TY - JOUR
T1 - Risk assessment of process systems by mapping fault tree into artificial neural network
AU - Sarbayev, Makhambet
AU - Yang, Ming
AU - Wang, Haiqing
PY - 2019/7
Y1 - 2019/7
N2 - Quantitative risk assessment is a crucial step in the safety analysis of process systems. The advancement of modern process systems has made a large volume of process data and information available for process safety analysis. This tendency urges the need for developing new risk assessment approaches. Fault tree (FT), a conventional risk analysis method, is found to be ineffective in dynamic risk analysis and data analytics due to its static nature and reliance on experts' judgment. Artificial Neural Network (ANN) is a structured model built upon data samples and learning algorithms to process complex input/output data in the way that it is being trained. The application of ANN can help to overcome some of the limitations of FT. The data-driven nature, independency on prior information on events relationships, and less reliance on experts’ judgment are the advantages of ANN over FT. The use of ANN in risk assessment is not a new concept. However, there is limited work on the development of ANN-based risk assessment models using conventional methods such as FT as an informative base. This study proposes a methodology for mapping FT into ANN to support the convenient and practical application of ANN in risk assessment. The proposed method is demonstrated through its application to the analysis of a system failure in the Tesoro Anacortes Refinery accident. The results have shown that the ANN model mapped from the FT is an effective risk assessment technique.
AB - Quantitative risk assessment is a crucial step in the safety analysis of process systems. The advancement of modern process systems has made a large volume of process data and information available for process safety analysis. This tendency urges the need for developing new risk assessment approaches. Fault tree (FT), a conventional risk analysis method, is found to be ineffective in dynamic risk analysis and data analytics due to its static nature and reliance on experts' judgment. Artificial Neural Network (ANN) is a structured model built upon data samples and learning algorithms to process complex input/output data in the way that it is being trained. The application of ANN can help to overcome some of the limitations of FT. The data-driven nature, independency on prior information on events relationships, and less reliance on experts’ judgment are the advantages of ANN over FT. The use of ANN in risk assessment is not a new concept. However, there is limited work on the development of ANN-based risk assessment models using conventional methods such as FT as an informative base. This study proposes a methodology for mapping FT into ANN to support the convenient and practical application of ANN in risk assessment. The proposed method is demonstrated through its application to the analysis of a system failure in the Tesoro Anacortes Refinery accident. The results have shown that the ANN model mapped from the FT is an effective risk assessment technique.
KW - Artificial neural network
KW - Fault tree
KW - Process safety
KW - Process systems
KW - Risk assessment
UR - http://www.scopus.com/inward/record.url?scp=85065606875&partnerID=8YFLogxK
U2 - 10.1016/j.jlp.2019.05.006
DO - 10.1016/j.jlp.2019.05.006
M3 - Article
AN - SCOPUS:85065606875
VL - 60
SP - 203
EP - 212
JO - Journal of Loss Prevention in the Process Industries: the international journal of chemical and process plant safety
JF - Journal of Loss Prevention in the Process Industries: the international journal of chemical and process plant safety
SN - 0950-4230
ER -