee and reducing the cost of use and guarantee. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] The drawings described herein are used to provide a further understanding of the embodiments of the present invention, constitute a part of this application, and do not constitute a limitation of the embodiments of the present invention. In the drawings:

[0043] Figure 1 It is a schematic diagram of the external interface between the nuclear control power supply device and the reactor control, protection and other devices;

[0044] Figure 2 It is a schematic diagram of the TTF distribution calculation process based on PIML of the present invention;

[0045] Figure 3 It is a flow chart of the TTF distribution calculation method of the nuclear-controlled power supply device based on the fusion model of the present invention;

[0046] Figure 4 This is a schematic diagram of the PINN principle of the present invention;

[0047] Figure 5 This is a schematic diagram of the Deep Operator Net principle of the present invention;

[0048] Figure 6 This is a schematic diagram of the process of Example 1 of the present invention;

[0049] Figure 7 This is a system framework diagram of the failure data acquisition device of the present invention;

[0050] Figure 8This is a technical flow chart of the aging failure mechanism analysis of key components based on the failure physics sub-model of the present invention;

[0051] Fig. 9 A technical roadmap for selecting aging characteristic parameters of a driving module based on a data-driven sub-model of the present invention;

[0052] Fig.10 The structure block diagram of the TTF distributed calculation system of the nuclear-controlled power supply device based on the fusion model of the present invention. DETAILED DESCRIPTION

[0053] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with embodiments and drawings. The exemplary embodiments of the present invention and their description are only used to explain the present invention and are not intended to limit the present invention.

[0054] At present, the more perfect TTF calculation technology adopts the "fusion" technology combining the above two methods, combining the failure sign monitoring and reasoning method and the failure physics (PoF) model method for prediction, so as to use their respective advantages to achieve all the goals of the computing system and overcome the limitations of using a single method. The failure sign monitoring and reasoning method can provide a certain equipment life status assessment when the amount of data is sufficient; while the PoF model method helps to determine the root cause of failure, and provides the threshold and failure definition of system parameters, thereby supporting the evaluation of TTF. Finally, the two prediction results are weakly competitively decided to give a comprehensive TTF evaluation report. This "fusion" method combines the strengths and weaknesses of the two methods and makes a comprehensive decision. It has certain technical completeness and engineering application value, but the accuracy of the results in the reliability prediction part is low and generally too conservative. The fundamental reason is that the experience knowledge and model laws accumulated in the industry over the long history have not been truly organically integrated with the existing historical failure data and real-time monitoring data information. The two prediction technology routes are basically carried out independently, and there is a serious overlapping coupling and complementary relationship between the input information of the two.

[0055] In the final analysis, all our knowledge comes from empirical data. In the past, people obtained a large-scale knowledge model system through analysis, deduction, fitting, induction and other methods. Later, in the process of machine learning development, they fully explored the information that was usually considered useless and messy in the past, and had further learning and cognition of the system. Now, we need a more complete learning machine to integrate and re-integrate the knowledge from two sources, and give it valuable prior knowledge, laws and connotations from human society at the beginning of the establishment of the learning model, so as to realize the true integration of knowledge model-based methods and data-driven technologies, so that TTF computing technology has the ability to face complex nonlinear, failure mechanism coupled systems and even chaotic systems.

[0056] In order to ensure that the nuclear control power supply device can complete specific functions within the expected life span, ensure the safe and reliable operation of the equipment, reduce the operating risk and maintenance cost of the reactor, and also to meet the urgent needs of the equipment users to improve the safety level and guarantee capability, it is urgent to carry out reliability analysis and prediction of reactor instrumentation and control system equipment represented by nuclear control power supply devices. By collecting and collating the failure data of nuclear control power supply devices, analyzing the failure mechanism of equipment under different working conditions, forming a set of reliability prediction methods suitable for reactor instrumentation and control systems, calculating the probability distribution of the time to failure (TTF) of nuclear control power supply devices, providing decision support for the active operation and maintenance system, reducing the operating risk and maintenance cost of the reactor, and improving the safe operation level and guarantee capability.

