Bayesian Network Exploration for Airplane Engine Fault Diagnostics

Author: Archana Devasia (Rochester Institute of Technology, Department of Electrical Engineering)

Abstract

Fault diagnosis and prediction is a pertinent problem in today's times. Bayesian Networks provide an attractive solution to such problems. This paper puts forward a method for exploiting Bayesian Networks for airplane engine fault diagnostics. A distributed Particle Swarm Optimization (PSO) approach is explored in order to construct the best Bayesian Network from a large dataset comprising of raw data taken from the sensors of airplane engines during actual flights. The inherent parallelism of the PSO technique has been exploited with the algorithm being implemented on a cluster of 16 processors using Message Passing Interface (MPI) in Linux. The proposed approach involves an initial preprocessing of the available data by means of an equal frequency binning algorithm which is fed as an input to the cluster for generating the Bayesian Network. Expert information is integrated only after the network is conceptualized resulting in a more effective model of the system. The work presented here involved repeated iterations in order to converge onto a network that fitted the data the best. This resulted in the generation of a network that could successfully detect faults in test datasets.

Bacterial Foraging Approach to Classic Traveling Salesman Problem

Author: Archana Devasia (Rochester Institute of Technology, Department of Electrical Engineering)

Abstract

A traveling salesman has to leave home, stop in "n" number of cities once, and returns home by traveling the shortest distance between all the cities. Finding the shortest route though is a very complicated problem. The number of different routes that can be taken exponentially increase as the number of cities increase making it very hard, if not impossible, to find the shortest one. For this reason, different AI techniques have been implemented to find the optimal or near-optimal path. Particle Swarm Optimization, Genetic Algorithms, and Ant Foraging are a few of the optimization methods that have been used in the past to find a solution. Bacterial Foraging will be applied to the problem and the results will be compared with the past methods used to see if Bacterial Foraging can generate a more-optimal solution.