Fusion reactor dynamic equipment cooperative fault diagnosis method based on multiple agents

Abstract

The invention relates to the technical field of fault diagnosis and intelligent manufacturing, in particular to a fusion reactor dynamic equipment collaborative fault diagnosis method based on multiple agents. According to the technical scheme, the method comprises the following steps that a multi-agent system is constructed, and a vibration agent, a temperature agent and an electromagnetic agent are deployed for fusion reactor moving equipment and used for collecting vibration data, temperature data and electromagnetic data respectively; performing signal conditioning on the acquired data by each agent and extracting universal features; based on a federated learning framework, multi-modal distributed training is carried out, and each agent independently trains a model to obtain a local gradient; each agent encrypts the local gradient and then uploads the local gradient to a cloud server, and secondary joint training of the model is carried out; and the cloud server encrypts and issues the model parameters after joint training to each agent, and through multi-agent collaboration, multi-modal data fusion and federated learning privacy protection, the fusion reactor moving equipment fault diagnosis accuracy, comprehensiveness and security are significantly improved.

Description

Technical Field

[0001] The present invention relates to the technical fields of fault diagnosis and intelligent manufacturing, and particularly relates to a collaborative fault diagnosis method for fusion reactor moving equipment based on multi-agent. Background Art

[0002] At present, during the operation of a nuclear fusion reactor, moving equipment (such as superconducting magnet systems, cryogenic pump sets, divertor drive mechanisms, etc.), as the core devices for maintaining plasma confinement and energy conversion under extreme conditions, are long-term subjected to the multiple coupling effects of high-intensity electromagnetic fields, thermal stress shocks, and complex mechanical vibrations. Traditional fault diagnosis methods rely on single sensor data sources for analysis and are difficult to effectively capture the abnormal signs of the interaction of multiple physical fields. For example, in the early stage of magnet quench, it may only be manifested as the distortion of the magnetic field distribution and the slight offset of the local temperature field, while the wear of the mechanical transmission mechanism may present high-frequency harmonic characteristics in the vibration spectrum. These cross-modal weakly correlated signals are easily ignored by conventional single-dimensional analysis methods. To solve the problems existing in the prior art, the present invention provides a collaborative fault diagnosis method for fusion reactor moving equipment based on multi-agent. Summary of the Invention

[0003] The object of the present invention is to address the problem that traditional fault diagnosis methods rely on single sensor data sources for analysis and are difficult to effectively capture the abnormal signs of the interaction of multiple physical fields in the background art, and propose a collaborative fault diagnosis method for fusion reactor moving equipment based on multi-agent.

[0004] The technical solution of the present invention: A collaborative fault diagnosis method for fusion reactor moving equipment based on multi-agent, comprising the following steps:

[0005] Construct a multi-agent system, and deploy vibration agents, temperature agents, and electromagnetic agents for the fusion reactor moving equipment, which are respectively used to collect vibration data, temperature data, and electromagnetic data;

[0006] Each agent performs signal conditioning on the collected data and extracts common features;

[0007] Based on the federated learning framework, perform multi-modal distributed training, and each agent independently trains a model to obtain local gradients;

[0008] Each agent encrypts and uploads the local gradients to the cloud server for secondary joint training of the model;

[0009] The cloud server encrypts and distributes the model parameters after joint training to each agent for multi-agent collaborative fault diagnosis.

[0010] Optionally, the vibration agent collects vibration data based on an IEPE acceleration vibration sensor and performs signal conditioning using an adaptive resonance wavelet packet decomposition algorithm. The adaptive noise threshold update formula is as follows:

[0011] , where is the noise threshold at the current moment , is the noise threshold at the next moment , is the step size, is the gradient value of the mean square error function;

[0012] The calculation formula for the denoised signal DS( ) is:

[0013] , where is the original acceleration signal collected by the vibration sensor, is the vibration signal value at time , is the sign function, is the noise threshold, is the exponential function.

[0014] Optionally, the temperature agent collects temperature data based on distributed optical fiber and thermal infrared imaging technology and constructs a three-dimensional temperature field reconstruction model. The model expression is:

[0015] , where is the reconstructed temperature value at any point in the three-dimensional space, is the three-dimensional single-point coordinate of the th measurement point, is the temperature coefficient of the th measurement point, is the measured temperature value at this location, is the temperature error value at this location, is the total number of points in the temperature field.