[0057] The technical route of the present invention is:

[0058] First, the data parameters before and after the failure of the nuclear power supply device and the historical usage data of the nuclear power supply device should be fully monitored, collected and recorded; the knowledge of the physical model laws of electronic components, printed circuit boards, through holes, connecting wires, etc. at all levels in the system should be fully mastered to provide necessary information for the establishment of models and algorithms for the detection and prediction of the status of the equipment;

[0059] Secondly, the sensor system processes and stores the data collected above, provides a processing method, and performs real-time analysis and feature extraction on the sensor data by integrating the data-driven method and the PoF method.

[0060] Then, through data monitoring and analysis, when abnormal performance is detected, the parameters causing the abnormality are isolated. The associated parameters that reflect or cause changes in equipment performance are found. The established PoF model method can be used to determine failures and isolate parameters. Then, based on the extracted features, the fault baseline is compared with the features of the monitoring data to detect anomalies.

[0061] Furthermore, a database is established by combining isolation parameters, monitoring data, and historical data to provide data support for the subsequent DeepONet and embedded physical mechanism neural network (PINN) learning models; on the other hand, the knowledge of physical model rules at all levels is integrated and collected to establish a PoF-based knowledge base to provide physical information embedding for the subsequent embedded physical mechanism machine learning (PIML) model. The DeepONet-based learning model can supplement the PoF knowledge base with models or parameters based on the input of the database, and this method can be used to fill in gaps or remodel for different failure mechanisms or situations that exceed the failure threshold defined by the selected PoF model.

[0062] Finally, comprehensively determine the failure signs and characteristic parameters, use the failure threshold setting and the above two categories of methods (rich PoF knowledge base and PINN model) to estimate the trend function of important parameters and time, and use the time (TTF) when the parameter does not exceed the failure threshold to estimate the remaining life of the equipment. Repeat the above evaluation method many times to fit the TTF probability distribution of the equipment. And set an alarm based on the calculated TTF distribution to provide a reasonable repair or replacement time for the equipment. The schematic diagram of the TTF distribution calculation process based on PIML is as follows: Figure 2 shown.

[0063] The technology of the present invention combines the latest research results in artificial intelligence and data science for dealing with partial differential equation problems. It is a deep integration of the two major technical routes in the existing TTF calculation field. It has the potential to develop to the ultimate complete solution of such TTF calculation problems, and has great research value and field development connotation. At present, the research direction of deep integration of physical models and data science is relatively blank in the field of TTF prediction and calculation of domestic electronic products. It introduces the existing knowledge model physical information accumulated in the industry into the architecture design of the learning model, which can improve the algorithm prediction performance of TTF calculation technology in the case of insufficient data volume or poor data quality, difficult analysis of physical model laws by traditional methods, difficult decoupling of interaction of multiple failure mechanism models in the system, and inapplicability of existing models of complex systems. It can also realize the expansion of the original knowledge model library, improve the ability to fill the missing gaps of the existing knowledge model and the ability to establish analytical degradation models of complex coupled systems such as board-level system-level. At present, the difficulties in the practical application of the above-mentioned TTF calculation technology exist not only in the field of reactor instrumentation and control equipment, but also in various industries.

[0064] In the equipment operation and maintenance guarantee activities, the calculation method has a high degree of fit between the physical test data and the historical operation and maintenance data, and the prediction effect is better, so that the TTF reliability prediction index estimation of this type of equipment is no longer a fuzzy fixed value estimation. According to certain data and model inputs, the degradation curves of multiple sets of characteristic parameters and the TTF life distribution of the equipment can be obtained, which is conducive to users to master the failure information of the equipment throughout its life, understand and predict when the equipment failure will occur, and carry out active equipment operation and maintenance. When an unexpected failure occurs, a simple maintenance activity is triggered, thereby realizing autonomous guarantee and reducing the cost of use and guarantee.

