Hot cathode filament full life prediction device and method based on deep learning

Abstract

The invention is suitable for the field of health detection and life prediction of a hot cathode ionization vacuum gauge filament in a full life cycle, and provides a hot cathode filament full life prediction device and method based on deep learning, and the method comprises the steps: carrying out the preprocessing of historical data related to the performance of a hot cathode iridium-yttrium oxide filament; carrying out convolution and pooling operation on the preprocessed data through a convolutional neural network, carrying out modeling on a time sequence dependency relationship in filament performance data by utilizing a bidirectional long-short-term memory network, discarding part of useless data through a forgetting gate, introducing an attention mechanism, and carrying out the time sequence dependency relationship modeling; adopting a particle swarm optimization algorithm to optimize hyper-parameters in the convolutional neural network, the bidirectional long-short-term memory network and the attention mechanism; and the features processed by the modules are input to an output layer, key performance indexes of the thermionic cathode iridium-yttrium oxide filament are predicted, efficient fusion of multi-dimensional features and model performance optimization are realized, and the key performance indexes of the thermionic cathode filament are accurately predicted.

Description

Technical Field

[0001] The present invention belongs to the field of health detection and life prediction of hot cathode ionization vacuum gauge filaments throughout their life cycle, and in particular to a hot cathode filament full life prediction device and method based on deep learning. Background Art

[0002] In the field of vacuum technology, hot cathode filaments are the core components of many vacuum devices. Their performance and lifespan directly determine the reliability and service life of the entire device. Therefore, accurately predicting the full lifespan of hot cathode filaments has important theoretical significance and practical application value.

[0003] The current hot cathode filament lifecycle prediction technology still has many shortcomings. There is an urgent need to develop a lifecycle prediction device and method that can comprehensively consider multiple influencing factors, establish a high-precision prediction model, and adapt to new filament materials and application scenarios, so as to improve the reliability and service life of vacuum devices and reduce maintenance costs. Summary of the Invention

[0004] The purpose of the present invention is to provide a device and method for predicting the full life of a hot cathode filament based on deep learning, aiming to solve the problems raised in the above background technology.

[0005] The present invention is implemented as follows: on the one hand, a method for predicting the full life of a hot cathode filament based on deep learning, the method comprising:

[0006] Preprocessing historical data related to the performance of a hot cathode iridium-yttrium oxide filament, the historical data including experimental data of cathode voltage, cathode current, cathode bias, grid electron current, tube wall temperature, and vacuum chamber pressure, the preprocessing including data cleaning and normalization;

[0007] The pre-processed data is convolved and pooled using a convolutional neural network to obtain the attenuation characteristic information of the hot cathode filament resistance.

[0008] A bidirectional long short-term memory network is used to model the temporal dependencies in filament performance data. Use a forget gate to discard some useless data and capture forward and backward time series information.

[0009] The attention mechanism is introduced to assign different weights to the features of different time steps output by the bidirectional long short-term memory network, highlighting the key features that have a greater impact on the prediction target;

[0010] Particle swarm optimization algorithm is used to optimize the hyperparameters of convolutional neural networks, bidirectional long short-term memory networks, and attention mechanisms;

[0011] The features processed by the above modules are input into the output layer to predict the key performance indicators of the hot cathode iridium-yttrium oxide filament.

[0012] As a further solution of the present invention, the vacuum chamber is a KF25 tee, one end of which is connected to a cold cathode ionization vacuum gauge, and the other end is connected to a hot cathode ionization vacuum gauge.

[0013] As a further solution of the present invention, the molecular pump group is fixedly connected to the inflation vacuum valve.

[0014] As a further solution of the present invention, the convolutional neural network is used to extract spatial features from input image data and automatically learn local patterns, textures, and shape information in the image.

[0015] As a further solution of the present invention, the bidirectional long short-term memory network is used to process the feature sequences extracted by the convolutional neural network, capture the temporal dependencies of the feature sequences, and understand the long-term patterns and evolution trends in the sequence data.

[0016] As a further solution of the present invention, the particle swarm optimization algorithm is used to automatically search and optimize key hyperparameters in convolutional neural networks, bidirectional long short-term memory networks, and attention mechanisms, and to find the optimal model configuration in the parameter space by simulating the foraging behavior of bird flocks.

