Fault detection method for refueling system outside reactor based on Bayesian network

Abstract

The invention provides a fault detection method for an out-of-pile refueling system of a reactor based on a Bayesian network. The fault detection method comprises the following steps: extracting a time sequence of a target variable in test work data of a mechanical arm and a transport vehicle; training a long short-term memory network based on the time sequence of the target variable, and obtaining a target feature vector corresponding to the target variable output by the long short-term memory network; taking the target feature vector and variables in the test work data as topological nodes of a Bayesian network, and constructing a fault detection model based on the Bayesian network; and inputting the current working data of the mechanical arm and the transport vehicle into the fault detection model so as to determine the fault type of the out-of-pile refueling system of the reactor. The dependency relationship between the change trend of the target variable and other variables in the test work data is effectively established, the probability of fault occurrence can be identified according to the change trend of the variables in the test work data in time, and the generalization ability of a fault detection model to unexperimented scenes is improved.

Description

Technical Field

[0001] The invention relates to the field of nuclear reactors, and in particular to a method for detecting faults in an external refueling system of a reactor based on a Bayesian network. Background Art

[0002] The nuclear reactor off-core refueling system is an extremely important component in the operation of nuclear reactors. It is responsible for removing the spent fuel assemblies that have been "burned" in the reactor and loading new fuel assemblies into the reactor to ensure the continuous operation of the reactor, and for hoisting, transporting, storing and monitoring new and spent nuclear fuel assemblies to ensure the safety of nuclear fuel during transportation and storage outside the reactor.

[0003] The related technology uses a manipulator to cooperate with a transport vehicle loaded with nuclear fuel assemblies to complete the refueling. Due to the harsh working environment of the refueling system, which involves complex mechanical, electrical and hydraulic systems, various faults are prone to occur during operation, especially "loading synchronization failure" and "premature unloading failure". Loading synchronization failure means that when the manipulator loads the nuclear fuel assembly, the transport vehicle has started before the manipulator completely grabs the nuclear fuel assembly onto the transport vehicle; or when the nuclear fuel assembly is still being loaded, the airlock controlling the manipulator begins to close. Premature unloading failure means that the manipulator releases the nuclear fuel assembly and unloads it before reaching the unloading position of the transport vehicle; or the manipulator releases the gripper to unload the nuclear fuel assembly before it reaches the designated preheating box unloading position. The above two faults seriously affect the progress of the nuclear reactor, which may not only cause the reactor to shut down, but also cause serious safety problems, posing a huge threat to personnel and the environment. Summary of the invention

[0004] In view of this, the purpose of the present invention is to provide a reactor off-core refueling system fault detection method based on Bayesian network, which solves the problem that the nuclear reactor off-core refueling system in the prior art lacks the recognition of "loading synchronization failure" and "premature unloading failure".

[0005] In a first aspect, the present application provides a method for detecting a reactor off-core refueling system fault based on a Bayesian network, comprising:

[0006] Extract the time series of the target variable in the test work data of the robot arm and the transporter;

[0007] Training a long short-term memory network based on the time series of the target variable, and obtaining a target feature vector corresponding to the target variable output by the long short-term memory network;

[0008] Using the target feature vector and the variables in the test working data as topological nodes of a Bayesian network, and building a fault detection model based on the Bayesian network;

[0009] The current working data of the robot arm and the transport vehicle are input into the fault detection model to determine the fault type of the reactor off-core refueling system.

[0010] In one embodiment, the target variables in the test work data include the hoisting pressure, running time and stroke of the manipulator in the experimental test, and the total bearing pressure of the transport vehicle, the start time node of carrying the nuclear fuel assembly, and the start time node; the time series of the target variables in the test work data of the manipulator and the transport vehicle are extracted, specifically including:

[0011] Determine a pressure change time series and a speed change time series of the robotic arm based on the hoisting pressure, the operating time, and the stroke;

[0012] The time series of the change in the carrying capacity of the transport vehicle is determined based on the total bearing pressure of the transport vehicle, the start time node of carrying the nuclear fuel assembly, and the start time node.

[0013] In one embodiment, the training of the long short-term memory network based on the time series of the target variable and obtaining the target feature vector corresponding to the target variable output by the long short-term memory network specifically includes:

[0014] Dividing the time series into a first training set and a first test set according to a preset ratio;

[0015] Training the preset initial long short-term memory network based on the first training sample set, and obtaining a trained long short-term memory network;

[0016] The first test set is input into the long short-term memory network for training to obtain a predicted value of the target variable, and the target feature vector is obtained based on the predicted value of the target variable and the output of the hidden layer of the long short-term memory network.