[0065] Example 1

[0066] like Figure 3 and Figure 6 As shown, the TTF distribution calculation method of the nuclear control power supply device based on the fusion model of the present invention includes:

[0067] Step 1, obtaining first data and second data; the first data is data parameters of the nuclear-controlled power supply device before and after failure, and the second data is historical usage data of the nuclear-controlled power supply device;

[0068] Step 2: construct a failure physics sub-model and a data-driven sub-model based on the first data and the second data; and fuse the failure physics sub-model and the data-driven sub-model to obtain a data-driven model with an embedded fault mechanism, i.e., a fusion model;

[0069] Step 3: Based on the fusion model, a TTF distribution calculation model is constructed and trained; based on the TTF distribution calculation model, a TTF probability distribution of the nuclear control power supply device is obtained by fitting calculation.

[0070] Step 4: According to the TTF probability distribution, an alarm is set to provide the repair or replacement time of the nuclear control power supply device.

[0071] The present invention uses key equipment such as nuclear control power supply devices in the reactor instrumentation and control system as the carrier for design, collects data from the system instrumentation and control equipment test and operation and maintenance process, develops a failure data collection device, uses data-driven and failure physical models as analysis inputs, and constructs a dynamic, closed-loop instrumentation and control product reliability index prediction technology, grasps the failure laws of nuclear control power supply devices and their key components, and provides technical support for active operation and maintenance and reliability improvement of nuclear control power supply devices.

[0072] In this embodiment, the present invention is based on a set of data acquisition devices for nuclear control power supply devices to carry out the collection of fault data. Considering both software and hardware aspects, the acquisition device automatically collects and records according to the preset configuration, judges the correctness of the test results, and can generate corresponding analysis tables for the test results. The data stored in the data acquisition device can be exported through storage media or communication networks, which can meet the requirements of application scenarios such as factory, on-site, and portable testing. Failure data of equipment during operation and maintenance is collected, and accelerated life tests of equipment, boards and components are carried out. The acquisition device is used to monitor and record the failure and aging process of key nodes of the equipment in real time, including failure data measurement of key components. The current front-end acquisition is realized by Hall sensors, the voltage signal front-end acquisition is realized by the precision resistor voltage divider method, and the temperature signal acquisition is realized by NTC temperature sensors. The system framework of the data acquisition device is as follows: Figure 7 shown.

[0073] The data acquisition device realizes the following functions: (1) temperature measurement and real-time monitoring of full-bridge IGBT devices; (2) bus voltage and current measurement; (3) output three-phase AC power supply phase voltage and current measurement; (4) full-bridge IGBT control signal measurement and IGBT drive signal phase difference measurement; (5) current and voltage measurement of optocouplers and electrolytic capacitors; (6) temperature, voltage, current and control signal waveform generation, and the waveform status can be recorded through the host computer software.

[0074] Based on the above, we can know that:

[0075] The data parameters of the present invention include temperature data, voltage and current data, and control signal data; wherein the temperature data is the temperature data measured for the full-bridge IGBT device; the voltage and current data include the voltage and current data measured for the bus, the phase voltage and current data measured for the output three-phase AC power supply, and the current and voltage data measured for the optocoupler and the electrolytic capacitor; the control signal data is the phase sequence difference data of the IGBT drive signal measured for the full-bridge IGBT control signal.

[0076] The historical usage data of the present invention includes environmental parameters and parameters of the nuclear-controlled power supply device under various working conditions.

[0077] In this embodiment, for the nuclear control power supply device system, each functional subsystem, key circuit components, and core component levels, a system functional principle model is established according to the hierarchical composition of the nuclear control power supply device system. Based on this, reliability analysis based on the system-level model is carried out, and potential failure modes of the system are searched according to multiple failure clues. The functional relationship of each unit is analyzed, the key components and their failure modes that have a significant impact on the system reliability are determined, and a system-level reliability analysis model of the nuclear control power supply device is established.

[0078] Combined with the typical mission environment of nuclear power control device, the hierarchical composition of subsystems and circuit components such as functional control chassis, low-frequency power device chassis, main transformer chassis, low-frequency signal unit, isolation drive unit, protection unit, DC unit, D automatic unit, rod speed unit, reverse insertion unit, shutdown unit, main circuit, etc. is sorted out from top to bottom, and the possible working loads during the mission are analyzed. The material / information / energy modeling method is used to establish the functional principle model of nuclear power control device based on system composition, functional principle, input and output information.

