Underwater robot health management system based on Bayesian network
By using a health management system based on Bayesian networks, which combines underwater robot node information and a lifespan model, the problem of insufficient utilization of multi-source data in existing technologies is solved. This enables accurate assessment of the health status of underwater robots and fault prediction, thereby improving the system's real-time monitoring and prediction capabilities.
Patent Information
- Application Number
- CN202111367676.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-11-18
- Publication Date
- 2026-11-17
- Estimated Expiration
- 2041-11-18
AI Technical Summary
Existing technologies have failed to fully leverage multi-source data and similar sample information in the health management of underwater robots, making it difficult to accurately assess the health status of the system and predict faults, especially under the influence of factors such as alternating environmental temperatures.
A health management system based on Bayesian networks is adopted. By constructing Bayesian networks and lifetime models, and combining underwater robot node information and lifetime information, the system can achieve quantitative assessment and prediction of node health status and lifetime indicators.
It enables accurate assessment of the health status of underwater robots and fault prediction, reduces dependence on modeling errors, and improves the system's real-time monitoring and prediction capabilities.
Smart Images

Figure CN116136926B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of fault diagnosis and health management, specifically to a health management system for underwater robots based on Bayesian networks. Background Technology
[0002] Fault diagnosis is a relatively new discipline that has developed over the past 40 years. It mainly utilizes various detection methods to assess the operating status and abnormal conditions of a system, and to diagnose the type, location, and cause of the fault. It is an interdisciplinary field that integrates multiple sciences to meet practical engineering needs. Its theoretical foundation lies in modern control theory, computer technology, mathematical statistics, signal processing, pattern recognition, artificial intelligence, artificial neural networks, and other relevant applied disciplines.
[0003] In practical applications such as deep-sea salvage, space exploration, rubble rescue, battlefield bomb disposal, underground mining, nuclear waste cleanup, and polar scientific expeditions, human intervention on-site often requires substantial financial and material resources, or the conditions for human entry or operation may simply be unavailable. In these situations, robots are undoubtedly the best choice to replace humans in performing dangerous tasks. Robots can monitor and control their own behavior in real time, and can automatically handle system malfunctions promptly. In other words, if robot malfunctions can be detected and predicted in a timely manner, and if a malfunctioning robot can be restored to normal operation, its performance will be significantly improved.
[0004] Currently, the development of technologies such as machine learning has been applied to practical engineering, greatly improving the intelligence level of fault diagnosis methods. Given its strong practicality, intelligent diagnosis technology has become a global research hotspot in the field of fault diagnosis.
[0005] Existing, relatively mature data-driven health management methods in other fields have not fully utilized multi-source data and similar sample information in practical applications of underwater robot control. Existing research indicates that, on the one hand, alternating environmental temperature is a significant factor causing performance degradation of key components in underwater robot systems; telemetry data can accurately reflect the real-time operating status and performance changes of the system. On the other hand, data from the development, testing, and experimentation of control systems, as well as the lifespan information of similar systems and decommissioned systems, can provide crucial information for studying the remaining lifespan of the current system. Current methods have not fully explored the effective information implicit in these multi-source data, and the scarcity of statistical control system lifespan data makes it difficult to directly apply existing methods to satellite control systems. Therefore, integrating health management systems with real-time data has become a research focus for underwater robots. This invention combines a health management system with real-time data to achieve quantitative analysis and expression of different health levels, using a health assessment method to represent these levels in the form of scores. Summary of the Invention
[0006] The purpose of this invention is to provide a health management system based on underwater robots to overcome the above-mentioned defects.
[0007] The technical solution adopted by the present invention to achieve the above objectives is as follows:
[0008] A Bayesian network-based underwater robot health management system includes:
[0009] The health assessment submodule is used to obtain Bayesian information files and underwater robot node information from the database, construct a Bayesian network based on the Bayesian information files, input the underwater robot node information into the Bayesian network, obtain the probability of the node's health status, and then obtain health indicators to describe the health status of the node.
[0010] The fault prediction submodule is used to obtain underwater robot node lifespan information files, construct a lifespan model, and use the underwater robot node lifespan information files as input to the lifespan model to obtain lifespan indicators that describe the aging degree of the node.
[0011] Each component of the underwater robot is treated as a node in a Bayesian network, and the Bayesian information file includes the node's reliability and influence information.
[0012] The node information includes which child nodes the node contains, as well as the weight of the child nodes, the update status and scoring method, the usage time, and the number of times they are used.
