Proposal of fault prediction system for underground installations

The absence of intelligent systems for fault prediction in components and equipment in underground installations does not allow a correct structure for maintenance programs to reduce intervening maintenance actions. Thus, it is necessary to devise a computer system that can conjugate potential causes and factors of failure as input data that can correctly inform output data, which represents a component or equipment failure prediction. The aim of this paper is to present a proposal system to support maintenance analyses, where sensors of temperature, water level and vibration, can provide real time data to analyze the isolation condition of transformers and switches within underground chambers. The paper also presents a new use of a neural network to alarm partial discharges (PD) activities. A case is presented where hardware and software are developed and an underground chamber is used as a real test case. In this research, cost, performance, environmental hostility, and the availability of installed equipment in the Brazilian market were considered. After the methodology was verified, the cost was compared with imported equipment, and the result was approximately a 50% cost reduction. Issues of telemetry, hub specification, and sensor installation were also evaluated.