[0017] In one embodiment, the preset initial long short-term memory network is defined by the following method:

[0018] The number of neurons in the input layer, the number of neurons in the LSTM layer and the input shape parameter, the number of hidden layers and the activation function of the initial long short-term memory network are determined according to the characteristic dimension of the time series.

[0019] In one embodiment, the target feature vector and the variables in the test working data are used as topological nodes of a Bayesian network, and a fault detection model is constructed based on the Bayesian network, specifically including:

[0020] Discretize the target feature vector, and compare the data of the target feature vector after discretization with the variables in the test working data to obtain a discrete variable data set;

[0021] An initial Bayesian network structure is established based on a discrete variable data set, and an edge modification operation is performed on the initial Bayesian network structure to generate a new Bayesian network structure. The Bayesian network structure that meets the preset conditions is used as the optimal Bayesian network structure for constructing a fault detection model.

[0022] In one embodiment, the discretization of the target feature vector specifically includes:

[0023] The data of the target feature vector is divided into a plurality of discretization intervals, each of which can accommodate a specified amount of data.

[0024] In one embodiment, the edge modification operation is performed on the initial Bayesian network structure to generate a new Bayesian network structure, and the Bayesian network structure that meets the preset conditions is used as the optimal Bayesian network structure for constructing the fault detection model, specifically including:

[0025] Using the initialized Bayesian network structure as the current Bayesian network structure, performing edge modification operations on the current Bayesian network structure and generating a modified Bayesian network structure;

[0026] defining a scoring function for evaluating the Bayesian network structure, evaluating the score of the modified Bayesian network structure based on the scoring function, and selecting the modified Bayesian network structure with the highest score as the current Bayesian network structure;

[0027] Repeat the edge modification operation on the current Bayesian network structure, and use the current Bayesian network structure that meets the preset iteration termination condition as the optimal Bayesian network structure for building the fault detection model.

[0028] In one embodiment, the construction of a fault detection model based on a Bayesian network specifically includes:

[0029] Establishing a fault node in the Bayesian network structure, estimating the conditional probability parameters of the evidence variables in the test work data for the fault node; the evidence variables include the hoisting pressure, running time and stroke of the manipulator in the experimental test, and the total bearing pressure of the transport vehicle, the start time node of carrying the nuclear fuel assembly, the start time node, and the number of the transport vehicle carrying stations;

[0030] Based on the conditional probability parameters, calculating the joint probability distribution of all the evidence variables in the Bayesian network;

[0031] Selecting the evidence variables of the test working data that the fault node exists or does not exist to construct a second training set and a second test set;

[0032] The fault detection model is trained through the second training set based on the joint probability distribution, and the fault detection model is tested through the second test set to obtain a trained fault detection model.

[0033] In a second aspect, the present application provides a reactor off-core refueling system fault detection system based on a Bayesian network, comprising a processor and a memory; wherein the memory stores a computer program, and the computer program is used to be loaded by the processor and execute the reactor off-core refueling system fault detection method based on a Bayesian network as described in any one of the first aspects.

[0034] In a third aspect, the present application provides a computer-readable storage medium storing instructions for a processor to load and execute a Bayesian network-based reactor off-site refueling system fault detection method as described in any one of the first aspects.

[0035] In the reactor off-core refueling system fault detection method based on Bayesian network in the present embodiment, the temporal change trend of the target variable when the reactor off-core refueling system fails in various test scenarios is first effectively captured through the long short-term memory network, and the target feature vector and the test work data generated based on the long short-term memory network are used as the topological nodes of the Bayesian network to effectively establish the dependency relationship between the change trend of the target variable and other variables in the test work data. When it comes to "loading synchronization failure" and "premature unloading failure", the probability of the occurrence of the fault can be identified according to the temporal change trend of the variables in the test work data, which improves the generalization ability of the fault detection model for untested scenarios, effectively detects the occurrence of reactor off-core refueling system failures, ensures the smooth progress of the nuclear reactor process, and improves the safety of relevant personnel. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0037] Figure 1 It is a flow chart of a method for detecting faults in a reactor off-site refueling system based on a Bayesian network in an embodiment of the present application. DETAILED DESCRIPTION

[0038] The specific embodiments of the present invention will be described in detail below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. Based on the description of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0039] In the description of the present invention, unless otherwise clearly specified and limited, the terms "set", "install", "connection" and the like should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium. For ordinary technicians in this field, the specific meanings of the above terms can be understood according to specific circumstances.

[0040] The directions or positional relationships indicated by terms such as “upper”, “lower”, “left”, “right”, “front”, “back”, “top”, “bottom”, “inside” and “outside” are based on the directions or positional relationships shown in the accompanying drawings, or are the directions or positional relationships in which the inventive product is usually placed when used. They are only for the convenience of description and simplified description, and do not indicate or imply that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore should not be understood as a limitation on the present invention.

