A Novel Configuration-Driven Data Mining Framework for Health and Usage Monitoring Systems

Abstract

Health and Usage Monitoring Systems (HUMS) and Condition Based Maintenance (CBM) systems are closely related systems since the data collected by HUMS can be effectively used by CBM systems to generate condition and health indicators of air-crafts in order to find out the components on which maintenance is required. Efficient and comprehensive automation of such CBM systems is, therefore, of utmost importance to take an advantage of the massive amount of data continuously collected by HUMS. In this chapter, we present a novel framework for automating the CBM systems, based on supervised learning practices established in UH-60 Condition Based Maintenance (CBM) manual [3]. The proposed framework is based upon a concept of building a configurationdriven data mining tool in which the maintenance decisions of aircrafts’ components are driven by configuration metadata. We seek to address a key design goal of building a generic framework that can be easily instantiated for various CBM applications. This design goal, in turn, raises the challenges of building a system that features modularity (i.e., the software structure is based on a composition of separate modules, which jointly incorporate with each other.), extendibility (i.e., the software should be easy to extend.), and maintainability (i.e., the software should be easy to maintain and update.). To meet this design goal, we adopted a layered architecture for the proposed framework. The framework consists of three layers; 1) Storage layer, which concerns the storage of source data and configuration metadata used by the system; 2) Extraction layer, in which the source data and configuration metadata are extracted and passed to the upper (processing) layer; and 3) Processing layer, which is responsible for executing mining algorithms and reporting necessary maintenance actions. This layered architecture uses two prominent techniques, namely XML-based metadata storage, and dynamic code generation and execution. The XML-based metadata storage provides a generic platform for storing configuration meta-data, whereas dynamic code generation and execution allows applications to be extended with non-compiled source code, which is stored within configuration metadata. Both the generic storage platform and the on-the-fly code generation features help significantly reduce the cost of the software