The health condition of power transformers plays a vital role in ensuring reliable power transmission. As the most valuable substation asset, representing up to 60% of investment, its failure can result in significant operational and financial losses. PLN has adopted a Health Index (HI) framework based on 30 diagnostic parameters, which often require both online and offline testing. However, routine access to complete testing data can be limited in practice. This study proposes a simplified yet effective method to estimate the transformer’s Health Index using more accessible parameters, including Dissolved Gas Analysis (DGA), oil quality, load profiles, and environmental conditions. Supervised machine learning techniques, Multiple Linear Regression (MLR) and Elastic Net Regression were used to develop predictive models for HI. The Elastic Net model outperformed MLR, achieving higher accuracy and better generalization, particularly in handling multicollinearity and sparse data. The predicted Health Index was not intended to replace PLN’s official HI but used it as a training baseline. The resulting HI predictions were then utilized to compute the Probability of Failure (PoF) through logistic regression, estimate the Apparent Age via multivariate modeling, and calculate the Remaining Useful Life (RUL) using a validated Weibull distribution. These integrated outputs form a comprehensive and interpretable framework for condition-based maintenance and risk-informed asset management. The proposed approach provides utilities with a practical tool to estimate transformer condition, even in cases of limited testing access, and supports more targeted, data-driven maintenance planning.

Keyword: Power transformer, health index, load profile, climate, probability of failure, apparent age, remaining useful life