Liquid-Metal-Based All-Soft Pressure Sensor and Machine Learning Application for Multidirectional Detection

Electronic skin needs to be able to sense different types of pressures, such as radial pressures, which require the development of a soft multidirectional pressure sensor. Although previous studies have developed multidirectional pressure sensors, the investigation into discerning the direction and magnitude of applied pressure remains an unexplored domain in multidirectional pressure sensor research. Multidirectional pressure sensors have been extensively developed in literature due to their vast application opportunities. Recently, the trend has shifted towards developing soft sensors with a focus on the design of the structure and sensing materials. The four-element sensing structure is the most commonly used structure for multidirectional sensors. In the case of the four-element sensing structure, the four sensing element is positioned at 90-degree intervals from one other, forming a cube-like structure. The four-element sensing structure facilitates the measurement of multidirectional forces. Wide-ranged 3D bump structures are used at the top of the four-element sensing materials to identify the applied multidirectional forces effectively. Lee et al. demonstrated trapezoid structure and CNT array as the sensing material. Jung et al. proposed a multidirectional force sensor with the sensing materials placed on the sidewalls of the 3D bump structure. Vogt et al. and Kim et al. presented liquid-metal-based multidirectional sensors with rigid pressure transmission structures on the top of the liquid metal channels. In these sensors, the multidirectional force is generated by applied shear pressure to the four-sensing structure of the sensor. Nevertheless, extracting directional information from a four-sensing element-based sensors’ setup when exposed to applied radial pressure remains a formidable challenge, because the four-sensing element method is primarily designed for shear force-based applications.<br/>In the present study, we present a new solution to address the limitations of existing multidirectional sensors. Our approach involves the development of a soft sensor using liquid metal with a dome-shaped design that enables discrimination of directional pressures. Additionally, we leverage machine learning algorithms to identify the direction and magnitude of applied pressure. The microchannels of the soft pressure sensor were created using fused deposition modeling (FDM). A 3D printed mold made of poly(vinyl alcohol) (PVA) was embedded into the elastomer and dissolved with water to create the microchannels. The empty microchannels were filled with liquid metal, and a vacuum process was applied to create the dome structure for multidirectional pressure sensing. The microchannels within the liquid-metal-based soft pressure sensor exhibit a transient response when pressure is applied. Specifically, each direction of the sensor indicates a distinct transient response when the pressure is applied, which helps to differentiate between different directions of pressure. The machine learning technique, we have implemented is based on dual-task 1D convolutional neural networks (CNN). This approach allows for real-time classification and regression of the magnitude of pressures applied on different directions of the sensor. We achieved direction identification in real-time using the transient response data from the soft pressure sensor. The developed liquid-metal-based soft pressure sensor can detect the direction (classification accuracy of 99.1%) and magnitude (regression error of 20%) of the multidirectional pressure in real-time pressure prediction through a 1D CNN algorithm. The effectiveness of the sensor and algorithm is demonstrated with a human-machine interface application where the sensor is used to control an RC model car. The machine learning predictions, including the direction and magnitude of the pressure, are used to control the vehicle's real-time movement.