At Boston University’s College of Engineering, an autonomous research system combines automation and machine learning to accelerate the pace of creating the most efficient energy-absorbing shape.

What is Energy Absorption Efficiency?

Keith Brown, Associate Professor of Mechanical Engineering, Physics, and Materials Science & Engineering at Boston University, explained to us that “energy absorption efficiency is a metric that is defined in the context of protecting a specific system and is used to define how much energy can be absorbed before that system is damaged.”

This is an important metric for a wide range of products – including bike helmets, packaging materials, etc. Applied to automobiles, vehicles have what is referred to as “crumple zones” which are areas of the vehicle designed to deform during a collision to absorb the impact energy, protecting the passengers.

The energy absorbing efficiency defines how much weight or volume the crumple zone must have in order to keep occupants safe. A high energy absorption efficiency means the protective components can be lighter or smaller.

Up until recently, the energy absorption efficiency record across all tested designs had stood at 71.8% – but in January of 2023, the KABlab created a structure that broke the record at 75.2% efficiency.

Video credit Boston University

How Automation and Machine Learning Helped Them Do It

The KABlab is a laboratory within Boston University’s College of Engineering that studies hierarchically structured soft materials including polymers and smart fluids. As part of these studies, they are interested in the mechanical properties of polymer structures produced using additive manufacturing.

Illustration of the MAMA BEAR self-driving lab
Illustration of the MAMA BEAR self-driving lab.¹

Within the KABlab is a robot named MAMA BEAR (Mechanics of Additively Manufactured Architectures Bayesian Experimental Autonomous Researcher) which is a self-driving lab that combines multiple 3D printers, an Instron universal testing machine, a collaborative robot (cobot), and machine learning to autonomously create and test structures. In 2021 this system was put to work with the ambitious goal of creating a shape capable of absorbing the most energy possible – and it’s been running near-continuously ever since, resulting in over 25,000 tested 3D printed structures.

Consistency is the main challenge in this type of work. From this perspective, our Instron system has been exceptional. Keith Brown, Associate Professor of Mechanical Engineering, Physics and Materials Science and Engineering at Boston University. 

The Setup

Using 11 independent geometric parameters (which allow for more than a trillion unique design possibilities), MAMA BEAR 3D prints a small plastic structure – recording the shape and size. Then a cobot moves the structure through the system to the Instron universal testing machine where it undergoes a compression test. From this testing, the system measures how much energy the structure absorbed and how its shape changed after being compressed.

The process then starts over with the next 3D printed structure – but its design and dimensions are modified slightly by MAMA BEAR’s machine learning algorithm that automatically interprets the data from previous experiments and uses it to select subsequent designs to test.

Each maximum energy absorption efficiency measured over the first 21,500 experiments
Each maximum energy absorption efficiency measured over the first 21,500 experiments.¹

How’s the Data Being Used?

The project was sponsored by the US Army and the research is currently supporting the development of high-performance structures for use in helmet pads. KABlab has also made the data available for all to see and use at kablab.org/data.

Even though MAMA BEAR already broke the record with 75.2% energy absorption efficiency, KABlab is keeping the system running to see if they can continue to improve upon their own record. After all, there are still more than a trillion unique design possibilities to test.

The Instron System is very user friendly. Training new students is a fast process owing to the simple nature of the Bluehill software. Keith Brown, Associate Professor of Mechanical Engineering, Physics and Materials Science and Engineering at Boston University. 

We’d like to thank Keith Brown, Associate Professor of Mechanical Engineering, Physics, and Materials Science & Engineering at Boston University, for discussing this work with us which helped inform this article.

Additional Reading

References

¹ Snapp, K.L., Verdier, B., Gongora, A.E. et al. Superlative mechanical energy absorbing efficiency discovered through self-driving lab-human partnership. Nat Commun 15, 4290 (2024). https://doi.org/10.1038/s41467-024-48534-4