A millimeter wave security inspection instrument health management method and system

Abstract

This invention discloses a health management method and system for millimeter-wave security scanners, comprising: acquiring monitoring data on the working status and performance of each subsystem of the millimeter-wave security scanner and generating real-time operating status messages; constructing a digital model of the millimeter-wave security scanner based on the real-time operating status messages, fusing the working status and performance information features using an attention mechanism, and updating the parameters of the millimeter-wave security scanner digital model in real time based on a reinforcement learning algorithm; and completing fault detection of the millimeter-wave security scanner based on the residual between the predicted and measured values ​​output by the digital model. This invention not only monitors the working status of the security scanner but also its performance. By fusing the working status and information features using an attention mechanism and mapping them into a digital twin model in real time, and updating the model parameters in real time based on improved reinforcement learning, it achieves an upgrade from passive maintenance to proactive health management, providing highly reliable and low-cost operation and maintenance support for security inspection tasks.

Description

Technical Field

[0001] This invention relates to the field of millimeter-wave security inspection technology, specifically to a health management method and system for millimeter-wave security inspection equipment. Background Technology

[0002] To improve the efficiency of security checkpoints, millimeter-wave security scanners are evolving from linear arrays to area arrays. As the number of millimeter-wave security scanner channels increases, the system complexity and integration also rise significantly. Existing health management methods for millimeter-wave security scanners have significant shortcomings in terms of data acquisition during scanner operation, specifically:

[0003] 1) Conventional millimeter-wave security scanners only monitor the overall operating status of the machine, such as temperature and humidity, lacking dynamic monitoring of performance indicators. Traditional health management models based on single threshold criteria and periodic maintenance only monitor the operating status of individual subsystems, lacking dynamic monitoring of performance indicators. Suboptimal states in multiple subsystems can lead to system indicator degradation, resulting in increased false negative and false positive rates, making it difficult to ensure the reliable completion of security inspection tasks.

[0004] 2) The parameters of the digital twin model of conventional millimeter-wave security inspection equipment rely on manual experience and expert experience for calibration. It lacks an adaptive online update mechanism for model parameters and has disadvantages such as subjectivity, poor real-time performance, and high limitations. This affects the dynamic adaptability and accuracy of the digital twin model and makes it difficult to meet the high reliability requirements of security inspection tasks.

[0005] In the prior art, patent application publication number CN118828145A discloses a video summarization method and related apparatus. After obtaining the multimodal video representation of each video segment in the target video, the process of extracting the optimal combination of video segments from multiple video segments in the target video is modeled as a Markov decision process, taking into account both the local importance and global coherence of video segments. A reinforcement learning algorithm is used to solve the optimal strategy of the Markov decision process, extracting a more concise and coherent optimal combination of video segments composed of important video segments. This generates an accurate video summary based on the optimal combination of video segments, improving the quality of the video summary. Furthermore, a multi-head attention mechanism is applied to perform multimodal splicing and fusion of each video segment. However, in this patent, the use of reinforcement learning to solve the optimal strategy of the Markov decision process is only for obtaining the optimal combination of video segments in a multimodal video. Also, the multi-head attention mechanism in this patent only fuses multi-source data with the same attribute but different modal representations. Summary of the Invention

[0006] The technical problem to be solved by this invention is that conventional millimeter-wave security inspection instruments only monitor the overall working status of the machine and lack dynamic monitoring of working performance indicators, as well as the synergistic monitoring of working status and working performance indicators.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0008] A health management method for millimeter-wave security scanners includes:

[0009] Acquire monitoring data on the working status and performance of each subsystem of the millimeter-wave security inspection device, and generate real-time operating status messages;

[0010] Based on real-time operation status messages, a digital model of the millimeter-wave security scanner is constructed by fusing working status and performance information features using an attention mechanism, and the parameters of the millimeter-wave security scanner digital model are updated in real time based on a reinforcement learning algorithm.

