Eco-Friendly Plastic Discovery: Robotics and AI Revolutionize Material Design

By Dr Silpaja Chandrasekar, PhDReviewed by Susha Cheriyedath, M.Sc.Mar 27 2024

In a paper published in the journal Nature Nanotechnology, researchers unveiled an innovative solution to combat the environmental burden of petrochemical plastics. They proposed a novel approach that merged robotics and machine learning (ML) to expedite the discovery of biodegradable plastic alternatives with tailored properties.

By employing automated experimentation and predictive modeling, they successfully developed all-natural substitutes that mimicked the performance of traditional plastics. This groundbreaking methodology promised accelerated progress in eco-friendly material design, utilizing safe building blocks from established databases.

Accelerated Plastic Innovation

Petrochemical plastics pose significant environmental challenges due to their low recyclability and high pollution rates. Addressing this issue, researchers developed a pioneering approach, integrating robotics and ML, to expedite the discovery of biodegradable plastic substitutes with tailored properties.

By utilizing automated experimentation and predictive modeling, they successfully identified all-natural alternatives matching the performance of traditional plastics, offering a promising solution to reduce plastic waste. This hybrid methodology streamlines the research process, enabling efficient exploration of eco-friendly materials sourced from safe databases, and holds potential for widespread adoption in sustainable material design.Active Learning Loops

The construction of the prediction model involved initiating active learning loops by preparing mixtures using the OT-2 robot, resulting in 10 nanocomposite films for each loop. These films underwent optical, fire-resistant, and mechanical characterizations, with their properties recorded as labels. Through data augmentation and user input principles, virtual and actual data points were synthesized and used for training the artificial neural network (ANN) model, which predicted the unfamiliarity level of targeted data points for subsequent loops.

During the study, researchers carried out fourteen active learning loops, fabricating 135 all-natural nanocomposite films, and they visualized data points using Voronoi tessellation diagrams. Researchers assessed the accuracy of the multi-property prediction and selected the ANN model with the lowest mean relative error (MRE) as the champion model. This model accurately predicted various nanocomposites' optical, fire-resistant, and mechanical properties, facilitating the discovery of high-strength structural materials using natural building blocks.

Model Expansion Strategy

Researchers employed a model expansion strategy to introduce chitosan as an additional building block to broaden the range of available all-natural plastic alternatives. It involved conducting three active learning loops guided by the prediction model to integrate chitosan loading into the existing model architecture.

One hundred thirty-three experiments were performed during this expansion phase, significantly reducing the MRE from 107% to 21%. Incorporating chitosan notably enhanced the mechanical properties of the substitutes, increasing ultimate strains. Leveraging this expanded model, researchers successfully produced two new all-natural plastic substitutes with improved mechanical properties, catering to specific applications such as clear file folders and transparent air pillows.

Subsequently, researchers utilized the champion model to facilitate the inverse design of all-natural plastic substitutes with customizable physicochemical characteristics. This approach enabled the automated generation of substitutes tailored to replace various plastic products, ensuring optical transparency, fire retardancy, and mechanical resilience in line with diverse design requirements.

The model-recommended compositions facilitated the large-scale production of these substitutes, demonstrating their potential for widespread adoption. Biodegradability tests confirmed the effectiveness of these all-natural substitutes. At the same time, model interpretation techniques like Shapley additive explanations (SHAP) analysis provided insights into the complex composition-property relationships, further enhancing the understanding of the design process and the influence of individual components on material properties.

Experimental Validation: Confirming Trends

The investigation focused on understanding the strengthening mechanisms between cellulose nanofibril (CNF) chains and montmorillonite (MMT) nanosheets through molecular dynamics (MD) simulations, examining CNF-only, MMT-only, and MMT/CNF models. Findings revealed distinct failure mechanisms, with the MMT/CNF model demonstrating higher tensile strength due to localized tensile deformation.

Experimental validation via thin-film samples supported these findings. Sensitivity analyses explored the impact of various building block attributes on end-product properties, highlighting significant influences of gelatin source and MMT size on optical properties. This comprehensive approach provided insights into the complex relationships between building block attributes and end-product characteristics, facilitating the design of all-natural nanocomposites with tailored properties.

Materials and Preparation Processes

Researchers used various materials received in the methods, including MMT, northern bleached softwood kraft (NBSK) pulp, 2,2,6,6-tetramethylpiperidine-1-oxyl (TEMPO), sodium bromide, sodium hypochlorite solution, sodium hydroxide, gelatin, and glycerol.

Researchers prepared MMT and CNF dispersions by subjecting them to ultrasonication and centrifugation, respectively, while they dissolved gelatin and glycerol in water to create solutions. They fabricated an automated pipetting robot all-natural nanocomposite film with different MMT/CNF/gelatin/glycerol ratios.

The quality of these films was assessed through detachment and flatness testing. The authors evaluated support vector machine (SVM) and ANN-based prediction models for accuracy, and various characterizations were conducted, including film thickness, transmittance spectra, fire resistance, and mechanical properties. They performed biocompatibility tests on cultured cells. Researchers conducted MD simulations using the reactive force field (ReaxFF) potential in large-scale atomic/molecular massively parallel simulators (LAMMPS) to study molecular behavior under tension.

Conclusion

To sum up, researchers developed an unconventional design platform, integrating automated robots, machine intelligence, wet-lab experiments, and simulation tools to discover a diverse library of all-natural nanocomposites serving as biodegradable plastic alternatives with customizable optical, fire-resistant, and mechanical properties.

Despite significant progress, challenges persisted, including the need for fully automated robotic systems, ensuring consistent quality of natural building blocks, integrating cost and life cycle analyses into the design process, and addressing end-of-life processing considerations for the all-natural substitutes. Addressing these challenges further enhanced the efficiency and sustainability of the artificial intelligence (AI)/ML-integrated workflow for accelerated material design.