Prognostic Strategy of Boiler Pipe Overheating Failure Through Data Science

The Unit 1–2 Paiton Power Plant, a coal-fired power plant owned by PT PLN Nusantara Power, plays a critical role in the East Java regional 500 KV system and the 500 KV Java Madura Bali (JAMALI) system. Despite its significance, the plant experienced substantial loss output from 2019 to 2021, primarily due to boiler pipe leakage, with five occurrences during this period. These failures, predominantly caused by overheating failure. Moreover, the existing methods for RLA were insufficient for predictive maintenance. This paper propose a Boiler pipe Overheating-Failure Prognostic Strategy leveraging a Data-Science. This approach involves developing algorithms and utilizing anomaly detection models. Additionally, a Real-time Calculated Remaining Useful Life Dashboard is implemented to predict the lifespan of Boiler pipe panels. The methodology encompasses the collection of maintenance, testing and sensor data. This data is processed and analyzed to create algorithmic models. The models are trained and evaluated to ensure optimal performance. Anomaly detection uses a CNN autoencoder model, setting a threshold based on reconstruction error (MAE ± 2.5 x STD), covering about 98.8% of normal data. During training, the MAE values for each variable converged after several epochs, averaging below 0.5. The results of the realtime RUL calculation dashboard have provided a realtime report via TelegramBot and Anomaly Detection model have provided an early-warning dashboard of anomalies before overheating occurs during operation. Thus both results are helpful in resolving the problem of maintenance and operation strategies for Overheating-Failure of Boiler pipes.