[0079] Based on the system functional principle model, using the failure mode search method that considers operating conditions and loads, the potential failure modes of each level of the system are sorted out according to different operating conditions and loads and mission scenarios, based on failure clues such as loss of function, discontinuity of function, and incomplete function.

[0080] Based on the results of the system's failure mode analysis, according to the functions of the nuclear control power supply device itself and external influences, the severity categories of failure modes at each level are defined, the list of Class I and Class II failure modes and the list of key components and key parts that have a significant impact on system reliability are determined, and a system-level fault evolution model of the nuclear control power supply device is established.

[0081] Based on the above system-level fault modeling analysis, a failure physical sub-model and a data-driven sub-model are constructed. Specifically, starting from the current state of the equipment, combined with the current environmental conditions, relevant parameters and historical data, relevant algorithms are used to predict, analyze and judge the failure and operation of the equipment, determine the nature, category, degree and cause of the failure, and based on the functional principles and reliability relationships obtained from the above system-level reliability analysis, a method based on failure physical models and failure data-driven is used to predict the development trend of the equipment status in the next period of time. Among them, various existing failure physical models or DeepONet and other methods are used to establish a device-level life analysis model, and a data-driven method is used to select the aging characteristic parameters of key components and set the fault threshold, and experiments are designed to verify the accuracy of the corresponding models.

[0082] Specifically, the construction process of the failure physics sub-model is as follows:

[0083] Analyze the failure mechanism of the drive module and key components of the nuclear power supply device, study the aging failure mechanism of the drive module and key components under the optimal characteristic parameter failure mode, and determine the failure signs and characteristic parameters;

[0084] Based on the failure mechanism, a multi-failure mechanism physical evolution model of key components and drive modules (including Coffin-Manson model, LESIT model, Bayerer model, capacitor capacitance degradation model, etc.) is established and used as a failure physics sub-model;

[0085] The effectiveness of selecting characteristic parameters of the aging process of each device is verified by simulation, and the effectiveness of failure mechanism analysis of key components is verified by conducting accelerated aging experiments on IGBTs, capacitors, and optocouplers;

[0086] Among them, the technical process of aging failure mechanism analysis of key components based on failure physics sub-model is as follows: Figure 8 shown.

[0087] Specifically, the construction process of the data-driven sub-model is as follows:

[0088] Clarify the aging failure characteristic parameters of the drive module and its key components, build a database of aging failure characteristic parameters of the drive module and key components based on artificial intelligence, data screening and other methods, and analyze the coupling relationship between each key aging characteristic parameter and the aging characteristic parameter of the driver board to obtain a data-driven sub-model. Based on the data-driven sub-model, select the aging failure characteristic parameters and fault thresholds of the driver board.

[0089] Among them, the technical route for selecting the aging characteristic parameters of the driver module based on the data-driven sub-model is as follows: Fig. 9 shown.

[0090] In this embodiment, in view of the fact that the interaction relationship between the failure mechanism models of multiple IGBTs, electrolytic capacitors and other devices in the nuclear control power supply device is difficult to decouple, the existing models of complex systems are difficult to characterize, or the model parameters have the characteristics of evolving over time, a PINN model with an embedded fault mechanism is established based on the established PoF model of the corresponding device module and physical laws such as electrothermal force and empirical formulas, and the known physical laws and mathematical characteristics of the model are introduced into the design of the neural network architecture.

[0091] Specifically, a TTF distribution calculation model is constructed and trained, including:

[0092] First, integrate the Coffin-Manson model, LESIT model, Bayerer model, capacitor capacitance degradation model and other fault mechanism models of IGBT devices, and find the operating boundaries and model initial values ​​of these complex failure devices; based on this, construct the boundary residual terms and model residual terms of the fusion model; then, based on the results of the above data selection and collection, construct the key feature data residual terms, including the temperature, current, voltage and other characteristic parameters of the key nodes.

