The Future of Wearables: Material That Can Recognize Patterns
New research attempts to achieve a hybrid material system which can sense, communicate, and compute.
Researchers at the University of Pittsburgh Swanson School of Engineering have designed materials which can perform simple pattern recognitions. The research attempts to achieve a hybrid material system which can sense, communicate, and compute.
The materials the researchers worked with, which utilize the energy from chemical reactions and do not need an external electrical power supply, can be used in smart clothing to monitor whoever's wearing it. As another application, a robot can use this technology to have a form of sensory skin. In addition, since the new material can expand the functionality of everyday objects, we can expect numerous other potential applications that we have not thought about yet.
The invention, which aims to recognize black-and-white pixels in the shape of numbers, is a network of coupled BZ-PZ units. Each of these BZ-PZ units consists of a piezoelectric (PZ) layer over a self-oscillating gel which exhibits Belousov-Zhabotinsky (BZ) oscillatory reaction.
An animated .gif of the Belousov-Zhabotinsky (BZ) oscillatory reaction.
The concept of oscillation in electronics engineering is amazing to behold. However, as an electronics engineer, we somehow cannot appreciate this because we have been exposed to the concept so many times.
There is oscillatory behavior in other fields too. As an example, a Belousov-Zhabotinsky (BZ) reaction is a chemical reaction which is far from equilibrium and remains so for a significant length of time. A BZ gel shows chemomechanical oscillations. Theoretically, the importance of a BZ reaction is that its chaotic evolution gives an example of non-equilibrium thermodynamics.
Boris Belousov noticed this oscillatory behavior in 1950s. He discovered that the color of a solution of bromine and an acid oscillated between yellow and a colorless solution.
A visual example of the oscillation taking place in a solution.
Around a decade later, Anatol Zhabotinsky, a graduate student, explained this reaction in greater detail.
A BZ-PZ Unit
The new research done at the University of Pittsburgh employs the rhythmic expansion and contraction of the BZ gels to deflect the PZ cantilevers. The result is that the chemical reaction produces a periodic electrical signal.
Figure 1 shows two of these units connected by electrical wires. In this figure, the chemical reactions deflect the PZ layers and produce an electrical voltage, U, across the red and black lines.
Note that the inverse PZ and BZ effects are also possible and when we apply a voltage to the PZ layer, the chemomechanical oscillations in the gels are modified. As a result, the two BZ-PZ units in Figure 1 are interacting with each other.
Figure 1 Two BZ-PZ oscillators interacting with each other. Image courtesy of Science Advances.
In other words, these two units are communicating with each other and, through this communication, the oscillators can reach a synchronized state after some time. The research shows that it is possible to use these oscillators, proposed as so-called “oscillator-based computing”, to perform simple calculations and achieve simple pattern recognition.
Another study done by this research team examined theoretical and numerical models of networking BZ-PZ oscillators. For example, as shown in Figure 2, it is possible to connect these units in parallel or in series.
Figure 2 The BZ-PZ oscillators can be connected in series or in parallel. Image courtesy of Science Advances.
Pattern Recognition with a Network of Oscillators
A number of other research groups have previously proposed a network of oscillators to perform pattern recognition. One example of these studies, the oscillatory neural network (ONN), inspired the research team to apply the concept to a network of BZ-PZ oscillators.
To this end, as shown in Figure 3, the new scheme places several BZ-PZ oscillators in series and chooses the positive polarity for a white pixel and the negative polarity for a black pixel. Hence, each oscillator represents a single pixel of the image. By changing the polarity (flipping the connecting wires as shown in the figure), the sign of the voltage produced by the oscillator will change. In this way, it is possible to store a pattern in the network.
Figure 3 The polarity of the oscillators can be used to model the white and black pixels of the pattern. Image courtesy of Science Advances.
Based on the extensive numerical simulations and a linear stability analysis, the study conjectures that with multiple BZ-PZ units in series, all the oscillators with the same polarity will exhibit in-phase synchronization and the oscillators of the opposite polarity will reach the antiphase synchronization.
As explained above, the stored pattern is defined by the polarity of the oscillators. The question is how we can apply an input pattern to this system? It is possible to use the initial phase of the BZ gel in each oscillator to define the input to the system. Since a BZ gel is chemo-, photo-, and mechanoresponsive, we can use chemical stimulation, light, or pressure to force the oscillators to have a specific initial phase.
The study utilizes a normalized phase value which varies between 0 and 1. When inputting a pattern, the initial phase of an oscillator which corresponds to a black pixel is forced to be zero and that of a white pixel is forced to be 0.5.
Now, we let the system evolve the phases into a stable state. Interestingly, the stable state of the network of BZ-PZ oscillators is the one which was stored in the system by choosing the polarity of the oscillators. In other words, no matter what we give the system as the input pattern, it will give us its stored pattern as the output of the system. However, obviously, when the input pattern is similar to the one stored in the network, the time required to evolve to the final stable state will be significantly shorter.
Figure 4 illustrates the operation of three different networks of BZ-PZ oscillators. The stored pattern of these networks are “0”, “1”, and “2”, respectively. However the input pattern is a distorted pattern of 1 for all three networks. The patterns on the right are used to illustrate how the pattern represented by the system evolves with time. The diagrams on the left show that all of the oscillators will be in either in-phase or anti-phase synchronization at the end of the process.
As shown in this figure, the final state is simply the pattern which was previously stored in the system. However, the system storing the pattern of “1” evolves significantly faster (in 53 units of time) to its final state. This shows that the input and stored patterns may be compared easily by measuring how long it took for the final state to be achieved.
Figure 4 When the input pattern is similar to the stored one, the system evolves significantly faster to its final state. Image courtesy of Science Advances.
According to Yan Fang, a grad student in the ECE department and lead author of the paper, the future goal of the research is to design materials which can recognize grayscale and more complicated images and shapes.
Unfortunately, a network of BZ-PZ oscillators is not as fast as a traditional computer and it takes the system several minutes to complete the pattern recognition. This is mainly due to the slow period of BZ oscillations.
Despite that this technology may never approach modern computing levels, it could find use in the medical field, particularly as wearables.
For a detailed explanation of this study please read the research paper "Pattern recognition for materials that compute" published in Science Advances.