Abstract

Spectral measurements offer real-time insights into the composition and concentration of species in process samples. However, they are sensitive to external factors such as temperature, pressure, and particle size distribution, all of which have a significant impact on the precision of spectroscopic measurements. In this study, we introduce an integrated discrete-time nonlinear model considering the dynamic aspects of the process, alongside a physics-based sensor model. Additionally, we propose an innovative application of two alternative Prediction Error Adaptive Kalman Filters to estimate both concentrations and sensor model parameters. The simulation of a simple ternary mixing process enables a comparison of the key characteristics of the proposed adaptive Kalman filters with a standard Extended Kalman Filter. The simulation results show that both Prediction Error Kalman filters can estimate concentrations and sensor parameters with minimal error, even in the presence of temperature variations and measurement noise. However, the proposed filters offer advantages in terms of ease of tuning and convergence.