Real-Time Estimation of Slag Chemical Composition in Direct-To-Blister Flash Furnace Using a High-Temperature Optical Probe

This chapter explores the implementation and validation of an optical probe system supported by deep learning models to monitor slag copper content in real-time within a direct-to-blister flash-smelting process. The study highlights the challenges faced in copper smelting, including high temperatures, corrosive gases, and dust, which complicate the use of traditional measurement techniques. The optical probe system employs Planck’s radiometry to calculate slag surface irradiance, temperature, and spectral emissivity, which are then used as inputs for deep learning models. The system was validated in a real-case scenario at Olympic Dam’s direct-to-blister flash furnace, demonstrating its effectiveness in providing real-time estimates of slag copper content. The integration of process data, such as copper concentrate feed rate, further enhances the model's accuracy, reducing prediction errors. The chapter also discusses the design and installation of the optical probe, ensuring it can withstand the harsh operating conditions. The results show that the pure radiometric model has a mean absolute error of 1.07% with a standard deviation of 1.37%, while the feed rate-assisted model improves accuracy with a mean absolute error of 0.90% and a standard deviation of 1.18%. This study underscores the potential of combining optical sensing and deep learning to improve process control and efficiency in copper smelting.