[0011] Based on the residual between the predicted and measured values ​​output by the digital model of the millimeter-wave security scanner, fault detection of the millimeter-wave security scanner is completed.

[0012] Technical Effects: Based on the application scenarios of millimeter-wave security scanners in airports, customs, and other locations requiring long-term, high-reliability operation, this invention achieves the following through dynamic fusion detection of operating status and performance: First, it eliminates the limitations of single-sensor data, improving fault detection accuracy and enabling precise location of complex faults. Second, by utilizing multi-dimensional data dynamic fusion, it can obtain the nonlinear dynamic correlation between operating status and performance in real time. This invention significantly improves fault detection accuracy and greatly reduces the misjudgment problems caused by data fragmentation in traditional methods.

[0013] In this embodiment, the real-time running status message is generated, including:

[0014] Working status monitoring data X i,s (t) includes at least temperature information X i,s1 Humidity information X i,s2 Salt spray information X i,s3 Vibration information X i,s4 Where t is time and i is the i-th millimeter-wave security inspection subsystem;

[0015] Performance monitoring data X i,p (t) includes at least the channel amplitude-phase consistency parameter X. i,p1 Inter-channel amplitude and phase consistency parameter X i,p2 Noise figure parameter X i,p3 Local oscillator phase noise parameter X i,p4 Simulated target self-test parameters X i,p5 ;

[0016] The real-time running status message is [X] i,s1 ,X i,s2 ,X i,s3 ,X i,s3 ,X i,s4;X i,p1 ,X i,p2 ,X i,p3 ,X i,p4 ,X i,p5 ] T , where T is the transpose.

[0017] In this embodiment, the construction of a digital model for a millimeter-wave security scanner includes:

[0018] Working state features are extracted based on fully connected neural networks respectively. and working performance characteristics

[0019] Based on the attention mechanism, the working status characteristics of each subsystem of the millimeter-wave security scanner are analyzed. and working performance characteristics Perform feature fusion to obtain the fused feature vector h. f ;

[0020] The fused feature vector h f Map the data to the output space to obtain the digital model of the millimeter-wave security scanner.

[0021] In this embodiment, the fused feature vector h is obtained. f ,include:

[0022]

[0023]

[0024]

[0025] In the formula, Here, n represents the attention weights for the working status characteristics and performance characteristics of the millimeter-wave security scanner. The attention scores are defined for work status features and work performance features, where v is the attention vector and W is the attention score. a Let b be the attention weight matrix. a Let be the attention bias vector, tanh(·) be the hyperbolic tangent function, and exp be the exponential function.

[0026] Technical effect: By integrating working status and performance characteristics, the digital model of the millimeter-wave security scanner can reflect the health status of each component from real-time performance indicators, and also estimate the performance degradation from the health data of each component, effectively ensuring the reliable execution of security inspection tasks.

[0027] By updating the weighting coefficients of operational status and performance data in real time, the core capabilities of the millimeter-wave security scanner's digital model, such as dynamic adaptability, robust fault tolerance, and high sensitivity, can be enhanced. This mechanism not only improves the accuracy of data fusion of operational status and performance of various components in the security scanner but also enhances its practicality and reliability in complex security inspection scenarios.

[0028] In this embodiment, the expression for the digital model of the millimeter-wave security scanner is:

[0029]

[0030] In the formula, y(t) represents the measured value of the digital model of the millimeter-wave security scanner. The predicted value of the millimeter-wave security scanner digital model is W, where Activation Function(·) is the activation function. m b m These are the weight matrix and bias vector corresponding to the digital model of the millimeter-wave security scanner.

