Accurate detection and differentiation of human motion are crucial for various applications, including human-machine interaction, entertainment, and rehabilitation. Existing methods like camera-based motion capture systems and inertial sensors have limitations. Camera systems are computationally intensive and spatially restricted, while inertial sensors' rigidity limits their real-world applicability. More significantly, neither approach effectively measures the individual forces and stresses contributing to motion. Flexible strain sensors offer a promising alternative, but most existing designs fail to address the multimodal nature of joint movements, which involve a complex interplay of tension, bending, shear, and twisting forces. These sensors typically cannot differentiate these forces from their electrical signals. For example, wrist motion involves multiple degrees of freedom (bending, twisting, rotation), and conventional sensors produce electrical signals upon deformation, but they cannot identify the individual forces involved and their contribution to the overall motion. This paper addresses this critical limitation by proposing a novel concept of unimodal strain sensors capable of detecting and differentiating individual forces/stresses by decoupling their effects. This allows for the identification of the contributions of specific individual forces/stresses in complex motions, enabling accurate motion pattern recognition. The proposed unimodal strain sensors utilize uniaxially drawn piezoelectric poly L-lactic acid (PLLA) films. These sensors are then integrated into an integrated unimodal sensor (i-US) to demonstrate their ability to detect and differentiate the four primary deformation modes: tensioning, bending, shearing, and twisting. The i-US's capabilities are demonstrated through its integration into a sleeve and glove for capturing wrist and finger movements, demonstrating its potential for significant advancements in motion recognition and control systems.

Existing technologies for human motion capture include camera-based systems and inertial measurement units (IMUs). However, camera-based systems are computationally expensive and limited by spatial constraints. IMUs suffer from rigidity issues, hindering their use in daily-life scenarios. Both lack the ability to isolate and quantify individual forces/stresses within complex motions. Flexible and lightweight strain sensors, such as piezoelectric, resistive, capacitive, and triboelectric sensors, present a compelling alternative for motion monitoring. However, the complexity of human joint movement, involving multiple degrees of freedom (tension, bend, shear, twist), necessitates multiple sensors to capture the nuanced forces at play. Current sensor networks, while offering some solutions, compromise the naturalness of movement due to their size and quantity. Critically, these sensor networks often lack the ability to differentiate individual forces/stresses from the aggregated electrical signals they produce. The lack of this crucial differentiation capability hampers the accurate understanding of motion dynamics.

The research employed a multi-stage methodology, beginning with the theoretical design and simulation of unimodal strain sensors. Four types of unimodal sensors (tension, bend, shear, and twist) were designed using uniaxially drawn piezoelectric poly L-lactic acid (PLLA) films. The piezoelectric properties of the PLLA films were manipulated by controlling the cutting angle (CA) during fabrication. Finite element simulations (ABAQUS) were used to predict the piezoelectric response of the PLLA films under different deformation modes. Mesh refinement procedures ensured the accuracy and independence of simulation results from discretization resolution. The simulations verified the hypothesis that by strategically combining two PLLA films with specific CAs, unimodal sensors could be created, responding primarily to only one type of deformation while exhibiting minimal response to others. The fabrication process involved preparing uniaxially drawn piezoelectric PLLA films with varying drawing ratios (DRs) to optimize piezoelectric response. The optimal DR was determined through characterization techniques such as 2D-WAXD, DSC, and SEM, evaluating crystallinity and crystal orientation. The chosen PLLA film was then used to fabricate the four types of unimodal sensors according to the design concept: two PLLA film segments with strategically chosen CAs were bonded together to create each sensor. The piezoelectric responses of the fabricated unimodal sensors were extensively tested under controlled conditions (tensioning, bending, twisting, and shearing). Statistical analysis (P-value < 0.001) was used to evaluate the sensor's sensitivity and specificity to the target deformation mode. The stability and repeatability of the sensors were assessed through cyclical testing. The impact of electromagnetic (EM) interference and motion artifacts was evaluated by comparing shielded and unshielded sensor configurations. Subsequently, the four unimodal sensors were integrated into an integrated unimodal sensor (i-US) for testing. This i-US was then integrated into wearable devices (sleeve and glove) for real-world applications in measuring wrist and finger movements. The data collected from the wearable sensors were processed and analyzed to demonstrate the i-US's capability to differentiate complex movements. For finger motion classification and finger-air-writing, a LeNet-5 based convolutional neural network (CNN) was employed. The effectiveness of the system was evaluated through classification accuracy metrics.

The finite element simulations confirmed the feasibility of the unimodal sensor design, demonstrating that the selected PLLA film cutting angles resulted in sensors highly sensitive to their designated deformation mode (tension, bend, shear, or twist) and insensitive to other modes. The experimental results showed that each unimodal sensor exhibited a significantly higher piezoelectric response when subjected to its corresponding deformation, compared to other deformation types (P-value < 0.001). The integrated unimodal sensor (i-US) successfully differentiated tensioning, bending, twisting, and shearing deformations in controlled experiments and under manual manipulation. The i-US demonstrated a significant improvement in piezoelectric response compared to conventional single-layer PLLA sensors, particularly for bending and twisting. The self-shielding design of the unimodal sensors effectively mitigated EM interference and motion artifacts, leading to improved signal quality in real-world conditions. The i-Sleeve, incorporating the i-US, effectively differentiated complex wrist motions involving multiple degrees of freedom. The i-Glove, incorporating the i-US, accurately classified various finger motions (bending, shearing, turning, flexion and extension) with a mean classification accuracy of 90.2%. The finger-air-writing application using the i-Glove and a CNN algorithm achieved an 89.4% classification accuracy for 13 characters, significantly improving upon existing finger motion capture systems using similar approaches. The system's triboelectric effect was negligible due to effective grounding. The key advantage of this system compared to others is it only requires a single i-US to achieve the same or better results, compared to other systems that require many distributed sensors.

The findings demonstrate the successful development and application of flexible, unimodal strain sensors for the first time. The i-US provides a significant advancement in motion capture technology by overcoming the limitations of existing methods that cannot reliably differentiate individual forces/stresses within complex motions. The ability to isolate and quantify these individual forces is crucial for improving the accuracy and sophistication of motion control systems. The high classification accuracy achieved in finger-air-writing suggests potential applications in virtual and augmented reality interfaces. The improved signal quality due to the self-shielding design and the absence of significant triboelectric effects highlight the practical advantages of this approach. The results suggest that the unimodal sensor design principles could be extended to other piezoelectric materials, expanding the range of applications.

This research successfully demonstrated a novel concept of flexible unimodal strain sensors using piezoelectric PLLA films. The i-US, integrating these sensors, effectively detects and differentiates individual forces/stresses in complex motions, representing a significant step forward in motion recognition and control. Future work could explore the application of these sensors in other areas, such as robotics, prosthetics, and medical diagnostics. Investigating other piezoelectric materials and sensor configurations could further improve the performance and expand the capabilities of these sensors.

While the study demonstrated high accuracy in controlled environments, further testing in diverse real-world scenarios is needed to evaluate robustness and generalizability. The current CNN model could be optimized further to improve classification accuracy and reduce computational requirements. The size and fabrication complexity of the i-US might limit its scalability for certain applications. Long-term stability and durability of the sensors under extensive use require more investigation.