A vacuum leak automatic detection system and method

Abstract

The application discloses a kind of vacuum leakage automatic detection system and method, including fusion device plasma discharge vacuum chamber, pumping system, quadrupole mass spectrometer, fusion device vacuum chamber monitoring system and neural network model for vacuum leakage monitoring.The application utilizes the pumping pipeline arranged in the fusion device vacuum chamber, collects the gas partial pressure data in the vacuum chamber by the quadrupole mass spectrometer and transmits to the fusion device vacuum chamber monitoring system, monitors the gas composition in the vacuum chamber in real time and displays in graphical form on the computer interface;And use partial pressure data as input, use neural network model to judge the leakage type, timely and alarm when vacuum leakage occurs.The application can quickly detect the vacuum leakage occurred in nuclear fusion device, help to timely handle leakage, to reduce the influence of leakage on plasma experiment, ensure the safety and stability of fusion device.

Description

Technical Field

[0001] This invention relates to the field of vacuum technology for fusion devices, and specifically to an automatic vacuum leak detection system and method. Background Technology

[0002] Due to its outstanding advantages in terms of resources, safety, and cleanliness, nuclear fusion energy is an ideal energy source for the future of human society. Tokamak-type magnetic confinement fusion devices have the scientific feasibility to realize the utilization of fusion energy and are a possible pathway to first commercialize fusion energy.

[0003] The vacuum system is one of the fundamental support systems for fusion devices, and its main functions are as follows: 1) to provide good vacuum insulation conditions for the normal operation of superconducting magnets; 2) to provide a good vacuum environment for plasma discharge. Vacuum leakage is extremely dangerous. Once a vacuum leakage occurs, impurity gases will enter the inner vacuum chamber, making it difficult for plasma to build up, degrading plasma confinement performance, causing the discharge to terminate, and also damaging cryogenic pumps, molecular pumps, and other system components on the device.

[0004] However, the residual gas composition characteristics detected by quadrupole mass spectrometer vary significantly depending on the location, medium, and wall conditions of the leak within the plasma discharge vacuum chamber, increasing the difficulty of analyzing the leak type. Traditional vacuum leak detection methods rely on individual worker experience, are time-consuming, labor-intensive, inaccurate, and cannot promptly detect or analyze leaks during manual operation. This results in a lag in vacuum leak detection and recovery, significantly delaying leak handling and hindering the restoration of the vacuum environment and wall conditions.

[0005] In recent years, the rapid development of artificial intelligence has led to highly efficient and accurate results in fault diagnosis and prediction technologies based on machine learning. Machine learning technology has been widely applied in the field of fusion, particularly in plasma fracture prediction and tomographic imaging. Therefore, machine learning methods can be used for vacuum leak monitoring, analyzing the types of leaks in the vacuum chamber of fusion devices and issuing timely alarms.

[0006] To address the hazards of leaks in fusion devices and the challenges of analyzing leak types under different conditions, it is crucial to explore an automatic vacuum leak detection system and method for real-time monitoring of residual gas partial pressure in fusion devices and timely detection of vacuum leaks. This is of great significance for restoring the vacuum environment of the vacuum chamber. Summary of the Invention

[0007] To address the challenges posed by the significant variations in residual gas composition under different leakage conditions in complex environments of fusion devices, and the shortcomings of manual monitoring, this invention provides an automatic vacuum leak detection system and method that monitors the gas composition of the vacuum chamber of a fusion device in real time and automatically triggers an alarm when a vacuum leak occurs.

[0008] This invention is achieved through the following technical solution:

[0009] An automatic vacuum leak detection system includes a plasma discharge vacuum chamber of a fusion device, a pumping system, a quadrupole mass spectrometer, a vacuum chamber monitoring system for the fusion device, and a neural network model for vacuum leak detection.

