复杂机电系统贝叶斯网络性能模型结构学习优化

As mechanical electronic hydraulic integrated complex system, diesel engine needs work in cold, hot, altiplano, rainy or drought environment, its fault has the characteristic of random ,uncertainty, coupling and hierarchy because of the complexity of structure and motion ,incomplete of information, which poses a challenge to the traditional fault diagnosis methods..This study aims at exploring the theories and methods of diesel engine fault diagnosis based on varying weight rapid clustering Bayesian networks, exploring the correlation of fault samples , diagnosis rule and failure symptom in extreme environments. The varying weight rapid clustering algorithm will be improved to simplify structure learning of fault diagnosis Bayesian networks although the fault diagnosis in multiple extreme environments..The study mainly contains establishing of fault samples database and diagnosis rule base, changing the model and strategy weight tuning algorithm, simplifying the fault diagnosis networks, reducing calculation and enhancing the accuracy of fault diagnosis..Research results will improve diesel engines maintains, ensure the equipments in good condition, craftwork technological capabilities, gain maximum military effectiveness of weaponry, and it will serve as a theoretical underpinning for fault diagnostics and prognostics of complex mechatronic systems with uncertain and incomplete fault information.Research results will afford effective support for complex mechatronic systems fault diagnosis in theory, technology and engineering application, it is also a solid step toward fault diagnosis based on BNs in theoretical and technological innovation.

柴油发动机作为机、电、液一体的复杂机电系统，需要工作在高温、高寒高海拔、潮湿多雨或干旱等恶劣环境，加上发动机结构及运动复杂性导致其故障具有多层次性、耦合性和随机性及故障信息不完整性等特点，这给传统故障诊断手段提出了挑战。.研究旨在探索复杂运动控制环境下柴油发动机故障传播机理，建立复杂参数环境中以电控系统数据为基础的性能分析和故障诊断模型，深入研究基于变权值快速分簇的贝叶斯网络结构学习优化算法。研究内容包括：对实际环境中获取的发动机数据统计分析，建立故障样本库和诊断规则库；运用统计与推理方法实现故障样本、诊断规则和故障表征之间的关系规律研究，从中总结出故障传播机理；调整贝叶斯网络的分簇权值优化模式和策略，简化分析诊断模型，通过贝叶斯网络结构学习优化，提发动机故障诊断系统的诊断效率和精度。研究成果将丰富军民用复杂机电系统故障诊断理论，对巩固国防和提高工农生产具有较高的理论和工程意义。

针对复杂机电系统贝叶斯网络性能模型结构学习优化研究，以变压器系统性能分析为背景，以油中溶解特征气体含量、电量、油温、振动、微水等监测数据，构建贝叶斯推理网络分析其运行状态，预测内部是否发生潜在故障的概率。变压器工作在恶劣环境和自身工作过程的复杂性导致其故障具有多层次性、耦合性和随机性及故障信息不完整性等特点，探索不同环境下变压器故障传播机理，并建立复杂环境中变压器性能分析模型、贝叶斯网络结构学习优化算法。通过优化节点间的权值调整策略，构建优化变压器故障的贝叶斯故障诊断网络优化模型。. 对实际环境中获取的变压器数据统计分析，建立初步故障样本库和诊断规则库；运用专家知识、概率统计与推理的方法实现故障样本、诊断规则和故障表征之间的规律研究，从中总结出故障传播机理。优化BNs网络的结构学习算法，简化性能分析、故障诊断模型复杂程度，提高推理的准确率；探索了动态贝叶斯网络的性能推理模型构建与优化。完成算法仿真的基础上，完成性能分析推理、故障诊断的系统软件设计并测试其准确性。研究成果将丰富复杂机电系统性能分析与故障诊断，对各类复杂系统开展故障预测、性能推理具有一定的理论和工程应用意义。