The Intersection of AI and Mechanical Engineering – My lab at CMU is focusing on bringing machine learning and artificial intelligence into all areas of mechanical engineering. Traditionally, if you are using mathematics and, basically, physics in order to describe many physical phenomena like fluid mechanics, stress and strain analysis, control problems. We have four areas of thrust in our lab.
One of them is bringing, basically, machine learning algorithms in order to infer, simulate, and predict transport phenomena. If you think about it, things like turbulence are very difficult to learn and, basically, to predict. Sometimes mathematical foundations of these works cannot be sufficient in order to describe the phenomena. So, what we do is that we use data, ample data, and we need to train, basically, on the data that we have from turbulence, and then learn, basically, the physics of turbulence and predict even a new situation or condition.
The second area of thrust in my lab is using machine learning and artificial intelligence for discovering new materials and new molecules. What we do is that we train on multiple data that are generated either via simulations or via experiments, and then we make a predictive model that if you give me this molecule or material, what will be the properties? So, we are using deep neural networks or graph convolutional networks in order to be able to model this functional map.
The third area of thrust in my lab is bringing artificial intelligence into robotics field. There is a rise of this beautiful algorithm called the reinforcement learning and deep reinforcement learning, which is really the biggest, I think, part of AI these days. So that is the line of research we are pursuing here. And it has two folds. One of them is in manipulation, which robot manipulation is very important.
If you want to bring robots to our daily life, you need to have, basically, intelligence in manipulation and handling the objects and recognizing many tasks. And the second part is giving intelligence to drones. We are trying to use deep neural networks, combined with reinforcement learning, which we call a deep reinforcement learning, in order to give intelligence to the drones to use only vision to come up with, basically, decisions, to find their path, plan their path, and basically do a task without the need for other sensors fusions.
So we are decreasing or we are making them independent. A human doesn’t have, basically, doesn’t need to connect with GPS all the time. Right, I mean they can plan their way, or they can find, basically, to go and avoid obstacles by only vision. You don’t use your phone to just avoid, for example, some obstacle. The fourth area of thrust in my lab is using AI and machine learning in health and biology. If you think about it, many phenomena in mechanical engineering and in chemical engineering, they are not as complex as biology.
Biology is very complex. But I think AI and machine learning has this power, basically, to create this connection between, for example, patients, health status, and the structure of biology. The impact in the area of health obviously is, can I find an antibody or a vaccine that can cure HIV or, basically, Ebola virus. My research is also trying to pursue is how we can shorten this 15-years timescale for drug discovery using AI and machine learning.
The impact of my research is going to be bringing machine learning and AI to the traditional fields of mechanical engineering and making the impossible, possible.
Web enthusiast. Thinker. Evil coffeeaholic. Food specialist. Reader. Twitter fanatic. Music maven. AI and Machine Learning!