[0016] Optionally, the electromagnetic agent measures the magnetic field strength using three-dimensional vector fluxgate technology. The magnetic flux calculation formula is:

[0017] , where is the magnetic flux at time is the magnetic induction intensity, is the area perpendicular to the magnetic field direction.

[0018] Optionally, the three-modal data is normalized by Z-score and then converted into an image signal for training. Among them, the vibration signal is converted into a time-frequency diagram through time-frequency analysis, and the temperature and electromagnetic three-dimensional data are converted into standard images.

[0019] Optionally, during the federated learning process, an improved differential privacy algorithm is used to encrypt and transmit local features. By adding privacy noise that satisfies the Laplace probability distribution, differential attacks are prevented. The expression of the feature gradient output distribution is:

[0020] , where is the feature gradient output distribution, is the query function, is the privacy noise, is the feature data of the vibration agent, is the feature data of the temperature agent, is the feature data of the electromagnetic agent. The privacy noise satisfies the Laplace probability distribution:

[0021] , where is the Laplace probability distribution, is the interval range parameter.

[0022] Optionally, the cloud and each agent deploy a homomorphic encryption algorithm to achieve the encrypted transmission of local gradients. The gradient parameter encryption process is:

[0023] , where is the plaintext, is the ciphertext, is the encryption operation, is to randomly select a positive integer less than , is a random number less than , is the public key, ;

[0024] The gradient parameter decryption process is:

[0025] , where is the decryption operation, is the private key, is a specific function, specifically , is the independent variable, is the public key, is the random number.

[0026] Compared with the prior art, the present application includes at least one of the following beneficial technical effects:

[0027] By deploying three types of agents for vibration, temperature, and electromagnetic field to collect and analyze multi-physical field data, weak cross-modal correlation signals can be effectively captured, addressing the deficiencies of traditional single-dimensional analysis methods.

[0028] Based on federated learning, a collaborative analysis encryption mode is designed to encrypt and upload local gradients instead of raw data, avoiding the leakage of enterprise equipment operation data and ensuring data privacy and security.

[0029] Adaptive resonance wavelet packet decomposition, three-dimensional temperature field reconstruction, three-dimensional vector fluxgate technology, and improved differential privacy and homomorphic encryption algorithms are adopted to improve the signal processing accuracy and the security of gradient transmission, enabling joint training and dynamic updating of the diagnostic model.

[0030] Through multi-agent collaboration, multi-modal data fusion, and federated learning privacy protection, the present invention significantly improves the accuracy, comprehensiveness, and security of fault diagnosis for moving equipment in fusion reactors. Brief Description of the Drawings

[0031] Figure 1 It is a flowchart of a collaborative fault diagnosis method for moving equipment in a fusion reactor based on multi-agents. Detailed Embodiments

[0032] The technical solutions of the present invention will be further described below in conjunction with the drawings and specific embodiments.

[0033] Embodiment

[0034] Refer to Figure 1 : A collaborative fault diagnosis method for moving equipment in a fusion reactor proposed by the present invention constructs multi-agents for data collection and analysis scheduling for the moving equipment in the fusion reactor. Considering the particularity of the moving equipment in the fusion reactor, vibration agents, temperature agents, and electromagnetic agents are deployed. Data collection and signal conditioning are performed through the three types of agents to extract their respective general features. Based on the design idea of federated learning, multi-modal distributed training is carried out. By encrypting and uploading local gradients, secondary training of the model is performed to achieve the function of multi-agent collaborative diagnosis.

[0035] Among them, the vibration agent collects data based on an IEPE acceleration vibration sensor, and the signal conditioning is based on the adaptive resonance wavelet packet decomposition algorithm. The acceleration sensor signal is , the noise threshold is set as N, the noise threshold at the current moment is N(t), and the noise threshold at the next moment is N(t + 1). Then the adaptive threshold relationship is

[0036] , where is the noise threshold at the current moment and is the next moment The noise threshold is the step size is the gradient value of the mean square error function

[0037] After adaptive resonance wavelet decomposition, the denoised signal DS( ) is

[0038] , where is the original acceleration signal collected by the vibration sensor is the time The vibration signal value at is the sign function is the noise threshold is the exponential function