[0093] Then, a multi-input and multi-output neural network is constructed, and the activation function of the neural network is defined as tanh and normal Glorot initialization; after defining the optimization domain of each residual, an overall loss function is constructed including boundary residual terms, model residual terms and key feature data residual terms; based on this, the constraint functions of the mathematical models of multiple failure mechanisms, the residual functions of the target values ​​and the actual estimated values, the constraint functions of the initial conditions and boundary conditions, and other related physical system information are substituted into the loss function of the neural network through operators;

[0094] Then, the Jacobi matrix and Hessian matrix modules are constructed using automatic differentiation technology, and high-order differential calculations are achieved through nesting; the optimizer mode, learning rate, number of iterations, and loss weight are set for each item, and the fusion model is compiled; and the neural network is trained by the gradient descent method and the loss function is minimized to find the optimal value until the loss drops below the set value, thereby obtaining a trained fusion model.

[0095] The neural network trained to convergence is a near-optimal solution for the prediction of the comprehensive data model information. The final TTF calculation value is obtained by inputting the characteristic parameters in the domain. In this way, a special neural network is obtained that embeds the existing failure physics sub-model, empirical formula laws and constraints. This type of embedding method has strong versatility and can basically realize the embedding of any physical model with an analytical expression, and adjust the training process of the neural network so that it can learn and master the embedded physical laws. The schematic diagram of the latter general PINN model principle is shown in the figure. Figure 4 shown.

[0096] In order to solve the problem that the failure mechanism model of components such as transformers and optocouplers is missing or the failure physical model of a certain mechanism is difficult to establish, a deep operator network (Deep Operator Net) is established to identify the system (parameters) of the model, and the temperature, voltage ratio, current ratio and other data related to the unknown field are used for training. The empirical physical formulas such as the CTR degradation of the optocoupler with known related mechanisms are input into the discretized Branch net, and the expected but unknown mechanism information (initial or boundary conditions in the calculation domain) is input into the Trunk net that ensures the continuity property, thereby avoiding the discretization of the output of the traditional neural network. Finally, the output of the two parts of the neural network is dot-producted (or other operations) to obtain the basis function expansion of the estimation model, and a higher system level (such as board-level system level) or failure physics (PoF) continuous analytical model that is difficult to establish with traditional methods is established, so as to enrich and supplement the existing incomplete PoF knowledge base. In the case of interactive coupling of multi-level physical laws, DeepM&Mnet can also be further established based on this method to solve the difficult problem of modeling multi-physics complex systems. The schematic diagram of the model principle of Deep Operator Net is shown in the figure. Figure 5 shown.

[0097] In this embodiment, based on the TTF distribution calculation model, the TTF probability distribution of the nuclear-controlled power supply device is obtained by fitting and calculation, including:

[0098] Determine failure signs and characteristic parameters, and use failure threshold setting and TTF distribution calculation model to perform data trend analysis and Trend function estimation on important parameters and time;

[0099] The remaining life of the nuclear control power supply device is estimated by using the time when the parameter does not exceed the failure threshold; and the TTF distribution calculation model is repeated multiple times to fit the TTF probability distribution of the nuclear control power supply device.

[0100] In the calculation process of the above TTF probability distribution: according to the system-level fault evolution model, failure symptoms and characteristic parameters of the nuclear power control device, the failure threshold of the system is set. Based on the established database, model library and device-level data model fusion drive model, the PINN model of the fusion failure mechanism is used to estimate the trend function of important parameters and time, and the time (TTF) when the parameter does not exceed the failure threshold is used to estimate the remaining life of the device level. The random process principle of equipment failure is embedded through the above embedding method, including equations such as Markov process and martingale process, and the device-level TTF probability distribution is output. The above evaluation method can also be repeated multiple times to directly obtain the TTF probability distribution of the device using a constrained fitting algorithm. Finally, the calculation results of the device level are input into the system-level fault evolution model, and the system-level TTF distribution is output. The operation and maintenance personnel set a reasonable expected repair time and maintenance opportunity based on the calculated TTF distribution of the nuclear power control device.

