NARX-model-based parameter estimation for batchwise rotary drying processes

For rotary drying processes in batch production, setting appropriate dryer parameters at the beginning of a batch is critical to the product quality. This paper proposes a data-driven parameter estimation framework based on a nonlinear autoregressive with exogenous input (NARX) model to find optimal parameter settings for each batch. The framework utilizes the online measurements at the head stage to acquire additional information on the current batch and applies dimensionality reduction techniques to reduce the requirement for the training data size. The framework was realized with the neural network and XGBoost algorithms offline, and the model with the smallest estimation error was selected and deployed on-site to estimate the parameters for a real rotary dryer in a food production line. Results show a reduction of the parameter estimation error by about 27.6% against manual-setting strategies and improved product quality by the proposed framework.