Tapping beam recognition and measurement with infrared camera : Optimisation of an industrial process using an thermal imaging camera and computer vision algorithms

Abstract

This thesis presents the research and development of an application for recognition and measuring of a ferrosilicon tapping beam using an Infrared (IR) camera. Ferrosilicon production is a highly endothermic process, requiring large amounts of energy to operate; thus, the need for increased efficiency is constant. Using an IR camera for non-intrusive measurements of the tapping beam and developing computer vision algorithms that can identify distinctive process parameters are investigated in this project. Today a videostream from an IR camera overlooking the tapping area at Elkem Thamshavn is directly displayed in the tapping control room with no form of warning labels or data logging. Thus, the motivation for this research is derived from the need for more direct feedback on the tapping beam to the on-site human operators and data logging for the metallurgists. Developing a system that can detect temperature, the shape of the tapping beam and alert operators of unwanted situations is set as the main goals for this project. An application that connects the IR camera to a series of algorithms written in Python and publishes the resulting data to a web-based Human Machine Interface (HMI) is developed and tested on an NVIDIA JetsonNano single-board computer with an Optris IR camera. It incorporates OpenCV, NumPy, Numba, SQLite database, and GStreamer as computing and data communication tools. With testing on a tapping beam test stand, where the molten ferrosilicon is replaced with warm water; the application has proven to give feedback on the beam shape, temperature and status of the tapping beam through the web-based HMI.