Neural networks as "software sensors" in enzyme engineering

Area of Science:

• Biotechnology
• Enzyme Technology
• Process Control

Background:

• Industrial enzyme production faces challenges in on-line process control due to biological system variability.
• Lack of suitable on-line sensors for critical process variables and quality attributes hinders real-time monitoring.

Purpose of the Study:

• To demonstrate the efficacy of feedforward backpropagation neural networks for on-line control of enzyme production processes.
• To address the limitations of traditional methods in managing biological system uncertainties.

Main Methods:

• Development and implementation of neural network programs using Microsoft Visual C++ for Windows on a personal computer.
• Training feedforward backpropagation neural networks with one hidden layer.
• Utilizing the coefficient of determination (R2) to assess the goodness of fit to reference data.

Main Results:

• Successfully demonstrated that trained neural networks can effectively control enzyme production processes.
• Neural networks provided a solution without needing a priori knowledge of process variable relationships.
• Case studies included beta-galactosidase, glucoamylase, lipase, and xylanase production.

Conclusions:

• Feedforward backpropagation neural networks offer a viable solution for on-line control in industrial enzyme production.
• This approach mitigates challenges posed by biological system uncertainties and sensor limitations.
• The methodology is applicable to various enzyme production processes, enhancing efficiency and control.