[0013] The underwater robot node lifetime information file includes: lifetime decay function parameters and lifetime decay function correlation characteristics.
[0014] The lifetime model is a lifetime decay function L.
[0015]
[0016] Where a0, λ, and a1 are all constants, λ and a0 define the decay rate of the lifetime index, a1 defines the initial decay state of the lifetime index, and L start This is the initial value for the lifespan.
[0017] After obtaining the health and lifespan indicators of a certain component of the underwater robot, a step-by-step detection method is used to obtain the overall health and lifespan indicators of the underwater robot.
[0018] The step-by-step detection method is specifically as follows:
[0019] The health and lifespan indicators of a certain component of an underwater robot are used as inputs to the health assessment submodule and the fault prediction submodule, respectively, to obtain the health and lifespan indicators of the subsystem to which the component belongs.
[0020] The health and lifespan indicators of the subsystem are used as inputs to the health assessment submodule and the fault prediction submodule, respectively, to obtain the health and lifespan indicators of the underwater robot.
[0021] A Bayesian network-based underwater robot health management method includes the following steps:
[0022] The health assessment submodule obtains Bayesian information files and underwater robot node information from the database, constructs a Bayesian network based on the Bayesian information files, inputs the underwater robot node information into the Bayesian network, and obtains the health indicators of the node.
[0023] The fault prediction submodule obtains the underwater robot node lifespan information file, constructs a lifespan model, and uses the underwater robot node lifespan information file as input to the lifespan model to obtain the lifespan index of the node.
[0024] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the underwater robot health management method based on Bayesian networks.
[0025] The present invention has the following beneficial effects and advantages:
[0026] This invention utilizes Bayesian networks for probabilistic reasoning, which has a solid theoretical foundation. Furthermore, the Naive Bayes inference algorithm is mature, yields accurate results, and has low requirements for prior probability. It eliminates the need for a precise mathematical model of the system, avoiding the influence of modeling errors and related uncertainties. It also allows for full utilization of database data, better meeting the needs of practical engineering applications. Attached Figure Description
[0027] Figure 1 This is a diagram of the health assessment module of the present invention;
[0028] Figure 2 This is a diagram of the fault prediction module of the present invention;
[0029] Figure 3 This is a system configuration diagram of the present invention;
[0030] Figure 4 This is a schematic diagram of the health management module hierarchy of the present invention;
[0031] Figure 5 This is a schematic diagram illustrating the node computer functions and outputs of the present invention;
[0032] Figure 6This is a schematic diagram illustrating the functions and outputs of the management decision-making computer of the present invention;
[0033] Figure 7 This is the logic flowchart of the present invention. Detailed Implementation
[0034] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments.
[0035] A health management system based on underwater robots, the system comprising two modules: health assessment and fault prediction, including the following:
[0036] (1) The parameter file is read by the file reading submodule to obtain the parameters for constructing the Bayesian network. The Bayesian network is constructed by the health assessment submodule based on the Bayesian network parameters. The reliability information and node influence information of the nodes are read from the file. Each component is a node in the Bayesian network. The influence of the node on other nodes is realized through node patterns. Each pattern has a corresponding reliability value, which serves as the probability value in the Bayesian network.
[0037] (2) The health assessment submodule reads the patterns of the components and infers the health status of the components based on the patterns. For the formed Bayesian network, an inference engine is built to infer based on the obtained information;
[0038] (3) The health assessment submodule continuously obtains the pattern information of the component, passes the pattern information as input to the Bayesian network, obtains the current pattern of the component subsystem through reasoning, obtains the health index of the component through calculation, and outputs the health index and component health information.
[0039] (4) The fault prediction submodule first reads the life information file, obtains the parameters required by the life model, and establishes the component life model.
[0040] (5) Changes in component patterns are input information for fault prediction. The fault prediction submodule calculates the component's life index from the component pattern; the subsystem's life index is calculated based on the life weighting parameters and the component's life index.
[0041] The overall design of the system health management software is as follows:
[0042] The health management system assesses the health status of the system by inputting the status information of components and subsystems, and predicts the failures of components, subsystems and the system based on life models and failure information.
[0043] The system comprises two modules: health assessment and fault prediction. The health assessment module monitors the health status of each level of the system / subsystem based on data information, achieving health assessments at each level through the construction of inference networks and weighted functions. Based on the lifespan model and extracted fault characteristics, the system / subsystem performs fault prediction and forecasts future aging trends.
