A Virtual Sensing approach for approximating nonlinear dynamical systems using LSTM networks

In this contribution, we introduce a hybrid model for virtual sensing applications which combines frequency response function with Long Short-Term Memory network. It estimates the behavior of non-linear dynamic systems multiple input and output channels by generating predictions on short subsequences signals recombining them using windowing technique. The approach is tested an experimental dataset composed measurements from 3-component servo hydraulic fatigue test bench. parameterized noise data, while serviceloads variable amplitudes are used validation testing.