[0031] In this embodiment, the parameters W of the millimeter-wave security scanner digital model are dynamically updated based on reinforcement learning. m b m ;

[0032] Let the parameters of the millimeter-wave security scanner digital model be in the form of θ, such that θ = [W m b m ] T ;

[0033] Define state s t , representing the overall state of the millimeter-wave security scanner at time t;

[0034] Define action as a t Δθ represents the update amount of the digital model of the millimeter-wave security scanner. t ;

[0035] Define reward r t , representing the error between the measured and predicted values ​​of the digital model of the millimeter-wave security scanner;

[0036] Define strategy π θ (a t |s t ), indicating that in state s t Next, select action a. t The probability distribution;

[0037] Construct the objective function J(θ), representing the policy π. θ (a t |s t ) Expected cumulative discount rewards;

[0038] Combined with reward r t Discount Factor Υ t and state s t Construct the time-series difference error δ t ;

[0039] Based on the time difference error δ t Calculate the policy gradient Δ of the objective function. θ J(θ);

[0040] policy gradient Δ θ J(θ) is updated using the gradient ascent method until the strategy converges, thus obtaining the updated parameters of the millimeter-wave security inspection digital model.

[0041] In this embodiment, the timing difference error δ t It can be expressed by the following formula:

[0042] δ t =r t +Υ t V π (s t+1 )-V π (s t );

[0043] In the formula, V π (·) represents the value function.

[0044] In this embodiment, the policy gradient Δ of the objective function θ J(θ) is obtained through the following formula:

[0045]

[0046] In the formula, For the expected value, Δ θ logπ θ (a t |s t ) represents the logarithmic gradient of the policy, and T represents the total time.

[0047] In this embodiment, the policy convergence criterion is:

[0048] ||θ k+1 -θ k ||2≤∈;

[0049] In the formula, θ k+1 represents the parameters of the millimeter-wave security scanner digital model in the (k+1)th iteration; ∈ represents the convergence threshold.

[0050] Technical Effects: The millimeter-wave security scanner digital model achieves dynamic correction and prediction of model parameters through an improved reinforcement learning algorithm. Compared with traditional reinforcement learning algorithms, the optimization strategy adopts the 2-norm convergence criterion for model parameters. The 2-norm convergence criterion can significantly improve the stability, convergence speed, generalization ability and reliability of the model, and significantly improve the stability and adaptive adjustment ability of model parameter optimization.

[0051] This invention also provides a millimeter-wave security scanner health management system, which applies the above-described millimeter-wave security scanner health management method, including:

[0052] The dynamic monitoring module is used to acquire monitoring data on the working status and performance of each subsystem of the millimeter-wave security inspection instrument and generate real-time operating status messages;

[0053] The digital twin module is used to construct a digital model of the millimeter-wave security scanner based on real-time operating status messages, integrate working status and performance information features based on an attention mechanism, and update the parameters of the millimeter-wave security scanner digital model in real time based on a reinforcement learning algorithm.

[0054] The health management module completes fault detection of the millimeter-wave security scanner based on the residual between the predicted and measured values ​​output by the digital model of the millimeter-wave security scanner.

[0055] Compared with the prior art, the beneficial effects of the present invention are:

[0056] This invention not only monitors the working status of security scanners but also their performance. It uses an attention mechanism to fuse the working status and information into a digital twin model in real time. Based on improved reinforcement learning, it updates the model parameters in real time, achieving an upgrade from passive maintenance to proactive health management, and providing highly reliable and low-cost operation and maintenance support for security inspection tasks.

[0057] This invention is applied to the field of millimeter-wave security scanner management technology. Based on an attention mechanism, this invention fuses working status and performance data, enabling the millimeter-wave security scanner's digital model to acquire the nonlinear dynamic correlation between working status and performance in real time. This overcomes the pain point of misjudgment caused by fragmented data in traditional methods.

[0058] The reinforcement learning algorithm of this invention is applied to the real-time updating of digital model parameters of millimeter-wave security inspection equipment. It clearly defines important modules such as state, reward, and convergence strategy in reinforcement learning. Based on the reinforcement learning model, the convergence criterion is optimized. The L2 norm convergence criterion of model parameters can significantly improve the stability and reliability of the model. Attached Figure Description

[0059] Figure 1 This is a flowchart of a health management method for a millimeter-wave security scanner according to an embodiment of the present invention.

