Direct Smoothing Method for Detection of Fault Patterns

Abstract

Process knowledge can be utilized to detect occurrences of faults that leave a fault signature or a pattern in the process data being monitored. Here, I assume that the process knowledge is in the form of a specified function for the fault pattern but the parameters of this function are unknown. Furthermore, unknown function parameters may be constant or varying over time. Hence,an efficient method is required for estimation of the fault pattern parameters for use in the fault detection algorithms. Although, the least squares is a well known and proven method for estimation, on-line parameter estimation calculations may get cumbersome as the number of observations increase. Therefore, I propose the use of direct smoothing method, which is recursive and based on the discounted least squares criterion. Other than the computational efficiency, direct smoothing is also robust in estimation fault patterns with varying parameters.