Harnessing the power of machine learning and physics: Conversation with Dr. Sai Nethra Betgeri on advancing civil engineering solutions

Dr Sai Nethra Betgeri, an Assistant Professor at the University of Louisville, has been at the forefront of integrating machine learning (ML) with civil engineering to tackle real-world problems. In this interview, she shares insights into her groundbreaking research and how combining ML with physics could revolutionize the field.

Your work focuses on merging machine learning (ML) with physics, particularly in civil engineering. Can you tell us more about the challenges civil engineering faces with traditional ML models?

Civil engineering applications often struggle to implement machine learning models successfully, especially when these models are simulated in a lab but fail to perform in real-world scenarios. This discrepancy arises primarily due to what we call "data shift"—a mismatch between training data and real-world data. Standard ML models, which are often data-driven and not physics-based, may work well in controlled conditions but fail to generalize to the complexities of real-world applications. This is where integrating physics into the ML models becomes a game-changer.

What exactly is a physics-based ML model, and how does it differ from traditional ML approaches?

A physics-based ML model combines data with physical laws, typically expressed as partial differential equations (PDEs) or mathematical models. These equations describe the underlying physics of the system being studied. By incorporating this physical knowledge into the machine learning algorithms, the model can better handle real-world variations and data shifts. Essentially, it acts as a bridge between theoretical physics and practical data-driven techniques. This approach is particularly useful in civil engineering, where many problems, such as structural modeling, material behavior, and environmental impact, are governed by complex physical principles.

Could you give us an example of how you have applied physics-based ML in civil engineering?

One of the most exciting applications I've worked on is using machine learning for pipe maintenance scheduling. In traditional approaches, maintaining infrastructure like water pipes involves scheduling based on historical data and predictive algorithms. However, these methods can often lead to inefficient schedules or even costly maintenance oversights. I developed a framework using K-Nearest Neighbors (K-NN) to schedule pipe maintenance faster and more accurately, leading to significant cost savings. The physics-based ML model helped account for factors like pipe stress, material degradation, and environmental conditions, making the predictions far more reliable and efficient.

You’ve mentioned a roadmap for advancing research in this area. Can you explain what gaps exist in the current research and what future directions might look like?

The current research on physics-based ML in civil engineering is still in its early stages. While there have been reviews of ML applications in the field, very few studies focus on combining physics and machine learning. The main gap lies in synthesizing these two domains in a way that can be applied across various civil engineering challenges. Future research needs to focus on improving the integration of these models, testing them in diverse real-world settings, and developing generalized frameworks that can be adapted to different types of civil engineering problems. My vision is to see this approach used more broadly to address critical challenges like environmental impact prediction, infrastructure resilience, and sustainable development.

In your opinion, what role will physics-based ML play in the future of civil engineering?

I believe that merging ML with physics will be invaluable in advancing civil engineering. As the field evolves, the complexity of problems we're tackling grows exponentially. By leveraging both the data-driven insights of ML and the foundational principles of physics, we can build more accurate, robust, and scalable models. These models will be able to predict behaviors more reliably, optimize designs, and offer solutions that are both cost-effective and environmentally sustainable. It's a marriage of two powerful tools that will enable engineers to solve some of the most pressing challenges of our time.

Finally, what advice do you have for aspiring researchers or professionals looking to explore this intersection of machine learning and civil engineering?

My advice would be to stay curious and interdisciplinary. Understanding the fundamentals of both machine learning and civil engineering is key, but don't be afraid to push boundaries and look for ways to integrate concepts from different fields. As technology advances, the opportunities at this intersection will grow, and the ability to blend expertise in data science with practical engineering knowledge will be incredibly valuable.