[0041] The terms "first", "second", "third", etc. are merely used to distinguish elements of similar nature, and do not indicate or imply relative importance or a particular order.

[0042] The terms "comprises," "comprising," or any other variations thereof, are intended to cover a non-exclusive inclusion of the elements listed and may also include additional elements not expressly listed.

[0043] The reactor off-site refueling system usually includes refueling operation equipment, lifting equipment, transport vehicles and storage facilities. Refueling operation equipment such as rotary worktables and refueling mechanisms are used to realize the grabbing, loading and unloading and position conversion of fuel assemblies; in pressurized water reactors, refueling mechanisms are usually composed of remotely controlled grippers, refueling containers, shielding turnstiles, etc. Lifting equipment such as bridge cranes are equipped with manipulators and hanging baskets to remove spent fuel assemblies from the reactor and send them to the transport vehicle, and then transport them to the fuel element storage pool by the transport vehicle, and lift new fuel assemblies from the storage location to the reactor for loading operations. Storage equipment includes spent fuel pools, etc., which are used to temporarily store spent fuel assemblies unloaded from the reactor, so that they can be cooled and shielded in water to reduce the radioactivity.

[0044] This embodiment relates to the replacement of new fuel assemblies and the replacement of spent fuel assemblies. For new fuel assemblies, they need to be moved into the preheating box through the preheating box basket gripper and the preheating box basket for preheating, and then the preheated new fuel assemblies are transported to the new fuel transport vehicle through the preheating box basket gripper, transported to the reactor by the new fuel transport vehicle and loaded by the gripper of the new fuel assembly. For spent fuel assemblies, they are taken out from the core by the spent fuel gripper, transported to the transfer room by the transfer room crane and the transfer room gripper, and after the position recording and coding are completed in the transfer room, they are transported to the cleaning room by the spent fuel transport vehicle for cleaning, and then the cleaned spent fuel assemblies are taken out by the cleaning room gripper and transported to the spent fuel storage pool for storage.

[0045] like Figure 1 As shown, this embodiment provides a method for detecting a reactor off-core refueling system fault based on a Bayesian network, comprising:

[0046] Step S10: extracting the time series of the target variable in the test work data of the robot arm and the transport vehicle;

[0047] Step S20: training a long short-term memory network based on the time series of the target variable, and obtaining a target feature vector corresponding to the target variable output by the long short-term memory network;

[0048] Step S30: using the target feature vector and the variables in the test working data as topological nodes of a Bayesian network, and constructing a fault detection model based on the Bayesian network;

[0049] Step S40: inputting the current working data of the robot arm and the transport vehicle into the fault detection model to determine the fault type of the reactor off-core refueling system.

[0050] In the reactor off-core refueling system fault detection method based on Bayesian network in the present embodiment, the temporal change trend of the target variable when the reactor off-core refueling system fails in various test scenarios is first effectively captured through the long short-term memory network, and the target feature vector and the test work data generated based on the long short-term memory network are used as the topological nodes of the Bayesian network to effectively establish the dependency relationship between the change trend of the target variable and other variables in the test work data. When it comes to "loading synchronization failure" and "premature unloading failure", the probability of the occurrence of the fault can be identified according to the temporal change trend of the variables in the test work data, which improves the generalization ability of the fault detection model for untested scenarios, effectively detects the occurrence of reactor off-core refueling system failures, ensures the smooth progress of the nuclear reactor process, and improves the safety of relevant personnel.

[0051] Step S10: extracting the time series of the target variable in the test work data of the robot arm and the transport vehicle.

[0052] In this embodiment, the robotic arm includes a preheating box basket gripper, a preheating box basket, a transfer room crane and a transfer room gripper, a cleaning room gripper, and a spent fuel gripper, and the transport vehicle includes a new fuel transport vehicle and a spent fuel transport vehicle.

[0053] The target variables in the test work data include the lifting pressure, operating time and stroke of the robot arm in the experimental test, as well as the total bearing pressure of the transport vehicle, the start time node of carrying the nuclear fuel assembly, and the start time node.

[0054] Lifting pressure refers to the pressure borne by the lifting structure of the robot arm. Pressure sensors are installed at key stress-bearing parts of the lifting device of the robot arm (such as the lifting arm, rope connection point, etc.). When the robot arm performs lifting operations, the pressure sensor senses the pressure generated during the lifting process in real time. The control system of the robot arm usually has a built-in pressure monitoring module, and the lifting pressure data can be directly read from the parameter setting interface or data output port of the control system.

[0055] Using the clock function inside the robot control system, start timing when the robot starts and end timing when it stops, and determine the running time of the robot by reading the timestamp of the system clock. This time data can be obtained from the log file of the control system, or it can be displayed and recorded in real time by the host computer software that communicates with the control system.

