An ultralight, flexible, and biocompatible all-fiber motion sensor for artificial intelligence wearable electronics

Human body motion signals convey key physiological information related to movement, respiration, pulse, cardiovascular status, gestures, and symptoms. Reliable, convenient acquisition and processing of these signals are crucial for continuous daily healthcare. Advances in flexible electronics and intelligent medical devices enable real-time, portable, flexible, and reliable motion monitoring that can improve safety and quality of life. Skin-interface devices (E-skins, textile and hydrogel sensors) have been explored for motion monitoring, disease treatment, drug delivery, and analyses such as glucose or blood pressure. Despite progress using liquid metals, hydrogels, graphene aerogels, functional fibers, and piezoelectrics, many devices still suffer from rigid components, relatively large weight, lower comfort, and processing constraints that limit daily use. Safety for long-term skin contact requires high moisture and gas permeability to avoid inflammation or short circuits under humidity; preventing silver ion release is also critical. Furthermore, classifying subtle or similar motion signals remains challenging with traditional frequency-domain methods. Machine learning can extract high-dimensional features from irregular signals and, combined with advanced sensors, promises accurate motion tracking for next-generation AI healthcare. In this study, the authors propose an ultralight, flexible, and biocompatible all-fiber motion sensor (AFMS). The functional material—radial anisotropic porous silver fibers (RAPSF)—is fabricated via a modified blow-spinning technique with environmental control to induce phase separation and temperature-controlled grain growth, followed by UV photoreduction. The RAPSF film exhibits single-fiber conductivity of ~7.96 × 10³ S m⁻¹ and shape-dependent resistance tunability (80–213 Ω). A multilayer all-fiber sensor assembled from RAPSF, PAN, and AgNWs shows ultralight weight (68.7 mg cm⁻³), high moisture and gas permeability, biocompatibility, and excellent motion recognition, including AI-based throat motion identification with accuracy above 85%, highlighting potential for wearable electronics and intelligent medical technology.

The paper reviews recent developments in wearable, skin-interfaced sensors for motion monitoring and healthcare, including E-skins, textile sensors, and hydrogels. Materials explored include liquid metals, hydrogels, graphene aerogels, functional fibers, and piezoelectrics, enabling flexible devices but often with trade-offs in weight, comfort, rigidity, and processing constraints. Biocompatibility and safety concerns for long-term skin contact emphasize high moisture/gas permeability to prevent skin inflammation and device failure under sweat or humidity, as well as minimizing metal ion (e.g., Ag⁺) release. Traditional signal classification methods (e.g., DFT) are effective for large-amplitude, distinct-frequency motions (finger, elbow, knee bending) but struggle with subtle, similar, or noisy signals (e.g., throat motions). Machine learning has shown promise in domains such as optical recognition, object classification, and tactile sensing in robotics, offering high-dimensional feature extraction for more accurate classification of irregular biosignals, motivating its integration with wearable sensors.

Fabrication of RAPSF: A modified blow-spinning system with a liquid–gas two-channel six-needle module, compressed-air temperature control, and a heated collector was used. A precursor solution was prepared by dispersing 0.25 g PVP in 2.10 g acetonitrile (stir 5 h), adding 1.50 g AgNO₃ and 50 µL fluorocarbon surfactant (FS-3100), then introducing dichloromethane (0–10 wt%) to induce phase separation. The precursor was blow-spun onto a 250 °C metal collector at 1.5 mL h⁻¹ with 3.0 m³ h⁻¹ heated air (to ~80 °C). Rapid evaporation of dichloromethane during jetting created surface porosity, while the hot collector promoted unilateral prenucleation and grain growth. Photo-reduction was performed by UV irradiation (two 250 W lamps) for 3 h to form metallic silver particles within a PVP skeleton, yielding radial anisotropic porous fibers with heterogeneous top (granular islands) and bottom (continuous film) surfaces and internal porosity (confirmed by SEM, XRD, XPS, BET). Single-fiber conductivity and film resistance were evaluated under varying bending radii; FEA analyzed stress and conductive path changes during bending. Assembly of AFMS: A PAN fiber substrate was produced by blow spinning (PAN 0.14 g in 100 g DMSO; 70 °C; injection 2.0 mL h⁻¹; airflow 4.5 m³ h⁻¹) and dried. AgNWs were synthesized via a polyol method (PVP in EG; AgNO₃; CuCl₂·2H₂O; 130 °C, 3 h; washed; dispersed in PVP/ethanol). AgNWs/PVP/alcohol dispersion was dip-coated locally onto PAN to form electrodes. RAPSF film was hot-pressed onto PAN (60 °C, 750 Pa). The top PAN layer was spray-coated with PVB alcohol dispersion as adhesive, forming a multilayer all-fiber device. Characterization and testing: Micromorphology (FE-SEM with EDS), crystallography (XRD), surface chemistry (XPS), thermal analysis (TGA/DTA), BET surface/pore analysis, and IRT imaging were performed. Electrical measurements used a probe station and electrochemical workstation (IV/CV) on single fibers and films (mounted on flexible PET), with I–V testing under different bending radii and cyclic bending durability. Breathability was assessed by moisture permeability and gas permeability (automatic filter material tester). Silver ion release was quantified by ICP-MS after 24 h saline stirring at 600 rpm. Oxidation stability was tested in air at 80 °C for up to 96 h. In vivo biocompatibility was assessed by subcutaneous implantation in mice (C57BL/6) with histology at 1, 5, 10, 15 days. Motion sensing: AFMS was applied to fingers, elbow, knee, and throat; response and recovery times were measured. Signals for gestures (five-finger array) and throat activities (speaking “Hello,” “Goodbye,” swallowing, coughing) were collected. Signal analysis and AI classification: For distinct motions, DFT distinguished signal frequencies. For throat signals, three ML models (XGBoost, KNN, SVM) were evaluated with 5-fold cross-validation, using an 80/20 train-test split and varying frame lengths; performance and speed were compared, and a trained XGBoost model classified a 58 s continuous mixed signal.