[0060] Figure 2 This is a block diagram of a millimeter-wave security inspection device health management system according to an embodiment of the present invention.

[0061] Figure 3 This is a block diagram of the dynamic monitoring module in an embodiment of the present invention.

[0062] Figure 4 This is a block diagram of the digital twin module according to an embodiment of the present invention.

[0063] Figure 5 This is a block diagram of the health management module according to an embodiment of the present invention. Detailed Implementation

[0064] To facilitate understanding of the technical solution of the present invention by those skilled in the art, the technical solution of the present invention will now be further described in conjunction with the accompanying drawings.

[0065] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0066] Please see Figure 1 and Figure 2 As shown, this embodiment provides a health management method for millimeter-wave security scanners, including:

[0067] S10 acquires monitoring data on the working status and performance of each subsystem of the millimeter-wave security scanner and generates real-time operating status messages.

[0068] Please see Figures 1 to 3 As shown in this embodiment, the millimeter-wave security scanner's subsystems include a millimeter-wave transceiver array, a frequency synthesizer and synchronization module, a millimeter-wave imaging module, and a hazardous materials detection module. The millimeter-wave transceiver array primarily utilizes a millimeter-wave radar array. The frequency synthesizer and synchronization module performs frequency synchronization processing on the millimeter-wave signals from the radar array. The millimeter-wave imaging module constructs the object's outline based on the frequency synthesizer and synchronized millimeter-wave signals by analyzing the time delay, intensity, and Doppler effect of the reflected signals. The hazardous materials detection module is primarily used for detecting hazardous materials.

[0069] In this embodiment, the working status monitoring data X i,s (t) includes at least temperature information X i,s1 Humidity information X i,s2 Salt spray information X i,s3 Vibration information X i,s4 ;

[0070] Working status monitoring data X i,s (t) is: [X i,s1 ,X i,s2 ,X i,s3 ,X i,s3 ,X i,s4 ,...] T ;

[0071] In the formula, t represents time, i represents the i-th millimeter-wave security inspection subsystem, and T represents transpose.

[0072] In this embodiment, the performance monitoring data X i,p (t) includes at least the channel amplitude-phase consistency parameter X. i,p1 Inter-channel amplitude and phase consistency parameter X i,p2 Noise figure parameter X i,p3 Local oscillator phase noise parameter X i,p4 Simulated target self-test parameters X i,p5 .

[0073] Then the performance monitoring data X i,p (t) is: [,X i,p2 ,X i,p3 ,X i,p4 ,X i,p5 ,...] T ;

[0074] The real-time running status message is [X] i,s1 ,X i,s2 ,X i,s3 ,X i,s3 ,X i,s4 ;X i,p1 ,X i,p2 ,X i,p3 ,X i,p4 ,X i,p5 ] T .

[0075] S20: Based on real-time operating status messages, the system integrates working status and performance information features using an attention mechanism to construct a digital model of the millimeter-wave security scanner, and updates the parameters of the digital model in real time using a reinforcement learning algorithm.

[0076] Please see Figures 1 to 4 As shown in this embodiment, before constructing the digital model of the millimeter-wave security scanner, data preprocessing is first performed. The real-time operating status messages are then denoised and normalized.

[0077]

[0078] In the formula, X′ s (t) represents the working status monitoring data after preprocessing; X′p (t) represents the performance monitoring data after preprocessing; μ s μ p : represents the working status monitoring data and the mean of the working status monitoring data, respectively; σ s σ p These represent the standard deviation of the working status monitoring data and the working status monitoring data, respectively.

