Improved ANN AISI and Its Application in Erythromycin Fermentation Process

The dynamic soft-sensing method that we proposed in Xianzhong Dai (2004) is improved by amending the so-called "assumed inherent sensor" (AIS) that plays a central role in this approach, so as to achieve more accurate estimation of some immeasurable crucial variables in erythromycin fermentation process. Different from Xianzhong Dai (2004), the improved AIS enables its inversion (AISI) to avoid the feedback of derivatives of the to-be-estimated variables that can incur amplifications of estimation errors. The above AISI is composed of a nonlinear function and a set of differentiators and can be used as a dynamic soft-sensing model or soft-sensor. Furthermore, to overcome the difficulty in constructing the AISI by analytic means as well as to enhance its robustness to model errors, a static ANN is used to approximate the nonlinear function appearing in AISI, then obtaining the desired dynamic soft-sensing model (i.e. ANN AISI). The ANN AISI has been used to estimate such immeasurable variables in erythromycin fermentation process as mycelia concentration, sugar concentration and chemical potency. The field results show its validity and improvement