Case studies

Ammonia Production Potential Powered by Codvo.ai's Technological Leap Forward

About the client

The client is well known for its pioneering approach to sustainability and innovation, the company oversees one of the most sophisticated ammonia production facilities worldwide. In pursuit of elevating operational efficiencies and amplifying output, the client sought a revolutionary solution to optimize its ammonia production process in real-time.

This ambition to blend technological advancement with environmental stewardship underscores the client's commitment not just to leading the energy sector, but to doing so in a manner that is both efficient and sustainable.

Overview

Codvo.ai collaborated with a global energy titan to revolutionize its ammonia production plant through a Real-Time Process Optimization Solution. Faced with the challenge of enhancing efficiency and maximizing output, Codvo.ai crafted a solution leveraging advanced machine learning regression models and optimization frameworks. This innovative approach focused on identifying optimal setpoints for control valves, ensuring peak production levels. Utilizing a robust tech stack including Python, Gurobi, and Scikit-learn, the project aimed to significantly reduce infrastructure costs and minimize downtime, setting a new benchmark in operational excellence within the ammonia production in the Energy sector.

Business Challenge

The client, faced the challenge of optimizing their ammonia production plant's efficiency in real time. The core task was to design a system that could accurately model the ammonia production process using machine learning models, with a goal to maximize output. This involved a complex interplay of variables, requiring precise feature engineering and analysis to ensure the models accurately reflected the plant's operating conditions.

Our Approach and Solution

Our team embarked on a systematic approach to tackle the client's challenges:

  • Analysis and Modeling: We commenced by meticulously analyzing the relationship between various valves and sensor measurements to identify potential independent and dependent variables for machine learning models. This phase was crucial for understanding the plant's operational dynamics.
  • Machine Learning Model Selection: Our team iteratively experimented with different combinations of regression models. This rigorous experimentation led to the final selection of 16 regression models that accurately modeled the plant's process flow.
  • Optimization Framework Development: We formulated a linear programming optimization problem aimed at maximizing the ammonia production rate. This involved considering the independent control inputs (setpoint) and subjecting them to the constraints formed by the machine learning regression models.  
  • Utilization of Frameworks: To accommodate both linear and non-linear regression models, we utilized the GurobiML and OMLT frameworks with custom auxiliary variable equality constraint logic.
  • Real-Time Optimization: The designed optimization framework operates on a set frequency of smoothed sensor data, enabling it to suggest optimal plant set points in real time.

Tech Stack

Tech stack used: Python, Gurobi, Scikit-learn, Pandas, OMLT

Highlights

Business Impact

Substantial reduction in infrastructure costs, leading to increased operational efficiency and cost-effectiveness.
Drastic decrease in application downtime, enhancing reliability and productivity of the ammonia production process.
Achieved optimal setpoints for control valves in real-time, maximizing ammonia production rates and ensuring consistent quality.
Fostered a sustainable production process by optimizing resource use and minimizing waste, contributing to environmental sustainability goals.
Strengthened the client's competitive edge in the global energy market by setting new benchmarks in operational efficiency and technological innovation.