[0039] The temperature agent collects temperature data based on distributed optical fiber and thermal infrared imaging technology and constructs a three-dimensional temperature field reconstruction model. Among them, for a single point in three-dimensional space, the temperature field model reconstructed based on the temperature data is

[0040] , where The reconstructed temperature value of any point in three-dimensional space is the Three-dimensional space single-point coordinates of the th measurement point is the Temperature coefficient of the th measurement point is the measured temperature value at this place is the temperature error value at this place is the total number of points in the temperature field

[0041] The electromagnetic agent uses three-dimensional vector fluxgate technology to measure the magnetic field strength of the equipment in the fusion environment, and the magnetic flux value obtained is

[0042] , where is the magnetic flux at time t is the magnetic induction intensity is the area perpendicular to the magnetic field direction

[0043] After the three-modal data is normalized by Z-score, it is all converted into image signals for training. Among them, the vibration agent converts the vibration signal into a time-frequency diagram based on the time-frequency analysis method. The temperature agent and the electromagnetic agent convert the three-dimensional data into standard images

[0044] The three agents obtain their respective local gradients through independent training models. Subsequently, based on the federated learning framework, the agents improve the differential privacy algorithm and encrypt and transmit the local features of the agents to the cloud server for joint training. Among them, in order to prevent differential attacks, a query protection mechanism is designed for multiple agents. By adding noise, the probability of querying the same value is approximated.

[0045] , where, is the output distribution of the feature gradient, is the query function, is the privacy noise. is the feature data of the vibration agent, is the feature data of the temperature agent, is the feature data of the electromagnetic agent. The privacy noise satisfies the Laplace probability distribution

[0046] , where, is the Laplace probability distribution, is the interval range parameter.

[0047] The cloud and each agent deploy the homomorphic encryption algorithm to achieve the encrypted upload and download of the local gradients. Among them, the encryption process of the gradient parameter is:

[0048] , where, is the plaintext, is the ciphertext, is the encryption operation, is to randomly select a positive integer less than , is less than random number, is the public key, .

[0049] The decryption process of the gradient parameter is:

[0050] , where, is the decryption operation, is the private key, is a specific function, specifically , is the independent variable, is the public key, is the random number.

[0051] Through encrypted communication of the local gradients of multiple agents, after joint training in the cloud, it is encrypted and distributed to each agent to achieve collaborative training of data and privacy protection. The present invention designs three types of special agents, namely vibration agents, temperature agents, and electromagnetic agents, which are respectively used to collect and analyze the vibration, temperature, and electromagnetic multi-physical field data of the moving equipment of the fusion reactor. Through multi-modal data fusion and collaborative diagnosis, cross-modal weak correlation signals (such as magnetic field distortion and temperature offset in the early stage of magnet quench, high-frequency vibration harmonics of mechanical wear) that are easily overlooked by traditional single-dimensional analysis methods can be effectively captured, solving the problem that traditional methods rely on a single sensor data source and are difficult to reflect the abnormal signs of the coupling effect of multi-physical fields. The present invention adopts a federated learning framework to design a collaborative analysis encryption mode. Each agent only encrypts and uploads local gradient parameters (not the original data) to the cloud for joint training, avoiding the external leakage of the original equipment operation data of enterprises and meeting the data privacy protection requirements in industrial scenarios. Through the encrypted mutual transmission of gradient parameters and joint training in the cloud, a closed-loop for dynamic update of the diagnostic model is realized, solving the problem that the model cannot be trained online and the update lags due to data privacy restrictions.

[0052] Among them, the vibration agent adopts an adaptive resonance wavelet packet decomposition algorithm. By dynamically updating the noise threshold (formula ), and a non-linear denoising strategy, the noise interference under complex working conditions can be effectively suppressed, and the accuracy of vibration signal feature extraction can be improved.

[0053] In addition, the temperature agent constructs a three-dimensional temperature field model (formula ) based on distributed optical fiber and thermal infrared imaging technology to achieve a three-dimensional representation of the equipment temperature distribution and accurately locate local thermal anomalies.

[0054] It should be noted that the electromagnetic agent accurately measures the magnetic field intensity through three-dimensional vector fluxgate technology, combined with an improved differential privacy algorithm (adding Laplace noise) and a homomorphic encryption algorithm (formula and ) to ensure that the gradient parameters resist differential attacks and privacy leakage during transmission, realizing secure collaborative training of "data can be used but not visible".

