基于多准则场景缩减的“零停机”设备状态预测与维护方法研究

The unscheduled downtime of any equipment in operation may cause enormous economic losses and even cascading disasters. To improve continuity and reliability of equipments and eliminate unscheduled downtime, massive monitoring data under uncertainties need to be processed for condition-based prediction and then condition-based maintenance. .Focusing on multi-criteria scenario reduction, this project proposes a methodology for condition-based prediction and maintenance of mechanical equipment without unscheduled downtime including three layers, scenario construction, scenario reduction and maintenance decision-making. This project comprises four sections as follows. .① First, on the ground of monitoring data, all values of condition-based features are discretized with variable step-size in temporal and status scales, so as to generate the initial set of all possible scenarios. .② Then, association rules between scenario features and maintenance objectives are extracted using data mining technique, multi-criteria scenario reduction mechanism is developed based on probability measure and association rules, and a reduced scenario set with allowable solution errors and low computational complexity is obtained. .③ Scenario tree is formed based on the selected scenario set by forward selection heuristic. Condition-based maintenance strategy is deduced with balanced stability and reliability. .④Finally, utilizing condition-based prediction and maintenance methodology via multi-criteria scenario reduction techniques, guide bearing of water turbine and roller bearing of rolling mill are both hired to further testify the proposed methodology. .This research will enrich basic theory in health management of equipment, and promote operation and maintenance level of mechanical equipment.

设备运行时，若发生非计划故障停机，会产生较大经济损失，严重时会导致连锁灾难性事故。为保证设备连续可靠运行、实现“零停机”，需分析具有不确定的大监测数据，预测设备状态演变趋势，制定设备维护策略。.为此，提出以多准则场景缩减为核心，包括场景构建、场景缩减、维护决策三层次的“零停机”设备状态预测与维护方法，主要包括：①抽取状态特征，提出基于状态空间和时间尺度的双向变步长最优离散方法，构造涵盖全部不确定情形的场景集；②利用设备可靠性和维护成本生成目标函数，挖掘场景特征与目标函数间的关联规则，研制能保证预测精度和缩减效率、融合概率测度与关联规则两类准则的场景缩减机制；③利用前向选择法生成场景树，基于时序逻辑构造多时段设备维护模型，获得稳定性强、可靠性高的状态维护方案；④将研究成果用于水轮机水导轴承、轧机轧辊轴承等关键设备，展开实例验证。.本研究有利于丰富和发展设备健康管理理论，提高设备运行维护水平。

为保证“零停机”设备的连续可靠运行，需分析具有不确定的大监测数据，预测设备状态演变趋势，制定设备维护策略。.项目组以多准则场景缩减为核心，从场景构建、场景缩减、维护决策三层次，构建了“零停机”设备状态预测与维护策略框架。项目组首先围绕“零停机”设备/关键元件，辨识其外部和内部状态特征量，通过引入扰动参数对不确定型特征向量进行时序演化描述，基于状态空间和时间尺度的双向离散，构造状态预测初始全场景。在此基础上，挖掘状态时序演化场景特征和维护优化目标函数值之间的关联模式和规则，凝练出了电价时序特征与发电机组维修阈值之间三大类关联模式，轴承劣化演变时序特征与维修成本率之间四大类关联规则。然后，研制了融合概率测度和关联规则的多准则场景缩减方法，提出了基于目标函数值的场景缩减方法；然后，针对维修阈值为区间值的情形，研制了一种围绕目标函数值、最优维修阈值上下边界值，基于多维输出性能的场景缩减方法。在此基础上，研制了基于多准则场景缩减的状态维护决策，提出了基于劣化场景分析的多元件系统状态维护模型，研制了一种双重近似拟合的多元件系统寿命分布缩减方法，便于快速可靠地生成维护方案，提高设备运行和维护水平，实现“零停机”。本研究有利于丰富和发展设备健康管理理论。.上述研究成果发表在《Maintenance and Reliability》、《Mechanika》、《International Journal of Reliability Quality & Safety Engineering》、《机械设计与制造》等刊物及会议发表论文6篇，其中SCI/EI检索5篇；并参与制修订国家标准《GB/T 35023-2018 液压元件可靠性评估方法 》；申请软件著作权1项；培养了硕士研究生3名，博士研究生1名。