[0010] The plasma discharge vacuum chamber of the fusion device is the core of the device, serving as the container for plasma. The vacuum environment of this chamber directly affects the smooth operation of the fusion device's discharge and various physical experiments. The pumping system provides a vacuum environment for plasma discharge and the first wall treatment, promptly removing fuel and impurity particles generated during these processes and reducing the content of water and impurity gases. The quadrupole mass spectrometer is used to monitor the pressure and residual gas composition of the vacuum chamber in real time. The fusion device's vacuum chamber monitoring system includes LabVIEW leak monitoring software developed on the LabVIEW software platform. This software displays the residual gas partial pressure data collected by the quadrupole mass spectrometer as images on a computer and integrates a BP neural network model trained using historical fault data for vacuum leak analysis. The neural network model used for vacuum leak monitoring is trained based on historical vacuum chamber gas partial pressure data under normal and fault conditions. By promptly addressing leaks based on neural network models, repairs can be carried out, effectively reducing the likelihood of major accidents during fusion device operation. This enhances the safety of the vacuum system and the entire device, resolving the challenges of identifying vacuum leak types in complex environments and the untimely detection by manual operators. This is of great significance for the automated operation and automatic fault alarm of future fusion devices, providing safety support for future fusion devices.

[0011] The present invention also provides an automatic method for detecting vacuum leaks, comprising the following steps:

[0012] Step 1: During plasma discharge and physical experiments, a quadrupole mass spectrometer collects gas from the pumping pipe of the pumping system; the gas collected by the quadrupole mass spectrometer is ionized, mass analyzed, and signal processed to obtain the partial pressure data of the vacuum chamber gas corresponding to the relative molecular masses of 2, 3, 4, 14, 18, 28, 32, 40, 44, and 57, and the partial pressure data of the vacuum chamber gas is transmitted to the computer;

[0013] Step 2: After the vacuum chamber monitoring system of the fusion device on the computer acquires the partial pressure data of the gas in the vacuum chamber of the quadrupole mass spectrometer, it displays the changes in the partial pressure of the gas at each relative molecular mass in the form of a graph on the computer.

[0014] Step 3: The fusion device vacuum chamber monitoring system integrates a pre-trained neural network model through a deep learning module. While performing continuous spectrum scanning, it preprocesses the collected vacuum chamber gas partial pressure data and transmits the processed vacuum chamber gas partial pressure data as input to the neural network model. After calculation, the neural network model outputs the probability value of each leakage type. When the probability value of a certain leakage type reaches a certain threshold, the leakage type of that type is displayed on the computer and an alarm is triggered.

[0015] Furthermore, the real-time vacuum leak monitoring software designed using the LabVIEW software development platform mainly includes continuous spectrum scanning and full spectrum scanning functions, allowing users to select the appropriate scanning function according to their needs. Continuous spectrum scanning performs real-time continuous monitoring of several pre-set relative molecular masses of gas, displaying the results on the user interface, thus enabling real-time monitoring of leaks in the inner vacuum chamber. The full spectrum scanning function scans the partial pressure of gas components of each relative molecular mass within a selected range and displays it on the user interface. In addition, the software provides interfaces for external expansion, enabling communication with other control systems, such as acquiring plasma discharge firing data and status information of other vacuum equipment, facilitating the development of subsequent functions.

[0016] Furthermore, the machine learning model used to analyze leakage types employs a backpropagation (BP) neural network algorithm. Gas partial pressure data collected from the plasma discharge vacuum chamber of the fusion device under different leakage conditions and normal operating conditions are compiled into a sample set. This sample set is randomly divided into a training set and a test set. The training set is used to train the neural network model, while the test set is used to evaluate the model's performance.

[0017] Furthermore, a pre-trained neural network model is integrated into the real-time leak monitoring software developed using LabVIEW. In practical applications, during continuous spectrum scanning, the partial pressure data of the gas composition in the vacuum chamber of the fusion device is monitored in real time, and this data is used as input to the neural network model. The model outputs the leak type judgment result in real time, and when a leak is detected, the type of leak is promptly displayed on the interface, and an alarm is triggered.

[0018] The beneficial effects of this invention are as follows:

[0019] This invention designs and expands the functionality of a LabVIEW-based vacuum leak monitoring system, integrating a machine learning model. By introducing an artificial intelligence-driven self-learning approach, it analyzes the hidden internal relationships within the data, solving problems such as low efficiency in traditional leak detection, difficulties in leak analysis, and limitations of manual monitoring, thus providing technical assurance for the safe operation of fusion devices.