[0079] In this embodiment, after preprocessing, multi-dimensional data feature fusion is performed. Specifically, the working status and performance information features are fused based on the attention mechanism, and the working status and performance monitoring data of each subsystem of the millimeter-wave security scanner are effectively combined through dynamic weight allocation to generate a unified feature representation and construct a digital model of the millimeter-wave security scanner.

[0080] Please see Figures 1 to 4 As shown, in this embodiment, the construction of a digital model for a millimeter-wave security scanner includes:

[0081] S21, extracting working state features based on fully connected neural networks respectively. and working performance characteristics

[0082] In this embodiment,

[0083] In the formula, This represents the working status data of the i-th millimeter-wave security inspection subsystem; The weight matrix represents the monitoring data of the working status of the i-th millimeter-wave security inspection subsystem; Let represent the bias vector of the monitoring data for the working status of the i-th millimeter-wave security scanner subsystem; ReLU(·) represents the activation function. Undoubtedly, This is the preprocessed data.

[0084]

[0085] In the formula, This represents the performance monitoring data of the i-th millimeter-wave security inspection subsystem; The weight matrix represents the performance monitoring data of the i-th millimeter-wave security inspection subsystem; This represents the bias vector of the performance monitoring data of the i-th millimeter-wave security inspection subsystem.

[0086] S22, based on the attention mechanism, analyzes the working status characteristics of each subsystem of the millimeter-wave security scanner. and working performance characteristics Perform feature fusion to obtain the fused feature vector h. f .

[0087] In this embodiment,

[0088] In the formula, Here, represents the attention weights for operational status characteristics and performance characteristics, and n represents the subsystems of the millimeter-wave security scanner.

[0089]

[0090]

[0091] W represents the attention scores for work status features and work performance features, where v is the attention vector used to map features in the attention space to scalar attention scores. a Here is the attention weight matrix, used to map feature vectors to the attention space, b a Let be the attention bias vector, tanh(·) be the hyperbolic tangent function, and exp be the exponential function.

[0092] S23, fuse the feature vector h f Map the data to the output space to obtain the digital model of the millimeter-wave security scanner.

[0093] In this embodiment, the output parameter y(t) of the millimeter-wave security scanner digital model and the input fusion feature vector h f The equation is:

[0094]

[0095] In the formula, y(t) represents the measured value of the digital model of the millimeter-wave security scanner. The predicted value of the millimeter-wave security scanner digital model is W, where Activation Function(·) is the activation function. m b m These are the weight matrix and bias vector corresponding to the digital model of the millimeter-wave security scanner.

[0096] In this embodiment, the parameters of the millimeter-wave security scanner digital model are updated in real time based on a reinforcement learning algorithm. Specifically, the parameters W of the millimeter-wave security scanner digital model are dynamically updated based on reinforcement learning. m b m ,include:

[0097] S241, let the parameters of the millimeter-wave security scanner digital model be in the form of θ, such that θ = [W m b m ] T .

[0098] S242, Define state s t , representing the overall state of the millimeter-wave security scanner at time t.

[0099] In this embodiment, s t =h f (t).

[0100] S243, define action as a t Δθ represents the update amount of the digital model of the millimeter-wave security scanner. t This refers to the direction and magnitude of parameter adjustment.

[0101] In this embodiment, a t =Δθ t .

[0102] S244, Define reward r t , which represents the error between the measured and predicted values ​​of the digital model of the millimeter-wave security scanner.

[0103] In this embodiment,

[0104] S245, Define strategy π θ (a t |s t ), indicating that in state s t Next, select action a. t The probability distribution.

[0105] S246, Construct the objective function J(θ), representing the policy π. θ (a t |s t ) Expected accumulated discount rewards.

[0106] In this embodiment, the expression for the objective function J(θ) is:

[0107]

[0108] In the formula, Let Υ be the expected value, T be the total time, and Υ be the expected value. t The discount factor represents the importance of the current reward versus the future reward. The objective function J(θ) measures the policy π. θ Long-term performance.

