Early stage white etching crack identification using artificial neural networks

Abstract White Etching Cracks (WEC) in gearbox bearings is a major concern the wind turbine industry, which can lead to premature failure of gearbox. Though many hypotheses regarding generation WEC have been proposed over decades, answer still disputable. To trace back failures earlier stages before they occur, an innovative sensor-set has utilized on test rig monitor influencing factors that WEC. This paperwork seeks recognize abnormal patterns from recorded sensor data and derive statements sensible combinations early detection. A Long Short Term Memory (LSTM) network-based autoencoder for anomaly detection (AD) task. Employing auto-associative sequence-to-sequence predictor, model trained reconstruct normal time series without The reconstruction error testing evaluated determination its anomaly. results show specified LSTM framework qualitatively distinguish anomalies collected multivariate data. Moreover, score via reconstruction-error-based metrics discriminate behaviors study. investigation’s entail significant step towards risk more cost-efficient technology if this approach be further applied stream with plausible thresholds monitoring system.