(339p) Fault Detection and Diagnosis in Refinery Operations: A Case Study on Rotating Equipment and Continuous Catalytic Reforming Unit

Industry 4.0 brings in a new era of industrial evolution based on smart manufacturing through interconnectivity, automation, machine learning, and real-time data. By the emergence of intelligent systems, fault detection and diagnosis (FDD) algorithms offer opportunities for the early detection of faults in processes and determination of their root causes, which then yields improvement of manufacturing performance. In this work, we will share some key findings from a model-based approach for fault detection and diagnosis on a reciprocating compressor in the platformer unit in the largest oil refinery in Turkey. Forty six variables are monitored for the selected compressor including vibration and temperature readings at different locations. Various techniques are employed for both fault detection (such as Principal Component Analysis (PCA) and autoencoder neural networks) and diagnosis (such as Bayesian networks and transfer entropy). In fault detection studies as in statistical process control (SPC), two distinct phases take place. In Phase I, the off-line study of the data from a seemingly well-behaving process is performed for parameter estimation that will be used in developing the monitoring tool, often in the form of a control chart. Phase II refers to real-time monitoring of the process based on the monitoring tool developed in Phase I. In that sense the determination of the right data to be used in Phase I is quite crucial. In this study, we provide the most convenient approach for the determination of Phase I data. We present our monitoring tools including the benchmark Hotelling's T² multivariate control chart based on the selected principal components for Phase I and Phase II with the corresponding fault detection in Phase II. We also present our findings when the model is extended from being equipment-wise to plant-wise as we integrate the model to the operation of the Continuous Catalytic Reforming (CCR) unit. We expect determining the most feasible methodology for FDD that substantially contributes to process sustainability and reliability.