A Human-Machine Function Allocation Method for Armored Vehicle Cockpit Based on Dynamic Bayesian Networks

By decomposing the performance of the human-machine system using a dynamic Bayesian network model, the problem of dynamically adjusting the allocation of human-machine functions in the cockpit of armored vehicles was solved, enabling the prediction of future performance and the optimization of automation levels, thereby improving system efficiency.

CN117196183BActive Publication Date: 2026-11-10NORTHWESTERN POLYTECHNICAL UNIV
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202311030002.0
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2023-08-15
Publication Date
2026-11-10
Estimated Expiration
2043-08-15

AI Technical Summary

Technical Problem

The existing static human-machine interface allocation method in the driver's cab of armored vehicles cannot meet the dynamic needs of complex battlefield environments, cannot predict the future performance status of the human, and the performance improvement after the automation level is adjusted is not significant.

Method used

A dynamic Bayesian network model is adopted to decompose the performance of the human-machine system into human performance and machine performance. Based on the task attribute requirements, human performance is decomposed into accuracy and urgency performance. A fault tree model is established and transformed into a dynamic Bayesian network model. The node probability is confirmed by the fuzzy comprehensive evaluation method, and future performance changes are predicted and the automation level is adjusted.

Benefits of technology

It enables real-time performance prediction of changes in battlefield mission attributes, allowing for advance adjustment of automation levels, improvement of human-machine system performance, and adaptation to the demands of complex and ever-changing battlefield environments.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN117196183B_ABST
    Figure CN117196183B_ABST
Patent Text Reader

Abstract

This invention relates to the field of dynamic human-machine function allocation technology, specifically to a method for allocating human-machine functions in the cockpit of armored vehicles based on a dynamic Bayesian network. The method includes: dividing the performance of the human-machine system into human performance and machine performance, and further decomposing human performance into accuracy performance and urgency performance; obtaining a dynamic Bayesian network model; obtaining predicted values ​​for accuracy performance, urgency performance, and machine performance; obtaining predicted values ​​for the human-machine system performance; and comparing the adjusted human-machine system performance with the original human-machine system performance to confirm the final adjustment scheme. This invention utilizes a dynamic Bayesian model to realize a method for predicting the performance of the human-machine system in real time according to battlefield mission attributes. It can predict the performance changes of the human-machine system over a period of time and confirm the dynamic adjustment scheme for the automation level in advance, which has high application value for improving the effectiveness of the human-machine system in the transparent cockpit of future armored vehicles.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of dynamic human-machine function allocation technology, and specifically to a method for human-machine function allocation in the cockpit of an armored vehicle based on a dynamic Bayesian network. Background Technology

[0002] With the rapid development of information technology, the automation and intelligence levels of armored vehicle crew compartments have significantly improved. Faced with the complex working environment of the armored vehicle driver's cab, the constantly changing human states and machine performance render existing static human-machine function allocation methods inadequate for practical applications. Therefore, it is necessary to research dynamic human-machine function allocation methods to dynamically distribute functions between the human and the system based on changes in the human-machine system's state during system operation, thereby maintaining optimal system performance. Dynamic human-machine function allocation needs to address two issues: when to trigger and how to allocate functions after triggering.