[0109] S247, combined with reward r t Discount Factor Υ t and state s t Construct the time-series difference error δ t .

[0110] In this embodiment, δ t =r t +Υ t V π (s t+1 )-V π (st In the formula, V π (·) represents the value function, which represents the expected accumulated discount reward.

[0111] S248, based on the timing difference error δ t Calculate the policy gradient Δ of the objective function. θ J(θ).

[0112] In this embodiment,

[0113] In the formula, Δ θ logπ θ (a t |s t Let θ be the logarithmic gradient of the policy, representing the relationship between parameters θ and the action a. t The impact;

[0114] S249, regarding the policy gradient Δ θ J(θ) is updated using the gradient ascent method until the strategy converges, thus obtaining the updated parameters of the millimeter-wave security inspection digital model.

[0115] In this embodiment, θ is updated using the gradient ascent method as follows:

[0116]

[0117] In the formula, Let θ be the updated parameter, and α be the learning rate, which controls the step size for updating parameter θ.

[0118] In this embodiment, the 2-norm convergence criterion is used for policy convergence. Specifically, the policy convergence criterion is as follows:

[0119] ||θ k+1 -θ k ||2≤∈;

[0120] In the formula, θ k+1 represents the parameters of the millimeter-wave security scanner digital model in the (k+1)th iteration; ∈ represents the convergence threshold.

[0121] S30 completes fault detection of the millimeter-wave security scanner based on the residual between the predicted and measured values ​​output by the digital model of the millimeter-wave security scanner.

[0122] Please see Figures 1 to 5 As shown in this embodiment, the predicted values ​​of the output parameters of the millimeter-wave security scanner digital model are... The residual between the measured value y(t) and the fault detection is performed, i.e., a health assessment. If the residual exceeds a threshold, a fault is determined. After fault determination, the health status of the security scanner is identified using a deep learning model, including: lifespan prediction of each subsystem component and risk prediction of each performance indicator. In this embodiment, the method of fault diagnosis using deep learning is not limited. Furthermore, based on the results of fault diagnosis and fault mode recognition, maintenance strategies can be output according to corresponding contingency plans.

[0123] Please see Figures 1 to 5 As shown, the present invention also provides a millimeter-wave security scanner health management system, which applies the above-described millimeter-wave security scanner health management method, including:

[0124] The dynamic monitoring module is used to acquire monitoring data on the working status and performance of each subsystem of the millimeter-wave security scanner and generate real-time operating status messages.

[0125] The digital twin module is used to construct a digital model of the millimeter-wave security scanner based on real-time operating status messages, integrate working status and performance information features using an attention mechanism, and update the parameters of the millimeter-wave security scanner digital model in real time based on a reinforcement learning algorithm.

[0126] The health management module completes fault detection of the millimeter-wave security scanner based on the residual between the predicted and measured values ​​output by the digital model of the millimeter-wave security scanner.

[0127] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention, and no reference numerals in the claims should be construed as limiting the scope of the claims.

[0128] The above-described embodiments are merely examples of implementation methods of the invention. The scope of protection of the present invention is not limited to the above-described embodiments. For those skilled in the art, several modifications and improvements can be made without departing from the concept of the present invention, and these all fall within the scope of protection of the present invention.

