Essentially, power transformer was becoming the indicator of the early industrial revolution. The popularity and flexibility of the AC system are extensive, owing to its ability to transmit electrical energy beyond human imagination during this period. Therefore, the premature breakdown is not tolerable for power utility providers caused by insufficient maintenance or unsolved abnormality during the operation or manufacturing phase with condition-based maintenance regime. Recorded by CIGRE, 58% power transformer breakdown contributed by insulation failure that associated with partial discharge. In the power transformer, insulation parts are the indicator of power transformer health. According to the literature, no study has been conducted to create partial discharge (PD) detection utilizing a single microstrip patch antenna that is combined with a noise filter. Other researchers in the PD recognition literature analysis plainly focused on the kind of PD rather than PD severity. There appear to be two separate research gaps in this field, as well as the possibility to fix this problem. These practices endanger power transformer reliability and stability since power transformer PD activity is generally evaluated using either an integrated sensor or oil sample. This antiquated approach may be the cause of premature power transformer failure. In this thesis, the research targeted to develop a new partial discharge antenna at ultra-high frequency range between 1.5 until 2.6 GHz. The PD antenna designed and simulated using the Computer Simulation Technology Studio environment to gather important parameters before being fabricated using selected FR-4 substrates. The important modification on the antenna is on its ground plane, the feed leg, embedding with slots or cavity, and adding the final parasitic patch before characterized and verified on its important parameter and operational performance in the Anechoic chamber. The data collected from the fabricated antenna in lab scaled power transformer are then used to classify the of PD severity using Self Organizing Maps hybrid with statistic extraction features technique with advance U-Matrix PD mapping. As a result, the proposed antenna design is fabricated accordingly before being subjected to rigorous testing in the anechoic chamber to gather essential parameters using a vector network analyzer (VNA) and contrasting with the CST simulation results. The proposed PD antenna are verified in the lab-scale modelled power transformer, and the ability are compared to the conventional monopole antenna with both detected PD signal at approximate 68 μs. The new and compact PD antenna design that integrated with parasitic noise filter efficaciously generate improvement in detection ability with minimum 25 % in noise reduction. While, in the hybridization of SOM implementation with extracted statistical features, the recognition is faster for about 4 s compared to the conventional phase-resolved partial discharge method that achieved results within 15 s. To conclude, the proposed PD antenna successfully to be yield better results compare to conventional antenna method with better noise rejection with a suitable operational bandwidth, and PD SOM recognition produced better and fast classification results for novel PD severity assessment.