[0003] To achieve the above objectives, those skilled in the art establish a quantitative assessment model of task load based on task complexity and physiological signals, and determine an automation level adjustment scheme based on load balancing rules. Alternatively, they determine the operator's load status by obtaining the preset operator's workload and reaction time, and then confirm the automation level adjustment scheme. It is evident that those skilled in the art primarily predict human performance based on mental workload and task performance, and use the current state prediction results to determine whether a trigger has occurred. However, many indicators affect human performance in armored vehicle compartments, and relying solely on mental workload and task performance to confirm human performance is too one-sided. Moreover, the above methods can only determine whether a trigger has occurred based on the predicted current human performance, but cannot predict performance over a future period, and cannot provide advance prevention and adjustment based on the prediction results. In actual combat, changes in the battlefield situation may lead to changes in the urgency and accuracy requirements of the mission, and existing methods cannot determine whether human performance still meets the requirements when mission attributes change. Furthermore, for human-machine systems, there is a problem that the performance improvement after adjusting the automation level is not significant, requiring optimization of key indicators affecting human-machine system performance. However, existing methods can only predict human performance positively and cannot reversely identify the causal indicators. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention primarily focuses on the allocation of human-machine functions in the transparent cockpit of armored vehicles. Addressing the shortcomings of existing methods, three issues need to be resolved: (1) Existing methods employ overly simplistic performance indicators, especially in the context of armored vehicle combat, where the environment significantly impacts human and machine performance. A comprehensive and effective human-machine performance prediction model based on the combat environment is needed, considering human, machine, and environment aspects; (2) In the face of fluctuating demands for mission accuracy and urgency in battlefield environments, existing methods cannot determine the impact of mission attribute changes on human-machine system performance; (3) Existing methods can only predict the current performance status of the crew, not the future performance status, to facilitate preventative measures and adjustments. Furthermore, they do not consider solutions when the performance improvement of the human-machine system is not significant after automation level adjustments. This invention provides a method for allocating human-machine functions in the cockpit of armored vehicles based on dynamic Bayesian networks. This method utilizes a dynamic Bayesian model to realize a real-time prediction method for human-machine system performance that changes with battlefield mission attributes. It can predict the performance changes of the human-machine system over a future period and confirm the dynamic adjustment scheme for automation levels in advance. When changing the automation level does not significantly improve the performance of the human-machine system, the performance can be enhanced by optimizing the priority order of indicators and improving the status of these indicators. Addressing the complex and ever-changing battlefield environment and mission requirements, this invention has high application value in improving the effectiveness of the human-machine system in the transparent cockpit of future armored vehicles.

[0005] The purpose of this invention is to provide a method for allocating human-machine functions in the cockpit of an armored vehicle based on a dynamic Bayesian network, including:

[0006] The performance of human-machine systems is divided into human performance and machine performance, and human performance is further decomposed into accuracy performance and urgency performance; indicators affecting human performance and machine performance are obtained.

[0007] Based on the indicators of accuracy performance, urgency performance, machine performance, and human performance and machine performance, a dynamic Bayesian network model structure is established, and dynamic nodes are identified. The fuzzy comprehensive evaluation method is used to confirm the prior probability of the root node and the conditional probability of other nodes except the root node, and the transition probability of the dynamic nodes is confirmed. The dynamic Bayesian network model structure is quantified to obtain the dynamic Bayesian network model.

[0008] Predicted values ​​for accuracy performance, urgency performance, and machine performance are obtained based on the dynamic Bayesian network model;

[0009] Based on the predicted values ​​of accuracy and urgency performance, obtain the predicted value of human performance; based on the predicted values ​​of human performance and machine performance, obtain the predicted value of human-machine system performance.

[0010] Set a performance threshold for the human-machine system. When the predicted performance value of the human-machine system is lower than the threshold, the automation level will be adjusted.

[0011] Based on the performance of the adjusted human-machine system, compare it with the performance of the original human-machine system to confirm the final adjustment plan.

[0012] Preferably, the process of establishing the dynamic Bayesian network model structure includes: establishing a fault tree model based on indicators of accuracy performance, urgency performance, machine performance, and human performance and machine performance, and then transforming the established fault tree model into a dynamic Bayesian network model structure.

[0013] Preferably, the predicted value of human performance is obtained according to the following steps:

[0014] Obtain historical data on various task scenarios, and establish weights for accuracy and urgency performance under each task scenario using the fuzzy comprehensive evaluation method; based on the weights of accuracy and urgency performance under each task scenario and the predicted values ​​of accuracy and urgency performance, obtain the predicted value of human performance under each task scenario.

[0015] Preferably, the predicted values ​​of accuracy performance, urgency performance, and machine performance are obtained by inputting node evidence of the root node over a period of time into a dynamic Bayesian network model.

[0016] Preferably, the predicted performance value of the human-machine system is obtained according to the following steps:

[0017] Based on the current level of automation and the performance of humans and machines in various types of tasks, obtain the weights of human performance and machine performance;

[0018] Based on the weights of human performance and machine performance, as well as the predicted values ​​of human performance and machine performance, the predicted values ​​of human performance and machine performance are obtained.