[0017] As a further solution of the present invention, on the other hand, a hot cathode filament full life prediction device based on deep learning includes: a molecular pump group, a vacuum valve, a cold cathode ionization vacuum gauge, a vacuum chamber, a hot cathode ionization vacuum gauge, a controller, a host computer and a temperature control box.

[0018] The present invention provides a device and method for predicting the full life of a hot cathode filament based on deep learning. The device improves data quality through preprocessing, uses a convolutional neural network to extract resistance attenuation characteristics, and a bidirectional long short-term memory network to capture timing dependencies and filter invalid information. It combines the attention mechanism to focus on key features, and then optimizes hyperparameters through a particle swarm optimization algorithm to achieve efficient fusion of multi-dimensional features and optimization of model performance, accurately predict key performance indicators of hot cathode filaments, provide a scientific basis for filament life assessment and maintenance, and significantly improve prediction accuracy and reliability. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 This is a schematic diagram of the structure of a hot cathode filament full life prediction device based on deep learning.

[0020] Figure 2 This is a flow chart of the convolutional neural network-bidirectional long short-term memory network-attention mechanism-particle swarm optimization algorithm of the present invention.

[0021] Figure 3This is a graph of the measured time series values of the characteristic order parameters of the hot cathode filament throughout its life cycle extracted by the experimental device established by the present invention.

[0022] Figure 4 This is a database diagram consisting of 36,662 sets of measured experimental data sets required for the deep learning model training based on the convolutional neural network-bidirectional long short-term memory network-attention mechanism-particle swarm optimization algorithm proposed in this invention.

[0023] Figure 5 This is the loss function diagram of the convolutional neural network-bidirectional long short-term memory network-attention mechanism-particle swarm optimization algorithm deep learning algorithm proposed in this invention.

[0024] Figure 6 This is a diagram showing the hot cathode filament life prediction effect of the convolutional neural network-bidirectional long short-term memory network-attention mechanism-particle swarm optimization algorithm deep learning algorithm proposed in the present invention.

[0025] Labeling instructions: 1~2-molecular pump group; 3-vacuum valve; 4-cold cathode ionization vacuum gauge; 5-vacuum chamber; 6-hot cathode ionization vacuum gauge; 7-controller; 8-host computer; 9-temperature control box. DETAILED DESCRIPTION

[0026] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0027] The specific implementation of the present invention is described in detail below with reference to specific embodiments.

[0028] To make the purpose, technical solutions, and advantages of the examples of the present invention more clear, the technical solutions in the examples of the present invention are clearly and completely described below in conjunction with the examples of the present invention. Obviously, the examples described are only part of the examples of the present invention, not all of them. All other examples obtained by ordinary technicians in this field based on the examples of the present invention without making any creative efforts are within the scope of protection of the present invention.

[0029] like Figure 1The figure shows the principle of a method for monitoring the damage status of a hot cathode filament. The method comprises molecular pumps 1 and 2, a vacuum valve 3, a cold cathode ionization vacuum gauge 4, a vacuum chamber 5, a hot cathode ionization vacuum 6, a controller 7, a host computer 8, and a high-speed camera 9. The molecular pumps 1 and 2 are connected to the vacuum valve 3. One end of the vacuum chamber 5 is connected to the cold cathode ionization vacuum gauge 4, and the other end is connected to the hot cathode ionization vacuum gauge 6. The hot cathode ionization vacuum gauge 6 is connected to its controller 7 and the host computer 8 to monitor various factors of the filament. A temperature control box 9 maintains the ambient temperature.

[0030] In order to facilitate data recording and make data recording comprehensive, a host computer is used to collect data.

[0031] In order to control the temperature of the environment, a temperature control box is used to control the temperature of the environment.

[0032] like Figure 2 As shown in the figure, a flowchart of establishing a hot cathode filament life prediction model is shown. First, the filament operation data is preprocessed and feature engineered. Then, a convolutional neural network is used to extract local time features. Then, a bidirectional long short-term memory network is used to capture the bidirectional time dependency in the sequence. The attention mechanism is used to weight the importance of different time steps. Finally, a particle swarm optimization algorithm is used to search for the optimal hyperparameter configuration of the model. The trained model can predict and evaluate the remaining life of new filament operation data.

[0033] In order to better provide higher-level input for subsequent time series modeling, a convolutional neural network algorithm is used to automatically extract representative local patterns and short-term correlation features from time series data.

[0034] In order to better predict the lifespan trend of iridium-yttrium oxide, a bidirectional long short-term memory network algorithm is used to capture the long-term, bidirectional dependencies in time series data and understand the impact of historical information on future states.

