Fault detection and diagnosis using correlation coefficients

Accurate process fault detection and diagnosis (FDD) at an early stage of a chemical process is very important to modern chemical plant in overcoming challenges such as strict requirements on product quality, low consumption of utility, environmentally friendly and safe operation. The use of the Contribution Plots (CP) for fault diagnosis in previous methods in Multivariate Statistical Process Control (MSPC) is not suitable since it is ambiguous due to no confidence limit in the CP. This research is to formulate a FDD algorithm based on MSPC via correlation coefficients. A fractionation column from a palm oil fractionation plant is chosen as the case study and the model of the case study is simulated in Matlab. Data collected with a process sampling time, TMSPC, of 4.6 hours and following the normal distribution are used as Nominal Operation Condition (NOC) data. Normal Correlation (NC), Principal Component Analysis (PCA) and Partial Correlation Analysis (PCorrA) are used to develop the correlation coefficients between the selected key process variables with the quality variables of interest in the process from the NOC data. Faults considered in the research are sensor faults, valve faults and controller faults generated in the fault data (OC). Shewhart Control Chart and Range Control Chart together with the developed correlation coefficients are used for fault detection and diagnosis. Results show that the method based on PCorrA (overall FDD efficiency = 100%) is more superior than the method based on NC (overall FDD efficiency = 67.82%) and the two analysis methods based on PCA (overall FDD efficiency = 67.82%)