[0019] Preferably, when the level is below a threshold, the automation level is adjusted, including: when the human performance improvement is not significant after adjustment, a sensitivity analysis is performed to confirm the optimization priority of each node, and the decision-maker determines the optimization plan.

[0020] Preferably, the transition probability of the dynamic node is obtained based on the joint probability of the child node set and the parent node set.

[0021] Preferably, the prior probability of the root node is obtained by using triangular fuzzy numbers to perform a weighted average of expert evaluations and normalizing the result.

[0022] Preferably, the conditional probabilities of nodes other than the root node are obtained by using triangular fuzzy numbers to perform a weighted average of expert evaluations and normalizing the results.

[0023] Preferably, the indicators affecting human performance and machine performance are obtained from three aspects: human, machine, and environment; human performance is decomposed into accuracy performance and urgency performance based on task attribute requirements.

[0024] The present invention has at least the following beneficial effects:

[0025] This invention provides a method for allocating human-machine functions in the cockpit of armored vehicles based on dynamic Bayesian networks. This method decomposes human-machine system performance into human performance and machine performance. Based on task attribute requirements, human performance is further decomposed into accuracy performance and urgency performance. Indicators affecting human performance are identified from three aspects: human, machine, and environment. The dynamic Bayesian network model structure is confirmed based on a fault tree model. Fuzzy comprehensive evaluation is used to confirm the prior probability and conditional probability of the root node, and dynamic nodes and transition probabilities are confirmed based on expert experience. Node evidence predicting the root node over a certain period is confirmed based on task list information. This evidence is input into the constructed dynamic Bayesian network model to obtain predicted values ​​for crew accuracy performance, urgency performance, and machine performance. Possible task scenarios are summarized, and a weight library for accuracy and urgency performance is established using fuzzy comprehensive evaluation. This library is invoked based on real-time task conditions to obtain predicted human performance values. Human-machine system performance is calculated based on the current level of automation. A threshold for human-machine system performance is set; if the level falls below the threshold, dynamic adjustment of the automation level is triggered. When the improvement in human performance after adjustment is not significant, root cause analysis is used to confirm the optimization priority of each node, and the decision-maker determines the optimization scheme.

[0026] This invention utilizes a dynamic Bayesian model to implement a method for predicting the performance of human-machine systems (HMS) in real time, adapting to changes in battlefield mission attributes. It can predict HMS performance changes over a future period and proactively determine dynamic adjustment schemes for automation levels. When changing the automation level does not significantly improve HMS performance, performance can be enhanced by optimizing the priority order of indicators and improving their states. Addressing the complex and ever-changing battlefield environment and mission attribute requirements, this invention has high application value in improving the effectiveness of HMS systems in the transparent cockpits of future armored vehicles.

[0027] This invention comprehensively considers changes in both human and machine performance, establishing a comprehensive and effective human-machine performance prediction model based on the combat environment from three aspects: human, machine, and environment. This invention decomposes human performance into accuracy performance and urgency performance based on the accuracy and urgency requirements of the mission. It then calculates the current crew performance by applying the accuracy and urgency weights in real time according to the battlefield environment. This invention utilizes a human-machine system performance prediction model constructed using a dynamic Bayesian network model. By inputting predictive evidence from predictable nodes over a period of time, it can obtain predicted performance values ​​for the human-machine system over a future period, facilitating proactive preventative decision-making. Furthermore, when performance improvement is not significant after adjusting the automation level, sensitivity analysis can be used to determine the optimization order of each performance indicator, addressing the problem at its root. Attached Figure Description

[0028] Figure 1 Fault tree model for human-machine system performance.

[0029] Figure 2 DBN model for accuracy performance, urgency performance, and machine performance.

[0030] Figure 3 Membership degree of working hours.

[0031] Figure 4 This is a flowchart of a human-computer performance modeling method based on dynamic Bayesian networks.