[0056] A laser rangefinder is set up in the working area of ​​the robot arm. By measuring the time from the laser being emitted to the time it is reflected by the robot arm, the change in distance between the robot arm and the rangefinder is calculated to determine the stroke of the robot arm. It is understandable that encoders can also be installed at the joints of the robot arm, which can accurately measure the rotation angle and displacement of the joints. By collecting and processing the encoder data of each joint of the robot arm, combined with the kinematic model of the robot arm, the stroke of the end of the robot arm is calculated.

[0057] Multiple pressure sensors are installed at the connection between the frame and wheels of the transport vehicle, the bottom of the carriage, etc. to form a pressure sensor array. When the transport vehicle carries nuclear fuel assemblies, each pressure sensor measures the pressure it bears. The data acquisition system collects and calculates the data of each sensor to obtain the total load pressure.

[0058] A proximity switch is installed at a specific position on the transport vehicle for carrying nuclear fuel assemblies. When the nuclear fuel assembly approaches and triggers the proximity switch, the proximity switch sends a signal to the control system, and the control system records the current time to determine the start time node for carrying the nuclear fuel assembly.

[0059] A current sensor is installed in the starting circuit of the transport vehicle. When the vehicle is started, the current of the starting motor will change significantly. After the current sensor detects this current change signal, it transmits it to the time recording device, records the corresponding time, and thus determines the starting time node of the transport vehicle.

[0060] In step S10, the time series of the target variables in the test work data of the robot arm and the transport vehicle are extracted, specifically including:

[0061] Step S101: determining a pressure change time series and a speed change time series of the robotic arm based on the hoisting pressure, the operating time and the stroke;

[0062] Step S102: Determine the time series of changes in the carrying capacity of the transport vehicle based on the total carrying pressure of the transport vehicle, the start time node of carrying the nuclear fuel assembly, and the start time node.

[0063] In step S101, for the pressure change time series, the collected lifting pressure data are sorted in chronological order to form an ordered data set. In order to more clearly show the change trend of the lifting pressure over time, the sorted data can be plotted into a time series graph with time as the horizontal axis and the lifting pressure as the vertical axis. In the time series graph, each data point represents the lifting pressure value at a specific moment. By connecting these data points, the pressure change time series curve of the robot arm can be obtained.

[0064] For the speed change time series, the travel data is arranged in chronological order. The displacement increment △s can be calculated according to the travel corresponding to two adjacent time points. Then, according to the interval △t between two adjacent time points, the speed △v=△s / △t in each time interval can be calculated. Then, it is sorted in chronological order to form an ordered data set. In the speed change time series diagram, time is the horizontal axis and speed is the vertical axis. Each data point represents the speed value at a specific moment. Connecting these data points can obtain the speed change time series curve of the robot arm.

[0065] In step S102, for the time series of bearing capacity changes, the entire time process is divided into different stages according to the start time node and the start time node of the bearing nuclear fuel assembly. In each time stage, the total bearing pressure data is recorded at a certain time interval (for example, every second or every few seconds), and the total bearing pressure is arranged and sorted in chronological order to form an ordered data set. In the graph of the time series of bearing capacity changes, time is the horizontal axis and the total bearing pressure is the vertical axis, and the changing trend of the bearing capacity in different stages can be clearly seen, such as the rise, stability and possible small fluctuations of pressure. By analyzing the time series of bearing capacity changes, the bearing condition of the transport vehicle in different time stages can be evaluated.

[0066] Step S20: training a long short-term memory network based on the time series of the target variable, and obtaining a target feature vector corresponding to the target variable output by the long short-term memory network.

[0067] The step of training a long short-term memory network based on the time series of the target variable and obtaining a target feature vector corresponding to the target variable output by the long short-term memory network specifically includes:

[0068] Step S201: dividing the time series into a first training set and a first test set according to a preset ratio;

[0069] Step S202: training the preset initial long short-term memory network based on the first training sample set, and obtaining a trained long short-term memory network;

[0070] Step S203: input the first test set into the long short-term memory network for training to obtain a predicted value of the target variable, and obtain the target feature vector based on the predicted value of the target variable and the output of the hidden layer of the long short-term memory network.

[0071] In step S201, the time series can be divided into a first training set and a first test set at a ratio of 8:2 (which can also be changed to 7:3, 9:1, or other ratios according to actual needs), wherein the first training set accounts for 80% and the first test set accounts for 20%. The first training sample set is used to train the long short-term memory network, and the first test set is used to test the long short-term memory network.

