Remaining Useful Life Prediction Under Multiple Operating Conditions Based on a Novel Dual-Layer Temporal Convolutional Network

Modern mechanical equipment often works under multiple operating conditions. Ignoring the impact of conditions on the remaining useful life (RUL) of equipment will reduce the prediction accuracy. Based on this, a novel prediction method that properly integrates multidimensional state monitoring data with operating condition data is proposed to evaluate the impact of operating conditions on equipment RUL. First, to extract feature representations incorporating operational information, a dual-layer temporal convolutional network (TCN) structure is designed to simultaneously model the state data and operating condition data at the current time. A convolutional self-attention mechanism (CSA) is also constructed in this structure to fully capture spatiotemporal information in state data. Then, the time-varying future condition probabilities are computed based on the future condition data and integrated with the extracted features to mitigate the impact of future condition changes upon equipment RUL. Finally, extensive experiments conducted on two datasets demonstrate the excellent prediction performance of the proposed method under multiple operating conditions.