A thesis submitted in partial fulfillment of the requirements for the degree of Master of Engineering in Energy, School of Environment, Resources and Development

Thesis (M. Eng.) - Asian Institute of Technology, 2020

The power transformer is a crucial asset in the electrical power system since high cost and failure can lead to system failure. Power transformer maintenance management is asset management that can prevent unexpected failure and extend the power transformer's useful life, saving the replacement cost. This study presents reliability-centered maintenance that uses machine learning to predict future conditions and make the maintenance plan. LSTM (Long Short Term Memory) predicted 2020 power transformer component index from 2019 maintenance test record then use SVM, Decision tree, Random forest, and k-NN algorithm to classify to 2020 maintenance condition. Compared with the traditional technique such as time-based maintenance plan and reliability-center maintenance plan using the calculation in weighing and score technique. The cost of maintenance determined by combining routine maintenance cost 35,371 Baht and replacement cost 791,800 Baht, which calculates the depreciation expense of installing a new power transformer based on 20 years of useful life. In this study, the 2020 maintenance plan presents that Reliability-Centered maintenance with Long Short Term Memory prediction and random forest classification technique can predict the future power transformer conditions and classify them into suitable maintenance activities. These techniques make an effective maintenance plan that plans the suitable maintenance for power transformer, reduce maintenance costs, and extend the power transformer useful life.