[0072] Before dividing the test data into the first training set and the first test set, the test data needs to be cleaned and normalized. When cleaning the data, interpolation methods (such as linear interpolation and polynomial interpolation) can be used to fill missing values, and statistical methods (such as the 3σ principle) can be used to identify and process outliers. When normalizing the data, Min-Max normalization and Z-Score normalization can be used to normalize the time series to a specific range (such as [-1,1]), and the time series data can be divided into windows of fixed length, with each window as a sample.

[0073] In addition, when filling missing values ​​of data, the random forest algorithm can be used to randomly delete the data of the test working data, and multiple sets of deleted data sets to be restored after deletion can be established for the same data; then, the deleted data set can be restored by the recovery model established based on the diagnosed damage degree type to obtain a restored data set; and the restored data set can be compared with the deleted data set.

[0074] When calculating the similarity between the deleted dataset and the restored dataset, the abnormal feature difference between the deleted dataset and the restored dataset is constructed according to the change trend of the data in the time series combined with the cosine similarity, where the value range of the cosine similarity is between [-1, 1]. The similarity between the deleted dataset and the restored dataset is calculated. The closer the cosine similarity between the deleted dataset and the restored dataset is to 1, the more similar they are. However, if the cosine similarity between the deleted dataset and the restored dataset is close to -1, it means that the deleted dataset and the restored dataset are extremely dissimilar, and the restoration parameters provided by the restored dataset are inaccurate. The deleted dataset needs to be reconstructed according to the change trend of the data in the time series until the cosine similarity between the deleted dataset and the restored dataset is closer to 1.

[0075] In step S202, the preset initial long short-term memory network is defined by the following method: determining the number of neurons in the input layer, the number of neurons and input shape parameters of the LSTM layer, the number of hidden layers and the activation function of the initial long short-term memory network according to the characteristic dimension of the time series.

[0076] Among them, the number of neurons in the input layer of the initial long short-term memory network is usually directly determined by the characteristic dimension of the time series. In this embodiment, the characteristic dimensions of the time series include the pressure change time series, the speed change time series and the carrying capacity change time series, so the number of neurons in the input layer is 3. The number of neurons in the LSTM layer will be greater than the number of neurons in the input layer. In order to better capture the complex patterns in the time series, the number of neurons in the LSTM layer can be set to 32 or 64. The input shape parameters of the LSTM layer usually include three indicators (number of samples, time step, and characteristic dimension). The number of samples is the number of sample data in the time series set. The time step represents the number of time steps contained in an input sample, which can be customized according to the length of the acquisition time. The characteristic dimension is 2. Since the time series of this implementation has a high degree of complexity and nonlinear relationships, the number of hidden layers is set to 3. The Sigmoid function is used as the activation function of the input gate, the forget gate, and the output gate. This is because the Sigmoid function can map the output value to between 0 and 1, which is convenient for controlling the flow of information. At the same time, the tanh function is usually used as the activation function of the cell state and hidden state. The tanh function can map the output value to between -1 and 1, which helps to capture the positive and negative information in the time series.

[0077] When the initial LSTM network is trained with the first training sample set, the data in each time series is processed by the LSTM layer to generate the corresponding hidden state and output. The hidden state contains information from the previous time steps, and the flow of information is controlled by the gating mechanism (input gate, forget gate, and output gate), so that the network can effectively learn the dependencies between the pressure change time series, the speed change time series, and the carrying capacity change time series. In addition, the weights of the LSTM network can be updated by continuously iterative optimization using the backpropagation algorithm and the optimizer, and finally a trained LSTM network is obtained.

[0078] In step S203, the hidden layer usually outputs a hidden layer vector for predicting the change trend, which can be used as a basis for predicting the value of the next time step, and combined with the weight correction of the predicted value output by the first test set training, so that the target feature vector can have good predictive power in time change. The target feature vector can be obtained by weighted fusion according to the predicted value of the target variable and the output of the hidden layer of the long short-term memory network.

[0079] Step S30: taking the target feature vector and the variables in the test working data as topological nodes of a Bayesian network, and constructing a fault detection model based on the Bayesian network.

[0080] In this embodiment, based on the target feature vector and the test work data as the topological nodes of the Bayesian network, the dependency relationship between the change trend of the target variable and other variables in the test work data is effectively established. When dealing with "loading synchronization failure" and "premature unloading failure", the probability of the occurrence of the fault can be identified based on the temporal change trend of the variables in the test work data, thereby improving the generalization ability of the fault detection model for untested scenarios.

[0081] In step S30, the target feature vector and the variables in the test working data are used as topological nodes of a Bayesian network, and a fault detection model is constructed based on the Bayesian network, which specifically includes:

[0082] Step S301: discretize the target feature vector, and combine the data of the discretized target feature vector with the variables in the test working data to obtain a discrete variable data set;

[0083] Step S302: establishing an initial Bayesian network structure based on a discrete variable data set, performing edge modification operations on the initial Bayesian network structure to generate a new Bayesian network structure, and using the Bayesian network structure that meets preset conditions as the optimal Bayesian network structure for constructing a fault detection model.

