Detection and diagnosis of plant-wide oscillations using GA based factorization

The degradation in the performance of the plant is observed in form of oscillations in time trends of measurements. These disturbances propagate throughout the plant and also affect the performance of healthy loops. Thus, it becomes increasingly important to detect all the loops that lead to plant-wide oscillations. In this paper, spectral decomposition based on Evolutionary Algorithms is proposed for the detection of plant-wide oscillations. The key feature of the proposed technique is that it retains causal basis spectrum like shapes consisting of narrow band peaks by searching the solution space globally. Two industrial case studies are presented to demonstrate the efficiency of GA based Evolutionary Algorithms over existing techniques like independent component analysis (ICA) and non-negative matrix factorization (NMF) in detecting plant-wide oscillations. Results show that the proposed technique outperforms ICA and NMF with respect to reconstruction error.

► The work extends the concept of GA based factorization of high dimensional matrices for the analysis of process data.
► We propose a novel method based on Genetic Algorithm (GA) based factorization is proposed to deal with the problem of plant-wide oscillations and ensure global optimality in the search method.
► We validate of the approach has been done on two real industrial case studies.