MODELLING OF A RECURSIVE POLYNOMIAL ESTIMATOR WITH MACHINE LEARNING TO IMPROVE SAFETY INTEGRITY LEVEL OF PETROCHEMICAL PROCESS DESIGN

Subject Area: CONTROL SYSTEM ENGINEERING

This paper presents the application of machine learning technique and recursive polynomial estimator for improving the reliability of critical safety instrument system. The aim of the research is to improve the reliability of critical safety instrument system using machine learning technique and the main objective to develop a neuro logic solver and polynomial estimation model which monitoring the behaviour and distillation plant and control against system failure. To address this problem, literatures were reviewed, and a gap on the safety integrity was identified. This was done using methods such as risk assessment test, data collection, neurologic solver algorithm and error estimation algorithm and guided by the International Electrochemical Commission (IEC) 61508 and 61511 methodologies for the design and implementation of Safety Instrument System (SIS). The neurologic solver algorithm was developed using artificial neural network, tansigactivation function and gradient descent back-propagation algorithm, while the error estimation algorithm was developed with recursive polynomial functions. These algorithms were implemented with Simulink, evaluated and cross validated considering Mean Square Error (MSE), regression, PFD, Risk Reduction Factor (RRF) and Safety Integrity Level (SIL). The result of the neurologic solver MSE is 2.98E-09, Regression is 0.9978 and PFD is 9.00E-04. When the neurologic solver was integrated on the testbed and evaluated, the PFD is 8.52E-04, thus presenting a SIL of 4 as against 1.14E-03 in the test bed with neurologic PLC solver and hence SIL of 3. The overall neurologic-based SIS PFD is 6.44E-03 and RRF of 155.279 as against 6.72E-02 with RRF of 14.881 which is characterized with PLC based logic solver, recording a 33.8% improvement in reliability.