An Investigation of the Negative Selection Algorithm for Fault Detection in Refrigeration Systems

Abstract

Abstract. Supermarkets lose millions of pounds every year through lost trading and stock wastage caused by the failure of refrigerated cabinets. Therefore, a huge commercial market exists for artificially intelligent systems which are able to detect the early symptoms of faults. Previous work in this vein, using real-world data and now in the throes of being deployed commercially, has employed evolved neural networks to predict volumes of temperature and other alarms emerging from refrigeration system controllers, and also to predict likely refrigerant gas loss from such alarm patterns. In this work we turn to the use of in-cabinet temperature data which has recently become available, and the aim is to predict refrigeration system faults from the pattern of in-cabinet temperature over time. We argue that artificial immune system inspired technologies are particularly appropriate for this task. The negative selection algorithm is therefore investigated as a tool to detect anomalous patterns in temperature data. Using a simple AIS system in preliminary experiments to assess feasibility, we compare the performance of a simple matching rule based on Euclidean distance and one using traditional r contiguous bits matching. We also investigate a 'differential' encoding scheme which reduces the size of pattern space while retaining what we feel to be the essential elements of patterns in this application. We find that feasibility for AIS in this application is proven, with particularly good selfdetection rates, and a fault-detection rate adequate as a springboard for further development. The best AIS configuration of those examined here seems to be that which uses the novel differential encoding in conjunction with the r-bits matching rule. The differential encoding scheme is simple, yet seems powerfully suited to such time series pattern analysis tasks.

Immunity-based evolutionary algorithm for optimal global container repositioning in liner shipping

“…Recently, an increased interest has been found in formalizing and adapting the theories and underlying mechanisms of the natural immune system to solve complex engineering problems (Luh et al 2004). Immunological theories have been applied to produce solutions for these problems, including pattern recognition (Garain et al 2006;White and Garrett 2003;Nicosia et al 2001), fault detection (Guzella et al 2007;Taylor and Corne 2003), scheduling (Swiecicka et al 2006;Coello Coello et al 2003), and optimization (Omkar et al 2008;Chen and Mahfouf 2006;Coello Coello and Cortes 2002;Nicosia and Cutello 2002).…”

Intrusion Detection System using Artificial Immune Systems: A Case Study

“…A simple rule is used to compare bits in two such strings and decide whether a match has occurred. Such a match is equivalent to a match between lymphocyte and antigen [8].…”

A hybrid immune model for unsupervised structural damage pattern recognition

“…The novel characteristics of the immune system have inspired the development of artificial immune systems for various applications (Dasgupta, 2006;Hart & Timmis, 2008). The major application areas include data mining (Freitas & Timmis, 2007), pattern recognition (Watkins, Timmis, & Boggess, 2004;Zhong, Zhang, Gong, & Li, 2007), fault diagnosis (Dasgupta, KrishnaKumar, Wong, & Berry, 2004;Taylor & Corne, 2003), and medical classification problems (Polat, Gunes, & Tosun, 2006).…”