INFORMATION TECHNOLOGY

Create a “liquid fingerprint library” to “inject” taste into smart devices


Intelligent sensing devices have been used in many fields, but the ability of intelligent sensing devices to simulate the substances sensitized by human taste organs still faces challenges.

In order to solve this problem, the team of Academician Wang Zhonglin and researcher Wu Zhiyi of Beijing Institute of Nanoenergy and Systems, Chinese Academy of Sciences, started from the process of human tongue perception of taste, and developed an intelligent bisensory liquid sensing system based on droplet dynamic morphological changes and liquid-solid interface contact electrification. By combining the triboelectric feature fingerprint signal of liquid and the convolutional neural network deep learning algorithm, the prediction accuracy of common liquid is higher than 90% in five different application places. The addition of liquid visual information can further improve the perception ability of the taste sensing system of smart devices, and the accuracy of liquid recognition is increased to 96.0% in five application scenarios. This design of self-powered liquid sensing system that integrates taste-vision dual information, together with droplet-based taste sensors that can autonomously generate triboelectric signals, provides a technical direction for the development of efficient and low-cost liquid sensing for liquid food safety management. Recently, the relevant work was published in Nature Food.

Intelligent bisensory triboelectric taste sensing system for human-like taste receptors. Photo courtesy of interviewee

Nanogenerators are used as sensing probes

The human taste recognition system has a highly complex perception mechanism, and on the tongue of the main taste organ, there are thousands of taste receptors – taste buds. Taste buds complete the process of taste perception by performing the “generate and recognize signal-signal processing” task. Inspired by the multi-sensory interaction of the taste system, the research team designed a liquid recognition system that integrates the two modes of perception of taste and vision.

The researchers found that during the contact of the droplet with the polymer surface, its morphology alternates dynamically from spreading-contraction-spreading until it falls. The difference in electron affinity and physicochemical properties of different liquids makes the charge transfer on the electrode form different triboelectric signals, so friction nanogenerators can be used as probes for liquid sensing.

In order to more concisely and intuitively verify the feasibility of this liquid sensing strategy, the team leader proposed to use an open friction nanogenerator with two electrodes arranged in space as an experimental prototype. The advantage of this prototype is that a triboelectric signal can be actively generated without the need for an external power supply.

“We quantify design parameters by frictional charge transfer between the polymer film and the droplet.” Wei Xuelian, the first author of the paper and a doctoral student at the Beijing Institute of Nanoenergy and Systems, Chinese Academy of Sciences, explained, “When deionized water (removing various ions and electrolytes in water) water droplets come into contact with different friction layer materials, the induced currents fed back by different friction layer materials are different due to the difference in electron affinity and contact angle. The relationship between the tilt angle and the output current is relatively complex, in addition to the speed of the droplet in the inclined plane, the tilt angle also affects the contact area between the droplet and the polymer film, the interaction time between the droplet and the induction electrode, and the landing point position of the droplet on the polymer surface. These parameters have different response trends affected by the tilt angle, which in turn act together on the output current. ”

Therefore, the researchers investigated a series of variable parameters such as friction layer material, liquid type, inclination angle of droplet fall, initial droplet flow rate, droplet volume, etc., to optimize the design of triboelectric sensors.

Create a complete “liquid fingerprint library”

Feature extraction is an important and critical part of achieving taste perception.

“Part of the liquid signature comes from the double triboelectric signal generated by the droplet triggering two independent copper electrodes in turn, and the other part is extracted from the droplet image acquired by the image sensor.” Wu believes that this is the innovation of the work on feature extraction, “The change of liquid composition will cause a change in the current output signal, and if this change is more comprehensively quantified, it can be used as a ‘binary feature’ of the liquid, providing rational technical guidance for liquid sample identification.” ”

In addition, not limited to some shallow features such as the amplitude of the current extracted from the triboelectric signal waveform of the droplet, and the sliding form extracted from the droplet image, many subtle information that cannot be distinguished by the naked eye is worth digging into.

With the support of convolutional neural network deep learning technology, the research team effectively extracted more subtle features of liquid. The team figuratively refers to it as a “liquid fingerprint library” and created and further enriched it for more accurate and comprehensive liquid identification.

“Dual sense” synergistically enhances taste sensing ability

Taste buds on the tongue help people feel taste, and each taste bud has taste cells that can distinguish between different tastes. Although it is not possible to fully replicate the perceptual mechanism of the human taste system at this stage, the researchers have found that when different droplets slide through the sensing electrode, they can actively generate triboelectric signals with unique characteristics.

This characteristic difference depends on a series of “liquid phase” differences brought about by the liquid class, including the chargeability of the droplet, ion concentration, pH, composition change, viscosity, slip morphology, etc. At the same time, the complete characteristics of the liquid also need to consider visual aspects. If the two modes of perception of taste and vision can provide complementary liquid information, the analysis of liquids can be more comprehensive and accurate.

“For the five applications involved in this work, the intelligent recognition system using only triboelectric taste sensors has a prediction accuracy rate of more than 90%.” Wu Zhiyi added, “If there is an image sensor to cooperate, the perception ability of the sensing system can be further enhanced. The experimental results show that the accuracy of liquid identification in the five application scenarios increases by an average of 4.5 percentage points. This may be summarized as ‘when the feature data from each liquid is combined to form a multimodal feature, the recognition ability of the sensing system can be significantly improved’. (Source: Zhang Shuanghu, China Science News)

Related paper information:https://doi.org/10.1038/s43016-023-00817-7



Source link

Related Articles

Leave a Reply

Your email address will not be published. Required fields are marked *

Back to top button