Modelling and Flame Segmentation for Real-Time Monitoring of Rotary Kilns

Abstract

Rotary kilns are important industrial plants for the refinery of materials. Effective monitoring of rotary kilns is therefore crucial to the successful operation in order to optimise the flame and reduce waste. In this industrial application-driven work, we describe such a system recently developed and successfully deployed. Our work aims at efficient real-time performance and accuracy for given operating constraints. Our contributions include novel methods that provide accurate depth calibration and mapping, detect and segment the flame, and estimate the flame bed position accurately in challenging conditions. For preprocessing, we use a multi-frame approach and carefully select an optimal window size. Then, we derive a mathematical model for depth calibration and mapping of the kiln and demonstrate the accuracy of the model. Finally, we propose a comprehensive pipeline for segmenting and detecting flame and provide an in-depth analysis of our method. Our evaluation shows that the proposed system performs well and meet real-time requirements whilst using only low-cost hardware.