变风量空调阀门卡死故障检测阈值在线训练策略

An online strategy of threshold training for the rule-based fault detection used in damper and valve stuck faults

依据专家经验选取阈值的专家规则故障检测方法,容易受限于经验,且不能准确地反映空调系统的运行性能特征。设计了基于专家规则的空调故障检测方法,采用大小合适的阈值来判断系统是否发生故障。对调试成功后的空调系统,采用一种基于规则的阈值在线训练策略进行系统性能监控与故障检测。利用局域网在线读取云服务器中无故障运行数据,训练85%、90%、95%等3种不同置信度下的阈值,并以新风阀卡死和冷水阀卡死故障为例进行测试检验。研究结果表明:空气处理机组新风阀卡死和冷水阀卡死的故障检测率均大于80.0%;系统正常运行时的无故障检测率均大于75.0%;训练阈值时,作为外部输入的置信度越大,训练出的阈值越大,导致系统故障检测率较低而无故障检测率较高。

The threshold selected according to expert experience is usually employed to determine the fault of familiar air conditioning systems,but it can rarely reflect the operational performance characteristics of specific or unfamiliar air conditioning systems.The expert-rule-based fault detection method whose thresholds are usually selected based on expert experience,is easily limited by experience and thus cannot accurately reflect the operational performance characteristics of air conditioning systems.The rule-based fault detection method for air conditioning systems was designed,and the appropriate threshold was used to judge whether the system failed.For the air conditioning system after successful commissioning,an online training strategy of threshold training for the rule-based fault detection was adopted.The local area network was employed to acquire the fault-free operational data in the cloud server,and the threshold was trained under three different confidence levels,such as 85%,90%and 95%.The fresh air damper and the cooling coil valve stuck faults were taken as examples to test.The study indicates that the fault detection ratio of the AHU air handling unit damper and valve stuck faults are both more than 80.0%.When the system is normal,the no-fault detection ratio is more than 75.0%.The larger confidence level as an external input during threshold training is,then the lower fault detection ratio is,and thus the higher fault-free detection ratio is.