Implementation of Text Mining-Based Linear Programming in Maintenance Scheduling in Power Plant

Abstract

Weekly maintenance scheduling is an important activity in maintaining the reliability of operating facilities, especially in the Power Plant area in the oil and gas industry sector. This study aims to optimize the maintenance planning and scheduling process by utilizing the text mining approach through Latent Dirichlet Allocation (LDA) modeling to manage and group maintenance data, and integrate it with the Linear Programming (LP) model as the basis for the preparation of an optimal Work Order (WO) Scheduler. The LDA model is used to categorize work based on Fixed Reference Activities (FRA), resulting in a more structured classification of maintenance activities. The output of this category is then an input for the LP model that compiles labor allocation, duration, and work priorities according to the available weekly time limits. Sensitivity analysis was carried out on the parameters of the number of labor, work priority, and length of time horizon with variations of ±10%, ±20%, and ±30%. The results show an increase in the WO completion rate, a reduction in the backlog, and a more accurate understanding of labor utilization. The model has proven to be sensitive to changes in workforce capacity so that human resource management is the dominant factor in the successful implementation of schedulers.