Adaptive nonlinear system identification of separation unit for raw materials

Abstract

The separation process in combine harvesters is a complex, nonlinear dynamic system influenced by varying crop types and environmental factors. Optimal operation requires models that adapt to these changing conditions, posing a significant challenge. This paper presents an automation system designed to assist operators by identifying the optimal operating point using adaptive process models. A nonlinear online system identification method is developed, featuring a local model network (LMN) with local regularized finite impulse response (RFIR) models. The local model adaptation employs a regularized recursive least squares algorithm extended with a variable forgetting factor, ensuring robust and rapid adaptation across different excitation scenarios. A novel affine output transformation enhances the model’s global adaptability, enabling efficient updates in new field conditions. A two-phase adaptation mechanism is proposed, starting with a rough global adaptation followed by more detailed refinements induced through updating the local model parameters based on current conditions. Experimental validation using real-world data demonstrates the model’s superiority in adaptation speed, prediction accuracy, and robustness across diverse field scenarios. The study also emphasizes the crucial role of process excitation in enhancing model accuracy, with implications for automated system performance.