Statistical signatures used with principal component analysis for fault detection and isolation in a continuous reactor

Principal component analysis (PCA) is a technique widely used in industrial process control for data analysis and reduction. A score discriminant can be used in conjunction with a PCA model to differentiate between the normal operating condition and an abnormal condition. To illustrate application of these analytical techniques, raw data is collected from a high fidelity simulation of a continuous chemical reactor for both the normal operating condition, and several different fault conditions. A PCA model and score discriminant are applied to analyze the raw data, but this approach does not reliably differentiate between all process conditions. To improve the differentiation, an alternative model is developed using the statistical signatures of mean and standard deviation, such as are computed in some present day intelligent field devices. The new PCA score discriminant model based on statistical signatures produces a much clearer differentiation between all process conditions. Copyright © 2006 John Wiley & Sons, Ltd.