Multi-Tasking Flexible Sensor with Reservoir Computing Monitoring Water Droplet Dynamics

Water droplets impacting a solid surface show different behavior depending on the angle of impact or the volume of water droplets. If sensor that monitors this impact behavior with time can be developed, it may be able to obtain information on the impacted water droplets, applying as a rain sensor to acquire precipitation and wind velocity. To achieve this rain sensor based on water dynamics monitoring, this study proposes a simple, low-cost resistive type multi-tasking flexible sensor that can be placed on a variety of surfaces such as an umbrella, roof, and car.<br/>Fabrication process is briefly explained. First, Polyimide (PI) film was exposed by CO₂ laser to generate laser-induced graphene (LIG), which is multilayered and defective graphene confirmed by Raman spectroscopy. The LIG layer was then transferred from PI film to a polydimethylsiloxane (PDMS) elastomer film. The surface of PDMS/LIG and PDMS were textured by CO₂ laser to form microscopic irregular surfaces that reproduce the Lotus effect. The textured LIG/PDMS has superhydrophobic surface with a contact angle (CA) of 179±0.1°and a sliding angle (SA) of 2.8±0.5° while the textured PDMS has a CA of 176±3.9° and a SA of 9.2±0.6°. This superhydrophobic surface controls the impact behavior of water droplets and enables continuous measurement of their impact dynamics.<br/>The sensing mechanism for the impact behavior of water droplets was studied by the relationship between the impact behavior of water droplets and the resistance value. The water droplets hit on the sensor spread rapidly in a radial pattern, then receded and left out from the sensor surface. The resistance at the moment of impact of the water droplet is the lowest and then gradually increases. The minimum resistance value was found to be smaller as the drop height increased or the sensor tilt decreased. This result shows the resistance value decreases as the force acting in a vertical direction to the sensor surface increases. This force also increases the amount of water droplets sinking into the microscopic irregular surfaces of the sensor, increasing the contact area between the droplets and the sensor surface. As the result, electrical resistance is changed depending on the condition of water droplet, which is the sensing mechanism. This sensing mechanism makes it possible to measure different time series data depending on the impact behavior of water droplets due to changes in water volume, wind velocity, and sensor tilt.<br/>Time series data of resistance changes in the impact dynamics of water droplets were obtained for volumes from 10 μL to 40 μL, wind velocities from 0 m/s to 5.0 m/s, and sensor tilt angles from 0° to 60°. These time series data were analyzed using reservoir computing, which is one of the recurrent neural networks, to estimate the volume of water droplets and wind velocity at each sensor tilt. For each tilt, the volume and wind velocity estimation results were obtained with relatively low normal mean square error (NMSE) &lt;0.2 and &lt;0.25, respectively. This result indicates that this sensor has the potential to be used as a sensor to acquire precipitation information by attaching this sensor on a variety of objects. Importantly, this single sensor can estimate multiple information of volume and wind velocity as a multi-tasking simple sensor system.<br/>In conclusion, we developed the multi-tasking rain sensor to obtain rain perspiration and wind velocity from the dynamic change of water droplet behaviors. By developing a resistive-type flexible sensor and reservoir computing algorithm, water droplet volume and wind velocity at each sensing angle were successfully extracted with relatively high accuracy. Although there are a lot of things to apply to the real rain/weather sensor system, this may contribute to a new class of flexible sensor integrated with machine learning potentially for low power consumption due to less number of sensors.