[0084] In step S301, since the variables and conditional probability distributions in the Bayesian network need to be clearly defined, the target feature vector needs to be discretized when it is defined as a topological node of the Bayesian network. The data can be divided into several intervals according to the equal frequency binning method, so that the number of data in each interval is roughly the same.

[0085] The discretization process of the target feature vector specifically includes: dividing the data of the target feature vector and the variables in the test working data into a plurality of discretization intervals, each discretization interval being able to accommodate a specified amount of data.

[0086] In step S302, the discrete variable data set includes data variables in the pressure change time series, the speed change time series and the load-bearing capacity change time series, and the test work data includes the hoisting pressure, running time and travel of the manipulator in the experimental test, and the total pressure of the transport vehicle, the start time node of the nuclear fuel assembly, the start time node, and the number of transport vehicle load stations. Parent nodes and child nodes are constructed according to the relationship between variables to generate an initial Bayesian network structure. For example, when the parent node is a pressure change time series, it contains two child nodes: the hoisting pressure of the manipulator and the running time of the manipulator; when the parent node is a speed change time series, it contains two child nodes: the running time and travel of the manipulator.

[0087] The edge modification operation is performed on the initial Bayesian network structure to generate a new Bayesian network structure, and the Bayesian network structure that meets the preset conditions is used as the optimal Bayesian network structure for building the fault detection model, specifically including:

[0088] Step S3021: using the initialized Bayesian network structure as the current Bayesian network structure, performing edge modification operations on the current Bayesian network structure and generating a modified Bayesian network structure;

[0089] Step S3022: defining a scoring function for evaluating the Bayesian network structure, evaluating the score of the modified Bayesian network structure based on the scoring function, and selecting the modified Bayesian network structure with the highest score as the current Bayesian network structure;

[0090] Step S3023: Repeat the edge modification operation on the current Bayesian network structure, and use the current Bayesian network structure that meets the preset iteration termination condition as the optimal Bayesian network structure for constructing the fault detection model.

[0091] In step S3021, edge operations include adding edges, deleting edges, and changing the direction of edges. For example, in the initial Bayesian network structure, when the parent node is a pressure change time series, it contains two child nodes, namely, the lifting pressure of the robot and the running time of the robot. By adding a child node of the bearing capacity change time series, a new edge is added between the pressure change time series and the bearing capacity change time series; for another example, when the parent node is a speed change time series, it contains two child nodes, namely, the running time and the stroke of the robot. The association between the speed change time series and the running time is cancelled, and a new one is added between the lifting pressure of the robot and the running time, thereby changing the direction of the edge.

[0092] In step S3022, for each modification of the above edge operation, the Bayesian information criterion is used as a scoring function, and the one with the highest score is selected as the current Bayesian network structure.

[0093] The scoring function is

[0094] Indicates the probability of predicting data D under the current Bayesian network structure G. N is the total number of samples, K is the number of parameters in the network, , where m is the number of nodes, and node i has value, its parent node has For example, when initializing the Bayesian network structure, the data variable of the parent node including the pressure change time series, and the two child nodes of the hoisting pressure of the robot arm and the running time of the robot arm, then for k= .

[0095] In step S3023, when the maximum number of iterations is reached or the score is less than a preset threshold, the iteration is terminated and the current Bayesian network structure is used as the optimal Bayesian network structure. Taking the three variables of the robot's driving pressure, running time and stroke as an example, different connection methods can be tried, such as driving pressure → running time → stroke, running time → driving pressure → stroke, etc., and the score of each structure is calculated, and the structure with the highest score is selected as the optimal Bayesian network structure.

[0096] Step S40: inputting the current working data of the robot arm and the transport vehicle into the fault detection model to determine the fault type of the reactor off-core refueling system.

[0097] Specifically, the fault detection model is constructed based on the Bayesian network, including:

[0098] Step S401: Establishing a fault node in the Bayesian network structure, and estimating the conditional probability parameters of the evidence variables in the test work data for the fault node; the evidence variables include the hoisting pressure, running time and stroke of the manipulator in the experimental test, and the total bearing pressure of the transport vehicle, the start time node of carrying the nuclear fuel assembly, the start time node, and the number of the transport vehicle carrying stations;

[0099] Step S402: Based on the conditional probability parameters, calculate the joint probability distribution of all the evidence variables in the Bayesian network;

[0100] Step S403: Selecting the evidence variables of the test working data for the presence or absence of faults at the fault node to construct a second training set and a second test set;

[0101] Step S404: training the fault detection model through the second training set based on the joint probability distribution, and testing the fault detection model through the second test set to obtain a trained fault detection model.

