Extended PLS approach for enhanced condition monitoring of industrial processes

Abstract

IntroductionThe detection and diagnosis of abnormal situations in the operation of industrial processes is a problem of considerable challenge that is attracting wide attention in both academe Ž . and industry. Nimmo 1995 showed that the U.S.-based petrochemical industry alone could save up to $10 billion per year if abnormal situations could be detected, diagnosed, and appropriately dealt with. The consequences of not being able to detect such abnormal situations range from increased operational costs to costly plant shutdowns.Industrial processes often present a large number of process variables, such as temperatures, pressures, flow rates, compositions, which are typically sampled at a frequency of one minute. Identifying and troubleshooting abnormal operating conditions simply by observation are difficult tasks with such a large amount of data, particularly since the process Ž variables are usually highly correlated MacGregor et al.,. 1991 . However, the sampled data have embedded within them information for revealing the state of the process operation.

Artificial intelligence for monitoring and supervisory control of process systems

“…Studies on applications of PCA to process industries can be found in Akbaryan and Bishnoi (2001), Amand et al (2001), Kano et al (2001Kano et al ( , 2002, Kruger et al (2001), McAvoy (2002), Ü ndey and Cinar (2002), Wong and Wang (2003), Lee et al (2004), Miletic et al (2004).…”

“…A popular application of PLS is to include process variables in the predictor matrix and product quality data in the dependent matrix, which can include off-line measurement data (Kruger et al, 2001). Such inferential models (also known as soft sensors) can be used for on-line prediction of product quality data.…”

Multimode process monitoring based on Bayesian method

“…In particular, principal component analysis (PCA) and partial least squares (PLS) have been widely used for process monitoring and fault diagnosis [1][2][3][4][5][6]. Recently, independent component analysis (ICA) was employed for monitoring processes with non-Gaussian information, which can effectively extract high-order statistical information from process data [7][8][9][10][11][12].…”

Robust partial least squares regression—part III, outlier analysis and application studies

“…PLS is predominantly used for the modelling of intensity-based data that stem from infrared or Raman spectroscopy experiments, for example References [2,3]. The application of PLS to recorded data from the chemical industry is often concerned with statistical-based process monitoring of continuous [4] and batch processes [5].…”