Flexible and stretchable high-performance electronics have opened up new possibilities in various fields, including bio-integrated health monitoring, curvilinear electronics, and human-machine interfaces. Liquid metals (LMs), with their excellent electrical conductivity and deformability, are particularly attractive for skin-mounted, deformable electronics. LM-based flexible electronics offer precise perception of complex strain and pressure stimuli, making them promising for applications in health monitoring, wearable electronics, and human-machine interfaces. However, their high surface tension and the need for improved sensor sensitivity pose significant challenges. Recent research has focused on modifying LMs to improve flexible electronics fabrication. Modifications to wetting characteristics and viscosity simplify manufacturing processes, eliminating the need for complex lithography and reduction procedures. Incorporating metal particles into LMs, often with acid treatment to enhance dispersion and adhesion, has shown some success. Ultrasonic treatment is another approach used to increase viscosity; however, the limited viscosity improvement remains a challenge for rapid prototyping. Beyond the manufacturing process, LM-based sensors suffer from relatively low sensitivity. Most rely on the piezoresistive effect, where resistance changes with deformation. The gauge factor, which relates resistance change to strain, depends on both geometry change and resistivity change. Since the resistivity of LMs remains largely unchanged, the sensitivity is limited by the geometry change, typically resulting in low gauge factors (2-6). This limits their use in applications requiring high sensitivity, such as skin prosthetics and wearable health monitoring. Therefore, there is a strong need for high-sensitivity LM-based sensory systems. This paper presents a novel modification method to enhance the sensitivity and robustness of LM-based strain and pressure sensors.

The literature extensively covers the use of liquid metals in flexible electronics and sensors, highlighting their unique properties for wearable applications. However, challenges in terms of sensitivity, fabrication techniques, and integration with deep learning algorithms remain. Several studies have explored modifying the rheological properties of liquid metals by adding nanoparticles or undergoing surface treatments, primarily to improve printability and stability. While some progress has been made in enhancing the sensitivity of liquid metal-based strain sensors, achieving high gauge factors comparable to other sensor technologies remains a significant hurdle. Furthermore, the integration of these sensors with advanced signal processing techniques, such as deep learning, for complex motion recognition, is an area of ongoing research. This paper contributes to the existing literature by demonstrating a novel approach to address both the fabrication challenges and the sensitivity limitations of liquid metal sensors, while simultaneously integrating them with a deep learning system for a specific application.

This study introduces a novel rheological modification strategy for liquid metals (LMs) to improve the sensitivity and printability of LM-based sensors. The researchers used a eutectic gallium-indium alloy (EGaIn) as the conductive LM and mixed it with nonconductive SiO₂ microparticles of two different sizes (40 µm and 6 µm). The weight fraction of SiO₂ particles was varied to investigate its effect on the rheological properties of the LM composites. Rheological tests, including frequency sweeps, amplitude sweeps, and flow sweeps, were performed using an Anton Paar Physica MCR 301 rheometer to determine the elastic modulus, yield stress, and viscosity of the composites. The results showed that increasing the weight fraction of SiO₂ particles significantly increased all three parameters, particularly with the larger 40 µm particles. This enhanced rheological behavior allowed for the 3D printing of complex structures using a Hyrel 3D Hydra 640 modular 3D printer with different nozzles. To evaluate the sensing performance, patterned LM-SiO₂ composite wires were printed onto an Ecoflex substrate, encapsulated with Ecoflex, and connected to copper tapes to create strain and pressure sensors. The sensors' electrical and mechanical properties were characterized. The gauge factor (GF), a measure of sensor sensitivity, was calculated from the change in resistance with applied strain. Finite element analysis (FEA) using ABAQUS software was employed to simulate the mechanical deformation of the sensors and investigate the mechanism behind the enhanced sensitivity observed with the SiO₂ particles. The FEA models considered the elastic properties of the Ecoflex and the LM-SiO₂ composite, showing that SiO₂ particles induced strain redistribution within the LM wires, leading to localized narrowing of conductive channels and increased resistance change. The strain sensors' stability and durability were assessed through cyclic loading and unloading tests. The pressure sensing capability was investigated by applying varying pressure loads and measuring the resistance change. Finally, to demonstrate the application, arrays of these sensors were integrated into a flexible tactile glove. The glove's ability to detect fist-clenching postures and punching strength was evaluated. A convolutional neural network (CNN) was trained using data collected from the glove during different boxing punches (jab, swing, uppercut, and combination punches). The CNN's performance was evaluated using a confusion matrix, assessing its ability to classify different boxing movements.

