The integration of artificial intelligence (AI) into chemical engineering marks a transformative era, redefining traditional methodologies with AI-driven approaches. AI has emerged as a powerful ally in tackling complex problems once considered insurmountable. As chemical engineering grapples with increasingly complex systems and stringent sustainability targets, AI sets the stage for a new generation of solutions.
AI applications in chemical engineering rely on three primary mechanisms: data-driven modeling, knowledge-based systems, and hybrid approaches that combine both. Often, data-driven models dominate, leveraging machine learning algorithms to extract patterns and insights from large datasets. In contrast, knowledge-based systems incorporate domain expertise and first-principles understanding to guide AI decision-making. Some applications require a synergistic combination of both approaches, necessitating the careful integration of data-driven and knowledge-based methodologies. To address these challenges and explore innovative solutions, experts worldwide have been invited to contribute articles on this topic.
Mesoscience, a field pioneered by Jinghai Li and others, seeks to bridge the macroscopic and microscopic scales by focusing on mesoscale problems at various system levels. It addresses the common challenges in different disciplinary fields by analyzing the competition between dominant mechanisms within complex systems. Integrating mesoscience with AI has proven to be a promising approach to modeling complex systems effectively. The research group led by Li Guo has proposed and demonstrated the use of mesoscience-guided deep learning (MGDL) for modeling complex chemical systems. By integrating physical principles and mesoscopic insights into deep learning architectures, they have significantly improved model accuracy and interpretability. Their work demonstrates AI’s potential to bridge the gap between empirical data and theoretical understanding, enhancing the predictive capabilities of models in multiphase systems.
In chemical engineering and materials science, accurately predicting pure component properties is foundational to designing and optimizing chemical processes and developing novel materials. Historically, these properties have been estimated using empirical methods such as group contribution approaches, which rely on the additive contributions of molecular fragments to predict properties such as boiling points, vapor pressures, and solubilities. However, these methods can suffer from limitations in accuracy, especially with complex molecules for which the interactions between functional groups are non-additive. Focusing on estimating pure component properties, Chen and coworkers have developed an enhanced machine learning framework that leverages group contribution methods and Gaussian processes. By mapping discrete molecular structures into a continuous domain, their model improves the representation of complex molecular interactions, leading to more accurate predictions of physicochemical properties.
In the pursuit of sustainable chemistry and engineering, the development of novel solvents plays a critical role in advancing green processes and reducing environmental impact. Among the most promising innovations in this field are deep eutectic solvents (DESs). However, the rational design of DESs has been impeded by the lack of predictive models capable of accurately identifying suitable combinations of components and predicting the resulting properties of the solvent mixture. Qing Shao and his team address the challenge of discovering new DESs by employing machine learning models that identify unique hydrogen bond features. Their work has led to the development of 30 models using various algorithms, significantly advancing the rational design of non-ionic designer solvents.
Porous media are integral to many environmental and energy systems, where accurately predicting reactive transport is crucial. Traditional modeling approaches struggle with the complexity of such environments, often failing to adequately represent the heterogeneous nature of porous materials. The group led by Cheng Lian and Honglai Liu introduces Porous-DeepONet, an AI model that solves parametric reactive transport equations in porous media. By incorporating convolutional neural networks, they enhance the model’s ability to capture the intricate features of porous media, enabling accurate predictions of reactive transport phenomena.
Modeling dynamic chemical processes is essential in optimizing operations and ensuring safety. Traditional models often rely on first-principles equations, which can be complex and computationally expensive. The advent of machine learning-particularly deep learning techniques-has introduced more efficient and accurate ways to model these processes. Weifeng Shen’s group has developed the light attention-convolution-gate recurrent unit (LACG) architecture for chemical process modeling. This architecture, which combines convolutional and recurrent neural networks with a light attention mechanism, demonstrates superior performance in modeling dynamic chemical processes.
Discovering and developing new materials has traditionally relied heavily on rational design approaches, which involve a meticulous understanding of material properties and their underlying atomic structures. However, the inherent limitations of this method, which include its time-consuming nature and the difficulty of exploring vast chemical spaces, have spurred a shift toward more agile and efficient techniques. In this context, Jianjun Hu and coworkers explore the use of generative AI for materials discovery. By moving beyond traditional rational design approaches, they advocate for a data-driven strategy that can rapidly identify novel materials with exceptional properties. A review article from Xinyan Liu’s group highlights the role of machine learning in accelerating the discovery of heterogeneous catalysts. The review underscores the potential of AI in predicting surface reactivity with lower computational costs, offering a roadmap for the rational design of catalysts.
As AI continues to permeate various sectors of chemical engineering, the demand for transparent and accountable AI systems grows. AI models are increasingly employed for process optimization, materials discovery, and predictive maintenance. However, the complexity of these models often leads to a "black box" effect, in which the decision-making processes remain opaque to users. A review from Jesse Zhu’s group focuses on the concept of transparency in AI applications within chemical engineering. By emphasizing the importance of causality, explainability, and informativeness, the researchers advocate for responsible AI utilization. Their review showcases state-of-the-art applications that combine physical principles with AI, promoting a hybrid modeling approach that enhances the reliability and interpretability of AI models in chemical engineering.
While the potential of AI in chemical engineering is immense, several challenges remain, including the scarcity of high-quality data, the need for robust validation metrics, and the integration of AI with existing engineering practices. Moreover, ensuring the interpretability and transparency of AI models is imperative for building trust and facilitating the widespread adoption of such models in industry.
Despite these challenges, the future of AI in chemical engineering is promising, with ongoing advancements in AI techniques and increasing collaborations between academia and industry. The integration of AI into chemical engineering is set to redefine how we approach complex problems in our field. From enhanced materials discovery to process optimization, is poised to deliver significant advancements. We anticipate the development of more sophisticated AI models that can handle multimodal data, integrate domain knowledge, and provide real-time process optimization. The emergence of AI-driven design platforms and digital twins will greatly facilitate predictive maintenance, process optimization, and sustainable chemical production. Let us embrace this exciting journey as we harness the power of to drive innovation and excellence in chemical engineering.
PDF
(298KB)