[0055] In summary, the present invention significantly improves the accuracy, comprehensiveness, and security of fault diagnosis of the moving equipment of the fusion reactor through core technologies such as multi-agent collaboration, multi-modal data fusion, and federated learning privacy protection.

[0056] The above specific embodiments are only several alternative embodiments of the present invention. Based on the technical solutions of the present invention and the relevant inspirations of the above embodiments, those skilled in the art can make various alternative improvements and combinations to the above specific embodiments.

Claims

1. A collaborative fault diagnosis method for fusion reactor dynamic equipment based on multi-agent, characterized in that, Including the following steps: Construct a multi-agent system, and deploy vibration agents, temperature agents, and electromagnetic agents for the moving equipment of the fusion reactor, which are used to collect vibration data, temperature data, and electromagnetic data respectively; Each agent performs signal conditioning on the collected data and extracts common features; Based on the federated learning framework, multi-modal distributed training is carried out, and each agent independently trains the model to obtain local gradients; Each agent encrypts the local gradients and uploads them to the cloud server for secondary joint training of the model; The cloud server encrypts and distributes the model parameters after joint training to each agent for multi-agent collaborative fault diagnosis.

2. The collaborative fault diagnosis method for the fusion reactor dynamic equipment based on multi-agent according to claim 1, wherein The vibration agent collects vibration data based on an IEPE acceleration vibration sensor and uses an adaptive resonance wavelet packet decomposition algorithm for signal conditioning, where the adaptive noise threshold update formula is: Among them, is the noise threshold at the current moment , is the noise threshold at the next moment , is the step size, is the gradient value of the mean square error function; Denoised signal DS( ) is calculated by the formula: Among them, is the original acceleration signal collected by the vibration sensor, is the time of the vibration signal value, is the sign function, is the noise threshold, is the exponential function.

3. The collaborative fault diagnosis method for the fusion reactor dynamic equipment based on multi-agent according to claim 1, wherein The temperature agent collects temperature data based on distributed optical fiber and thermal infrared imaging technology and constructs a three-dimensional temperature field reconstruction model, and the model expression is: in, Any point in three-dimensional space The reconstruction temperature value of For the The three-dimensional space single-point coordinates of the measurement points, For the The temperature coefficient of each measuring point, is the measured temperature value at that location, is the temperature error value at this location, is the total number of points in the temperature field.

4. A collaborative fault diagnosis method for fusion reactor dynamic equipment based on multi-agent according to claim 1, characterized in that The electromagnetic agent measures the magnetic field strength using three-dimensional vector fluxgate technology, and the magnetic flux calculation formula is: Among them, is the magnetic flux at a moment, is the magnetic induction intensity, is the area perpendicular to the magnetic field direction.

5. A collaborative fault diagnosis method for fusion reactor dynamic equipment based on multi-agent according to claim 1, characterized in that The three-modal data is transformed into image signals for training after Z-score standardization. Among them, the vibration signal is transformed into a time-frequency diagram through time-frequency analysis, and the three-dimensional temperature and electromagnetic data are converted into standard images.

6. The collaborative fault diagnosis method for the fusion reactor dynamic equipment based on multi-agent according to claim 1, characterized in that During the federated learning process, an improved differential privacy algorithm is used to encrypt and transmit local features, and differential attacks are prevented by adding privacy noise that satisfies the Laplace probability distribution. The expression of the feature gradient output distribution is: Among them, is the feature gradient output distribution, is the query function, is the privacy noise, is the feature data of the vibration agent, is the feature data of the temperature agent, is the feature data of the electromagnetic agent, and the privacy noise satisfies the Laplace probability distribution: Among them, is the Laplace probability distribution, is the interval range parameter.

7. A collaborative fault diagnosis method for fusion reactor dynamic equipment based on multi-agent according to claim 1, characterized in that The cloud and each agent deploy a homomorphic encryption algorithm to achieve encrypted transmission of local gradients. The gradient parameter encryption process is: Among them, is the plaintext, is the ciphertext, is the encryption operation, is to randomly select a positive integer less than , is a random number less than , is the public key, ; The gradient parameter decryption process is: Among them, is the decryption operation, is the private key, is a specific function, specifically , is the independent variable, is the public key, is a random number.

Images