Kim Juhyeon, Ryu Jiae, Yang Qiang, Yoo Chang Geun, Kwon Joseph Sang-Ii
Artie McFerrin Department of Chemical Engineering, Texas A&M University, College Station, Texas 77845, United States.
Texas A&M Energy Institute, Texas A&M University, College Station, Texas 77845, United States.
Ind Eng Chem Res. 2024 Nov 19;63(48):20978-20988. doi: 10.1021/acs.iecr.4c02918. eCollection 2024 Dec 4.
While lignin has garnered significant research interest for its abundance and versatility, its complicated structure poses a challenge to understanding its underlying reaction kinetics and optimizing various lignin characteristics. In this regard, mathematical models, especially the multiscale kinetic Monte Carlo (kMC) method, have been devised to offer a precise analysis of fractionation kinetics and lignin properties. The kMC model effectively handles the simulation of all particles within the system by calculating reaction rates between species and generating a rate-based probability distribution. Then, it selects a reaction to execute based on this distribution. However, due to the vast number of lignin polymers involved in the reactions, the rate calculation step becomes a computational bottleneck, limiting the model's applicability in real-time control scenarios. To address this, the machine learning (ML) technique is integrated into the existing kMC framework. By training an artificial neural network (ANN) on the kMC data sets, we predict the probability distributions instead of repeatedly calculating them over time. Subsequently, the resulting ANN-accelerated multiscale kMC (AA-M-kMC) model is incorporated into a model predictive controller (MPC), facilitating real-time control of intricate lignin properties. This innovative approach effectively reduces the computational burden of kMC and advances lignin processing methods.
虽然木质素因其丰富性和多功能性而引起了广泛的研究兴趣，但其复杂的结构对理解其潜在的反应动力学和优化各种木质素特性构成了挑战。在这方面，已经设计了数学模型，特别是多尺度动力学蒙特卡罗（kMC）方法，以提供对分级动力学和木质素性质的精确分析。kMC模型通过计算物种之间的反应速率并生成基于速率的概率分布，有效地处理系统内所有粒子的模拟。然后，它根据这个分布选择一个反应来执行。然而，由于反应中涉及大量的木质素聚合物，速率计算步骤成为计算瓶颈，限制了该模型在实时控制场景中的适用性。为了解决这个问题，将机器学习（ML）技术集成到现有的kMC框架中。通过在kMC数据集上训练人工神经网络（ANN），我们预测概率分布，而不是随着时间反复计算它们。随后，将得到的ANN加速多尺度kMC（AA-M-kMC）模型纳入模型预测控制器（MPC），便于对复杂的木质素性质进行实时控制。这种创新方法有效地减轻了kMC的计算负担，并推动了木质素加工方法的发展。
