ABSTRACT
Data-driven modelling is considered as a promising approach to better understand the operation of EPB shield machines (EPBM) for ultimate use in automation. Recent advances in deep learning has made it possible to detect the faintest patterns of latent relations in EPBM operation data. This paper aims to apply deep learning for an automated recommendation of EPBM controllable parameters. Stacked autoencoders are employed for dimensionality reduction which can, automatically, extract the most effective information measures (features) from EPBM operational data. Extracted features, at any given time are used to predict the controllable parameters for the next time step. EPBM data from the University Link U230 project in Seattle is employed to validate the proposed framework. The results indicate that stacked autoencoder can effectively extract features that represent the entire dataset and the predictive data-driven framework can successfully detect the trends in EPBM controllable parameters.
