A Deep Neuronal Network-Based Sensor for the Detection of Oxygen inside the Carbonization Furnace

Abstract

Agricultural production generates biological surpluses that can be used to produce additional products through thermal and/or chemical transformations, such as the carbonization of vegetables. The oxygen on the vegetable material during the carbonization process inside a furnace is an important parameter that determines not only the success of the process but also the quality of the final product. It is difficult to measure the oxygen inside the furnace in real time, partly because of the working environment of the sensor, and partly because of the operating characteristics of the furnace (continuous rotation on its axis). The goal of this project is to develop a reliable measurement system capable of operating in real-time. For the continuous and precise detection of the temperature inside the furnaces of a carbonization plant, we propose a method based on the characterization of the image inside the furnace using a deep neuronal network. First of all, the images of the interior of the furnace are captured through a digital camera in front of the material and the axis of rotation of the furnace. Then, the area of interest in each frame of the video is determined by image processing. Oxygen content information is then obtained using a deep model trained for the same furnace with pattern oxygen level sensors. Experiments conducted on actual operating conditions of the furnace demonstrate that the proposed method for estimating the working oxygen provides reliable data for the control of plant operation. The proposed measurement scheme demonstrated high reliability against the many changes present inside the furnace, also, its low computational consumption makes it a viable strategy for embedded implementation and operation in real-time.