Presentation on data analysis at WCCE 2005

This work describes the application of a trend detection method from academia to practical problems. The method is based on the wavelet-based detection of temporal patterns in data, so-called polynomial trends. The notion of a trend as presented here comprises, as special cases, steady states and the linear evolution of process values. The method simplifies the screening of large data sets, even when many process variables are looked at in parallel. For subsequent analyses, only process periods found to be relevant need to be considered. This contribution highlights practical applications of the method to problems from member companies of the AixCAPE consortium. One task is screening of process values to determine the potential of a process while excluding dynamic artifacts. Another example is the extraction of steady states for model building to improve the accuracy of a soft sensor. Finally, the problem of drift detection is considered.

Contact us to obtain an electronic copy of the paper and the poster.