A Data-Driven Approach For The Process Monitoring And Early Fault Detecion Of The Reactor-Regenerator In The RFCC Unit

Technical process operations require more and more advanced supervision and fault detection in order to improve safety, economy and reliability. Incipient faults are faint on initial impact and their effects cannot be easily detected. Such faults demonstrate an effect that is small but overtime can become worse and lead to abrupt system failure. In this study the effect of this slow dynamic fault in a process and their adverse affect on operational cost and facility capacity are studied.The Residue Fluid Catalytic Cracking (RFCC) Unit remains a key unit in the refinery industry. The increasing demand of refinery products coupled with strict production standards and environmental constraints gives rise to the need for proper monitoring of operational activities. There are numerous problems encountered during the operation of the RFCC unit. Many of the process dangers are problems caused by human error, sensor misrepresentation and/or actuator command error. The focal point of this study is the use of process monitoring to evaluate product quality and quantity.The reactor-regenerator is the heart of the RFCC unit. Changes in the performance of the reactor directly influence product quality and quantity. The key variables considered in this study include first stage regeneration, medium pressure steam, raw oil, slurry, lean gas and second stage regeneration. The Distributed control system provides measurements in flow rate, temperature and pressure for these seven variables. The research perspective takes into consideration disturbances that cause conditions of poor reactor performance. Typically, any element that greatly influences coke and catalyst circulation rate will bring about a loss of converted product and selectivity. Thus the reactor is observed under two different abnormal states for effect on product quality and quantity.Medium pressure steam and lean gas (dry gas) can cause unfavorable conditions in the reactor. Changes in these variables were introduced as a slow dynamic fault. The data is acquired and de-sampled without much loss of the data authenticity and then used in PCA and CUSUM-PCA data analysis. Hence, this data driven approach incorporates a moving window for the accumulation of the process data averages and then it is further analyzed using conventional PCA where the normal operation data serves as a standard for determining a diversion from regular operations. After which, the abnormal situations are detected using Squared Prediction Error (SPE) and Hotelling T2 control charts. Upon completion of the fault detection analysis it is observed that CUSUM-PCA can provide earlier fault detection for slow faults of this nature. Results on the case study for abnormal levels of medium pressure steam and lean gas show that CUSUM-PCA detects the fault over an hour in advance as compared to conventional PCA. The statistical process monitoring scheme study in this research can be recommended for application to similar industrial operations of the same complexity.