Detection of multivariable trends in measured process quantities

This paper addresses the detection of a trend of a multivariable process. In particular, many (measurable) process quantities are considered simultaneously to identify a polynomial trend of a given order. A novel method is suggested to detect such trends in noisy measurements. The start time and the duration of the multivariable trend are computed by means of an efficient multi-scale algorithm. The method can be employed in different contexts. For example, a multivariable trend with a given duration and an endpoint at current time can be determined in on-line applications, or an off-line search for multiple multivariable trends with arbitrary start times and durations in a historic archive of plant measurements can be carried out.