基于动力学行为的船舶电力系统设备潜隐故障诊断及其传播预测方法研究

The health management technology and fault prediction diagnosis method of marine electrical power system are the problems that need to be solved urgently to improve the ship technology. Yet the research of the fault prediction diagnosis is not in-depth. Aimed to the diagnosis and prediction for potential faults of shipboard power equipments, research works of this project are validated by using theory analysis, software simulation, platform experiment and entity ship measured study. Firstly, the effective samples of potential faults are obtained by using the dynamic behavior of the working state evolution progress of power equipments. Applying the combination of Gaussian window function and wavelet transform, time-frequency analysis of fault signals of power equipment is carried out to extract weak features of potential faults. Considering the characteristics of multi-operating conditions and varying loads, a potential fault classifier based on the similarity analysis of fault time-frequency feature signals was constructed. On the above basis, we study the directed acyclic graph-based model for the interaction between potential/failure faults in power equipments, it can prognosis the follow-up power equipment sequence of potential or failure faults by combining with the above-diagnosis method of power equipment potential faults. The finding of this project will provide a key scientific basis for the ship to be maintained under good berthing conditions or working conditions.

船舶电力系统健康管理技术和故障预测诊断方法是提高船舶技术亟待解决的难题，然而故障预测诊断方法研究尚不深入。本项目针对船舶电力系统设备潜隐故障预测诊断，拟采用理论分析、软件仿真、平台试验和实体船舶实测验证开展研究。首先利用电力设备工作状态演变过程的动力学行为特征获取有效的潜隐故障样本；应用高斯函数和小波变换相结合的变换方法对电力设备潜隐故障信号进行时频分析，提取潜隐故障微弱的特征；考虑多工况、变负载等特点，构造基于故障时频特征信号相似度分析的潜隐故障分类器。在此基础上，研究电力设备潜隐故障间、与失效故障间交互性的有向无环图形式模型，然后与上述基于动力学行为的电力设备潜隐诊断方法有机结合来预测潜隐故障诱发船舶电力系统后续可能发生潜隐故障或失效故障的电力设备序列。本项目的研究结果将为船舶得以在良好的停泊条件或工作环境下进行维护提供关键的科学依据。

潜隐故障是介于安全和故障之间的工作状态，若能检测出处于失效之前潜隐故障状态的电力设备，进而在有利的工作条件中实现故障潜隐电力设备的修复，就可以避免船舶电力系统出现破坏性故障。本项目针对船舶电力系统设备潜隐故障预测诊断，研究电力设备潜隐故障产生、发展以及向失效故障演化的分岔动力学行为，提出基于动力学行为的船舶电力系统设备潜隐故障样本数据辨识方法；根据电力设备潜隐故障特征信号微弱、传感器采集信号与潜隐故障类型难以对应等难题，提出基于长短期记忆的电力设备故障预测诊断方法，实现了“端对端”模式诊断潜隐故障以及预测其发展；构建电力设备潜隐故障间、与失效故障间交互性的有向无环图形式模型，实现了对潜隐故障诱发船舶电力系统后续可能发生潜隐故障或失效故障的电力设备序列预测。在本项目资助下，培养了7名硕士，发表学术论文23篇，出版专著2部，获授权国家发明专利9项。本项目的研究结果将为船舶得以在良好的停泊条件或工作环境下进行维护提供关键的科学依据。