[0102] In step S401, two fault nodes, "loading synchronization fault" and "unloading premature fault", are established in the Bayesian network structure, and the values ​​of the two fault nodes are "faulty" and "no fault" respectively. Variables in the Bayesian network that may be related to the fault, such as the lifting pressure, running time, and travel of the robot arm, the total bearing pressure of the transport vehicle, the start time node of carrying the nuclear fuel assembly, the start time node, the number of bearing stations, etc., are used as evidence variables.

[0103] The prior probability of a "faulty" node, that is, the probability of a fault occurring without any evidence, can be estimated based on experimental data. For example, if the system has a 5% fault frequency within a certain period of time based on past records, then the prior probability of the "faulty" state of the "faulty" node can be set to 0.05, and the prior probability of the "no fault" state can be set to 0.95.

[0104] In step S402, the conditional probability distribution between the evidence variable and the "fault" node is estimated using the Bayesian estimation method based on the existing data and the Bayesian network structure.

[0105] According to the structure of the Bayesian network and the estimated probability parameters, the joint probability distribution of all variables is calculated. For a Bayesian network X with n variables 1 ,X 2 ,…,X n , the joint probability distribution can be expressed as ,in is a variable A collection of parent nodes.

[0106] In step S403 and step S404, the existing test work data is screened and divided into two groups according to the state of the faulty node (faulty or not). One group is data where the faulty node indicates that the system has a fault, and the other group is data where the faulty node indicates that the system has no fault.

[0107] From the two selected data sets, data are selected as the second training set and the second test set in a certain proportion (for example, the training set usually accounts for a larger proportion, such as 70%-80%, and the test set accounts for 20%-30%). The training set is used to train the fault detection model so that it can learn the relationship between the evidence variable and the fault state; the test set is used to evaluate the performance of the trained model and test the generalization ability of the model on unknown data.

[0108] The fault detection model is trained using the data from the second training set combined with the previously calculated joint probability distribution. During the training process, the model adjusts the internal parameters of the model (such as the conditional probability parameters in the Bayesian network) based on the input evidence variables and the corresponding fault status to minimize the error between the predicted results and the actual fault status.

[0109] When the model training is completed, the model is tested using the data from the second test set. The evidence variables in the test set are input into the trained model, and the model outputs the prediction results of the fault node status.

[0110] Based on the same inventive concept as the above embodiments, this embodiment also provides a computer-readable storage medium, which stores instructions, and the instructions are used by a processor to load and execute the above-mentioned Bayesian network-based reactor off-site refueling system fault detection method.

[0111] In the embodiments of the mobile terminal and computer-readable storage medium provided in the present application, all technical features of the above-mentioned control method embodiments are included, and the expansion and explanation content of the specification are basically the same as those of the above-mentioned method embodiments, which will not be repeated here.

[0112] The embodiment of the present application further provides a computer program product, which includes a computer program code. When the computer program code runs on a computer, the computer executes the methods in the above various possible implementation modes.

[0113] An embodiment of the present application also provides a chip, including a memory and a processor, the memory is used to store a computer program, and the processor is used to call and run the computer program from the memory, so that a device equipped with the chip executes the methods in various possible implementation modes as described above.

[0114] The serial numbers of the above-mentioned embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0115] In the present application, the same or similar terminology concepts, technical solutions and / or application scenario descriptions are generally described in detail only the first time they appear. When they appear again later, they are generally not repeated for the sake of brevity. When understanding the technical solutions and other contents of the present application, for the same or similar terminology concepts, technical solutions and / or application scenario descriptions that are not described in detail later, reference can be made to the previous related detailed descriptions.

[0116] In the present application, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0117] The various technical features of the technical solution of the present application can be arbitrarily combined. In order to make the description concise, not all possible combinations of the various technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of the present application.

[0118] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product is stored in a storage medium as above, including a number of instructions for a terminal device to execute the method of each embodiment of the present application. The above is only a preferred embodiment of the present application, and does not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made by using the content of the specification and drawings of this application, or directly or indirectly used in other related technical fields, is similarly included in the patent protection scope of the present application.

[0119] It should be noted that the various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0120] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed by the present invention should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the attached claims.

Claims

1. A method for detecting faults in a reactor off-core refueling system based on a Bayesian network, characterized in that: include: Extract the time series of the target variable in the test work data of the robot arm and the transporter; Training a long short-term memory network based on the time series of the target variable, and obtaining a target feature vector corresponding to the target variable output by the long short-term memory network; Using the target feature vector and the variables in the test working data as topological nodes of a Bayesian network, and building a fault detection model based on the Bayesian network; The current working data of the robot arm and the transport vehicle are input into the fault detection model to determine the fault type of the reactor off-core refueling system.

