Designing Macro- Molecules With ML

Abstract

Nowadays, polymer reaction engineers seek robust and effective tools to synthesize complex macromolecules with well-defined and desirable microstructural and architectural characteristics. Over the past few decades, several promising approaches, such as controlled living (co)polymerization systems and chain-shuttling reactions have been proposed and widely applied to synthesize rather complex macromolecules with controlled monomer sequences. Despite the unique potential of the newly developed techniques, tailor-making the microstructure of macromolecules by suggesting the most appropriate polymerization recipe still remains a very challenging task. In the current work, two versatile and powerful tools capable of effectively addressing the aforementioned questions have been proposed and successfully put into practice. The two tools are established through the amalgamation of the Kinetic Monte Carlo simulation approach and machine learning techniques. The former, an intelligent modeling tool, is able to model and visualize the intricate inter-relationships of polymerization recipes/conditions (as input variables) and microstructural features of the produced macromolecules (as responses). The latter is capable of precisely predicting optimal copolymerization conditions to simultaneously satisfy all predefined microstructural features. The effectiveness of the proposed intelligent modeling and optimization techniques for solving this extremely important ‘inverse’ engineering problem was successfully examined by investigating the possibility of tailor-making the microstructure of Olefin Block Copolymers via chain-shuttling coordination polymerization.

Keywords: microstructure, Kinetic Monte Carlo, living copolymerization, olefin block copolymers, artificial intelligence, ethylene, machine learning, genetic algorithms

1. Introduction

Polymers are ubiquitous in daily life, with polyolefins being the most widely used type. The development of chain-shuttling polymerization has expanded the property range of polyolefins, allowing for the production of block copolymers with unique properties. However, designing olefin block copolymers (OBCs) with specific properties is a complex task due to the numerous interacting factors involved in the polymerization process. Kinetic Monte Carlo (KMC) simulations have been used to gain an in-depth understanding of the microstructure of OBCs, but these simulations cannot suggest polymerization variables that can yield a certain desired microstructure. To address this challenge, the use of Artificial Intelligence (AI) modeling and optimization techniques is proposed. The aim of the current study is to develop unique and versatile modeling and optimization tools that can handle precise predictions and intricate manipulations of microstructural features of complex macromolecules by combining KMC simulation approaches with Computational Intelligence techniques.

2. Model Development

The Kinetic Monte Carlo (KMC) simulator was used to develop two intelligent tools: the Intelligent Modeling Tool (IMT) and the Intelligent Optimization Tool (IOT). The IMT combines KMC with an Artificial Neural Network (ANN) to predict microstructural features of olefin block copolymers (OBCs) from polymerization recipes and conditions. The IMT calculates multiple scenarios, trains ANNs, and creates black boxes that can predict microstructural patterns with high accuracy.

The IOT, on the other hand, uses a Genetic Algorithm (GA) to optimize polymerization recipes for desired microstructural features. The GA optimizer generates recipes, sends them to the KMC simulator for evaluation, and receives feedback to generate the next generation of recipes. This process continues until the IOT finds the optimal recipe for the target OBC.

Both tools were implemented in PASCAL programming language and compiled into 64-bit executable codes. The IMT uses ANNs as a black-box modeling technique, while the IOT uses the Non-dominated Sorting Genetic Algorithm (NSGA-II) to communicate with the KMC simulator. The tools were tested on a desktop computer with an Intel Core i7 processor and 32 GB of memory. The results show that the IMT and IOT can efficiently predict and optimize microstructural features of OBCs, making it possible to produce new grades of OBCs with preset characteristics without numerous trials.

Key features of the IMT and IOT include:

Fast and accurate prediction of microstructural features

Ability to optimize polymerization recipes for desired microstructural features

Use of ANNs and GAs for modeling and optimization

Implementation in PASCAL programming language

Testing on a desktop computer with high-performance hardware.

