442321 Pipeline Risk Assessment Using Artificial Intelligence: A Case from the Colombian Oil Network

Wednesday, April 13, 2016: 11:15 AM
372 A & D (George R. Brown )
Alexander Guzman Urbina, Graduate School of Technology Management, Ritsumeikan University, Osaka, Japan

Recently, as the world’s energy consumption has increased, the number of serious accidents related with the failure of energy infrastructure has significantly grown and most of them have had a large impact on people and environment. Those serious accidents have increased public awareness about the risk of failure of certain energy infrastructure, and have also increased the concern about the risk acceptability by governments, regulatory bodies and operators.

Then, regarding the concern about the safety of energy infrastructure, operators and regulators have been performing risk assessments in attempt to identify and evaluate accurately the probability of failure of the infrastructure as well as the consequences related with that failure. However, estimating accidental risks in critical energy infrastructure involves a substantial operational expenditure due to the complexity and lack of information. 

Thus, this study attempts to show an aplication case for oil pipelines of the Colombian Oil network. The case analysis is divided into three stages: 1. Fuzzification of  probabilities and consequences, 2. Fuzzy inference, and 3. Defuzzyfication of risk values.

In summary, this paper presents a fuzzy risk inference system in which it considers the conversion of crisp values of probability and consequence into grades of memberships for linguistic terms of fuzzy sets, such as very low, low, medium, high and very high. After that, the inference system considers 25 conditional rules extracted from a risk matrix stated by the Department of Defense of U.S.A., MIL-STD-882D [2000]. Finally, the inference system converts the fuzzy risk values into crisp values by using a centroid method of defuzzification which allows the interaction of all conditional rules to get a numerical risk value.

Extended Abstract: File Uploaded