Novel prediction and controlling software for improved BHK-21 cell culture process based on soft sensing

The density of living cells in BHK-21 suspension cultures is directly related to the production efficiency of foot-and-mouth disease vaccine (FMDV), whereby the cell density is mostly influenced by the concentration of key substrates. Cell density needs to be monitored by analysis of samples withdrawn artificially, which causes large time delays. To overcome this and allow real-time control, an intelligent algorithm called dynamic relevance vector machine (DRVM) based on neural computing theory was proposed for the key state variables. The dominant variables of the model were based on analysis of the process mechanisms. The corrcoef( ) function (belonging to the correlation coefficient command in MATLAB), used to analyze the correlation among environmental variables, and the auxiliary variables of the soft sensing model were further determined. The dynamic weighted relevance vector machine algorithm based on the angle of right triangle was proposed. The maximum likelihood estimate method was used to train the weights and parameters of the DRVM model. By combination of the developed DRVM model with the static RVM model, advantages were established such as no need for adjustment of normalized parameters and non-sensitive parameters, a kernel function that is not restricted by the Mercer condition, a sparse solution, etc. The DRVM model is based on a Bayesian framework; it combines the dynamic information of the industrial process with static modeling, thus improving the prediction accuracy of the model. In order to realize real-time prediction of the key state variables, a cell culture monitoring interface was developed using the MATLAB Script node on a LabVIEW virtual instrument platform. Finally, data exchange between the DRVM soft sensing modeling for the key state variables and the monitoring interface were enabled. The operation results show that the prediction accuracy of DRVM soft sensing is very high and the developed system allows user-friendly control of cell culture conditions for FMDV production.