Comparison of global nonlinear models and "model-on-demand" estimation applied to identification of a RTP wafer reactor

"Model on demand" (MoD) simulation of the temperature dynamics in a simulated rapid thermal processing (RTP) reactor is compared against various types of global models (ARX, semiphysical, combined semiphysical with neural net). The identification data is generated from an m-level pseudo-random sequence input whose parameters are specified systematically using a priori information readily available to the engineer. The MoD estimator outperforms the ARX model and a two semi-physical models, while matching the performance of a combined semi-physical with neural net model. This makes MoD estimation an appealing alternative to global methods because of its reduced engineering effort and simplified a priori knowledge regarding model structure.