Iterative Learning Estimation with Lean Measurements

Precise estimation of parameters has always been an arduous task and parameters estimated at lab scale fail to replicate the output performance at the industrial scale. Also, process models are formulated after a series of assumptions and hence they fail to replicate the plant. Hence, there is a need to refine the parameters using the real time data. Iterative Learning Estimation (ILE) has been proposed in the previous works which updates the parameters using the real time plant output data by minimizing the prediction error. But the real time data is frugal and all the states of the system may not be observable. Thus, we need to improve the established ILE methodology to refine the parameter estimates based on the states which are measurable. In this paper, we develop the ILE methodology for the cases where all the states are not measurable. The effectiveness of ILE has been demonstrated on two case studies; first being the adaptive state estimation of batch reactor and other being the refinement of kinetic parameters of a continuous auto-thermal reformer model. Simulations have been performed to establish the convergence of ILE for both the cases.