< Back to previous page

Project

Decision making in the presence of uncertainty: Industry 4.0 enabled preventive maintenance

Decision making, accompanied by uncertainty in the quality of available information and the decision’s eventual outcome, plays a central role indaily life. In this dissertation, we explore the value of additional information in maintenance decision making and how best we can utilize available data to enhance such decisions. Specifically, we add to the state-of-the-art in maintenance optimization as facilitated by Industry 4.0 technologies. Industry 4.0 promises reductions in maintenance costs through access to digital technologies such as cloud computing, data analytics and, key to this dissertation, the Internet of Things. Many of the promised benefits to maintenance are, however, dependent on the quality of the data obtained through sensors and related technologies. We therefore first consider the effect of access to different levels of deterioration data quality on preventive maintenance policies. In this case, the decision maker has access to costly external sensor information, and is tasked with determining when to stop the system to perform preventive maintenance and when to pay for external information. We analyze the structure of the optimal policy and show that it consists of at most four regions, based on the believed deterioration state of the system. We next investigate whether historical maintenance records and observed machine failures may be used to supplement and improve on existing periodic maintenance practices. In this case, the decision maker may deviate from a planned maintenance schedule and perform preventive maintenance early. We obtain policies that prescribe when to perform such early preventive maintenance and validate our model using an original equipment manufacturer data set. We find that our policies can improve on those currently applied by 5%, on average. Finally, we develop a case study to disseminate our research findings. We use data to analyze the optimal maintenance interval and introduce the reader to the concepts of predictive analytics and the value that it can add to a business setting. The research may be extended in future work by considering human-in-the-loop analytics to protect against bias in the data and to further develop optimal, interpretable decision making policies.

Date:9 Jan 2017 →  28 Jun 2021
Keywords:Preventive maintenance, Condition-based maintenance, Markov decision processes, Decision making, Partial information
Disciplines:Applied economics, Economic history, Macroeconomics and monetary economics, Microeconomics
Project type:PhD project