Diagnosis Transformator Menggunakan Machine Learning dengan Metode Indeks Kesehatan (Studi Kasus di PT. PLN Nusantara Power Muara Karang)

Abstract

Transformer oil is one of the most important parts which functions as an insulating material, coolant, and heat conductor from the hot parts (coil and core) to the tub walls. Transformer oil testing aims to determine the quality of transformer oil, using data from the test results as basic information to predict transformer maintenance. The condition of the transformer which is influenced by chemical, electrical, and mechanical factors results in different ratings depending on the type of test performed, resulting in different conclusions about the condition of the transformer. The soundness index method is a method that can diagnose transformers by calculating the parameters of dissolved gas, insulating transformer oil, and insulating furan (paper). The 3 tests carried out were: Dissolved Gas Analysis (DGA) testing using the Total Dissolved Combustible Gas (TDCG) method, water content testing, and breakdown voltage (BDV) testing. The objects studied were 3 transformers that had different voltages, namely the GT 2.1 Excitation Transformer, the GT 2.1 Main Transformer, and the GT 2.1 UAT Transformer. Where the results of these three tests are entered into Machine Learning which has been programmed based on calculations using the soundness index method to diagnose the state of the transformer. The DGA test results of the GT 2.1 Excitation Transformer using the TDCG method stated that the dissolved gas concentration in the transformer oil was in good condition, although there was C2H6 gas whose concentration exceeded the dissolved gas concentration limit, while the GT 2.1 Main Transformer and GT 2.1 UAT Transformer stated that The concentration of dissolved gases in transformer oil is in good condition and no dissolved gases cross the concentration limit. The results of testing the Water Content and Breakdown Voltage (BDV) of the three transformers stated that the water content and breakdown voltage of each transformer tested were in good condition or still within the permissible limits. The results of the Machine Learning test using the Health Index Method conducted on the three transformers stated that the final condition of the three transformers was in very good condition and the predicted age of the transformer was more than 10 years. In facts, based on data the age of the three transformers are more than 10 years.