ABSTRACT
If the statistical properties of the noise, together with the way it affects the plant outputs, are known then the fault detection and isolation problem can be formulated in the framework of statistical decision making. Testing the residuals for the purpose of fault detection usually involves testing for the zero mean hypothesis. The purpose of low-pass residual filtering is to improve the signal-to-noise ratio in the residual. Instead of applying separate filters to the residuals, filtering may be part of the original design of the residual generator. The residual generators are so designed that their fault-induced response is enhanced for isolation. All the test statistics used in detection and isolation involve the inverse of some form of the residual covariance matrix. The purpose of the isolation test is to decide which of the possible faults is present. This chapter outlines time correlatedness which may be removed by recursive "whitening filters," a technique.
