Simplifying Old Fluid Dynamics Through New Machine Learning

  1. Home
  2. Innovations
  3. Simplifying Old Fluid Dynamics Through New Machine Learning

Innovation in your Inbox. Sign up for the CRI’s Newsletter

  • This field is for validation purposes and should be left unchanged.

DSC 5323
Photo by Joshua Brown

For even the most knowledgeable of engineers, fluid dynamics can be a tricky topic to tackle. When it comes to non-Newtonian fluids (mostly anything that isn’t water, oil, alcohol, or similarly viscous liquids), this can be even tougher, requiring a specialist who understands how to do complex calculations. With fluid dynamics being important across a wide range of industries, this process can severely slow development processes. For one researcher, the opportunity to streamline this process was readily apparent.

Safa Jamali is an Assistant Professor of Mechanical and Industrial Engineering at Northeastern University. In 2021, Jamali received a $50,000 award for his RhIMLIIES project from the Center for Research Innovation’s (CRI) Spark Fund Awards. Working alongside his PhD candidate students, Jamali’s goal for the project is to create a machine learning program on fluid mechanics that would have wide, interdisciplinary, and cross-industrial applications. Jamali intends for RhIMLIES to be useful for as many companies as possible, from small start-ups to large industrial cornerstones.

For Jamali, the origins of his project required a brief history lesson. “If you look at the history of chemical and mechanical engineering, specifically in the 50s and 60s, there was no computational science – it was all either theory or it was experiments,” he notes. “Machine learning showed up late in the game, for a good reason: our computers didn’t have the capacity to process and store large amounts of data, but now we do.” These tools, while relatively new to the field, are powerful and have extraordinary potential, but engineers and others in related disciplines are not accustomed to using this type of technology. Jamali saw the opportunity, sparking the thought that started his project: “it’s the question of how can you grab this revolution in data-driven science and machine learning and tailor it to fluid mechanics?” Jamali’s project intends to take these complex applications that are typically done by hand and instead streamline the process with the help of machine learning.

Finding a method to start the RhIMLIES project was an in-depth thought process in itself. Jamali notes that there were some initial barriers to think through. “In a crude term, machine learning is like a black box. A lot of scientists and engineers are hesitant to use these tools because they don’t understand what is going on inside that black box.” For Jamali, it’s key to make that tool transparent, which he discovered could be done by implementing equations and physical laws directly in the machine learning. “You can embed those laws and questions in your machine learning, and that was the lightbulb moment.” Having set laws in the program itself removes doubt, giving the project the proverbial green light. With that in mind, Jamali and his team began to apply fluid mechanics laws to their machine learning platform. Working with his students, Jamali began to add algorithms one at a time to the program, each one growing the breadth and scope of what it is capable of. This compartmentalization of the software helps organize the program and makes it easier to work with both up front and behind the scenes.

Safa Jamali
Photo by Matthew Modoono/Northeastern University

Computational fluid dynamics (CFD) are how these fluid mechanics are typically calculated, but Jamali notes why that method needs an alternative. “CFD is tough – if you go on a job market and search CFD, you’ll see jobs that require extreme rigor and are really not for a fresh graduate.” This limitation is a major barrier for those new graduates; there’s an expectation that they need to have specific and advanced programming skills despite not learning those skills in their studies. Jamali’s project intends to bridge that gap. “It makes a very tough and intimidating part of science more friendly and accessible to the general mechanical engineering pool.” With this tool, these new engineers would only need a basic understanding of coding to make it work, which would in turn let them focus on their strengths instead.

Beyond supporting new engineers, Jamali has a second drive pushing him on this project. The versatility of his tool has garnered attention from a plethora of companies across different industries. He’s recently spoken to The Dow Chemical Company; he’s also participated in several conferences that host a number of other companies. The tool can be incredibly helpful for both large and small companies; with it, companies can avoid hiring specialists and present proof-of-concept in a much quicker timeframe and under much easier circumstances. For Jamali, being able to reach so many fields with a single tool helps to keep him motivated. “Actually making an impact on R&D in several different industries is very exciting to me, something that I never saw myself doing, but now it’s happening!”

Getting started on this project wasn’t easy for Jamali. In the early stages of the project, he was surprised to find that not many people had looked into making a project like this. “It’s very difficult to go to an agency to ask them for money for something you haven’t started, for something you’re not known for.” Jamali found his break with CRI’s Spark Fund, which helped get his project off the ground. “Now, having built two years of expertise, more companies and agencies are very interested, they want to support us.” Thanks to that kickstart form CRI, Jamali has been able to greatly grow both his student team and his project as a whole. With a full team and funding to back him, Jamali can now work unfettered towards his goal of making fluid dynamics more accessible to everyone, individuals and companies alike.

Article by Corey Ortiz

1 Comment. Leave new

  • Gladys McKie
    May 18, 2022 12:50 pm

    Corey, I like this quote by Assistant Professor Safa Jamali, “In a crude term, machine learning is like a black box. A lot of scientists and engineers are hesitant to use these tools because they don’t understand what is going on inside that black box.” This analogy can be used in other professions where professionals are hesitant to adopt ideas or technology “because they don’t understand what is going on inside that black box.”

    Reply

Leave a Reply

Your email address will not be published.

Fill out this field
Fill out this field
Please enter a valid email address.