[0020] This invention provides an automatic vacuum leak detection system and method, which introduces artificial intelligence to provide a new approach for rapid leak detection and timely handling of leaks in future fusion reactors. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the structure of an automatic vacuum leak detection system according to the present invention.

[0022] Figure 2 This is a schematic diagram of the BP neural network algorithm for the vacuum leak monitoring method in this invention. Detailed Implementation

[0023] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the protection scope of the present invention.

[0024] like Figure 1 As shown, an automatic vacuum leak detection system of the present invention includes a plasma discharge vacuum chamber of a fusion device, a pumping system, a quadrupole mass spectrometer, a fusion device vacuum chamber monitoring system, and a neural network model for vacuum leak monitoring. Preferably, the pumping system includes pumping pipes.

[0025] The plasma discharge vacuum chamber of the fusion device is the core of the device, serving as the container for plasma. The vacuum environment of this chamber directly affects the smooth operation of the fusion device's discharge and various physical experiments. The pumping system provides a vacuum environment for plasma discharge and the first wall treatment, promptly removing fuel and impurity particles generated during these processes and reducing the content of water and impurity gases. The quadrupole mass spectrometer is used to monitor the pressure and residual gas composition of the vacuum chamber in real time. The fusion device vacuum chamber monitoring system includes LabVIEW leak monitoring software developed on the LabVIEW software platform. This software displays the residual gas partial pressure data collected by the quadrupole mass spectrometer as images on a computer and integrates a BP neural network model trained using historical fault data for vacuum leak analysis. The neural network model for vacuum leak monitoring is trained based on historical vacuum chamber gas partial pressure data under normal and fault conditions.

[0026] An automatic vacuum leak detection method of the present invention includes the following steps:

[0027] Step 1: During plasma discharge and related experiments at the fusion device, a quadrupole mass spectrometer deployed in the vacuum chamber pumping pipe collects the gas in the inner vacuum chamber of the pumping system. After ionization, mass analysis, and signal processing, the gas composition and partial pressure data of the inner vacuum chamber are obtained and transmitted to the computer.

[0028] Step 2: The fusion device vacuum chamber monitoring system, developed using LabVIEW on the computer, receives data from the quadrupole mass spectrometer and displays the gas partial pressure data of the vacuum chamber in a graphical manner using either continuous spectrum or full spectrum, based on the scanning parameters preset by the user, for easy and intuitive viewing by the user.

[0029] Step 3: Based on machine learning training, a model for vacuum leakage is trained. The fusion device vacuum chamber monitoring system integrates a pre-trained neural network model through a deep learning module. While performing continuous spectrum scanning, the collected vacuum chamber gas partial pressure data is preprocessed, and the processed vacuum chamber gas partial pressure data is transmitted as input to the neural network model. After calculation, the neural network model outputs the probability value of each leakage type. When the probability value of a certain leakage type reaches a certain threshold, the leakage type of that leakage is displayed on the computer and an alarm is triggered.

[0030] Step three, which involves training a model for vacuum leakage based on machine learning, includes the following steps:

[0031] Step 1: Collect and organize the partial pressure data of gas in the vacuum chamber of the fusion device under normal operating conditions and under different types of leaks over the years using a quadrupole mass spectrometer, and label them accordingly. Compile the gas partial pressure data collected from the plasma discharge vacuum chamber of the fusion device under different types of leaks and normal operating conditions into a sample set.

[0032] Step 2: Perform some preprocessing on the gas partial pressure data obtained in the above steps, sort out and remove extreme points and missing data points in the sample set, normalize the vacuum chamber gas partial pressure data at each relative molecular mass, and divide it into training set and test set according to a certain ratio.

[0033] Step 3: Use the training set obtained in Step 2 as input to the neural network model, and optimize the model using the backpropagation algorithm. Use the accuracy and error on the test set as evaluation metrics for model training. The BP neural network algorithm is as follows: Figure 2 As shown.

