Studies to Predict Maintenance Time Duration and Important Factors From Maintenance Workorder Data

Published

September 26, 2019

Author(s)

Fnu Madhusudanan Navinchandran, Michael E. Sharp, Michael P. Brundage, Thurston B. Sexton

Abstract

Maintenance Work Orders (MWOs) are a useful way of recording semi-structured information regarding maintenance activities in a factory or other industrial setting. Analysis of these MWOs could provide valuable insights regarding the many facets of reliability, maintenance, and planning Information such as which maintenance activities consume the most work hours, identification of problem machines, and spare parts needs can all be inferred to some degree from well-documented MWOs. However, before one can derive insights, it is first necessary to transform the data in the MWOs (generally some form of natural language) into something more suitable for computer analysis. NIST previously developed a computer aided tagging system that allows for the quick identification of key concepts within the natural language of the MWOs, and a protocol for categorizing these concepts as problems, solutions, or items. Using this annotation method, this paper investigates machine learning methods to gain insights about work hours needed for various maintenance activities. Through these methods, it is possible to explain the factors captured in the MWOs that have the strongest relationship with the duration of maintenance actions. The workflow of this research is to first build strong data-driven models to classify the duration of any maintenance activity based on the language and concepts gathered from the associated MWO. Sensitivity analysis of the inputs to these classifiers can then be used to determine relationships and factors influencing maintenance activities. This paper investigates two machine learning models - a neural network classifier and a decision tree classifier. Input features for the classifier were the annotated concept tags for solutions, problems and items derived from MWOs of an actual manufacturer. This process for gaining insights can be generalized to various applications in the maintenance and Prognostics and Health Management (PHM) communities.

Proceedings Title

Proceedings of the Annual Conference of the Prognostics and Health Management Society 2019

Conference Dates

September 21-26, 2019

Conference Location

Scottsdale, AZ

Conference Title

Annual Conference of the Prognostics and Health Management Society 2019

Pub Type

Conferences

Download Paper

Local Download

Keywords

maintenance workorder, machine learning, predictive model, data analysis

Citation

Madhusudanan, F. , Sharp, M. , Brundage, M. and Sexton, T. (2019), Studies to Predict Maintenance Time Duration and Important Factors From Maintenance Workorder Data, Proceedings of the Annual Conference of the Prognostics and Health Management Society 2019, Scottsdale, AZ, [online], https://tsapps.nist.gov/publication/get_pdf.cfm?pub_id=928112 (Accessed May 10, 2024)