3. Results and Discussion

The Intelligent Modeling Tool (IMT) has been successfully used to model the microstructure of olefin block copolymers (OBCs). The IMT uses a combination of Kinetic Monte Carlo (KMC) simulations and Artificial Neural Networks (ANNs) to predict microstructural features of OBCs from polymerization recipes and conditions. The IMT has been tested on five randomly selected points in a master plot, and the results show that the IMT can accurately predict both architecture-related and property-related molecular features of OBCs.

The master plot provides a fast survey of the architecture-property topological trends of OBCs and allows for the identification of different classes of OBCs with unique blocky natures. The IMT can be used to develop new grades of OBCs suited to different applications by exploiting its unique capabilities.

The Intelligent Optimization Tool (IOT) has also been tested and has shown to be reliable in optimizing polymerization recipes for desired microstructural features. The IOT uses a Genetic Algorithm (GA) to generate and analyze billions of solutions to find the chain microstructure closest to the target OBC. The results show that the IOT can precisely predict the microstructural characteristics of OBCs with errors below 3.0% for almost all quantities.

The IOT has been used to synthesize two tailored OBCs, OBC1 and OBC2, with specific microstructural characteristics. The results show that the IOT can concurrently control multiple microstructural characteristics of the copolymer chains and find the optimal operational conditions to produce the target OBCs.

The approach introduced in this study is specific to the catalyst/chain-shuttling system for producing OBCs, but it can be adapted to many other polymerization schemes by modifying the reaction kinetics model in the KMC simulator. The IMT and IOT can be used to determine the features required to synthesize a polymer with the desired molecular structure or to visualize the factor space for an overview of the possible combinations of molecular features as a function of polymerization variables.

Key findings:

The IMT can accurately predict both architecture-related and property-related molecular features of OBCs.

The IMT can be used to develop new grades of OBCs suited to different applications.

The IOT can precisely predict the microstructural characteristics of OBCs with errors below 3.0% for almost all quantities.

The IOT can concurrently control multiple microstructural characteristics of the copolymer chains and find the optimal operational conditions to produce the target OBCs.

The approach can be adapted to many other polymerization schemes by modifying the reaction kinetics model in the KMC simulator.

4. Conclusions

Two powerful tools were introduced and successfully implemented to model and optimize the microstructural aspects of complex macromolecules. The newly developed computational techniques were established based on hybridization of molecular simulation approaches and Machine Learning techniques. The strategy made it possible to construct intelligent modeling and optimization tools capable of learning and decision-making. Undoubtedly, when applying these tools, polymer reaction engineers not only can effectively discover the complex inter-relationships between polymerization conditions and final architectural characteristics, but will also have the opportunity to adjust rather precisely the polymerization inputs in an attempt to synthesize predefined microstructures in detail. Chain-shuttling coordination copolymerization, an intricate polymerization system, has been chosen as the first test case to challenge the proposed intelligent modeling and optimization tools. The results obtained clearly showed that the IMT was capable of meticulously patterning the molecular landscape of OBCs in terms of operating conditions, including monomer molar ratio, catalyst composition, and CSA level. IMT was effectively put into practice to ‘crack’ the inter-relationship between operating conditions and micro-molecular characteristics and/or final properties of interest. By superimposing all microstructural information pieces together, a master diagram is obtained that provides a fast survey of the recipe-architecture-property topological trends. In contrast, the IOT was able to accurately predict the input/operating factors in response to predefined micro-molecular/architectural characteristics of the target OBC chains. To precisely evaluate the accuracy and performance of the proposed IOT, two target OBCs were designed first. Then, the IOT was implemented and applied to concurrently optimize three and seven molecular characteristics of the predefined OBC chains, respectively. The results obtained demonstrated that the IOT was able to successfully handle multi-objective optimization problems and simultaneously control various micro-molecular/architectural characteristics, resulting in negligible errors calculated for the objective functions.

Although the unique capabilities of the proposed techniques were successfully tested and examined with a complex living shuttling coordination copolymerization case study, they can be employed by both academic and industrial experts to model and optimize all types of macromolecular reactions and other reactive systems. Both the IMT and IOT, as intelligent computational tools, have the potential to guide polymer chemists and engineers towards the realization of advanced ‘living and thinking’ materials.