Developing a Cellular Neural Network based control system for a coal froth flotation process

摘要

A cellular neural network (CNN) is a massively parallel, locally connected neural network. Their structure can be utilised for quick and efficient image pre-processing operations based on neighbourhood properties. The washing of coal prior to its burning or gasification helps to remove most of the minerals attached to it providing a cleaner combustion whilst limiting the production of noxious gas. Conventionally the process is supervised by an operator who adjusts the flows of water, coal, air and surfactant The aim of this project is to assess the feasibility of a control system based on video and image analysis for frothing flow systems in industrial production. To do so, the froth flotation process is physically modelled as a bubble flow in a batch column using soap bubbles and an adjustable air-flow. The control system can be simply described as a video camera coupled to a frame grabber and a computer application designed to act as a controller. The PC based application uses CNNs to perform computationally efficient image processing operations. Once the image analysis phase is completed, the information extracted from the image is sent to the mathematical model and eventually adjusts the flow accordingly. The design of the fluid mechanics model is not the focus of the present article, so only an overview is presented. The purpose of this article is to show the problems encountered during the design of a computer application that deals with the inclusion of non-standard hardware, in this case a CNN board.