Intelligent Sensor for Thermal Process Control using Convolutional Neural Network

Abstract

A thermal combustion process involves three variables: the fuel (which oxidizes and gives off heat), the comburent (which accelerates combustion), and the heat source. The proportion of these three variables determines the behavior of the process, and in the case of industrial production, their control is fundamental to guarantee continuity, quantity, and quality of the product. In the case of carbonization of plant material, the control of oxygen is decisive to guarantee these production parameters. However, the measurement in real-time becomes a complex problem due to the process temperatures, the requirements of precision and accuracy of the measurement, and the characteristics of the combustion furnace that normally must keep the material in motion. This research proposes an intelligent sensor that allows its remote use and guarantees the constant and safe monitoring of the variable, as well as its conditioning and communication. The sensor is composed of a digital camera aligned with the flame capable of capturing video frames continuously and safely. These digital images are processed by a categorization module previously trained with a convolutional neural network, and the result is transmitted to the control unit. In tests on a real furnace, high performance and reliable operation sufficient for industrial implementation were proved.