Fault diagnosis of batch processes using discriminant model

Abstract

A new statistical online diagnosis method for a batch process is proposed. The proposed method consists of two phases: offline model building and online diagnosis. The offline model building phase constructs an empirical model, called a discriminant model, using various past batch runs. When a fault of a new batch is detected, the online diagnosis phase is initiated. The behaviour of the new batch is referenced against the model, developed in the offline model building phase, to make a diagnostic decision. The diagnosis performance of the proposed method is tested using a dataset from a PVC batch process. It has been shown that the proposed method outperforms existing PCA-based diagnosis methods, especially at the onset of a fault.

An extensive reference dataset for fault detection and identification in batch processes

“…Li and Cui [51] also employ 60 faulty Pensim (5 types) in their work on Feature Vector Selection FDA using Nearest Feature Lines (FVS-FDA-NFL). No information is provided by Cho and Kim [19,20], Cho [18], or Li and Cui [51] on the type of the employed process upsets, their magnitude, or their onset time.…”

“…Cho and Kim [19,20] proposed an FDA-based fault classification using data from a simulated PVC polymerization. Hereto, they generated a set of 44 normal batches, and 3500 faulty batches of 5 types because their approach requires a number of faulty batches for classifier training greater than the dimensionality of the batches (in their case, the number of monitored sensors times the number of time points).…”

Nonlinear feature extraction and classification of multivariate data in kernel feature space

“…. , P. Basically entire tasks of the on-line KFDA based fault diagnosis are similar to those of the linear FDA case of Cho and Kim (2004). When a fault is detected, the on-line diagnostic decision making process is executed iteratively during the operation of a new batch.…”

“…Then the task of fault diagnosis is to classify a new data into one of predefined fault groups when a fault is detected. FDA has been shown to be the good linear technique for fault diagnosis and to outperform PCA based diagnosis methods (Chiang et al, 2000;Cho & Kim, 2004). It is attributed to the fact that FDA seeks directions that are efficient for discrimination (but PCA for representation) (Müller, Mika, Rätsch, Tsuda, & Schö lkopf, 2001).…”

“…Many multivariate statistical techniques had been developed and applied to fault diagnosis such as principal component analysis (PCA), partial least squares (PLS), Fisher discriminant analysis (FDA), and discriminant partial least squares (DPLS) (for example, see Akbaryan & Bishnoi, 2001;Alsbergav, Goodacrea, Rowlandb, & Kella, 1997;Chiang, Russell, & Braatz, 2000;Cho & Kim, 2004;Kemsley, 1996;Kourti, Nomikox, & MacGregor, 1995;Raich & Cinar, 1996). Such data-driven techniques for fault diagnosis have been extensively utilized due to the recent advances in sensors and data measurement technology.…”