Fault detection and identification spanning multiple processes by integrating PCA with neural network

•A comprehensive model is proposed for quick and accurate fault detection and identification in health monitoring.•A principal component analysis scheme is designed to reduce dimensionality of data collected by COSMED K4b2 and improve efficiency in data transmission.•A new method which combines the K-means clustering and artificial neural networks is developed to classify data and accurately detect and isolate faults.•The mechanism provides the feasibility and increases the efficiency of the fault detection and identification for a system with multiple processes.

This paper proposes an effective fault detection and identification method for systems which perform in multiple processes. One such type of system investigated in this paper is COSMED K4b2. K4b2 is a standard portable electrical device designed to test pulmonary functions in various applications, such as athlete training, sports medicine and health monitoring. However, its actual sensor outputs and received data may be disturbed by Electromagnetic Interference (EMI), body artifacts, and device malfunctions/faults, which might cause misinterpretations of activities or statuses to people being monitored. Although some research is reported to detect faults in specific steady state, normal approach may yield false alarms in multi-processes applications. In this paper, a novel and comprehensive method, which merges statistical analysis and intelligent computational model, is proposed to detect and identify faults of K4b2 during exercise monitoring. Firstly the principal component analysis (PCA) is utilized to acquire main features of measured data and then K-means is combined to cluster various processes for abnormalities detection. When faults are detected, a back propagation (BP) neural network is constructed to identify and isolate faults. The effectiveness and feasibility of the proposed model method is finally verified with experimental data.

Graphical abstractFigure optionsDownload full-size imageDownload as PowerPoint slide