Researchers are developing a smart material, capable of adapting its physical properties according to the situation

⇧ [VIDÉO] You might also like this partner content (after ad)

Imagine an aircraft whose wings transform autonomously during flight to limit turbulence and ensure optimal efficiency. If the concept is science fiction for the moment, research by a team from the Department of Mechanical and Aerospace Engineering at the University of California, Los Angeles, could very well lead to this type of application. They have in fact developed an intelligent architectural material, capable of learning to adjust the rigidity of the connections between its components according to the surrounding constraints.

With the exception of some living tissue, few materials can autonomously learn to exhibit desired behaviors following prolonged exposure to unforeseen ambient loading scenarios, the researchers note in Science Robotics. Rarer still are materials that can continue to exhibit previously learned behaviors under changing conditions (such as increasing levels of internal damage, fluctuating external loads, etc.) while acquiring new behaviors better suited to the situation.

To fill this gap, researchers have developed what they call a mechanical neural network (or MNN for mechanical neural network), able to learn to adjust the stiffness of the connections between these physical components in the same way that artificial neural networks mimicking the human brain adjust their weights. ” This work lays the foundations of materials to artificial intelligence who can learn behaviors and properties “, they summarize.

A mechanical grating that automatically adjusts its shape

Artificial neural networks, which are the basis of many modern models of artificial intelligence, are inspired by the functioning of neurons in the human brain: thus, just as the brain learns new behaviors by reinforcing synaptic connections, neural networks Artificial devices learn by adjusting the numerical values ​​representing these connections. The mechanical neural network discussed here works on the same principle, except that the weight of the connections between the neurons is replaced by physical connections of varying stiffness.

We hypothesized that a mechanical network with physical nodes could be trained to take on certain mechanical properties by adjusting the stiffness of each connection », explains Ryan Lee, a doctoral candidate in mechanical and aerospace engineering at the University of California and first author of the study. In other words, instead of processing numerical data, the DMM processes the forces applied to it, twisting and changing its shape accordingly. ” If a MNN is damaged, cut to occupy another volume, or fixed differently, then it can relearn previously mastered behaviors and acquire new behaviors as needed », specifies the team in its article.

(A) Computer-aided design model and (B) photo of the mechanical neural network used in the experimental study. © RH Lee et al.

To put this concept into practice, Lee and his colleagues first built a computer model of this mechanical network: they defined the desired shape as well as the input forces, then using an algorithm, they adjusted the connection tensions so that the input forces produced the desired shape.

They finally built an array of about 45 by 60 centimeters, consisting of 21 adjustable electromechanical springs, arranged in a triangular shape — which proved to be the best possible shape to achieve all the desired configurations. The connections measure approximately 15 centimeters; each is equipped with a small linear motor, which can modify its rigidity. The material “derives its properties primarily from the geometry and specific features of its design rather than what it is made of,” Lee points out.

For increased longevity and efficiency of structures

Once the network was trained — meaning the stiffness of the connections was tuned so it could perform a set of tasks — the material was able to learn and change shape based on the forces applied to it. He was thus able to adapt to any changing or unfamiliar condition.

According to Lee, such a material would be ideal for the design of aircraft wings: in the face of internal damage accumulation, changes in the way the wing is attached to the cabin, or even a strong gale. unexpectedly, the mechanical neural network would strengthen its connections accordingly; the wing would become stronger over time (and adjustments) and the aircraft would achieve increased efficiency and maneuverability as it accumulates flight experience.

This type of material could have far-reaching applications for the longevity and efficiency of built structures. “said Lee. Not only could the aircraft wing gain in robustness, but it could be possible to take the concept even further, by training the MNN so that it completely changes the shape of the structure if necessary to maintain efficiency. maximum energy.

This prototype is only a proof of concept used to show the potential of MNNs. The team would now like to adapt their network in 3D, because computer modeling revealed that 3D networks would have a much greater capacity to learn and adapt (due to their greater number of connections). But for the moment, the mechanisms used are too complex to be transposable in 3D, specifies the researcher.

Source : RH Lee et al., Science Robotics

We would like to give thanks to the writer of this article for this remarkable web content

Researchers are developing a smart material, capable of adapting its physical properties according to the situation


Explore our social media profiles and other pages related to themhttps://www.ai-magazine.com/related-pages/