[0032] Figure 5 A flowchart for dynamic human-machine function allocation. Detailed Implementation

[0033] In order to illustrate the technical means and effects adopted by the present invention in order to achieve the intended purpose of the invention, the following detailed description is provided in conjunction with the embodiments.

[0034] This invention mainly considers the impact of mission attributes on the human-machine system of armored vehicle cockpit. Based on dynamic Bayesian theory, it establishes a human-machine system performance prediction model and a dynamic adjustment method for automation level, solving the problems of when to trigger and how to allocate after triggering.

[0035] A method for allocating human-machine functions in the cockpit of an armored vehicle based on a dynamic Bayesian network includes:

[0036] S1. Divide the performance of human-machine systems into human performance and machine performance, and further decompose human performance into accuracy performance and urgency performance; obtain the indicators that affect human performance and machine performance;

[0037] The indicators affecting human and machine performance are obtained from three aspects: human, machine, and environment; human performance is decomposed into accuracy performance and urgency performance based on task attribute requirements.

[0038] In this embodiment, the performance of the human-machine system is divided into human performance and machine performance. Based on the task attribute requirements, human performance is further decomposed into accuracy performance and urgency performance. The indicators affecting human performance and machine performance are identified from the perspectives of human, machine, and environment through expert opinions and literature research.

[0039] The performance indicators obtained from the three aspects of people, machines, and environment are as follows: People aspect: task load, crew capability, task complexity, task workload, task difficulty, time pressure, mental fatigue, working hours, trust level, and task performance; Machine aspect: failure frequency; Environmental aspect: temperature, vibration, and overall working environment. The specific relationships between these performance indicators and their impact are shown in Table 1.

[0040] Table 1: Performance Indicators of Human-Machine Systems and Their Impact

[0041]

[0042]

[0043] S2. Based on the indicators of accuracy performance, urgency performance, machine performance, human performance, and machine performance, establish the dynamic Bayesian network model structure and identify the dynamic nodes; use the fuzzy comprehensive evaluation method to confirm the prior probability of the root node and the conditional probability of all nodes, and confirm the transition probability of the dynamic nodes, quantify the dynamic Bayesian network model structure, and obtain the dynamic Bayesian network model.

[0044] The process of establishing the dynamic Bayesian network model structure includes: establishing a fault tree model based on indicators of accuracy performance, urgency performance, machine performance, human performance, and machine performance, and then transforming the established fault tree model into a dynamic Bayesian network model structure.

[0045] The transition probability of the dynamic node is obtained based on the joint probability of the child node set and the parent node set.

[0046] The prior probability of the root node is obtained by using triangular fuzzy numbers to perform a weighted average of expert evaluations and normalizing the result.

[0047] The conditional probabilities of nodes other than the root node are obtained by using triangular fuzzy numbers to perform a weighted average of expert evaluations and normalizing the results.

[0048] In this embodiment, fault tree models are established for accuracy performance, urgency performance and machine performance respectively, and then transformed into a dynamic Bayesian network model structure.

[0049] Specifically, fault tree models that achieve accuracy and urgency performance include... Figure 1As shown; in the fault tree model, the various indicators are independent of each other. When converting it to a Bayesian network model, directed arcs need to be added based on the causal relationships between the performance indicators; dynamic nodes refer to nodes whose state at the previous moment will affect the state at the next moment, specifically including: mental fatigue, trust level; the final accuracy performance model and urgency performance model are as follows. Figure 2 As shown.

[0050] In this embodiment, the fuzzy comprehensive evaluation method is used to determine the prior probability and conditional probability of the root node, the transition probability of the dynamic node is determined based on expert experience, and the dynamic Bayesian network model is constructed in GeNIe software.

[0051] Specifically, dynamic nodes are only related to the state of the previous time step, and their transition probabilities are calculated using the joint probability formula.

[0052]

[0053] Wherein, P(X) t+1 X is the joint probability of the variables; t and X t+1 These represent the child node sets at time t and time t+Δt, respectively; and Let represent the parent node sets at time t and t+Δt, respectively.

