Kinematic Optimization of a Four-Bar Linkage for Continuous Eggshell Membrane Extraction Using Multi-Objective Optimization and Reinforcement Learning

Abstract

Eggshell membrane is a valuable by-product of the egg-breaking industry, rich in bioactive proteins and polysaccharides used in pharmaceutical and cosmetic products. Conventional separation methods for extracting membranes from eggshells using water or acid baths are inefficient and unsuitable for continuous operations. To address this, a mechanical extraction method can be used for direct integration into egg-breaking machines. A design was proposed to use a four-bar linkage based on the Chebyshev-Lambda mechanism to insert the extraction tool into eggs moving on constant-velocity conveyors typically found on egg-breaking machines. This thesis focuses on optimizing the prototype's linkages through a structured study with five objectives: minimizing velocity fluctuations of the output link effector, achieving path-generation using Fourier harmonic decomposition, optimizing link length magnitudes, and improving the insertion-to-exit velocity ratio of the extraction tool on the effector. Reinforcement learning was evaluated alongside classical global optimization for path-generation. The multi-objective study produced results that included a configuration with an 87% reduction in velocity variation and a 40% increase in theoretical extraction rate. The results were evaluated in a correlation study to assess the relative competitiveness of the different objectives with respect to velocity fluctuation.