Optimizing Thermoplastic Starch Film with Heteroscedastic Gaussian Processes in Bayesian Experimental Design...

Area of Science:

• Materials Science
• Polymer Science
• Chemical Engineering

Background:

• Thermoplastic starch (TPS) films are key for sustainable plastics but traditional development methods are inefficient.
• Existing Bayesian optimization (BO) methods struggle with the inherent noise in physical experiments.

Purpose of the Study:

• To introduce a heteroscedastic Gaussian process (HGP) model within BO for improved TPS film development.
• To optimize TPS film composition for maximum elongation at break and tensile strength.

Main Methods:

• Developed a novel BO framework incorporating an HGP model to handle experimental data uncertainty.
• Prepared TPS films using potato starch, glycerol (plasticizer), and acetic acid (catalyst).
• Conducted tensile tests (ASTM D638) and analyzed results to identify optimal formulations.

Main Results:

• The HGP-BO model efficiently identified optimal TPS formulations within 30 experiments and five iterations.
• Optimal elongation at break (96.7%) achieved with 4.5 wt% plasticizer and 2.0 wt% starch.
• Optimal tensile strength (2.77 MPa) achieved with 0.5 wt% plasticizer and 7.0 wt% starch.

Conclusions:

• The HGP model within BO significantly enhances experimental efficiency for material development.
• This approach shows promise for optimizing TPS films and other material formulations.
• The developed method effectively addresses heteroscedastic noise in physical experiments.