Fault Precognition System for Remaining Useful Life Estimation in Bearing Systems Using Autoencoder-LSTM and Clustering Techniques

This paper proposes a fault precognition system designed for predictive maintenance in bearing systems aimed at improving Remaining Useful Life (RUL) estimation accuracy.This study makes use of the Pronostia-bearing dataset, a recognized standard for RUL prediction and predictive maintenance.It includes vibration data captured by accelerometer sensors along two axes (X and Y), which shows how bearings deteriorate under different operation circumstances.The extensive size of the dataset, which includes several bearings experiencing progressive deterioration, guarantees strong validation of the suggested fault precognition and RUL prediction system in actual maintenance situations.The system utilizes the Pronostia bearing dataset, employing time-domain feature extraction, automated feature ranking, and fault pattern classification through Kmeans clustering with Silhouette Coefficients.A core component of the system is an Autoencoder-LSTM model, which identifies early fault occurrences by analyzing reconstruction loss thresholds-quantitative measures of deviation between observed and reconstructed data.These thresholds serve as indicators of anomalous behaviour, distinguishing normal operations from fault-prone data clusters.The system then estimates RUL using various LSTM variants, including Vanilla LSTM, BiLSTM, CNN-LSTM, StackLSTM, ConvLSTM and Encoder-Decoder LSTM, with performance evaluated using Mean Squared Error (MSE) and R scores.The results demonstrate that incorporating fault precognition into the system significantly enhances prediction accuracy, facilitating proactive maintenance and improving operational reliability.