SELF-OPTIMIZED LONG SHORT-TERM MEMORY FOR PREDICTING REMAINING USEFUL LIFE OF MACHINES

Remaining Useful Life (RUL) is the period from the current time to the time when a machine fails to operate. Unexpected machine failure brings critical damages to the industry, such as loss of investment in assets and high unplanned maintenance costs. Machine failure and RUL of machines can be predicted by using Recurrent Neural Network- Long Short Term Memory (RNN-LSTM) to avoid such consequences. RNN-LSTM utilises the long period of time series data recorded by machines without any memory issues due to LSTM. Long-Short Term Memory
unit (LSTM) is an improvement of the Recurrent Neural Network (RNN) as RNN faces issues
predicting long-term dependencies. Issues such as vanishing and exploding gradients result
from backpropagating errors, taking place when the network is learning to store and relate
information over extended time intervals. LSTM selective memory mechanism only reserves
space for significant inputs. Thus, RNN-LSTM suits time-series predictions such as machine
failure prediction. The performance of RNN-LSTM, specifically accuracy, depends heavily on
the weight training algorithm and network hyperparameter topology.