Predictive Analytics for the Royal Canadian Navy Fleet Maintenance Facilities – An application of Data Science to Maintenance Task Completion Times

PDF

Authors
  1. Maybury, D.
Corporate Authors
Defence Research and Development Canada, Centre for Operational Research and Analysis, Ottawa ON (CAN)
Abstract
In November 2016, Cdr Timothy Gibel, Directorate of Naval Strategic Management (DNSM) requested the Centre for Operations Research and Analysis’s (CORA) assistance with creating a predictive model for actual work hours per task on Royal Canadian Navy (RCN) vessels at the Cape Scott and Cape Breton Fleet Maintenance Facilities (FMF). He also requested help in finding patterns in maintenance task completions at the vessel level. The FMF provided dataset contains information on 132,292 unique tasks separated into 43,731 order keys. Using regression trees with feature engineering for actual work hours, I find that a simple 14 terminal node tree explains 18% of the variance in the data. Gradient boosted stumps explain as much as 25% of the variance, but at the expense of an interpretable structure for schedule validation. Hidden Markov Modelling of monthly order key completion time series data reveals coastal differences between the CN’s two maintenance facilities, providing an objective motivation to discover the source.

Il y a un résumé en français ici.

Keywords
Data Analytics;Classification and Regression Trees;Data Science;Gaussian Mixture Models;Hidden Markov Models
Report Number
DRDC-RDDC-2018-R150 — Reference Document
Date of publication
01 Dec 2018
Number of Pages
56
DSTKIM No
CA048216
CANDIS No
808850
Format(s):
Electronic Document(PDF)

Permanent link

Document 1 of 1

Date modified: