Prognostics and Health Management of Electronics

6 Machine Learning: Anomaly Detection

Myeongsu Kang

University of Maryland, Center for Advanced Life Cycle Engineering, College Park, MD, USA

A product's health can be defined as the extent of deviation or degradation from its expected typical operating performance, where typical operation refers to the physical or performance‐related conditions expected from the product [1]. Accordingly, it is important to identify deviation from the nominally healthy behavior of the product and detect the onset of the product's potential faults for achieving prognostics and health management (PHM). The aforementioned process is called “anomaly detection” in the field. During the past several decades, numerous anomaly detection techniques have been studied for PHM of electronics, and thus the primary focus of this chapter is to offer a comprehensive overview of the research on anomaly detection and discuss challenges in anomaly detection.

6.1 Introduction

Anomaly detection refers to the problem of identifying patterns in data that do not conform to expected behavior, where the patterns are commonly referred to as anomalies and outliers [2]. In PHM, the importance of anomaly detection is due to the fact that anomalies in data translate to significant information about the product's health status. In general, most anomaly detection methods construct a profile of normal instances, then identify anomalies that do not conform to the normal profile. For example, in [3], the healthy ...