Article ID: | iaor19982627 |
Country: | United States |
Volume: | 43 |
Issue: | 6 |
Start Page Number: | 827 |
End Page Number: | 840 |
Publication Date: | Jun 1997 |
Journal: | Management Science |
Authors: | Miller David M., Gopalakrishnan Mohan, Ahire Sanjay L. |
Keywords: | programming: integer, statistics: regression |
The dynamic nature of an operating environment, such as machine utilization and breakdown frequency results in changing preventive maintenance (PM) needs for manufacturing equipment. In this paper, we present an approach to generate an adaptive PM schedule which maximizes the net savings from PM subject to workforce constraints. The approach consists of two components: (a) task prioritization based on a multi-logit regression model for each type of PM task, and (b) task rescheduling based on a binary integer programming (BIP) model with constraints on single-skilled and multi-skilled workforce availability. The task prioritization component develops a multi-logit regression for machine failure probability associated with each type of PM task at the beginning of the year, using historical data on machine utilization, PM, and machine breakdowns. At the start of each PM time-bucket (e.g. a month), we use the updated machine failure probability for each candidate PM task to compute its current contribution to net PM savings, which indicates its current priority. The task rescheduling BIP model incorporates the priorities in selecting tasks for the current bucket to maximize PM effectiveness subject to workforce availability, yielding an adaptive and effective PM schedule for each time-bucket of the master PM schedule. We examine the effect of using multi-skilled workforce on the overall PM effectiveness, and also provide an illustration from a newspaper publishing environment to explain the use of the approach. We have developed four heuristic algorithms to yield good solutions to large scale versions of this scheduling problem. The heuristics perform extremely well, and the best heuristic solution is within 1.4% of optimality on an average.