Sarawan Wongsa

Model structure selection strategy for Wiener model identification with piecewise linearisation

This paper presents a method for identifying the optimum structure of Wiener model with piecewise linearisation. The number of piecewise linear functions for estimating the static nonlinear and the maximum lag of the linear dynamic part of the Wiener model are selected by cross-validation based approach. The maximum lag and the number of partitions are selected in two subsequence steps. Three popular model selection criteria, i.e. FPE, PRESS, and CP, are considered and compared in the selection process. With the ultimate aim of compensation for nonlinearities in sensors, we have illustrated the feasibility of using the proposed method to compensate hard nonlinearities, such as discontinuous nonlinear and saturation. The results from this work can be used as a guideline of model selection for Wiener model identification and nonlinear compensations of sensor.

Fault detection and isolation of acid gas removal units in a gas separation process using PLS

Partial least squares (PLS) is a potential data-driven technique to deal with a huge number of measured variables and complex relationship. It could be used for process monitoring even for chemical processes that have nonlinear dynamic properties. In this paper PLS is applied to detect two fault types which occurred in real Acid Gas Removal Units (AGRU) of a gas separation process. The developed PLS-based fault detection and isolation (FDI) is implemented under a careful selection of input variables and key quality variables for FDI purposes. The results indicate that the developed approach can predict and diagnose fault cases efficiently prior to process upset and shutdown.

A PCA-based fault monitoring of splitter nozzles in gas turbine combustion chamber using exhaust gas temperature

In this article, a principal component analysis (PCA) based process monitoring approach is proposed for detecting and isolating damaged splitter nozzles in gas turbine combustion chamber. Damage in splitter nozzles can lead to unstable combustion process and had to start-up and result in unplanned shutdown. The present method detects any abnormal temperature behavior by monitoring the combined index obtained from the PCA model. The abnormal temperature sensors are further analysed by using the partial composition (PD) contribution. To avoid smearing effects, the PD contributions are normalized such that they are even across variables when there is no fault. The possible damaged splitter nozzles are identified using the mapping table. Results of both simulated and real datasets reveal the effectiveness of the proposed method.

An Automatic Selection of Segment Width for the Single Sensor Differential Thermal Analysis (SS-DTA) Based on Piecewise Linear Approximation Technique

Single Sensor Differential Thermal Analysis (SS-DTA) is a novel non-destructive testing technique for studying and detecting the phase transformations and structural changes in materials. It uses only one temperature sensor to measure the temperature in a particular point of interest in the material during actual and simulated thermal processing of the material. SS-DTA compares the temperature recorded in a tested specimen against a reference thermal profile which can be generated either by analytical formulae or piecewise linear approximation. The main advantage of piecewise linear approximation over the analytical formulae is that it does not need the knowledge of tested material and processing conditions to optimally estimate the parameters of reference thermal history. On the other hand, in order to apply the piecewise linear approximation technique we must specify the segment width which is normally fixed at a nominal value of 1.5 seconds. We have recently found that this nominal value might not be an optimal choice for the segment width as it does not guarantee to give the best detectability of phase transformation. Therefore, in this research work we proposed a technique to automatically select an appropriate value of the segment width. The performance of proposed method has been evaluated by investigating the phase transformations of welded stainless steel SUS 321and SUS 304. It was found that the appropriate segment width could be ranging from 1.25-1.75 seconds and by using this selection technique, we could detect the differential temperature more accurately than when using the nominal value.

Online fatigue detection of a SMA-based control valve using CUSUM algorithm

This paper describes a functional fatigue detection system for nonlinear SMA-based control valve. Any drift from the normal behavior of the valve is revealed using a model-based residual generator by means of a nonlinear auto-regressive with eXogenous input (NARX) model. Based on the optimisation property of cumulative sum (CUSUM), an online system for detecting changes in statistical properties of residuals due to fatigue of the valve is developed. The experimental results suggest that the variance of one step ahead (OSA) error and model predicted error (MPE) is a good indicator of fatigue states of the valve.