[0054] Due to the difficulty in obtaining data in this field, the prior probability and conditional probability of the root node need to be obtained using a fuzzy evaluation method based on industry expert assessments, as detailed below:

[0055] Step 1: Set the rating level set V = {very inconsistent, inconsistent, somewhat inconsistent, moderate, somewhat consistent, consistent, very consistent}, and the triangular fuzzy number is represented by M = (l, r, m), as shown in Equation (2). Each language variable can be converted into the corresponding triangular fuzzy number.

[0056]

[0057] Step 2: Using P ij,k =(l ij,k ,r ij,k ,m ij,k ) represents the opinion of the k-th expert on node X. i The triangular fuzzy number for evaluation in state j, assuming the weights of each expert are (a1, a2, ..., a...). n Then, the formula for calculating the mean of the fuzzy number is shown in equation (3):

[0058]

[0059] Step 3: The obtained triangular fuzzy numbers are further converted into definite probability values ​​using the area mean method:

[0060] Specifically, the prior probability can be calculated through steps one to three. The final probability value is then normalized so that the sum of the probabilities of each state of the root node is 1, which is the prior probability of the root node.

[0061] Specifically, conditional probability refers to the probability of a child node occurring given a parent node state. It can be calculated through steps one to three. The final probability value is then normalized so that the sum of the probabilities of each state of the root node is 1, which is the conditional probability of the root node.

[0062] It should be noted that both prior probability and conditional probability are calculated based on expert evaluation. For example, prior probability is the weighted average of the expert's evaluation of the prior state of the root node, and conditional probability is the weighted average of the expert's evaluation of the state of the child nodes given the state of the parent node. The calculation process for both is the same.

[0063] S3. Obtain the predicted values ​​of accuracy performance, urgency performance, and machine performance based on the dynamic Bayesian network model;

[0064] The predicted values ​​for accuracy, urgency, and machine performance are obtained by inputting node evidence for the root node over a period of time into a dynamic Bayesian network model.

[0065] In this embodiment, based on task list information, environmental prediction information, and node values ​​available at the current moment, evidence for predicting the root node over a certain period is confirmed. This evidence is then input into the constructed dynamic Bayesian network model to obtain predicted values ​​for accuracy performance, urgency performance, and machine performance. The task list information includes future task difficulty, workload, crew capability values ​​for the corresponding task type, crew cumulative working time, and time pressure based on task completion time requirements. The environmental prediction information includes values ​​for temperature, vibration, noise, and harmful gas concentration. The node values ​​available at the current moment include trust, task performance, and failure frequency.

[0066] In this embodiment, predictable nodes include: temperature (machine), temperature (environment), vibration (machine), vibration (human), noise, harmful gases, crew capacity, task workload, task difficulty, working time and time pressure. Nodes for which only evidence at the current moment can be obtained include: trust, task performance, and failure frequency.

[0067] Nodes for which specific values ​​can be obtained include: temperature (machine), vibration (machine), temperature (environment), vibration (human), working hours, noise, and harmful gases. Taking working hours as an example, based on triangular fuzzy numbers, there are five numerical ranges for working hours: less than 1 hour, 0.5-2 hours, 1.5-4 hours, 3-7 hours, and greater than 6 hours. The membership function and the corresponding scores for changes in working hours are obtained as follows: Figure 3 As shown. If the current working time is 6.2 hours, then the probability of a node with a long working time is 0.8, and the membership degree of a node with a short working time is 0.2.

[0068] The nodes whose prior probabilities cannot be calculated include: crew capacity, task workload, task difficulty and time pressure, and conditional probabilities all need to be confirmed using formulas (2) and (3) based on expert fuzzy comprehensive evaluation.

[0069] S4. Obtain the predicted value of human performance based on the predicted values ​​of accuracy performance and urgency performance; obtain the predicted value of human-machine system performance based on the predicted values ​​of human performance and machine performance.

[0070] Set a performance threshold for the human-machine system. When the predicted performance value of the human-machine system is lower than the threshold, the automation level is adjusted. When the performance is lower than the threshold, the automation level is adjusted, including: when the improvement in human performance is not significant after adjustment, sensitivity analysis is performed to confirm the optimization priority of each node, and the decision-maker determines the optimization plan.

