Unsupervised Fault Detection of Refrigeration Containers using a Mahalanobis Inverse Moment Matrix Polynomial

摘要

Refrigeration containers experiences ambient temperatures ranging from — 25° C to 40° C and humidities that fluctuate just as much. Furthermore the maintenance engineers despite doing their best, only apply hot-fixes to the systems when they’re serviced. This work addresses the idea of detecting errors on the refrigeration system, by using all available sensory input. The data is then applied to construct an estimator of the sample-distribution, which in turn can be used to determine if a sample can be classified as a failure or not. The approach builds on that of Lasserre and Pauwels (2016), but the extension out-performs the original, across all log-files the system is tested on. In conclusion, the method derived throughout this paper, is indeed viable for fault detection on refrigeration containers.