[0034] Furthermore, the described neural network is a multi-layer feedforward neural network employing error backpropagation, which can effectively fit any continuous function. The neural network classification model consists of an input layer, hidden layers, and an output layer. The training process of the model is divided into two parts: forward propagation and backward propagation. During forward propagation, the learning sample set is input into the neural network and calculated layer by layer to obtain the output value. Compared with the actual label of the learning set, if the error does not reach the expected value, the error is used as the target value for backpropagation. By calculating the derivative, the direction of error reduction is determined based on the calculation result. The weights and thresholds of each neuron are adjusted layer by layer according to the set learning rate until the expected error range is reached or the training number of times is reached, at which point training stops.

[0035] Furthermore, the feature is that the neural network model has N input neurons, K hidden layer neurons, and M output layer neurons, wherein the weight between the i-th neuron in the input layer and the j-th neuron in the hidden layer is w. ij The threshold is θ j The weight between the j-th neuron in the hidden layer and the l-th neuron in the output layer is v. jl The threshold is θ l f() is the activation function of the neuron, and z is the output of the hidden layer node. j and the output y of the output layer node l for:

[0036] z j =f(∑w ij x i -θ j )

[0037] y l =f(∑vjl z j -θ l )

[0038] To enable the model to output probability values ​​for each leakage type, the SoftMax activation function is used in the output layer, where M is the number of output neurons and exp() represents the exponential function.

[0039]

[0040] Figure 2 Chinese x a f represents the a-th input data. bc Let y represent the activation function of the c-th neuron in layer b. d This represents the d-th output data.

[0041] The aforementioned method utilizes gas partial pressure data collected by a quadrupole mass spectrometer under normal operating conditions and during leaks in the vacuum chamber of a fusion device. After preprocessing and analysis, this data is used as input to a BP neural network classification model. Through training and testing, a vacuum leak monitoring model is obtained. In application, the real-time gas partial pressure data collected by the quadrupole mass spectrometer is used as input to the model. After calculation, the model determines whether a leak has occurred in the vacuum chamber and the type of leak. Based on the model's output, the location of the leak can be quickly pinpointed, the cause of the leak can be rapidly identified, and the hazards caused by the leak can be reduced, thus providing a safety guarantee for the operation of the fusion device.

[0042] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. An automatic vacuum leak detection system, characterized in that, This includes the plasma discharge vacuum chamber of the fusion device, the pumping system, the quadrupole mass spectrometer, the vacuum chamber monitoring system of the fusion device, and the BP neural network model for vacuum leak monitoring; The plasma discharge vacuum chamber of the fusion device is used to contain plasma; the pumping system provides a vacuum environment for plasma discharge and the first wall treatment, and promptly removes fuel and impurity particles during plasma discharge and the first wall treatment process, reducing the content of water and impurity gases; the quadrupole mass spectrometer is used to monitor the pressure and residual gas composition of the plasma discharge vacuum chamber of the fusion device in real time; the fusion device vacuum chamber monitoring system displays the residual gas partial pressure data collected by the quadrupole mass spectrometer in the form of images on a computer, and integrates a BP neural network model trained using historical fault data to analyze vacuum leaks; the BP neural network model for vacuum leak monitoring is trained based on historical vacuum chamber gas partial pressure data under normal and fault conditions.

2. The automatic vacuum leak detection system according to claim 1, characterized in that: The pumping system includes an internal vacuum chamber, pumping pipes, various types of pump units, valves, vacuum gauges, and a power supply.

3. The automatic vacuum leak detection system according to claim 1, characterized in that: The quadrupole mass spectrometer collects gas from the pumping pipe of the plasma discharge vacuum chamber of the fusion device. After ionization, mass analysis, and signal processing, the gas composition and partial pressure data of the inner vacuum chamber are obtained and transmitted to the computer.

4. The automatic vacuum leak detection system according to claim 1, characterized in that: The fusion device vacuum chamber monitoring system has continuous spectrum scanning and full spectrum scanning functions, and is equipped with a communication interface to obtain plasma discharge gun number data and vacuum equipment status parameters.

5. The automatic vacuum leak detection system according to claim 1, characterized in that: The BP neural network model for vacuum leak monitoring first sets the neuron data and activation functions for each layer. During training, the backpropagation algorithm is used, and the model is trained with historical vacuum chamber gas partial pressure data under normal and fault conditions as input. This model is used to determine whether a leak has occurred in the plasma discharge vacuum chamber of the fusion device and to determine the type of leak.

