Pattern Recognition in Industry

A commercial-scale fluidized bed reactor was to be designed based on pilot plant experience. Experimenting with the pilot plant yielded operating conditions that resulted in very good performance. In order to scale up the pilot plant results, it was necessary to determine the flow regime associated with the successful operating conditions. In this case the multi-phase flow regime is characterized by the following five parameters:
liquid and gas dispersion coefficients;
liquid and gas (actual, rather than superficial) velocities;
two-phase interfacial area.
A hydrodynamic model was constructed to simulate the flow process in the reactor. The inputs to this model included these five parameters, and the model would then predict the concentration profiles of a tracer, introduced at the reactor inlet, as functions of time and location along the length of the reactor. The flow regime would be established by those values of the five parameters that produced the closest fit between the model's tracer predictions and the experimental results of actual measurements made of the tracer concentrations.
[50]Collaboration with R.M. Koros and P.R. Ponzi.
The simulation model's formulation involved very complex non-linear differential equations. The model's non-linearity and complexity precluded the possibility of inverting it so that the tracer results could be fed in as inputs and the values of the flow parameters obtained as outputs. The only way to obtain the flow parameter values would have been by trial and error; and it is an understatement to say that that is not an attractive...