[0044] The health assessment and fault prediction submodules run in the secondary node computer and the management decision computer, respectively, to assess and predict faults in components and subsystems.
[0045] Optionally, the system's secondary node computer module is designed as follows:
[0046] For node computers, the function of the health management system is to assess the health status of components and calculate lifespan models. The obtained component-level health information includes: 1. Component-level health indicators; 2. Component-level health information; 3. Component-level lifespan indicators; 4. Component-level lifespan information.
[0047] The health management module requires component-related data for health assessment. The fault assessment submodule obtains component health indicators through a Bayesian network and component failure mode change inference. The fault prediction submodule calculates lifespan indicator values and lifespan information based on the component's current failure mode using an established lifespan model.
[0048] The health assessment and fault prediction submodule directly interacts with components at the node computer layer. This submodule directly processes component-level information and outputs results. On one hand, the node computer directly processes the component's input information, which, after passing through the health assessment and fault prediction module, yields the component's health assessment and fault prediction results. On the other hand, the component-level results become the input information for subsystem-level health assessment and fault prediction in the management decision-making computer. Because a subsystem is composed of components, the subsystem's health and lifespan are obtained through a weighted average of the components.
[0049] The method for the system health assessment module is as follows:
[0050] Naive Bayes, its formula is:
[0051]
[0052] P(B) and the conditional probability P(A|B) are both given by the expert system, and P(A) can be derived from the law of total probability.
[0053] Lifetime decay function:
[0054]
[0055] Where a0, λ, and a1 are all constants, λ and a0 define the decay rate of the lifetime index, a1 defines the initial decay state of the lifetime index, and L start This is the initial value for the lifespan.
[0056] Figure 1 This is a diagram of the health assessment module of the present invention.
[0057] (1) The health assessment submodule reads the patterns of the components and infers the health status of the components based on the patterns. For the formed Bayesian network, an inference engine is built to infer based on the obtained information;
[0058] (2) The health assessment submodule continuously obtains the pattern information of the component, passes the pattern information as input to the Bayesian network, obtains the current pattern of the component subsystem through reasoning, obtains the health index of the component through calculation, and outputs the health index and component health information.
[0059] like Figure 2 This is a diagram of the fault prediction module of the present invention.
[0060] (1) The fault prediction submodule first reads the life information file, obtains the parameters required by the life model, and establishes the component life model.
[0061] (2) Changes in component patterns serve as input information for fault prediction. The fault prediction submodule calculates the component's lifespan index from the component patterns; the subsystem's lifespan model calculates the subsystem's lifespan index based on lifespan weighting parameters and component lifespan indexes.
[0062] like Figure 3 This is a system configuration diagram of the present invention.
[0063] This invention relates to a health management system for underwater robots. The required input information includes preset parameters and input data, resulting in output information. The parameters are stored in JSON files and read by a preprocessing program within the system. This invention employs Bayesian networks and lifetime decay functions. The health assessment module monitors the health status of each level of the system / subsystem based on data information, achieving health assessments at each level through the construction of inference networks and weighted functions. The fault prediction module predicts faults in the system / subsystem based on lifetime models and extracted fault characteristics, forecasting the future aging trend of the system.
[0064] like Figure 4 This is a schematic diagram of the health management module hierarchy of the present invention.
[0065] The health assessment and fault prediction submodule directly interacts with components at the node computer layer. This submodule directly processes component-level information and outputs results. On one hand, the node computer directly processes the component's input information, which, after passing through the status assessment and fault prediction module, yields the component's health assessment and fault prediction results. On the other hand, the component-level results become the input information for the subsystem-level health assessment and fault prediction in the management decision-making computer. Because a subsystem is composed of components, the subsystem's health and lifespan are obtained by weighting the components.
[0066] like Figure 5 This is a schematic diagram illustrating the node computer functions and outputs of the present invention.
[0067] For node computers, the health management module functions as Bayesian network inference and lifetime model calculation. The obtained component-level health information includes: 1. Component-level health indicators; 2. Component-level health information; 3. Component-level lifetime indicators; 4. Component-level lifetime information.
[0068] like Figure 6 This is a schematic diagram illustrating the functions and outputs of the management decision-making computer of the present invention.
