Evolutionary Tuning of Non-parametric Information Conversion Functions for Diagnostics

In aircraft health management applications, there are two types of data available, parametric and non-parametric data. Parametric data, such as sensor measurements from engine or airframe, have traditionally been the primary data source for proactive diagnostic and performance monitoring applications because they allow one to trend even subtle changes over time. Non-parametric data (e.g., error logs and fault messages, that are either of textual nature or binary flags) are more often used for event driven maintenance. Fundamentally, both information sources harbor diagnostic information that, when used together, might improve results for both applications. This paper employs a conversion function to parameterize non-parametric information so that it can be used together with parametric data to enhance the diagnosis of aircraft engine faults. Genetic algorithms are applied for fine-tuning of parameters of the conversion function. Experimental results from real engine data are presented