##
365412 Dynamic Simulation, Sensitivity and Uncertainty Analysis of a Demonstration Scale Lignocellulosic Enzymatic Hydrolysis Process

Lignocellulosic agricultural wastes are transformed into bioethanol following 4 major steps: pretreatment, enzymatic hydrolysis, fermentation and separation. During the pretreatment process, lignin is relocated allowing cellulose and hemicellulose to be exposed for liquefaction. Nowadays enzymes are capable of hydrolysing both hemicellulose and cellulose fibers by a complex biochemical competitive conversion mechanism, which was thoroughly described and dynamically modeled in [1]. The enzymatic hydrolysis kinetics have 26 parameters, indicating an over parameterized model. The sensitivity analysis is carried with respect to kinetic parameters in order to determine the most significant parameters for model fine tuning. The uncertainty analysis is performed with respect to both kinetic and feed parameters in order to assess the accuracy of the model predictions. Real measurements can be disturbed by non-zero mean noise, e.g. a NIR sensor which is not calibrated correctly, and uncertainty of feed parameters becomes important.

This study offers insights into the methodology applied to assess the sensitivity and uncertainty analysis of the enzymatic hydrolysis dynamic model from [1]. The sensitivity analysis follows the methodology described in [2], where the delta mean square of sensitivity measure of a parameter is computed. All parameters are ranked with respect to this measure and a subset with the most significant parameters is built. The model calibration procedure follows and is regarded as a fine-tuning around the nominal values of these parameters taken from [3]. The demonstration scale real data were collected in closed loop operation and not under experimental design. This is why the data are not suitable for comprehensive system identification.

The uncertainty analysis follows the methodology as described in [4]: (1) define input uncertainties with their range; (2) sampling of kinetics and feed parameters using the Latin hypercube sampling (LHS) with correlation control; (3) run Monte Carlo simulations with sampled values; (4) evaluate results. The last step also includes a sensitivity analysis using linear regression of Monte Carlo outputs, also known as the standardized regression coefficients (SRC). Two sources of uncertainty are considered: kinetic parameters and feed composition (mass). The kinetic parameter uncertainty is defined as in [4], while the correlation matrix is taken from [5].

The feed composition uncertainty is considered as the most significant sources of uncertainty which needs to be treated comprehensively. The feed composition is obtained from measurements extracted from a NIR sensor and a mass inflow indicator. The mass inflow indicator was known not to be reliable as it had many offsets in reality, reaching negative values quite often in the absence of any inflow, indicating a miss calibration. Cellulose and xylan content plus inflow measurements are considered as relevant feed parameters. The mean value, or the offset of these variables is assumed to follow a Gamma distribution while the standard deviation is assumed to have a Gaussian distribution. The offset is set to 5-10% of the nominal operational value. Therefore, the uncertainty on the feed composition was effectively characterised by appropriate statistical distribution functions as mentioned above. A sample on feed measurements is generated using LHS sampling technique on 2 parameters, i.e. the probability of the mean value and the probability of the standard deviation. The SRC method follows the standard procedure from [6].

The sensitivity analysis shows that the parameter set can be reduced to 11 significant parameters out of 26 in total. The reaction rates constants for cellulose to glucose, cellulose to cellobiose, and xylan to xylose are important together with the inhibition role of glucose and xylose formation on these rates. Enzyme adsorption onto cellulose was also found to be significant. Samples from the outstream were grabbed and analyzed in the laboratory every 6 hours. The HPLC data fit well within the 5th-95th percentile interval of model predictions. The SRC coefficients show that most of the output uncertainty is due to kinetic parameters, which makes sense in a demonstration scale setup with large reactors that buffers any feed variations.

References:

[1] Prunescu, RM, Sin G. Dynamic modeling and validation of a lignocellulosic enzymatic hydrolysis process - A demonstration scale study. Bioresource Technology, 2013; 150:393-403.

[2] Brun R, Reichert P, Künsch HR. Practical identifiability analysis of large environmental simulation models. Water Resources Research, 2001; 37:1015–1030.

[3] Kadam KL, Rydholm EC, McMillan JD. Development and validation of a kinetic model for enzymatic saccharification of lignocellulosic biomass. Biotechnology Progress, 2004; 20:698–705.

[4] Morales-Rodriguez R, Meyer AS, Gernaey KV, Sin G. Dynamic model-based evaluation of process configurations for integrated operation of hydrolysis and co-fermentation for bioethanol production from lignocellulose. Bioresource Technology, 2011; 102:1174–1184.

[5] Sin G, Meyer AS, Gernaey KV. Assessing reliability of cellulose hydrolysis models to support biofuel process design - identifiability and uncertainty analysis. Computers & Chemical Engineering, 2010; 34:1385–1392.

[6] Helton J, Davis F. Latin hypercube sampling and the propagation of uncertainty in analyses of complex systems. Reliability Engineering & System Safety, 2003; 81:23–69.

**Extended Abstract:**File Uploaded

See more of this Group/Topical: Computing and Systems Technology Division