top of page

Deep Learning Technologies for Operational Excellence in Chemical Industries

My research is dedicated to leveraging the capabilities of Artificial Intelligence to develop superior, industry-oriented models. These models not only excel in their predictive capabilities, but they also prioritize explainability and ensure alignment with the fundamental laws of nature for greater reliability and coherence.

Deep Chemometrics 
Predicting Key Crude Properties

In my research, I focus on utilizing the power of artificial intelligence to streamline the process of determining the physicochemical properties of crude oils, traditionally a time-intensive process. Using Fourier Transform Infrared (FTIR) spectrums and convolutional neural networks (CNNs), I've significantly reduced the prediction error for properties like viscosity. This approach not only enhances efficiency but also boosts interpretability by unveiling the importance of various spectral features. The implications of this work span both practical enhancements in crude oil processing and theoretical advancements in AI application within the field.

In my ongoing research, my key drive is to enhance the interpretability of deep learning models. I am striving to design highly accurate AI systems that are not just effective but also transparent and understandable. The focus on Explainable AI is crucial in boosting human trust and enabling more informed decisions in complex engineering systems. Through this work, I aim to contribute significantly to AI applications in engineering, particularly within chemical engineering, by exploring innovative deep learning methodologies and explanation techniques.

Explainable Artificial Intelligence
Explaining Model Decisions

Physics Guided Artificial Intelligence
Physically Consistent Models

A key facet of my research involves investigating Physics-guided machine learning - a novel approach that leverages domain-specific knowledge to improve AI models. This knowledge can come from various sources such as conservation laws, governing equations, boundary conditions, and kinetics. Specifically, I'm building on the foundations of Physics Informed Neural Networks (PINNs), with an aim to incorporate more complex physical insights into these networks. My work is also set to explore ways to integrate plant kinetics and heuristics into neural networks, which has yet to be accomplished.

bottom of page