What Went Wrong and When? Multivariate Fault Diagnosis of a High Purity Distillation Column

Distillation is a widely used industrial separation process. Process monitoring using multivariate data analysis for such capital intensive processes has the potential to reduce energy and resource requirements. Notwithstanding the large body of work in industrial process monitoring, little prior work has focused on monitoring ultra-high purity mulitcomponent distillation processes. This extreme situation significantly increases the difficulties, particularly when considering the underlying nonlinearities.

In this paper, principal component analysis (PCA) is employed to analyze the process variables in order to obtain a better understanding of the key relationships affecting the product quality. Using this data-driven approach, it was found that the reboiler steam flowrates were the most important factor affecting the top product ppm specification. These results were subsequently validated by the plant engineers who attributed the abnormal variation in the steam supply system to other unit operations. It provides a potential solution to guide the control system to maintain the top product ppm specification in a narrow variation. PCA results are trended in real time using novel visualisation techniques that can provide the operator with early-warning signals.