Claims

1. A health management method for a millimeter-wave security scanner, characterized in that, include: Acquire monitoring data on the working status and performance of each subsystem of the millimeter-wave security scanner, and generate real-time operating status messages, including: Working status monitoring data At least include temperature information Humidity information Salt spray information Vibration information ;in, For time, For the first A millimeter-wave security inspection subsystem; Performance monitoring data At least include channel amplitude and phase consistency parameters Inter-channel amplitude and phase consistency parameters Noise figure parameters Local oscillator phase noise parameters Simulated target self-test parameters ; The real-time running status message is ,in, For transpose; Based on real-time operational status reports, a digital model of the millimeter-wave security scanner is constructed by fusing operational status and performance information features using an attention mechanism. The parameters of this digital model are then updated in real-time using a reinforcement learning algorithm. , ,include: Let the parameters of the millimeter-wave security scanner digital model be: Form, making ;in, , For the weight matrix and bias vector corresponding to the digital model of the millimeter-wave security scanner; Define state This indicates that millimeter-wave security scanners are in use. The overall state at any given moment; Define the action as This indicates the update amount of the digital model of the millimeter-wave security scanner. ; Define rewards , representing the error between the measured and predicted values ​​of the digital model of the millimeter-wave security scanner; Define strategy , indicating the state Next, select an action. The probability distribution; Construct the objective function , indicating strategy The expected cumulative discount reward; Combined with rewards Discount Factor and state Constructing time-series difference error ; Based on timing difference error Calculate the policy gradient of the objective function. ; policy gradient Update using gradient ascent method The process continues until the strategy converges, at which point the updated parameters of the millimeter-wave security scanner digital model are obtained. Based on the residual between the predicted and measured values ​​output by the digital model of the millimeter-wave security scanner, fault detection of the millimeter-wave security scanner is completed.

2. The health management method for millimeter-wave security scanner according to claim 1, characterized in that, Constructing a digital model of a millimeter-wave security scanner, including: Working state features are extracted based on fully connected neural networks respectively. and working performance characteristics ; Based on the attention mechanism, the working status characteristics of each subsystem of the millimeter-wave security scanner are analyzed. and working performance characteristics Perform feature fusion to obtain the fused feature vector. ; fuse feature vectors Map the data to the output space to obtain the digital model of the millimeter-wave security scanner.

3. The health management method for millimeter-wave security scanner according to claim 2, characterized in that, Obtain the fused feature vector ,include: ； ； ； ； ； In the formula, , Attention weights for work status characteristics and work performance characteristics. For the various subsystems of the millimeter-wave security scanner, , Attention scores are assigned to work status characteristics and work performance characteristics. For attention vectors, This is the attention weight matrix. For attention bias vectors, It is the tangent function of a hyperbola. It is an exponential function.

4. The health management method for millimeter-wave security scanner according to claim 2, characterized in that, The expression for the digital model of the millimeter-wave security scanner is: ； In the formula, These are the measured values ​​from the digital model of the millimeter-wave security scanner. These are the predicted values ​​from the digital model of the millimeter-wave security scanner. For activation function, , These are the weight matrix and bias vector corresponding to the digital model of the millimeter-wave security scanner.

5. The health management method for millimeter-wave security scanner according to claim 4, characterized in that, Timing Differential Error It can be expressed by the following formula: ； In the formula, The table shows the value function.

6. The health management method for millimeter-wave security scanner according to claim 4, characterized in that, Policy gradient of the objective function It can be obtained through the following formula: ； In the formula, As expected, Let the logarithmic gradient of the policy be . This represents the total time.

7. The health management method for millimeter-wave security scanner according to claim 4, characterized in that, The criterion for policy convergence is: ； In the formula, Indicates the first Parameters of the next iteration of the millimeter-wave security scanner digital model; This represents the convergence threshold.

8. A health management system for a millimeter-wave security inspection device, characterized in that, The health management method for the millimeter-wave security scanner according to any one of claims 1-7 includes: The dynamic monitoring module is used to acquire monitoring data on the working status and performance of each subsystem of the millimeter-wave security inspection instrument and generate real-time operating status messages; The digital twin module is used to construct a digital model of the millimeter-wave security scanner based on real-time operating status messages, integrate working status and performance information features based on an attention mechanism, and update the parameters of the millimeter-wave security scanner digital model in real time based on a reinforcement learning algorithm. The health management module completes fault detection of the millimeter-wave security scanner based on the residual between the predicted and measured values ​​output by the digital model of the millimeter-wave security scanner.

Images