PROGNOSTICS OF HIGH-SPEED ROTATING MACHINERY USING AI TECHNIQUES

Abstract

Rolling bearings are integral components of rotating machinery, with their failure posing a significant risk of complete production shutdowns. A single fault in a bearing can cascade into substantial breakdowns, highlighting the necessity for an advanced and comprehensive approach for bearing health monitoring. The reliability and safety of modern industrial systems depend heavily on precise prognostic analysis of these high-speed components. This report reviews various signal processing methods, emphasizing the use of time-frequency analysis-derived features for accurate fault prognosis. Bearing degradation analysis was conducted using data from the IEEE PHM 2012 (PRONOSTIA platform). Raw signal data were collected, and both time-domain and frequency-domain features were extracted for prognostic evaluation. Features were meticulously selected based on Monotonicity and Trendability, ensuring the incorporation of relevant information for accurate Remaining Useful Life (RUL) predictions. Accurately predicting bearing degradation trends is essential for estimating RUL, yet this task is inherently challenging due to the complexities of long-term forecasting. To address these challenges, Long Short-Term Memory (LSTM) variants, both individually and in combination with Attention and Transformer mechanisms, were employed. Among the evaluated methods, the Stacked Bidirectional LSTM with Transformer mechanism demonstrated superior performance, offering the most accurate RUL predictions.