Generating and exploiting bayesian networks for fault diagnosis in airplane engines

Abstract

Bayesian Networks has been proven to be successful tool for fault diagnosis. There are a variety of approaches for learning the structure of Bayesian Networks from data. This learning problem has been proven to be NP-hard hence none of the approaches are exact when no prior knowledge about the domain of the variables exists. Our approach is based on searching the best network by using particle swarm optimization (PSO) technique. PSO is inherently parallel, works for large domains and does not trap into local maxima. This paper is an application of this technique to a real world problem; fault diagnosis of an airplane engine for oil related variables. It is implemented by our improved software written in C/C++ by using MPI on Linux. Our implementation has the advantages of being general, robust and scalable. Moreover neither expert knowledge, nor node ordering is necessary prior to the optimization. The datasets are generated by preprocessing oil related sensor readings of airplane engines taken during the approach phase of flights. Using this datasets and our software, we constructed Bayesian Networks of the oil related variables in an airplane engine for diagnostics and predictive purposes.

Particle Swarm Optimization Algorithm as a Tool for Profile Optimization

“…Profiling by using particle swarm optimization algorithm could be done for each predictive model based on different methods (neural networks, support vector machines, Bayesian networks). Bayesian network model for finding risk profiles should be constructed in way that target variable contains probabilities for j number of products (Janecek & Tan, 2011;Kawamura & Suzuki, 2011;Yavuz, 2006). Parental nodes contain behavioral characteristics recognized as crucial for risk evaluation in process of attribute relevance analysis.…”

Distributed Particle Swarm Optimization for Structural Bayesian Network Learning

“…Table 2 shows the list of engine oil failure variables under investigation that directly or indirectly influences Fault. The problem of coming up with the best BN that models the dependencies between the listed variables has been attempted in our previous work (Sahin et al, 2007;Yavuz et al, 2006). Here we incorporate expert information in order to make our model more accurate.…”

“…Heuristic searches generally assume that ordering of variables is known and many do not scale well with networks having a large number (more than five) of variables. Additional scaling difficulties arise while dealing with large datasets such as gene and census data (Sahin et al, 2007;Yavuz et al, 2006). In order to avoid the pitfalls of heuristic searches we use a PSO based approach, as it is highly compatible with large datasets and large networks.…”