[0101] The PIML method makes full use of the industry's knowledge accumulation over long time scales in the past and the data processing capabilities of the learning model. It still has good prediction performance when the data volume is insufficient or the data quality is poor, or when the physical model or law is difficult to analyze. This achieves a deep combination of the two main technical routes in this field - failure physics models and data-driven technologies, which helps us to deeply grasp and understand the degradation and aging process of equipment objects.

[0102] The present invention takes the nuclear control power supply device of the reactor as the research object, aims at engineering application, combines the feedback of equipment operation and maintenance experience and the continuous technical service tracking, analyzes the relevant methods such as data acquisition, data analysis and evaluation, and physical model verification of the nuclear control power supply device reliability model, develops a failure data acquisition device, proposes an equipment-level reliability analysis method, designs and studies the calculation technology of the probability distribution of the equipment-level reliability prediction index TTF, and obtains a high-fitting calculation estimate of the probability distribution of the TTF of the nuclear control power supply device. Through the analysis and research on the calculation of the TTF probability distribution curve, a set of theoretical system of reliability prediction methods with promotion significance is formed.

[0103] Example 2

[0104] like Fig.10 As shown, the difference between this embodiment and embodiment 1 is that this embodiment provides a TTF distribution calculation system for a nuclear-controlled power supply device based on a fusion model, and the system includes:

[0105] An acquisition unit, used to acquire first data and second data; the first data is data parameters of the nuclear-controlled power supply device before and after failure, and the second data is historical usage data of the nuclear-controlled power supply device;

[0106] A model building unit is used to build a failure physics sub-model and a data-driven sub-model based on the first data and the second data; and fuse the failure physics sub-model and the data-driven sub-model to obtain a data-driven model with an embedded fault mechanism, that is, a fusion model; and build and train a TTF distribution calculation model based on the fusion model;

[0107] The TTF calculation unit is used to obtain the TTF probability distribution of the nuclear control power supply device by fitting calculation based on the TTF distribution calculation model.

[0108] As a further implementation, the system also includes:

[0109] The alarm unit is used to set an alarm to provide the repair or replacement time of the nuclear control power supply device according to the TTF probability distribution.

[0110] Among them, the execution process of each unit can be executed according to the process steps of the TTF distribution calculation method of the nuclear-controlled power supply device based on the fusion model in Example 1, and will not be repeated one by one in this embodiment.

[0111] At the same time, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the above-mentioned TTF distribution calculation method of a nuclear-controlled power supply device based on a fusion model is implemented.

[0112] At the same time, the present invention also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the above-mentioned TTF distribution calculation method of a nuclear-controlled power supply device based on a fusion model is implemented.

[0113] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0114] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0115] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0116] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0117] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A TTF distribution calculation method for a nuclear power control device based on a fusion model, characterized in that: The method includes: Acquire first data and second data; the first data is data parameters of the nuclear-controlled power supply device before and after failure, and the second data is historical usage data of the nuclear-controlled power supply device; Based on the first data and the second data, construct a failure physics sub-model and a data-driven sub-model; and fuse the failure physics sub-model and the data-driven sub-model to obtain a data-driven model with an embedded fault mechanism, that is, a fusion model; Based on the fusion model, a TTF distribution calculation model is constructed and trained; based on the TTF distribution calculation model, the TTF probability distribution of the nuclear control power supply device is obtained by fitting calculation.

2. The TTF distribution calculation method of a nuclear-controlled power supply device based on a fusion model according to claim 1 is characterized in that: The method also includes: according to the TTF probability distribution, setting an alarm to provide a repair or replacement time for the nuclear control power supply device.

3. The TTF distribution calculation method of a nuclear-controlled power supply device based on a fusion model according to claim 1 is characterized in that: The data parameters include temperature data, voltage and current data, and control signal data; The temperature data is the temperature data measured on the full-bridge IGBT device; The voltage and current data include voltage and current data measured on the bus, phase voltage and current data measured on the output three-phase AC power supply, and current and voltage data measured on the optocoupler and electrolytic capacitor; The control signal data is IGBT drive signal phase sequence difference data measured for the full-bridge IGBT control signal.

