A fault diagnosis method for fermentation process

Process fault diagnosis requires the on-line information on process state variables that are often inaccessible in real-time for the processes like a fermentation process. A composite model is proposed, combining a kinetic model of the first principles and a neural network model that models the kinetic model parameters changes, to estimate on line the states. This composite model can retain and enhance the process knowledge, at the same time, avoid the complexity of modeling the entire process by kinetics. The estimated process states from the composite model are then fed to a wavelet network for fault detection and diagnosis. The proposed system is successfully applied to a glutamic acid fermentation process, demonstrating the feasibility and effectiveness of the proposed system.