DIAGNOSISOFPOWERTRANSFORMERSBASEDON ANALYSISOFOILDIELECTRICCHARACTERISTICS, DISOLVEDGASES,FURAN,OPERATINGLIFE,AND LOADINGFACTOR (CASESTUDYATPTRA P)

• Main Author: Benjamin Sibuea, Esosia
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/66812
• Institution: Institut Teknologi Bandung

ABSTRACT DIAGNOSIS OF POWER TRANSFORMERS BASED ON ANALYSIS OF OIL DIELECTRIC CHARACTERISTICS, DISSOLVED GASES, FURAN, OPERATING LIFE, AND LOADING FACTOR (CASE STUDY AT PT RAPP) By Esosia Benjamin Sibuea NIM: 23220070 (Master’s Program in Electrical Engineering) In power transmission and distribution network systems, power transformer is one of the vital and expensive electrical equipment. The failure of the transformer will cause disruption to the electricity supply, which impacts on the availability of electrical energy and operating costs. In the industrial world, the availability of electricity supply is critical to be able to achieve production targets. Therefore, utility or industry must maintain and improve the reliability of transformers by developing a diagnostic system to estimate the condition of transformers accurately. Transformer condition assessment using a health index (HI) approach is one tool that can help achieve this, enabling more efficient asset management through a fast, practical method for assessing and evaluating the overall condition of the transformer. Diagnosis transformer conditions through the HI approach with scoring and weighting method proposed and implemented to power transformer assets in one of the industries in Indonesia. Furthermore, the results of the calculation using the HI approach are compared with the actual health conditions of the transformer in the field to obtain the percentage of accuracy and error rate of the transformer health index. The HI approach with the proposed scoring and weighting method was then developed using multi-algorithm machine learning (ML) based method. The classifier algorithms used are Logistic Regression, Naïve Bayes, Artificial Neural Networks (ANNs), Stochastic Gradient Descent (SGD), Gradient Boosting, Tree, Random Forest, CN2 Rule Induction, Adaboost, Support Vector Machine (SVM) and k- Nearest Neighbor (kNN). In this study, 167 units transformer samples were used, consisting of 16 units of 150 kV transmission transformer and 151 units of 20 kV and 33 kV distribution transformers. Then, these data are used as data training in HI approach-based ML to build prediction and classification models. To test the model, 50 samples of transformers (150 kV voltage) were used as data testing obtained from different utilities. iv Furthermore, the degradation of the insulating paper will affect the characteristics of the oil, the formation of dissolved gases, and furans. Data related to this parameter have been found in several studies, but studies on the correlation between parameters are still limited, especially in distribution transformers. Therefore, in this study, the characteristic databases of dielectric oil, dissolved gases, furan, operating life, and loading factor were analyzed for their correlation to insulation paper degradation of distribution transformers statistically using a linear regression method. Keywords: power transformer, diagnosis, health index, machine learning, insulation paper degradation