An Empirical Application of Linear Regression Method and FIR Network for Fault Diagnosis in Nonlinear Time Series

Abstract

A fault diagnosis scheme for nonlinear time series recorded in normal and abnormal conditions is described. The fault is first detected from regression lines of the raw time series. Model for the normal condition time series is estimated using a Finite Impulse Response (FIR) neural network. The trained network is then used for filtering of abnormal condition time series. The fault is further confirmed/ analyzed using the regression lines of the predicted normal and inverse-filtered abnormal conditions time series. The described scheme is applied to two fault diagnosis problems using acoustic and vibration data obtained from rotating parts of an automobile and a boring tool, respectively