A weighted k-nearest neighbor process monitoring method based on statistical volume pattern analysis

Abstract Process monitoring technology has developed rapidly in response to the increasing demand for safer and more reliable systems modern process operations. Online plays an important role not only ensuring safety through timely detection of faults but also improving productivity product quality. One outstanding features manufacturing facilities, which are large scale highly complex, is that processes contain numerous variables operating under closed-loop control. Fully exploiting utilizing valuable information these will benefit early accurate fault diagnosis processes, minimizing downtime, plant operational safety, reducing costs. The development great importance ensure reliability, economy complex industrial processes. With continuous collection use data gradually increasing, data-driven multivariate statistical methods have significantly. A weighted k-nearest neighbor method based on SPA proposed problem multimodal properties statistics. When covariance difference between different modes large, characteristics original variable space retained statistic feature space. By introducing weights assigning distances samples, can regulate samples with larger those smaller same scale, overcoming limitations SPA-based detection.