The Internet of Things (IoT) has significantly advanced the development of electronic skin (e-skin), aiming to replicate or surpass human skin functionalities. A crucial function of human skin is thermoregulation, achieved primarily through sweat glands. Current e-skin thermoregulation methods often focus on either heat dissipation or insulation, lacking dynamic temperature adjustment. The need for dynamic thermoregulation in e-skin is paramount to ensure comfort and prevent health issues arising from extreme temperatures. Phase change materials (PCMs), such as paraffin, offer a promising solution due to their ability to absorb and release heat at a nearly constant temperature during phase transitions. Microencapsulation prevents paraffin leakage, ensuring the integrity and performance of the e-skin. Additionally, e-skin requires effective sensing capabilities, often implemented using triboelectric nanogenerators (TENGs) due to their low cost and simple structure. However, environmental factors can significantly affect TENG signals. Deep learning provides an effective solution to accurately identify and process these signals, achieving high accuracy in handwriting recognition, even surpassing traditional methods. This research integrates microencapsulated paraffin for thermoregulation with a TENG-based sensor and a deep learning model for handwriting recognition, creating a novel self-powered real-time handwriting display system.

The literature extensively covers various aspects of e-skin development, including flexible electronics, tactile sensing, air permeability, and thermoregulation. Existing thermoregulating e-skins often utilize porous materials for heat dissipation or thermal insulation materials to reduce heat transfer, both of which lack dynamic adjustment. Research on PCMs, particularly paraffin, demonstrates their high energy storage density and suitability for thermal management. Microencapsulation techniques have been used to address the leakage issues associated with PCMs in flexible devices. Various sensing mechanisms exist for e-skin, including capacitive, resistive, and triboelectric sensors, with TENGs gaining popularity due to their low-power consumption and simple design. Deep learning has significantly improved pattern recognition in various applications, showing promising potential in e-skin signal processing and real-time recognition of handwritten characters. However, integrating these technologies into a single, self-powered system for real-time handwriting recognition and dynamic thermoregulation remains a challenge addressed in this study.

This research involved several key steps: 1. **Paraffin Microencapsulation:** Paraffin was microencapsulated using a three-stage process. First, a water-soluble prepolymer solution was created through the reaction of urea, melamine, and formaldehyde. Second, paraffin was emulsified in the prepolymer solution. Finally, the pH was adjusted to induce polymerization and solidify the microcapsules. 2. **ME-skin Fabrication:** Microencapsulated paraffin (M-paraffin) was dispersed in ethanol and dropped onto a glass sheet. After ethanol evaporation, silicone elastomer was added and cured, forming a layered structure with M-paraffin concentrated on one side. This layer was carefully peeled off from the substrate to create the ME-skin. 3. **Characterization:** The prepared M-paraffin and ME-skin were characterized using scanning electron microscopy (SEM) to examine their morphology and structure, differential scanning calorimetry (DSC) to determine the phase transition temperature and enthalpy of M-paraffin, and a multifunctional tensile machine to assess the mechanical properties of the ME-skin. Temperature changes were monitored using a dual-channel temperature sensor. 4. **Electrical Output Measurement:** The ME-skin's triboelectric nanogenerator (TENG) properties were investigated by measuring open-circuit voltage, short-circuit current, and transferred charge using an oscilloscope and current preamplifier. The response of the TENG to various types of human motion was evaluated, including finger bending, wrist bending, and arm bending. 5. **Handwriting Signal Acquisition and Processing:** Handwritten letters were generated on the ME-skin, and the corresponding voltage signals were recorded. These signals were then pre-processed through techniques like normalization and resampling before feeding into the deep learning model. 6. **Deep Learning Model Training and Testing:** A 1D-Convolutional Neural Network (CNN) model was developed and trained to identify the unique patterns of voltage signals generated by different handwritten letters. The training dataset consisted of 80 sets of data per letter, while 20 sets were reserved for testing the model's accuracy. 7. **Real-time Handwriting System Integration:** The trained deep learning model was integrated into a real-time system that uses LabVIEW software for signal acquisition, processing, and display of recognized characters. This system connects the ME-skin, signal acquisition hardware, and the trained CNN model, providing real-time handwriting recognition and display.

The study successfully demonstrated a novel electronic skin (ME-skin) with integrated dynamic thermoregulation and real-time handwriting recognition capabilities. Key findings include: 1. **Effective Thermoregulation:** The incorporation of microencapsulated paraffin into the ME-skin resulted in a significant reduction in temperature fluctuation under both high and low ambient temperatures, effectively mimicking human skin's thermoregulatory function. The temperature difference between ME-skin and control silicone elastomer was approximately 2°C on average, reaching 4°C under certain conditions. 2. **Robust TENG Performance:** The ME-skin exhibited reliable triboelectric nanogenerator (TENG) properties, generating a significant open-circuit voltage (average peak of 228V) and short-circuit current (average peak of 3.82 µA) upon contact and separation. The generated signals showed consistent responses to various human motions (finger, wrist, and arm movements). 3. **High Accuracy Handwriting Recognition:** Using a deep learning-based 1D-CNN model, the system achieved remarkable accuracy in recognizing handwritten letters (98.13% accuracy for the 8-letter dataset and 90.71% for a more challenging 12-letter dataset), showcasing the potential for practical applications. 4. **Real-time Handwriting Display:** The developed system successfully integrated the ME-skin, signal acquisition, and deep learning model into a functioning real-time handwriting display, directly translating written letters into real-time display output. 5. **Material Stability and Performance:** The ME-skin demonstrated excellent mechanical flexibility and durability, retaining its functionality after repeated stretching, twisting, rolling, and bending. The encapsulated paraffin showed high thermal stability over multiple heating-cooling cycles.

The successful integration of dynamic thermoregulation and real-time handwriting recognition into a single e-skin device significantly advances the field of flexible electronics and human-machine interfaces. The high accuracy achieved by the deep learning model demonstrates the potential for practical applications in various fields, including virtual reality/augmented reality (VR/AR) systems, smart homes, healthcare monitoring, and personalized assistive technologies. The design of the e-skin is bio-inspired, utilizing a microencapsulated paraffin layer analogous to human sweat glands, further improving its compatibility for wearable applications. The self-powered nature of the system eliminates the need for external power sources, enhancing its portability and practicality. The results suggest that the deep learning model is robust enough to overcome variations in writing styles and environmental conditions, making the system highly adaptable for diverse users. Future research could focus on expanding the character set recognized by the system, improving the system's robustness in varied environmental conditions and exploring applications beyond handwriting recognition, such as gesture recognition or other types of sensory inputs.

This research successfully developed a novel electronic skin (ME-skin) that integrates dynamic thermoregulation and real-time handwriting recognition using a deep learning model. The achieved high recognition accuracy (98.13%) and self-powered operation demonstrate its promising applications in diverse fields such as human-computer interaction, VR/AR, healthcare, and smart homes. Future research will explore expanding the alphabet recognized, enhancing robustness against external factors, and applying the technology to other human-machine interface scenarios, such as gesture recognition.

While the study demonstrates significant advancements, certain limitations exist. The current system recognizes a limited set of characters (initially 8, expanding to 12), limiting its immediate applicability to broader contexts. The accuracy, while impressive, might be affected by significant variations in writing speed, pressure, and environmental conditions. Further research is needed to assess the long-term stability and durability of the ME-skin under continuous use. The impact of different skin types and body locations on the signal quality needs further investigation.