数据驱动与解析模型融合的制冷系统故障诊断方法及鲁棒性研究

Applying the fault diagnosis techniques to refrigeration systems is beneficial to eliminate faults, to enhance their energy efficiency, and to achieve the energy conservation. But the feasibility of diagnostic cost, the limitations of the individual methods, and the diagnostic robustness under incomplete information are the key issues of restricting the field implementation of fault diagnosis techniques for refrigeration systems. From the perspective of field applications, the new theories and new methods about fault diagnosis for refrigeration systems are explored in this project, in order to obtain an optimal diagnostic performance by using the lowest input cost of sensors. The research contents are as follows: mechanism models about faults are researched, and the sensitivity between faults and features are analyzed, and the feature extraction are researched to determine the minimum quantity of sensors based on mutual information and to optimally select the types of sensors based on optimization algorithm, thus to achieve low-cost fault diagnosis; Considering the complementarity between the individual methods, based on probability graph theory, the fusion mechanisms about the data-driven and analytical models are explored, and the hybrid diagnostic methods are developed to make their respective advantages complementary to each other and to overcome the limitations of individual methods used separately; Based on the theories of information fusion and missing data estimation, an enhanced strategy of diagnostic robustness under incomplete information is proposed. These theoretical results will be verified by the experiments. The fault diagnosis theories for refrigeration systems will be enriched by this project, and a set of new theories and new methods will be developed to address the challenges of restricting the field implementation of fault diagnosis techniques for refrigeration systems.

对制冷系统实施故障诊断技术，有利于排除故障、保持系统高效运行和实现节能。但诊断成本的可行性、单类方法的局限性和非完整信息下诊断的鲁棒性是制约制冷系统故障诊断技术现场应用的关键问题。本项目以最低成本的传感器投入获得最优的诊断性能为出发点，拟从现场应用视角探索制冷系统故障诊断新理论与新方法，研究内容包括：建立故障机理模型，进行故障特征敏感性分析，开展运用互信息技术确定传感器最少数量以及优化算法优选传感器种类的关键特征提取研究，实现故障诊断低成本化；根据单类诊断方法的优势互补原则，运用概率图理论，探索数据驱动与解析模型的融合机制，开展复合诊断方法研究，打破单类方法的局限性；基于信息融合与缺失数据修复理论，提出非完整信息下诊断的鲁棒性增强策略。同时，开展实验研究，验证理论成果。本项目将丰富制冷系统故障诊断理论，为突破制冷系统故障诊断技术现场应用难题提供一套有效、实用的新理论和新方法。

对制冷系统实施故障诊断技术，有利于排除故障、保持系统高效运行和实现节能。本项目尝试从现场应用角度出发，围绕诊断成本的可行性、单类方法的局限性和非完整信息下诊断的鲁棒性三个关键问题开展研究工作，致力于探索制冷系统故障诊断新理论与新方法，推动这一技术在现场大规模应用。形成如下创新成果：第一，针对制冷系统典型故障，提出了一种基于互信息和灰色系统理论融合的特征选择方法，选择少量获取成本低的特征表征故障，达到以最低成本的传感器投入获得最优的故障检测与诊断性能，从而节省诊断技术应用成本。同时，在仅使用现场普遍可获得的特征参数表征机组健康状态的前提下，比较分析了典型数据驱动型方法对制冷系统典型故障的诊断性能，结果可指导现场故障诊断方法的应用；第二，提出了一种基于融入距离拒绝机制贝叶斯网络的故障检测方法，通过将故障检测转化成一类划分问题，有效克服了需要大量故障数据训练模型的难题；第三，从方法论角度，给出了将基于模型与基于数据两类方法融合在一个诊断体系中的通用流程，并提出了一种二者相融合的复合诊断方法，实现了彼此优势互补，有效克服了基于单类方法的主要局限性；第四，提出了两种故障检测性能增强策略。一种是将主元分析法与贝叶斯网络相结合，用以增强基于贝叶斯网络方法的故障检测性能。另一种是将期望最大算法和贝叶斯网络进行结合，目的是增强面向缺失数据的故障检测性能；第五，本项目针对制冷系统关键组成部件，提供了一种模型构建简单、现场应用成本低、便于推广应用、方便快捷的性能劣化诊断方法。具体包括一种基于简化物理模型的制冷机组换热器性能劣化诊断方法和一种制冷机组压缩机性能劣化诊断方法。最后，基于已取得的研究成果，在研究计划内容之外，开发了1套冷水机组故障诊断云平台，实现了关键运行参数的实时动态监测、故障在线检测与诊断，以及模型自学习等功能。.本项目很好地完成了既定目标，研究成果丰富了制冷系统故障诊断理论，为促进制冷系统故障诊断技术现场应用具有重要意义，同时为实现国家“3060”双碳目标贡献力量。