(614a) OASIS-P: Operable Adaptive Sparse Identification of Systems for Fault Prognosis of Chemical Processing System

Fault prognosis is a predictive diagnosis of a process based on its current dynamics. The goal of prognosis is to predict the occurrence of a fault and propose action to maintain process safety. Among the several prognostic approaches proposed in the literature, data-driven methods are widely implemented in the industrial sector as these methods do not require any process knowledge, which is otherwise difficult to obtain in the case of complex processes. Nevertheless, the existing data-driven monitoring frameworks have to be improved further to address the challenges of modern chemical process systems [1]. One such important problem is simultaneously occurring multiple faults. In a practical scenario, several single faults may occur independently in a process at the same time instance and can together result in a greater degree of damage. In such cases, methods developed to predict and isolate a single fault may have a limited capability in managing these simultaneous faults. Although several fault diagnostic approaches [2, 3] have been proposed in the literature to deal with simultaneous faults, these methods have not been explored from the prognosis perspective.

In this work, we propose a fault prognosis framework named ‘OASIS-P’ that utilizes operable adaptive sparse identification of systems (OASIS) [4] and two-level contribution plots [5] to perform fault prediction and isolation, respectively. The OASIS algorithm provides a partially interpretable process model that can cope up with system changes in real-time. In the proposed OASIS-P framework, we use OASIS to forecast the future process behavior for a pre-defined time-horizon. Utilizing this prediction, we estimate and monitor operational risk dynamically to predict any impending faults. This ensures to signal a warning only when a fault is expected to impact the process safety. Once a fault is predicted, we initiate the isolation step to identify the faulty process variables. Unlike the classification-based isolation techniques that rely on historical fault data, contribution plots do not require a priori knowledge about the system, and hence, they are better for fault isolation. In OASIS-P, we implement a two-level contribution analysis. First, the process variables are grouped based on the physical meaning obtained from the OASIS model structure. Next, the contribution of each of these groups is used to model risk. The risk is evaluated to identify the maximum contributing group. The process variable with the maximum contribution to the estimated risk is identified as a faulty variable. In the case of simultaneous faults, when the characteristics of the constituent faults are coupled, it becomes difficult to identify them [6]. During the contribution analysis, it is difficult to conclude if the contribution observed is due to the presence of a fault in a variable or the effect from another faulty variable. In this context, physical interpretability provides an understanding of the connection between variables, which helps overcome the ambiguity in the contribution results by revealing the fault propagation pathway. Consequently, all the underlying faulty variables can be identified, thus enhancing isolation accuracy. For demonstration purposes, we apply the proposed OASIS-P method to perform the prognosis of a single fault and simultaneously occurring multiple faults in a reactor-separator system.

Literature cited:

[1] Zhong, K., Han, M. and Han, B., 2019. Data-driven based fault prognosis for industrial systems: a concise overview. IEEE/CAA Journal of Automatica Sinica, 7(2), 330-345.

[2] Yélamos, I., Graells, M., Puigjaner, L. and Escudero, G., 2007. Simultaneous fault diagnosis in chemical plants using a multilabel approach. AIChE Journal, 53(11), 2871-2884.

[3] Watanabe, K., Hirota, S., Hou, L. and Himmelblau, D.M., 1994. Diagnosis of multiple simultaneous fault via hierarchical artificial neural networks. AIChE Journal, 40(5), 839-848.

[4] Bhadriraju, B., Bangi, M.S.F., Narasingam, A. and Kwon, J.S.I., 2020. Operable adaptive sparse identification of systems: Application to Chemical Processes. AIChE Journal, 66(11), p.e16980.

[5] Luo, L., Bao, S., Mao, J. and Tang, D., 2017. Fault detection and diagnosis based on sparse PCA and two-level contribution plots. Industrial & Engineering Chemistry Research, 56(1), 225-240.

[6] Wu, Y., Jin, W., Li, Y. and Wang, D., 2021. A novel method for simultaneous-fault diagnosis based on between-class learning. Measurement, 172, p.108839.