[0069] The health management module, within the management decision-making computer's functions, uses component-level health indicators, health information, and key parameters to weight the health indicators of components within a subsystem based on the component's weight, thus obtaining the subsystem's health indicators. The health management module also calculates subsystem-level lifespan indicators through weighted analysis of component-level lifespan information. Simultaneously, it weights the impact of key information on lifespan indicators to obtain the subsystem's output results.
[0070] like Figure 7 This is the flowchart of the invention.
[0071] Includes the following:
[0072] (1) The parameter file is read by the file reading submodule to obtain the parameters for constructing the Bayesian network. The Bayesian network is constructed by the health assessment submodule based on the Bayesian network parameters. The reliability information and node influence information of the nodes are read from the file. Each component is a node in the Bayesian network. The influence of the node on other nodes is realized through node patterns. Each pattern has a corresponding reliability value, which serves as the probability value in the Bayesian network.
[0073] (2) The health assessment submodule reads the patterns of the components and infers the health status of the components based on the patterns. For the formed Bayesian network, an inference engine is built to infer based on the obtained information;
[0074] (3) The health assessment submodule continuously obtains the pattern information of the component, passes the pattern information as input to the Bayesian network, obtains the current pattern of the component subsystem through reasoning, obtains the health index of the component through calculation, and outputs the health index and component health information.
[0075] (4) The fault prediction submodule first reads the life information file, obtains the parameters required by the life model, and establishes the component life model.
[0076] (5) Changes in component patterns are input information for fault prediction. The fault prediction submodule calculates the component's life index from the component pattern; the subsystem's life index is calculated based on the life weighting parameters and the component's life index.
Claims
1. A health management system for underwater robots based on Bayesian networks, characterized in that, include: The health assessment submodule is used to obtain Bayesian information files and underwater robot node information from the database, construct a Bayesian network based on the Bayesian information files, input the underwater robot node information into the Bayesian network, obtain the probability of the node's health status, and then obtain health indicators to describe the health status of the node. The fault prediction submodule is used to obtain underwater robot node lifespan information files, construct a lifespan model, and use the underwater robot node lifespan information files as input to the lifespan model to obtain lifespan indicators that describe the aging degree of the node. The lifetime model is a lifetime decay function. , in , , All are constants. and The rate of decay of the lifetime index was defined. The initial decay state of the lifetime index is defined. This is the initial value for the lifespan.
2. The underwater robot health management system based on Bayesian networks according to claim 1, characterized in that, Each component of the underwater robot is treated as a node in a Bayesian network, and the Bayesian information file includes the node's reliability and influence information.
3. The underwater robot health management system based on Bayesian networks according to claim 1, characterized in that, The node information includes which child nodes the node contains, as well as the weight of the child nodes, the update status and scoring method, the usage time, and the number of times they are used.
4. The underwater robot health management system based on Bayesian networks according to claim 1, characterized in that, The underwater robot node lifetime information file includes: lifetime decay function parameters and lifetime decay function correlation characteristics.
5. The underwater robot health management system based on Bayesian networks according to claim 1, characterized in that, After obtaining the health and lifespan indicators of a certain component of the underwater robot, a step-by-step detection method is used to obtain the overall health and lifespan indicators of the underwater robot.
6. The underwater robot health management system based on Bayesian networks according to claim 5, characterized in that, The step-by-step detection method is specifically as follows: The health assessment and fault prediction submodule directly faces the components at the node computer layer. The submodule directly processes component-level information and outputs results. On the one hand, the node computer directly processes the component's input information, and after passing through the status assessment submodule and the fault prediction submodule, obtains the component's health indicators and life indicators. On the other hand, the component-level results become the input information for subsystem-level health assessment and fault prediction in the management decision computer, which weights the component's health indicators and life indicators to obtain the subsystem's health indicators and life indicators.
7. A Bayesian network-based underwater robot health management method, applied to the Bayesian network-based underwater robot health management system described in claim 1, characterized in that... Includes the following steps: The health assessment submodule obtains Bayesian information files and underwater robot node information from the database, constructs a Bayesian network based on the Bayesian information files, inputs the underwater robot node information into the Bayesian network, and obtains the health indicators of the node. The fault prediction submodule obtains the underwater robot node lifespan information file, constructs a lifespan model, and uses the underwater robot node lifespan information file as input to the lifespan model to obtain the lifespan index of the node.
8. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, implements the underwater robot health management method based on Bayesian networks as described in claim 7.
Citation Information
Patent Citations
Digital-twin-driven fault prognosis method and system for subsea production system of offshore oil
AU2020102863A4
