Control strategy of a Multiple Hearth Furnace enhanced by machine learning algorithms

摘要

An enhanced control strategy for a multiple hearth furnace for the purpose of kaolin production is developed and presented in this paper. Mineralogy-driven machine learning algorithms play a key role in the optimization strategy of the furnace. First, the capacity and temperature setpoints for furnace control are determined based on the feed ore mineralogy. Next, the capacity is optimized by combining the prediction of soluble alumina content and mullite content, while maintaining the quality of the product. The stabilizing control level compensates the disturbances with a feedforward control, which uses a spinel phase reaction rate soft sensor, aimed at minimizing the energy use of the furnace. The control concept is successfully tested by simulation using industrial data. Finally, a sampling campaign and software testing of the soft sensors and machine learning algorithms are performed at the industrial site. The results are presented and discussed in the paper.