Assessment of Overhaul Effectiveness and Usage-based Inference using Bayesian Networks

##plugins.themes.bootstrap3.article.main##

##plugins.themes.bootstrap3.article.sidebar##

Published Sep 24, 2018
Nenad G. Nenadic Christopher J. Valant Sean P. McConky Michael G. Thurston

Abstract

The process of assessment of effectiveness of the existing overhaul practices determined that the historical usage of assets provides valuable contextual information. Usage data is typically highly reliable, but not in legacy fleets, featuring older vehicles with missing, incomplete, inconsistent, and contradictory data. This paper describes two methods for usage estimation from noisy data by exploiting two data sources: 1) unreliable, manually-entered usage data and 2) part replacements. The first method employs a probabilistic model to reconcile missing and inconsistent data entries; the second is based on the replacement of consumable components. The probabilistic model, fully and uniquely specified by the probabilistic variables (with their distributions) and deterministic variables, is validated using synthetic datasets because the real ground truth associated with the field data does not exist. Disproportional impact of an incorrect initial data point is mitigated by training the model in both forward and reverse directions. The motivating hypothesis for usage estimation from part replacements is based on a plausible assumption that specific consumables, e.g. brake pads, have reasonably repeatable replacement patterns which can be related to usage. For many vehicles mean time between failures of a component was even longer than the average data collection time span. But for assets with sufficiently longer data records, the cumulative replacements of components are well-correlated with the probabilistic usage estimates, providing additional reinforcement for the inference.

How to Cite

Nenadic, N. G., Valant, C. J., McConky, S. P., & Thurston, M. G. (2018). Assessment of Overhaul Effectiveness and Usage-based Inference using Bayesian Networks. Annual Conference of the PHM Society, 10(1). https://doi.org/10.36001/phmconf.2018.v10i1.295
Abstract 327 | PDF Downloads 285

##plugins.themes.bootstrap3.article.details##

Keywords

usage, CBM, Probabilistic modeling, Bayesian

Section
Technical Papers