[0071] The predicted values ​​for human performance are obtained by following these steps:

[0072] Obtain historical data on various task scenarios, and establish weights for accuracy and urgency performance under each task scenario using the fuzzy comprehensive evaluation method; based on the weights of accuracy and urgency performance under each task scenario and the predicted values ​​of accuracy and urgency performance, obtain the predicted value of human performance under each task scenario.

[0073] The predicted performance value of the human-machine system is obtained according to the following steps:

[0074] Based on the current level of automation and the performance of humans and machines in various types of tasks, obtain the weights of human performance and machine performance;

[0075] Based on the weights of human performance and machine performance, as well as the predicted values ​​of human performance and machine performance, the predicted values ​​of human performance and machine performance are obtained.

[0076] In this embodiment, possible task scenarios are summarized, and a performance weight library for accuracy and urgency under each scenario is established using the fuzzy comprehensive evaluation method. This library is then invoked based on the real-time status of the task to obtain human performance prediction values.

[0077] Accuracy and urgency are categorized into seven levels: very high, high, relatively high, medium, low, low, and very low. Based on triangular fuzzy numbers, the membership degrees corresponding to these seven levels are determined to be 0.975, 0.875, 0.7, 0.5, 0.3, 0.125, and 0.025, respectively. Let T be the membership degree of task accuracy and urgency. 准 T 紧 The accuracy calculation formula is shown in formula (4), and the urgency formula is shown in formula (5).

[0078]

[0079]

[0080] It should be noted that when the accuracy requirement is "high" and the urgency requirement is "low", the accuracy performance weight W... 准 for: Urgency performance weight W 紧 for:

[0081] Human performance HMS 人性能 The calculation formula is:

[0082] HMS 人性能 =W 紧 ×HMS 紧迫性 +W 准 ×HMS 准确性 (6)

[0083] In this embodiment, based on the current level of automation and the performance of humans and machines on various types of tasks, the weights of human and machine performance are determined, thereby obtaining the predicted performance value of the human-machine system. The overall process is as follows: Figure 4 As shown; the automation level needs to be divided according to the task type. Assuming there are 4 automation levels, when at level 3, the human performance automation level weight is W. 人自动化 =0.33, the weight of the machine automation level is W 机器自动化 =0.67; Through comparison of human and machine capabilities, expert evaluation obtains the capability scores of humans and machines under various task types. Let the capability score for task A be W. 人能力 =0.8, Machine Ability Score W 机能力 =0.4; then the performance weight of the person at this time is:

[0084] W 人 =Min(W 人自动化 W 人能h )=0.33 (7)

[0085] The weights for machine performance are:

[0086] W 机 =Min(W 机自动化 W 机能力 )=0.4 (8)

[0087] The formula for calculating the performance of a human-machine system is:

[0088] HMS 人机系统 =W 人 ×HMS 人性能 +W 机 ×HMS 机性能 (9)

[0089] The performance threshold of the human-machine system needs to be determined by experienced experts. Let the threshold be 0.5. When the performance of the human-machine system is lower than 0.5, the automation level is adjusted. For each alternative adjustment scheme, the adjusted performance of the human-machine system is calculated by formulas (7)-(9), and compared with the original performance of the human-machine system to confirm the final adjustment scheme.

[0090] S5. Based on the adjusted performance of the human-machine system, compare it with the original performance of the human-machine system to confirm the final adjustment plan.

[0091] In this embodiment, the optimization priority of each root node is calculated as follows, assuming the sensitivity of index A is S. A The distance between the current value and the optimal value of indicator A is T. A T A If the potential for improvement of indicator A is represented by the optimization priority P of indicator A, then the optimization priority P of indicator A is:

[0092] P = S A ×T A (10)

[0093] Assuming that after adjusting the current automation level, the HMI performance remains below 0.5, and vibration optimization has the highest priority, then the impact of vibration on the HMI can be reduced to improve efficiency by lowering the driving speed and replanning the route. The overall process is as follows: Figure 5 As shown in Table 2. Specific implementations of S3-S5 are as follows: Sheridan classifies automation levels from fully manual to fully automatic into 10 levels, with higher levels representing greater participation and decision-making authority of the automated system. Referring to this automation level classification table and the actual situation of armored vehicles, the automation level for each type of task is determined, as shown in Table 2.

