Representation learning for inferential sensor development in an electric arc furnace

Overview Stable smelter operation is critical for successful production of base metals from particulate ore. This work studies the operation of an industrial direct current electric arc furnace that operates as a smelter in a large-scale metallurgical process. Specifically, unexpected loss of the plasma arc is an important unresolved problem with a significant impact on production efficiency and profitability of the mining operation. To address this faulty operation a predictive inferential sensor is proposed to identify high risk operating regimes. Once a high risk situation is identified the alarm instructs operators to take corrective actions to avoid the loss of arc. A comprehensive representation learning framework involving latent variable methods is proposed in this work for preliminary development of the inferential sensor. Large quantities of historical industrial process data are analyzed under the proposed framework to perform fault detection, diagnosis, modeling and predictive classification.