Monitoring Method Based on Nonlinear Multi-way ICA for Batch Process

Most batch processes generally exhibit the feature of nonlinear variation. A nonlinear multi-way independent component analysis (MICA) technique was proposed that is multi-way kernel independent component analysis based on feature samples (FS-MKICA) method. This approach first makes three-way datasets of normal batch processes unfolded to be two-way and then chooses feature samples from the large two-way input training datasets. The nonlinear feature space is then transformed to high-dimensional linear space via kernel function and independent component analysis (ICA) model is established in the linear space. FS-MKICA not only extracts the nonlinear feature of batch processes, but also reduces the computational cost based on whole input samples. The simulation results in monitoring fed-batch penicillin fermentation show that FS-MKICA method is effective.