Accidentsare likely to occur in process industries due to the involvement of largenumber of hydrocarbons at high operating conditions. Reliability and safetyanalysis of complex process systems is a challenging task. Nowadays numerous techniquesare available along with their advantages and disadvantages.

Among themBayesian network analysis is most popular technique due to its structure whichhelps to model a wide variety of accidentscenarios. Toestablish the conditional dependencies among events in BN, traditional logic gates such as OR & AND gates has been used forpast several years in accident modeling which are sometimes too naïve toestablish cause-effect relationship and does not necessarily reflect the realinteractions. Thus, it introduces uncertainty into the model. As BN proves theability to manipulate the conditional distributions various canonicalprobabilistic models are becoming important because they allow building the conditional distribution from lessnumber of parameter. Therefore, it becomes easy to establish the distribution andrequired less computational time. In thisstudy, we only try to apply Noisy-OR and leaky Noisy-AND relaxationassumptions, since these two provide the median condition for their respectivelogic gates. Another source of uncertainty comes into play through data uncertainty.BN requires crisp probabilities to assess the likelihood of basic events andsafety system which are difficult to find.

Even if available, the data is subjectto incompleteness and imprecision. Addressing these issues is an important taskotherwise BN will provide a false imprecision of assessing the risk therebyundermining the credibility of entire process. Evidence theory is used forhandling partial ignorance and vagueness of the data.Applicationof these approaches helps to predict significantly improved accidentprobability. Thus, it renders useful information for taking early safetymeasures to ensure that risk is below acceptance criteria i.e. ALARP (As Low AsReasonably Practicable).

A case study has been used to illustrate the executionand effectiveness of the approaches and compare the results with deterministicapproach.