6. An automatic vacuum leak detection system according to any one of claims 1-5, characterized in that: During plasma discharge and physical experiments in the fusion device, the quadrupole mass spectrometer transmits the partial pressure data of the gas in the vacuum chamber to the computer after obtaining the data. The vacuum chamber monitoring system of the fusion device displays the data in the form of continuous spectrum or full spectrum images.

7. A detection method for an automatic vacuum leak detection system according to any one of claims 1-6, characterized in that, Includes the following steps: Step 1: During plasma discharge and physical experiments, a quadrupole mass spectrometer collects gas from the pumping pipe of the pumping system; the gas collected by the quadrupole mass spectrometer is ionized, mass analyzed, and signal processed to obtain the partial pressure data of the vacuum chamber gas corresponding to the relative molecular masses of 2, 3, 4, 14, 18, 28, 32, 40, 44, and 57, and the partial pressure data of the vacuum chamber gas is transmitted to the computer; Step 2: After the vacuum chamber monitoring system of the fusion device on the computer acquires the partial pressure data of the gas in the vacuum chamber of the quadrupole mass spectrometer, it displays the changes in the partial pressure of the gas at each relative molecular mass in the form of a graph on the computer. Step 3: Based on machine learning training, a model for vacuum leakage is trained. The fusion device vacuum chamber monitoring system integrates a pre-trained neural network model through a deep learning module. While performing continuous spectrum scanning, the collected vacuum chamber gas partial pressure data is preprocessed, and the processed vacuum chamber gas partial pressure data is transmitted as input to the neural network model. After calculation, the neural network model outputs the probability value of each leakage type. When the probability value of a certain leakage type reaches a certain threshold, the leakage type of that type is displayed on the computer and an alarm is triggered.

8. The detection method according to claim 7, characterized in that, Step 3, training a model for vacuum leakage based on machine learning, includes the following steps: Step 3.1: Collect and organize the vacuum chamber gas partial pressure data collected by the quadrupole mass spectrometer under normal operating conditions and under different types of leakage of the fusion device over the years, including the vacuum chamber gas partial pressure data corresponding to relative molecular masses of 2, 3, 4, 14, 18, 28, 32, 40, 44, and 57, and label them accordingly. Step 3.2: Preprocess the gas partial pressure data obtained in Step 3.1, sort out and remove extreme points and missing data points in the sample set, normalize the vacuum chamber gas partial pressure data under each relative molecular mass, and divide it into training set and test set according to a certain ratio. Step 3.3: Use the training set obtained in Step 3.2 as the input to the neural network model, and use the backpropagation algorithm to optimize the model. Use the accuracy and error on the test set as the evaluation metrics for model training.

9. The detection method according to claim 7, characterized in that, The neural network described is a multi-layer feedforward neural network employing error backpropagation. The neural network model consists of an input layer, hidden layers, and an output layer. The training process of the neural network model is divided into two parts: forward propagation and backward propagation. During forward propagation, the learning sample set is input into the neural network and then passed layer by layer to calculate the output value. The output value is compared with the actual label of the learning set. If the error does not reach the expected value, the error is used as the target value for backpropagation. By calculating the derivative, the direction of error reduction is determined based on the calculation result. The weights and thresholds of each neuron are adjusted layer by layer according to the set learning rate until the expected error range is reached or the training number is reached, at which point training stops.

10. The detection method according to claim 9, characterized in that, ... The neural network model has N input neurons, K hidden layer neurons, and M output layer neurons, where the weight between the i-th neuron in the input layer and the j-th neuron in the hidden layer is w. ij The threshold is θ j The weight between the j-th neuron in the hidden layer and the l-th neuron in the output layer is v. jl The threshold is θ l f() is the activation function of the neuron, and z is the output of the hidden layer node. j and the output y of the output layer node l for: z j =f(∑w ij x i -θ j ) y l =f(∑v jl z j -θ l ) To enable the model to output probability values ​​for each leakage type, the SoftMax activation function is used in the output layer, where M is the number of output neurons and exp() represents the exponential function.

Images