[0035] In order to enable the model to learn and focus on the time steps or features that are more important for the filament life prediction task, thereby improving the accuracy and interpretability of the prediction, an attention mechanism is added. The attention mechanism first solves the matrices Q, K, and V by performing a linear transformation on the input:

[0036] ; (1)

[0037] ; (2)

[0038] ; (3)

[0039] Where, I, w q (or w k 、wv ) represent: input time data, randomly generated weight matrix. Perform dot product operation on Q and K to get the attention score s i :

[0040] ; (4)

[0041] Then for s i Normalize and map to the data range of 0 to 1 to obtain the attention weight matrix a i :

[0042] ; (5)

[0043] Where L is the input step size. The final weighted output c is:

[0044] ; (6)

[0045] In order to improve the generalization ability and prediction performance of the model and avoid the blindness of manual parameter adjustment, the algorithm incorporates the particle swarm optimization algorithm, which efficiently searches for the optimal or near-optimal hyperparameter combination in the complex model parameter space. The particle velocity and position update corresponding to the particle swarm optimization algorithm in the convolutional neural network is expressed as:

[0046] ; (7)

[0047] ; (8)

[0048] in, 、 、 、 、 They represent: the velocity of particle i in the jth dimension of the tth generation, the influence of the previous iteration velocity on the current velocity, the particle's response to individual experience during the search, the group experience, the individual's historical optimal position, and the global historical optimal position. The inertia weight adjusts the particle's search range in the solution space, thereby balancing the particle's global and local search capabilities. The individual learning factor c1 and the group learning factor c2 adjust the step size of the particle's learning towards its own historical optimal position and the global optimal position, respectively.

[0049] The present invention provides a method for predicting the life of an iridium-yttrium oxide filament, which comprises the following steps in order:

[0050] Step (1): Solder the removed filament into the hot cathode ionization vacuum gauge 6 and connect it to Figure 2In the device, start the temperature control box 9, start the molecular pump group 1-2, open the vacuum valve 3, start the controller 7 and the host computer 8, observe the changes in some parameters of the filament during the change process, and wait for the pressure to reach the negative four order of magnitude of ten.

[0051] Step (2): The vacuum chamber pressure reaches 10 -4 After the Pa level is reached, the data is recorded using the host computer 8.

[0052] Step (3): Collect, clean, preprocess the filament operation data, perform normalization / standardization, and generate time series samples using the sliding window method.

[0053] Step (4): Convolutional neural network feature extraction, that is, inputting the time series samples into the convolutional neural network and extracting local time features through operations such as convolution and activation.

[0054] Step (5): Bidirectional long short-term memory network temporal modeling, that is, inputting the feature sequence extracted by the convolutional neural network into the bidirectional long short-term memory network, and using its bidirectionality to capture the long-term temporal dependency and contextual information in the sequence.

[0055] Step (6): Weighted attention mechanism, that is, inputting the hidden state of the bidirectional long short-term memory network into the attention mechanism layer, learning the importance of different time steps, and generating a weighted context vector to highlight key information.

[0056] Step (7): Particle swarm optimization algorithm parameter optimization, that is, using the particle swarm optimization algorithm to search for the hyperparameters of the model (including the convolutional neural network, the bidirectional long short-term memory network and the attention mechanism layer), the goal is to minimize the prediction error on the validation set, and the particle swarm optimization algorithm finds the optimal or approximately optimal hyperparameter configuration.

[0057] Step (8): Model training: Use the hyperparameter configuration optimized by the particle swarm optimization algorithm to train the entire convolutional neural network-bidirectional long short-term memory network-attention mechanism model. Usually, optimization algorithms such as gradient descent are used to update the weight parameters within the model.

[0058] Step (9): Life prediction and model evaluation, that is, inputting new filament operation data into the trained model to obtain the prediction result of the remaining life of the filament; and using an independent test set to evaluate the prediction performance of the model.

