Prediction of Remaining Useful Life of Bearings using a Parallel Neural Network

This study advocates the utilization of a parallel neural network (PNN) architecture for the estimation of remaining useful life (RUL) of bearings. The use of conventional machine learning and deep learning techniques has been inadequate in terms of accuracy and computation time, because of huge input data sizes and the time-dependent nature of the output. To address this limitation, the PNN architecture incorporates multiple parallel processing paths with multiple input neurons that take in data from condition detectors of mechanical machines and output neurons that predict RUL. The PNN structure provides better accuracy and computation time by efficiently handling vast amounts of data and integrating both spatial and temporal information simultaneously. Additionally, time-transformer and recurrent neural network (RNN) are used to handle complex time series data. Improvement methodologies like positional encoding with self-attention mechanism and ConvLSTM neural network are utilized to leverage multidimensional time-frequency data to process spatial and temporal dependencies present in the extracted features, further increasing model's efficiency. A case study is conducted on XJ-SY rolling element-bearing dataset to validate the proposed methodology, where PNN performed exceptionally in terms of accuracy and efficiency. It is concluded that PNNs exhibit potential for predicting RUL of bearings and can be applied to other machinery types.

Keywords: Remaining Useful Life (RUL), Parallel Neural Networks (PNNs), Bearings, Time-transformer, ConvLSTM, Positional encoding, Self-attention.

Download PDF