Fault detection and diagnosis of brahmanbaria gas processing plant using artificial neural network analysis

Abstract

Natural gas (NG) plays an important role in different sectors such as power generation, fertilizer, industrial and commercial sector in Bangladesh. There are many gas producing fields in Bangladesh. Although process safety technology has been gradually implemented over the years, several accidents have happened in gas field industry in Bangladesh like BGFCL, Niko, Occidental, Tullow, and Chevron. Massive blowout took place in Occidental operated field and the similar incident happened at Tengratila field of Niko. Major gas leakage was found in Titas Gas Field (BGFCL). Gas processing plants associated with the gas fields have encountered process industries and outage due to lack of proper monitoring. Bangura gas plant was shut down for lacking proper monitoring and the Bibiyana gas plant was shut down to repair a gas leaking . Those incidents lead to emphasis on fault detection and diagnosis. Advanced process control (such as supervisory control and data acquisition (SCADA) and Distributed control) systems help to operate the plant more reliably. However, operator is saturated by alarms due to disturbances in a chemical process, need tool to rapidly identifying root cause of fault and to rapidly intense to mitigate consequences. To reduce the frequency and consequences of accidents, several techniques of hazard identification and fault diagnosis have been developed and implemented. Over the last few years, several studies were carried out on detection and diagnosis of process plant disturbances using NN based Fault Diagnosis Technique. In this thesis, an attempt has been made to study the fault detection and diagnosis of gas processing plant using NN based system. Firstly, the steady state model of the gas processing plant was developed using HYSYS and be validated using Brahmanbaria gas plant data. Secondly Dynamic model is developed within Aspen HYSYS to study the transient behaviour and different states (normal and abnormal) of the plant. Thirdly, Different states of the process plant were generated using dynamic model.

Finally, a multi-layered feed forward NN based fault detection and diagnosis model has been developed to identify the fault (disturbance) and no fault (normal) operation. The developed NN based fault detection and diagnosis system has been trained using back propagation algorithm. The NN based fault detection and diagnosis system has been trained, validated and tested using the dynamic model data. Several neural networks with different configurations and various learning strategies has employed in the training process to obtain the optimum NN architecture for fault detection and diagnosis. Preliminary results shows that NN based method successfully detect the faults of Gas processing plant. It is expected that ANN based fault detection and diagnosis tool will be popular in petrochemical process due to its simplicity to develop.