Wearable flexible pressure sensors (FPSs) are crucial for continuous biophysical monitoring in healthcare and fitness, enabling early warning and rapid rehabilitation. While progress has been made with piezoresistive, piezocapacitive, piezoelectric, and triboelectric sensors, FPSs combining ultrahigh linear sensing range (near 1 MPa) and sensitivity (over 10 kPa⁻¹) remain rare. The need for high durability, especially during strenuous activity or emergencies involving high impact forces, further complicates the challenge. Iontronic piezocapacitive (IPC) sensors, with their high signal-to-noise ratio and broad sensing range, offer a potential solution. These sensors leverage the electrical double-layer (EDL) effect to achieve high capacitance, which increases with applied pressure due to enlarged surface area and reduced ion transport distance. Previous research has demonstrated IPC sensors with high sensitivity and wide sensing range using elastic ionic-electronic interfaces and graded intrafillable architectures. However, stretchable iontronic pressure sensors (SIPSs) maintaining both high sensitivity and a broad sensing range remain a significant challenge. The integration of deep learning with SIPS is expected to enhance biophysical signal monitoring and patient rehabilitation after surgery. In-sensor deep learning allows for real-time classification and prediction, as demonstrated in previous studies on ocular motion detection and hand gesture recognition. This study focuses on the development of an on-skin SIPS with an ultrabroad linear working range for biophysical monitoring and in-sensor dynamic deep learning for knee rehabilitation. The goal is to create a sensor capable of handling high impact forces while maintaining accuracy and sensitivity, facilitating advanced AI-aided rehabilitation strategies.

Existing literature showcases significant advancements in flexible pressure sensors based on various sensing mechanisms, including piezoresistive, piezocapacitive, piezoelectric, and triboelectric effects. While these technologies have yielded promising results, the simultaneous achievement of both ultrahigh linear sensing range and high sensitivity remains a significant hurdle. Iontronic piezocapacitive (IPC) sensors have emerged as a potential solution due to their ability to achieve high signal-to-noise ratio and broad sensing range through the electrical double-layer (EDL) effect. Studies have demonstrated IPC sensors with improved sensitivity and wide sensing range by employing elastic ionic-electronic interfaces and innovative structural designs like graded intrafillable architectures. However, the development of stretchable iontronic pressure sensors (SIPSs) that maintain both high sensitivity and a broad sensing range is still in its nascent stages. The field also recognizes the potential of integrating deep learning techniques with flexible wearable sensors for enhanced biophysical monitoring and personalized rehabilitation. Several studies have demonstrated the effectiveness of in-sensor deep learning algorithms for tasks such as ocular motion detection and hand gesture recognition. This underscores the potential of using AI-driven methods in conjunction with advanced sensor technologies for improved healthcare applications.

This study developed a novel on-skin SIPS with an ultrabroad linear working range for monitoring biophysical features and implementing in-sensor dynamic deep learning for knee rehabilitation. The SIPS comprises an encapsulated soft silicone layer, serpentine polyimide (PI) substrate electrodes coated with a gold and carbon-graphite flake (C-GF) composite, and a nanostructured ionic membrane (PVA-KOH). The serpentine electrode design enhances stretchability, enabling conformal attachment to various body surfaces. The ionic membrane contributes to the EDL effect, enabling high capacitance and sensitivity. A subtractive manufacturing approach was used for creating the electrode patterns. The sensor's performance was characterized by analyzing its sensitivity (defined as δ(ΔC/C₀)/δ(ΔP), where C₀ is the initial capacitance and ΔC and ΔP are capacitance and pressure variations), linear range, response time, relaxation time, and long-term durability. Finite element analysis (FEA) was used to simulate the sensor's mechanical behavior under various deformations. The SIPS data was integrated with a neuro-inspired fully convolutional network (FCN) algorithm for real-time analysis and prediction of knee joint postures during rehabilitation exercises. The FCN was trained on data collected from the SIPS during various knee movements. The performance of the FCN was evaluated based on its accuracy in predicting knee joint angles.

The fabricated SIPS achieved an ultrabroad linear sensing range of 1 MPa and high sensitivity (12.43 kPa⁻¹ from 325 kPa to 1 MPa, and 23.10 kPa⁻¹ below 325 kPa). High pressure resolution (6.4 Pa) and rapid response/relaxation times (14.2 ms and 13.9 ms, respectively) were also observed. The sensor maintained its functionality even under significant stretching (up to 20%) and high impact forces, demonstrating its anti-impact capabilities. Long-term durability tests showed no significant capacitance decay after 12,000 cycles. The SIPS successfully monitored various biophysical signals including pulse waves, muscle movements, and plantar pressure. Importantly, integration with the FCN algorithm enabled accurate prediction of knee joint postures, offering a valuable tool for AI-aided knee rehabilitation. Control experiments using pure PDMS and copper electrodes yielded significantly lower sensitivities, highlighting the synergistic effects of the chosen materials and design.

The high sensitivity, ultrabroad linear range, and excellent stretchability of the SIPS address the limitations of existing pressure sensors. The sensor's ability to function under high impact forces and its integration with the FCN algorithm for deep learning provide a powerful combination for various applications. The results demonstrate the potential of the SIPS for monitoring various biophysical signals and facilitating AI-driven rehabilitation strategies. The superior performance compared to control experiments underscores the importance of material selection and design optimization. The accurate prediction of knee joint postures using the FCN algorithm shows significant promise for personalized rehabilitation protocols, potentially leading to improved patient outcomes and faster recovery after orthopedic surgery. The combination of a highly sensitive and robust sensor with in-sensor deep learning represents a significant advancement in the field of wearable sensors for healthcare.

This study successfully developed a highly sensitive, stretchable, and anti-impact SIPS with an ultrabroad linear range. The sensor's performance characteristics and integration with a deep learning algorithm showcase its potential for advanced biophysical monitoring and AI-aided rehabilitation. Future research could focus on further miniaturization of the device, exploring alternative materials for improved biocompatibility, and expanding the applications to other rehabilitation contexts. Investigating different deep learning architectures and exploring wireless data transmission would enhance the device's practicality and clinical usability.

While the SIPS demonstrates excellent performance, there are some limitations. The long-term stability of the PVA-KOH ionic film under extreme conditions requires further investigation. The current study focused on knee rehabilitation; future studies should explore the applicability of the SIPS to other joints and movement monitoring scenarios. The FCN model was trained on a specific dataset and may require further training for improved generalization to different individuals and clinical situations. The study is limited by a relatively small number of subjects involved in the knee rehabilitation experiments.