Neural networks are being broadly explored for the identification of Industrial Cyber Physical Systems (ICPS) models from data sequences. However, learned representations typically lack explainability, representing nowadays a major challenge of deep learning. Interpreting the information structured across the synaptic links is particularly challenging for recurrent neural networks (RNN), encoding input features and observed system dynamics within a continuous latent space. In this work, we investigate the representation built within the RNN while learning behavioral models of a class of discrete dynamical systems. To this end, we propose a method to extract the symbolic knowledge structured by the continuous state, based on Gaussian Mixture Model clustering. Experiments are performed on a pilot remanufacturing plant, by learning the model of a conveyor controller from process data. We show the capability of the RNN to achieve accurate predictions while providing a Moore Machine representation of the latent activations, consistent with the target system.
Extracting finite state representations from recurrent models of Industrial Cyber Physical Systems
Brusaferri A;Spinelli S;Vitali A
2020
Abstract
Neural networks are being broadly explored for the identification of Industrial Cyber Physical Systems (ICPS) models from data sequences. However, learned representations typically lack explainability, representing nowadays a major challenge of deep learning. Interpreting the information structured across the synaptic links is particularly challenging for recurrent neural networks (RNN), encoding input features and observed system dynamics within a continuous latent space. In this work, we investigate the representation built within the RNN while learning behavioral models of a class of discrete dynamical systems. To this end, we propose a method to extract the symbolic knowledge structured by the continuous state, based on Gaussian Mixture Model clustering. Experiments are performed on a pilot remanufacturing plant, by learning the model of a conveyor controller from process data. We show the capability of the RNN to achieve accurate predictions while providing a Moore Machine representation of the latent activations, consistent with the target system.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.