2. The method for detecting faults in a reactor off-core refueling system based on a Bayesian network according to claim 1, characterized in that: The target variables in the test work data include the hoisting pressure, operation time and travel of the manipulator in the experimental test, as well as the total pressure of the transport vehicle, the start time node of carrying the nuclear fuel assembly, and the start time node; The extracting of the time series of the target variable in the test work data of the robot arm and the transport vehicle specifically includes: Determine a pressure change time series and a speed change time series of the robotic arm based on the hoisting pressure, the operating time, and the stroke; The time series of the change in the carrying capacity of the transport vehicle is determined based on the total bearing pressure of the transport vehicle, the start time node of carrying the nuclear fuel assembly, and the start time node.

3. The method for detecting faults in a reactor off-core refueling system based on a Bayesian network according to claim 1, characterized in that: The step of training a long short-term memory network based on the time series of the target variable and obtaining a target feature vector corresponding to the target variable output by the long short-term memory network specifically includes: Dividing the time series into a first training set and a first test set according to a preset ratio; Training the preset long short-term memory network based on the first training set to obtain a trained long short-term memory network; The first test set is input into the long short-term memory network for training to obtain a predicted value of the target variable, and the target feature vector is obtained based on the predicted value of the target variable and the output of the hidden layer of the long short-term memory network.

4. The method for detecting faults in a reactor off-core refueling system based on a Bayesian network according to claim 3, characterized in that: The preset long short-term memory network is defined by the following method: The number of neurons in the input layer, the number of neurons in the LSTM layer and the input shape parameter, the number of hidden layers and the activation function of the long short-term memory network are determined according to the characteristic dimension of the time series.

5. The method for detecting faults in a reactor off-core refueling system based on a Bayesian network according to claim 1, characterized in that: The method of using the target feature vector and the variables in the test working data as topological nodes of a Bayesian network and constructing a fault detection model based on the Bayesian network specifically includes: Discretize the target feature vector, and compare the data of the target feature vector after discretization with the variables in the test working data to obtain a discrete variable data set; An initial Bayesian network structure is established based on a discrete variable data set, and an edge modification operation is performed on the initial Bayesian network structure to generate a new Bayesian network structure. The Bayesian network structure that meets the preset conditions is used as the optimal Bayesian network structure for constructing a fault detection model.

6. The method for detecting faults in a reactor off-core refueling system based on a Bayesian network according to claim 5, characterized in that: The discretization process of the target feature vector specifically includes: The data of the target feature vector is divided into a plurality of discretization intervals, each of which can accommodate a specified amount of data.

7. The method for detecting faults in a reactor off-core refueling system based on a Bayesian network according to claim 5, characterized in that: The edge modification operation is performed on the initial Bayesian network structure to generate a new Bayesian network structure, and the Bayesian network structure that meets the preset conditions is used as the optimal Bayesian network structure for building the fault detection model, specifically including: Using the initial Bayesian network structure as the current Bayesian network structure, performing edge modification operations on the current Bayesian network structure and generating a modified Bayesian network structure; defining a scoring function for evaluating the Bayesian network structure, evaluating the score of the modified Bayesian network structure based on the scoring function, and selecting the modified Bayesian network structure with the highest score as the current Bayesian network structure; Repeat the edge modification operation on the current Bayesian network structure, and use the current Bayesian network structure that meets the preset iteration termination condition as the optimal Bayesian network structure for building the fault detection model.

8. The method for detecting faults in a reactor off-core refueling system based on a Bayesian network according to claim 1, wherein the fault detection model is constructed based on a Bayesian network, specifically comprising: Establishing a fault node in the Bayesian network structure, and estimating conditional probability parameters of the evidence variables in the test work data for the fault node; The evidence variables include the lifting pressure, operation time and travel of the manipulator in the experimental test, the total pressure of the transport vehicle, the start time node of carrying the nuclear fuel assembly, the start time node, and the number of the transport vehicle carrying stations; Based on the conditional probability parameters, calculating the joint probability distribution of all the evidence variables in the Bayesian network; Selecting the evidence variables of the test working data that the fault node exists or does not exist to construct a second training set and a second test set; The fault detection model is trained through the second training set based on the joint probability distribution, and the fault detection model is tested through the second test set to obtain a trained fault detection model.

9. A reactor off-core refueling system fault detection system based on Bayesian network, characterized in that: It comprises a processor and a memory; wherein the memory stores a computer program, and the computer program is used for the processor to load and execute the reactor off-core refueling system fault detection method based on Bayesian network as described in any one of claims 1-8.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores instructions, and the instructions are used by the processor to load and execute the reactor off-core refueling system fault detection method based on Bayesian network as described in any one of claims 1-8.

Images