基于主元分析的传感器故障检测盲区预测

Blind zone prediction for PCA-based sensor fault detection

传感器故障极易发生但不易察觉,其测量数据是制冷空调系统安全运行和优化节能的必要条件。分析了以Q统计量为故障检测边界的基于主元分析的传感器故障检测流程,建立了一种预测传感器故障检测能力的指标——故障检测盲区,用于预测训练数据集中各个传感器的故障检测能力,从而分析、评价和优化建模用训练数据的质量。采用工程数据、实验数据分别开展算法验证,结果表明故障检测盲区能有效预测选定数据集的相关传感器的故障检测结果。

Sensor faults occur unavoidably but cannot be detected easily. Sensor measurement is fundamental for safe operation and optimal energy conservation of refrigeration and air-conditioning systems. After analyzed procedure for sensor fault detection based on the principal component analysis（PCA） with Q-statistics as fault detection boundary, blind zone prediction was established as an index to estimate sensor fault detectability. The index was used to assess fault detectability of each sensor in training dataset and to analyze, evaluate, and optimize dataset quality of training model. Results of analyzing in-site and laboratory datasets of water-cooled chillers at different introduced fault levels show that the blind zone can effectively predict fault detection outcome for sensors by selected datasets.