Fuzzy entropy used for predictive analytics

Abstract

Process interruptions in (very) large production systems are difficult to deal with. Modern processes are highly automated; data is collected with sensor technology that forms a big data context and offers challenges to identify coming failures from the very large sets of data. The sensors collect huge amounts of data but the failure events are few and infrequent and hard to find (and even harder to predict). In this article, our goal is to develop models for predictive maintenance in a big data environment. The purpose of feature selection in the context of predictive maintenance is to identify a small set of process diagnostics that are sufficient to predict future failures. We apply interval-valued fuzzy sets and various entropy measures defined on them to perform feature selection on process diagnostics. We show how these models can be utilized as the basis of decision support systems in process industries to aid predictive maintenance.