Prediction of mass transfer performance of microchannel dialyzers using deconvolution of impulse-response experiments

Type
Thesis
Year of Publication
2009
Authors
Eric K. Anderson
Date Published
Jan. 1, 2009
Publisher
Oregon State University
Abstract

The parallel implementation of a large number of functional units is necessary for any industrial scale microfluidic process. The concept of a 'numbering up' strategy where a single highly optimized functional unit that has a low individual production is replicated a large number of times to create a device that has the necessary output. Designing a system using this strategy assumes that the final device will have the performance characteristics of the individual unit. In reality there will be a distribution of operating conditions clustered around the set point, and this will impact the performance of the overall device. The main operating condition of the microfluidic device under consideration here is the average fluid velocity in each channel. Most techniques that could measure the fluid velocity in each channel require an optical path to the measurement point. For a device with a large number of channels, it is highly unlikely that every channel will be accessible for observation. Even if they were, it would be extremely time consuming to measure each channel individually. Another approach would be to use an impulse response test to infer the velocity distribution; if an adequately narrow input pulse would yield a output pulse that would be a reasonable approximation of the system response function. In the case at hand, the input pulse is too broad to be able to directly infer the velocity distribution from the output pulse. A numerical deconvolution technique was applied to the data to be able to effectively remove the error associated with the input pulse. For sufficiently accurate impulse response data, this method would yield an accurate estimate of the system response function. Once an estimate of the velocity distribution is known, a method for inferring the performance impact is needed. Two approaches were used: 1. A stochastic simulation that directly generates possible device internal states and then calculates the performance; and 2. A theoretical approach based on the performance surface and and assumed velocity distribution form. Both methods require knowledge of the performance surface of an individual channel with respect to the local velocity. To generate this surface a finite volume was developed in FORTRAN that directly simulates a single microchannel pair. The stochastic model predicted a negative performance impact with increasing velocity distribution variance. A theoretical model was developed that calculates the difference between the real and ideal case using the covariance matrix and Hessian as well as provides a framework for predicting the sign of the deviation in advance.