面向潜伏期故障的数模联动并行补偿诊断方法研究

The latent period of fault is the key stage to achieve early fault prediction and prevent severe accident. Therefore, the study of fault diagnosis on latent fault has important theoretical significance and momentous engineering value. In order to address the problem that the data imbalance and model mismatch influence the accuracy of latent fault diagnosis results, the feasible fault diagnosis theory and method centered on the strategy of Data Model Cooperation and Parallel Compensation (DMC-PC) are deeply researched in the project. The main research contents include: identification-oriented data preprocessing of imbalanced sample; adaptive fault diagnosis based on recursive subspace model; Bayesian Network (BN) fusion for DMC-PC based fault diagnosis. The remarkable characteristics of the research are as follows. 1. The DMC-PC based diagnostic strategy is proposed. The strategy not only avoids the result distortion caused by serial diagnostic mechanism, but also solves the problems (data imbalance and the lack of model self-improvement ability) during the application of single diagnostic method in latent fault diagnosis. Such proposed method can improve the accuracy of fault diagnosis results. 2. The updating mechanism of event-triggered receding-horizon is proposed, which can accurately realize the dynamic description of system objective status and improve the capability of early fault forecast. 3. The BN-based PC approach is used to deduce the probability confidence of fault diagnosis results. Such formulation solves the quantitative evaluation problem of latent fault. Additionally, it contributes to fault trend forewarning and timely maintenance schedule establishment.

故障潜伏期是故障早期预报和防范严重事故的关键阶段，因此面向潜伏期的故障诊断研究具有重要的理论意义和很高的工程价值。针对数据不平衡和模型不匹配影响潜伏期故障诊断结果准确性的问题，研究以数模联动并行补偿为核心的潜伏期故障诊断理论与方法。主要研究内容包括：面向辨识导向的不平衡样本数据预处理；基于递归子空间模型的自适应故障诊断；面向数模联动并行补偿诊断的贝叶斯网络演绎推理。研究的显著特点是：1.提出数模联动并行补偿诊断策略，避免了串行诊断机制因数据不平衡而导致的结果失真，解决了单一方法在潜伏期故障诊断中遇到的数据不平衡和模型不能自我完善的问题，提高了故障诊断的准确性；2.提出基于事件触发滚动时域的更新机制，实现了系统客观状态的准确动态描述，提高了故障的早期预报能力；3.采用基于贝叶斯网络的演绎推理方法，诊断结果以概率置信度描述，解决了潜伏期故障难以量化评价的问题，有助于故障趋势预警、制定维修计划。

故障潜伏期是故障早期预报和防范严重事故的关键阶段，面向潜伏期的故障诊断研究具有重要的理论意义和很高的工程价值。本项目研究从典型工业失效过程的历史数据分析入手，总结归纳了大部分工业装备失效数据中工况样本的三个不平衡特性，即样本标记不平衡、样本信息不平衡和样本类别不平衡。针对数据不平衡和模型不匹配影响潜伏期故障诊断结果准确性的问题，研究了以数模联动并行补偿为核心的潜伏期故障诊断理论与方法。提出了面向具有数据样本不平衡特性的工业过程故障诊断的特征提取、数据重构和诊断决策方法，主要算法成果包括：面向辨识导向的不平衡样本数据预处理算法；基于递归子空间模型的自适应故障诊断算法；面向数模联动并行补偿诊断的贝叶斯网络演绎推理算法，并设计了监测数据分析软件系统。研究成果的显著优点在于：提出的数模联动并行补偿诊断策略，避免了串行诊断机制因数据不平衡而导致的结果失真，解决了单一方法在潜伏期故障诊断中遇到的数据不平衡和模型不能自我完善的问题，提高了故障诊断的准确性；提出的基于事件触发滚动时域的更新机制，实现了系统客观状态的准确动态描述，提高了故障的早期预报能力；所采用的基于贝叶斯网络的演绎推理方法，诊断结果以概率置信度描述，解决了潜伏期故障难以量化评价的问题，有助于故障趋势预警。研究成果有利于实现复杂工业装备早期故障数据的特征提取与安全性评估决策，为生产过程的高效稳定运行提供了一种基础理论和关键技术方法。