[0059] The experimental device developed by the present invention can effectively collect cathode current, cathode voltage, bias voltage, grid electron current, vacuum chamber pressure, reference gauge (i.e., full-scale cold gauge) pressure, and experimental gauge (i.e., hot cathode ionization gauge) pressure to obtain a real-time data set of the characteristic order parameters of the hot cathode filament under actual working conditions, such as Figure 3In addition, the present invention is based on the proposed deep learning model based on convolutional neural network-bidirectional long short-term memory network-attention mechanism-particle swarm optimization algorithm. The total amount of data required for training is 36662 sets of measured experimental data. The data of the filament life cycle of the hot cathode ionization gauge under actual working conditions has been systematically extracted: ionization gauge startup-ionization gauge degassing-ionization gauge gear switching-ionization gauge stable operation-filament natural aging-filament melting, as shown in the figure. Figure 4 In addition, the convolutional neural network-bidirectional long short-term memory network-attention mechanism-particle swarm optimization algorithm proposed in the present invention can ensure that the training loss and verification loss of the deep learning model converge after multiple rounds of training by combining the local feature extraction capability of the convolutional neural network, the long-range dependency modeling capability of the bidirectional long short-term memory network, the key information focusing capability of the attention mechanism, and the global parameter optimization capability of the particle swarm optimization algorithm. Figure 5 As shown, the filament degradation law and remaining life are accurately extracted, such as Figure 6 shown.

[0060] In order to enable the above-mentioned method and system to be loaded and run smoothly, in addition to the various modules mentioned above, the system may also include more or fewer components than described above, or a combination of certain components, or different components, for example, it may include input and output devices, network access devices, buses, processors and memories, etc.

[0061] The processor may be a central processing unit, other general-purpose processors, digital signal processors, application-specific integrated circuits, off-the-shelf programmable gate arrays or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor. The processor is the control center of the system, connecting various components using various interfaces and lines.

[0062] The technical features of the above-mentioned embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above-mentioned 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 this specification.

[0063] The above-described embodiments merely illustrate several implementations of the present invention, and while their descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art would be able to make numerous variations and improvements without departing from the spirit of the present invention, all of which fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be determined by the appended claims.

[0064] The above description is only 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 in the scope of protection of the present invention.

Claims

1. A hot cathode filament life prediction method based on deep learning, characterized in that: The method comprises: Preprocessing historical data related to the performance of a hot cathode iridium-yttrium oxide filament, the historical data including experimental data of cathode voltage, cathode current, cathode bias, grid electron current, tube wall temperature, and vacuum chamber pressure, the preprocessing including data cleaning and normalization; The pre-processed data is convolved and pooled using a convolutional neural network to obtain the attenuation characteristic information of the hot cathode filament resistance. A bidirectional long short-term memory network is used to model the temporal dependencies in filament performance data. Use a forget gate to discard some useless data and capture forward and backward time series information. Introducing the attention mechanism to assign different weights to the features of different time steps output by the bidirectional long short-term memory network; Particle swarm optimization algorithm is used to optimize the hyperparameters of convolutional neural networks, bidirectional long short-term memory networks, and attention mechanisms; The features processed by the above modules are input into the output layer to predict the key performance indicators of the hot cathode iridium-yttrium oxide filament.

2. The hot cathode filament life prediction method based on deep learning according to claim 1, characterized in that: The vacuum chamber is a KF25 tee, one end of which is connected to a cold cathode ionization vacuum gauge, and the other end is connected to a hot cathode ionization vacuum gauge.

3. The hot cathode filament life prediction method based on deep learning according to claim 1, characterized in that: The molecular pump group is fixedly connected to the inflation vacuum valve.

4. The hot cathode filament life prediction method based on deep learning according to claim 1, characterized in that: The convolutional neural network is used to extract spatial features from input image data and automatically learn local patterns, textures, and shape information in the image.

5. The hot cathode filament life prediction method based on deep learning according to claim 1, characterized in that: The bidirectional long short-term memory network is used to process the feature sequences extracted by the convolutional neural network, capture the temporal dependencies of the feature sequences, and understand the long-term patterns and evolution trends in the sequence data.

6. The hot cathode filament life prediction method based on deep learning according to claim 1, characterized in that: The particle swarm optimization algorithm is used to automatically search and optimize key hyperparameters in convolutional neural networks, bidirectional long short-term memory networks, and attention mechanisms, and to find the optimal model configuration in the parameter space by simulating the foraging behavior of bird flocks.

7. A hot cathode filament life prediction device based on deep learning, characterized in that: The method for predicting the full life of a hot cathode filament based on deep learning as described in any one of claims 1 to 6 is applied, and the device includes: a molecular pump group, a vacuum valve, a cold cathode ionization vacuum gauge, a vacuum chamber, a hot cathode ionization vacuum gauge, a controller, a host computer and a temperature control box.

Images