A general Bayesian predictive maintenance methodology
J. Cullum
1
1. No Affiliation
Abstract
The development of new and the improvement of existing Health and Usage Monitoring Systems
remains an active and multi-disciplinary area of research. Applications of these systems span
numerous applications including Defence assets, infrastructure, utilities, aviation and
transportation; sharing the common aim of improving our understanding of overall condition or
‘health’ of the application using data. Systems can be developed on the basis of empirical data,
theoretical formulae or machine learning algorithms and may also include elements of reasoning
abstracted from human psychology. The present work provides a methodology which can be
applied to create a Bayesian predictive maintenance system for a general component. The
methodology incorporates any number of faults or failure scenarios and produces a discrete set of
outcomes which are human-readable and actionable.
Considering that maintenance is a decision taken in the context of uncertainty, or ‘risk’, the
methodology describes how sensor data can be transformed using a Bayesian machine learning
algorithm and integrated with a multi-attribute utility value system to produce these outcomes. A
predictive maintenance system developed using this framework is expected to deliver the greatest
impact when supporting high-value applications such as those within the aerospace and Defence
industries; where investment in maintenance is needed to achieve high reliability requirements and
prevent often extreme consequences of failure. A theoretical application to the hull of a Reusable
Launch Vehicle is presented as it is anticipated that predictive maintenance will play a role in
Australia’s future as an emerging Space power. By presenting this methodology, it is intended that
future research develops and evaluates its value and accuracy toward improving our understanding
of equipment health and function.
Keywords: Bayesian, maintenance, predictive, risk, Reusable Launch Vehicle, rocket.
Introduction
Data-driven maintenance, applied as a Health and Usage Monitoring System, is driven by the desire
to understand, how a piece of equipment is functioning. This understanding is based upon the
widespread understanding that failure mechanisms as well as maintenance can impact equipment
health, which was formalised as Reliability-Centered Maintenance [1]. New insights in this area
can also be gained though applications of Artificial Intelligence [2], the Internet of Things [3] and
Industry 4.0 [4, 5] concepts, as well as through the development of a Digital Twin [6].
In principle, applications of data-driven maintenance can lead to improved asset reliability. While
this is beneficial in any industry, high-value assets in Defence stand to benefit the most since
typically high through-life sustainment costs can be reduced [7].
Space is a Defence domain. It is currently an expensive, technically challenging and time-
consuming endeavour for a vehicle, manned or unmanned, to travel there. It is common for
equipment failure or non-ideal conditions to delay or prevent a launch. It is less common, though
