Sustainable Process Engineering
We collaborated with an AI-driven green engineering company that focused on addressing environmental and safety concerns in process engineering. Their goal was to detect and mitigate environmental disturbances during the early stages of chemical processes, ensuring eco-friendly and sustainable practices. The company sought advanced AI and machine learning solutions to accurately predict environmental impacts, improve safety, and enhance their green engineering efforts.
The Challenge
The company faced the challenge of identifying and mitigating environmental risks from new chemical processes before they escalated. Accurate early-stage predictions were crucial to ensure sustainable development. They needed advanced machine learning models to forecast potential safety risks and environmental impacts, especially with the introduction of new chemical reactions that could interact negatively with ecosystems. The solution had to be both robust and capable of handling complex chemical data.
The solution
We developed a cutting-edge solution using Rust's Axum framework to power machine learning model inference for generating environmental impact analysis metrics and reports. The back end, built with Axum, ensures high performance and reliability by leveraging Rust’s efficiency and speed. The system integrates 20 ONNX-based machine learning models, managed by a dynamic model manager that seamlessly loads and switches between models. This architecture optimizes inference speed and simplifies multi-model deployment, enabling rapid innovation and high-performance decision-making across various applications.
The features
Our high-performance backend, built in Rust, combined with a dynamic frontend using HTML5 and Java8, ensures responsive design. We integrated Python’s ML libraries with PostgreSQL for data management while maintaining security best practices throughout the development lifecycle.
We applied Recursive Feature Elimination (RFE) to refine models, used polynomial features and interaction terms to enhance datasets, implemented LASSO regression for accuracy, and deployed genetic algorithms to identify optimal feature sets for better model performance.
We used Graph Neural Networks (GNNs) to predict chemical properties, geometric networks for 3D molecular structure predictions, and fine-tuned Large Language Models (LLMs) for chemical reaction predictions, literature reviews, and patent analysis.
We developed quantum mechanical scripts using Density Functional Theory (DFT) and integrated these advanced computational tools into the web application for precise chemical analysis.
We created specialized datasets by curating and labelling chemical engineering texts. Through transfer learning, we fine-tuned LLMs for specific tasks to improve performance in chemical research and analysis.
A RESTful API was developed to integrate LLMs into a web app, enabling real-time query processing with features like auto-completion, context-aware suggestions, and semantic search. This system supports process scale-up and Life Cycle Analysis.
Trained models were converted to ONNX format for seamless deployment across platforms and integrated them into the web app to ensure efficient and scalable machine learning inference.
We leveraged Hadoop and Spark for processing large datasets and used statistical techniques to gain insights from smaller, nuanced datasets, delivering high accuracy and relevance in data analysis.
The result
Our solution enabled the company to enhance its AI-driven green engineering efforts, providing more accurate predictions of environmental and safety risks in chemical processes. The integrated web application and machine learning models helped them streamline environmental monitoring, improve process safety, and ensure sustainable development from the outset. This resulted in a more efficient and eco-friendly approach to process engineering.
More Case Studies