4. The TTF distribution calculation method of a nuclear-controlled power supply device based on a fusion model according to claim 1 is characterized in that: Construct failure physics sub-model and data-driven sub-model, including: The construction process of the failure physics sub-model is as follows: Analyze the failure mechanism of the drive module and key components of the nuclear power supply device, study the aging failure mechanism of the drive module and key components under the optimal characteristic parameter failure mode, and determine the failure signs and characteristic parameters; Based on the failure mechanism, a multi-failure mechanism physical evolution model of key components and drive modules is established and used as a failure physics sub-model; The effectiveness of selecting characteristic parameters of the aging process of each device is verified by simulation, and the effectiveness of failure mechanism analysis of key components is verified by conducting accelerated aging experiments on IGBTs, capacitors, and optocouplers; The construction process of the data-driven sub-model is as follows: Clarify the aging and failure characteristic parameters of the drive module and its key components, build a database of aging and failure characteristic parameters of the drive module and key components, and analyze the coupling relationship between each key aging characteristic parameter and the aging characteristic parameter of the driver board to obtain a data-driven sub-model.

5. The TTF distribution calculation method of a nuclear-controlled power supply device based on a fusion model according to claim 1 is characterized in that: Build and train the TTF distribution calculation model, including: Integrate the fault mechanism model of IGBT devices and find the operating boundary and model initial value of the failed device; construct the boundary residual term, model residual term and key feature data residual term of the fusion model; the key feature data residual term includes the temperature, current and voltage characteristic parameters of the key nodes; Construct a multi-input and multi-output neural network, and define the activation function of the neural network as tanh and normal Glorot initialization; after defining the optimization domain of each residual, construct an overall loss function including the boundary residual term, model residual term and key feature data residual term; accordingly, the constraint function of the mathematical model of multiple types of failure mechanisms, the residual function of the target value and the actual estimated value, the constraint function of the initial condition and the boundary condition and other related physical system information are substituted into the loss function of the neural network through operators; The Jacobi matrix and Hessian matrix modules are constructed using automatic differentiation technology, and high-order differential calculations are realized through nesting; the optimizer mode, learning rate, number of iterations and loss weight are set, and the fusion model is compiled; and the neural network is trained by the gradient descent method and the loss function is minimized to find the optimal value until the loss drops below the set value, thereby obtaining a trained fusion model.

6. The TTF distribution calculation method of a nuclear-controlled power supply device based on a fusion model according to claim 1 is characterized in that: Based on the TTF distribution calculation model, the TTF probability distribution of the nuclear-controlled power supply device is obtained by fitting and calculating, including: Determine failure signs and characteristic parameters, and use failure threshold setting and the TTF distribution calculation model to perform data trend analysis function estimation on important parameters and time; The remaining life of the nuclear-controlled power supply device is estimated by using the time when the parameter does not exceed the failure threshold; and the TTF distribution calculation model is repeated multiple times to fit the TTF probability distribution of the nuclear-controlled power supply device.

7. A TTF distributed computing system for nuclear power control devices based on a fusion model, characterized in that: The system includes: an acquisition unit, configured to acquire first data and second data; the first data being data parameters of the nuclear-controlled power supply device before and after failure, and the second data being historical usage data of the nuclear-controlled power supply device; A model building unit, configured to build a failure physics sub-model and a data-driven sub-model based on the first data and the second data; and to fuse the failure physics sub-model and the data-driven sub-model to obtain a data-driven model with an embedded fault mechanism, i.e., a fusion model; and to build and train a TTF distribution calculation model based on the fusion model; The TTF calculation unit is used to obtain the TTF probability distribution of the nuclear-controlled power supply device by fitting calculation based on the TTF distribution calculation model.

8. The TTF distributed calculation system for nuclear-controlled power supply device based on fusion model according to claim 7 is characterized in that: The system also includes: The alarm unit is used to set an alarm to provide a repair or replacement time for the nuclear control power supply device according to the TTF probability distribution.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the TTF distribution calculation method of the nuclear-controlled power supply device based on the fusion model as described in any one of claims 1 to 6 is implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the TTF distribution calculation method of a nuclear-controlled power supply device based on a fusion model as described in any one of claims 1 to 6 is implemented.

Images