The incorporation of SiO₂ microparticles into the liquid metal significantly improved its rheological properties, increasing its viscosity, modulus, and yield stress. This modification enabled the 3D printing of complex structures using a layer-by-layer dispensing process. The resulting printed LM-SiO₂ composite sensors demonstrated enhanced sensitivity and robustness compared to sensors made with pure liquid metal. The gauge factor (GF) increased significantly with increasing SiO₂ concentration and particle size, reaching a GF of 23.91 for strain ranges between 200% and 300% with 0.6 mm SiO₂ particles – a tenfold increase compared to pure LM. This enhancement was attributed to strain redistribution around the SiO₂ particles, causing localized narrowing of the conductive LM channels. Finite element analysis confirmed this mechanism. The sensors exhibited excellent mechanical flexibility and durability, withstanding twisting, curling, stretching, and significant pressure loads. The cyclic loading-unloading tests showed reversibility and repeatability with negligible hysteresis, indicating good stability and reliability. The sensors were successfully integrated into a tactile glove for real-time monitoring of fist-clenching postures and punching strength. The ability to distinguish between correct and incorrect clenching postures demonstrated the sensor's sensitivity to subtle finger movements. The real-time monitoring of punching strength showed a clear correlation between impact force and sensor response. The integrated tactile glove, when combined with a trained convolutional neural network (CNN), achieved high accuracy (90.5%) in classifying different types of boxing punches. The CNN processed the signals from the middle finger sensor for simplicity. This accuracy demonstrates the potential for using this technology in advanced sports training and motion capture applications. The study showcases the success of the rheological modification strategy in significantly enhancing sensitivity, robustness, and the potential for integrating liquid metal sensors into sophisticated wearable applications.

This study successfully addressed the limitations of existing liquid metal-based sensors by demonstrating a novel approach that combines rheological modification with strain redistribution mechanics. The integration of SiO₂ particles not only enhanced the printability of the liquid metal but also significantly improved the sensitivity of the resulting sensors. The high gauge factor achieved, especially at large strains, surpasses many previously reported LM-based strain sensors, opening up possibilities for applications requiring high sensitivity, such as advanced prosthetics and humanoid robotics. The robustness and stability of the sensors under cyclic loading and pressure further enhance their practical applicability for wearable devices. The successful integration of these sensors into a tactile glove for boxing training demonstrates the potential of the technology in smart sports training, providing real-time feedback for athletes and coaches. The use of deep learning to classify complex boxing movements showcases the synergy between advanced materials and sophisticated signal processing, resulting in a multifunctional system with diverse applications.

This work demonstrated a high-sensitivity, printable liquid metal sensor with enhanced mechanical properties. The rheological modification using SiO₂ particles allowed for 3D printing of complex structures and significantly increased the gauge factor due to strain redistribution. Integration into a tactile glove successfully decoded clenching postures and punching strength, and deep-learning algorithms achieved high accuracy in recognizing boxing punches. This technology shows great promise for applications in smart sports training, soft robotics, and human-machine interfaces. Future research could focus on integrating sensor arrays with signal processing circuits and wireless transmission modules for improved usability and expanding applications to other complex movements and activities.

While the study demonstrated high accuracy in recognizing boxing punches, the current setup uses data from only the middle finger sensor for simplification of the deep learning model. The generalizability of the model to other individuals and boxing styles needs further investigation. Additionally, the long-term stability and durability of the sensors under prolonged and varied usage conditions require further evaluation. The current design might need further optimization to address potential challenges related to signal noise, sensor drift, and robustness in diverse environments. Future work should also explore miniaturization and integration of the sensors with wireless data transmission for truly seamless wearable applications.