[0094] Table 2: Classification of Automation Levels for Various Task Types

[0095]

[0096] Assume that a passenger will need to perform a task list on a human-machine automation system in the near future, and the passenger's human-machine ability level is as shown in Table 3:

[0097] Table 3: Task List

[0098]

[0099]

[0100] Based on battlefield conditions, predicted values ​​for cabin temperature, cabin vibration, overall environmental conditions, mission workload, mission difficulty, time pressure, crew capabilities, and working hours can be obtained for a future period. Values ​​for current trust levels, mission performance, and failure frequency can also be obtained, as shown in Table 4. Based on these predictions, crew and machine performance are forecasted for a future period to determine whether dynamic human-machine interface allocation should be triggered.

[0101] Table 4: Evidence from the Root Node

[0102]

[0103] The accuracy requirements for each task are obtained based on formulas (2)-(3), as shown in Table 5.

[0104] Table 5: Accuracy Requirements

[0105]

[0106] Driving missions prioritize timeliness, with operational precision having little impact on the driving process. In contrast, reconnaissance, decision-making, strike, and assessment missions prioritize accuracy. The weighting ratios of accuracy and urgency performance, derived from formulas (4)-(5), are as follows:

[0107] Table 6: Weighting Ratios

[0108]

[0109] Based on the dynamic Bayesian network, the node evidence in Table 4 is input, and the human performance is obtained according to formula (6). The real-time capability prediction results of the human-machine system are obtained according to formulas (7)-(9). The performance prediction results under different urgency conditions are then obtained as shown in Tables 7 and 8:

[0110] Table 7: Performance Prediction Results with Time Pressure of 0.1

[0111]

[0112] Table 8: Performance Prediction Results at Time Pressure 0.9

[0113]

[0114] Suppose that when the real-time capability level of the human-machine system is lower than 0.5, the automation level is adjusted. The reliability of human-machine performance after the automation level is increased or decreased can be calculated according to formulas (7)-(9). The reliability of human-machine performance before the change is compared with that before the change to determine whether to increase or decrease the automation level. If there is no significant improvement after the adjustment, the key factors affecting human-machine performance can be identified through sensitivity analysis, and then a human decision can be made to confirm the improvement plan. As shown in Table 8, when the time pressure of the task is small, it can be found that the real-time capability of the human-machine system at time T5 of the strike phase is less than 0.5, and the automation level needs to be adjusted to improve the overall system efficiency. The performance of the human-machine system after increasing or decreasing the automation level is shown in Table 9:

[0115] Table 9: Human-Machine System Performance Before and After Automation Level Adjustment

[0116]

[0117] Analysis of Table 9 shows that when the time pressure is low, the system is already at the optimal automation level at time T5, but the real-time capability value of the human-machine system is still less than 0.5. This means that simply adjusting the automation level of the human-machine system cannot improve its performance. Root cause optimization is required. The optimization priority of the root node is obtained according to formula (10) as shown in Table 10.

[0118] Table 10: Root Node Optimization Priority

[0119]

[0120]

[0121] Table 10 shows that at time T5, vibration will significantly affect the real-time capabilities of the human-machine system. The impact of vibration can be reduced by adjusting the route or reducing the speed.

[0122] This invention provides a method for allocating human-machine functions in the cockpit of armored vehicles based on dynamic Bayesian networks. This method effectively solves the problem of dynamic human-machine function allocation in the cockpit of armored vehicles, and establishes a relatively accurate human-machine system performance model. This addresses the issues of when to trigger functions and how to adjust them after triggering, demonstrating a certain degree of innovation. Furthermore, the proposed method is universal and provides new research ideas for human-machine function allocation in other fields. The above description is merely a preferred embodiment of this invention and is not intended to limit the invention. Any modifications, equivalent substitutions, or improvements made within the principles of this invention should be included within the scope of protection of this invention.

Claims

1. A method for allocating human-machine functions in the cockpit of an armored vehicle based on a dynamic Bayesian network, characterized in that, include: The performance of human-machine systems is divided into human performance and machine performance, and human performance is further broken down into accuracy performance and urgency performance. To obtain metrics that affect human and machine performance; Based on the metrics of accuracy, urgency, and machine performance, as well as human and machine performance, a dynamic Bayesian network model structure is established, and dynamic nodes are identified. The fuzzy comprehensive evaluation method is used to confirm the prior probability of the root node and the conditional probability of other nodes except the root node, and to confirm the transition probability of dynamic nodes. The structure of the dynamic Bayesian network model is quantified to obtain the dynamic Bayesian network model. Predicted values ​​for accuracy performance, urgency performance, and machine performance are obtained based on the dynamic Bayesian network model; The predicted values ​​of human performance are obtained based on the predicted values ​​of accuracy performance and urgency performance; Based on the predicted values ​​of human performance and machine performance, obtain the predicted values ​​of human-machine system performance. Set a performance threshold for the human-machine system. When the predicted performance value of the human-machine system is lower than the threshold, the automation level will be adjusted. Based on the performance of the adjusted human-machine system, compare it with the performance of the original human-machine system to confirm the final adjustment plan; The process of establishing the dynamic Bayesian network model structure includes: establishing a fault tree model based on the indicators of accuracy performance, urgency performance, machine performance, as well as human performance and machine performance, and then transforming the established fault tree model into a dynamic Bayesian network model structure. The transition probability of the dynamic node is obtained based on the joint probability of the child node set and the parent node set; The predicted values ​​for human performance are obtained by following these steps: Obtain historical data on various tasks, and establish weights for accuracy and urgency performance for each task scenario using the fuzzy comprehensive evaluation method; based on the weights of accuracy and urgency performance for each task scenario and the predicted values ​​of accuracy and urgency performance, obtain the predicted value of human performance for each task scenario. The predicted values ​​for accuracy performance, urgency performance, and machine performance are obtained by inputting node evidence of the root node over a period of time into a dynamic Bayesian network model. The indicators of human performance include: task load, crew capability, task complexity, task workload, task difficulty, time pressure, mental fatigue, working hours, trust level, and task performance.

2. The method for allocating human-machine functions in the cockpit of an armored vehicle based on a dynamic Bayesian network according to claim 1, characterized in that, The predicted performance value of the human-machine system is obtained according to the following steps: Based on the current level of automation and the performance of humans and machines in various types of tasks, obtain the weights of human performance and machine performance; Based on the weights of human performance and machine performance, as well as the predicted values ​​of human performance and machine performance, the predicted values ​​of human performance and machine performance are obtained.

3. The method for allocating human-machine functions in the cockpit of an armored vehicle based on a dynamic Bayesian network according to claim 1, characterized in that, When the level falls below the threshold, the automation level is adjusted, including: when the improvement in human performance is not significant after adjustment, sensitivity analysis is performed to confirm the optimization priority of each node, and the decision-maker determines the optimization plan.

4. The method for allocating human-machine functions in the cockpit of an armored vehicle based on a dynamic Bayesian network according to claim 1, characterized in that, The prior probability of the root node is obtained by using triangular fuzzy numbers to perform a weighted average of expert evaluations and normalizing the result.

5. The method for allocating human-machine functions in the cockpit of an armored vehicle based on a dynamic Bayesian network according to claim 1, characterized in that, The conditional probabilities of nodes other than the root node are obtained by using triangular fuzzy numbers to perform a weighted average of expert evaluations and normalizing the results.

6. The method for allocating human-machine functions in the cockpit of an armored vehicle based on a dynamic Bayesian network according to claim 1, characterized in that, The indicators affecting human and machine performance are obtained from three aspects: human, machine, and environment; human performance is decomposed into accuracy performance and urgency performance based on task attribute requirements.

Citation Information

Patent Citations

  • Aircraft cockpit man-machine function distribution method based on load balancing

    CN111767611A

  • Ground attack unmanned aerial vehicle autonomous ability evaluation method based on Bayesian network

    CN114444201A