- Materials and structure: RAPSFs exhibit a porous, radial anisotropic architecture with heterogeneous top (granular islands) and bottom (continuous film) surfaces and internal porosity; increased pore/surface area vs. non-porous silver fibers (BET). Overall RAPSF film density is ultralow at 68.7 mg cm⁻³; sample weight 7.95 mg. - Electrical performance: Single-fiber conductivity ~7.96 × 10³ S m⁻¹. RAPSF film resistance tunable from 80 to 213 Ω by bending radius; linear relation R = −8.73 r + 214.78 (r in mm), R² ≈ 0.991, with stable, reversible response due to polymer skeleton and bottom continuous silver layer. Durability: >10,000 bending cycles at r = 8 mm with stable performance. Response time ~45 ms; recovery time ~62 ms. - Breathability and biostability: High moisture permeability 59.7 g m⁻² h⁻¹ (≈1432.8 g m⁻² day⁻¹) and low gas resistance 9.8 × 10⁻³ Pa at 31.6 L min⁻¹, comparable to nickel foam and superior to PI, PET, and PDMS films. Silver ion release after 24 h saline stirring is low (0.1359 mg L⁻¹), indicating strong bonding and structural stability. Oxidation stability: ΔR/R increases only ~0.9% after 96 h at 80 °C. - Biocompatibility: In vivo implantation in mice shows initial transient inflammatory response resolving by day 10; by day 15, no irreversible tissue necrosis and overall structure comparable to control, indicating good biocompatibility. - Motion sensing: Clear, repeatable signals for finger, elbow, and knee motions; five-sensor finger array distinguishes gestures (“good,” “yes,” “OK”). - AI classification: For throat motions (speaking “Hello,” “Goodbye,” swallowing, coughing), traditional DFT insufficient; machine learning applied with 5-fold cross-validation. XGBoost achieves the highest accuracy and stable performance for frame lengths >180 points, with overall throat state identification accuracy above 85%. The trained model accurately classifies a 58 s continuous mixed signal containing four throat motion types and noise.

The study addresses the need for ultralight, flexible, stable, and biocompatible wearable motion sensors by introducing RAPSF-based all-fiber devices. The anisotropic porous silver architecture enables shape-dependent resistance modulation with high sensitivity while maintaining mechanical stability and durability through a flexible PVP skeleton and continuous bottom silver layer. The all-fiber assembly with PAN substrates and AgNW electrodes preserves high moisture and gas permeability, improving comfort and safety for long-term skin contact and minimizing risks such as sweat-induced short circuits and silver ion exposure. The AFMS captures a wide range of body motions with fast response and recovery and demonstrates robust performance in gesture recognition. Critically, integrating machine learning overcomes the limitations of traditional frequency-domain analysis for subtle, similar throat motions, achieving high (>85%) classification accuracy with XGBoost and effective real-time identification in continuous signals. These results validate the approach of combining engineered anisotropic conductive fibers with AI to realize practical, imperceptible, and safe wearable sensors for next-generation intelligent healthcare.

The work presents a modified blow-spinning and UV photoreduction route to fabricate radial anisotropic porous silver fibers with tunable, shape-dependent electrical properties and exceptional lightness. Layer-by-layer assembly with PAN and AgNWs yields an all-fiber motion sensor combining ultralow density, high breathability, mechanical robustness, biocompatibility, and fast response. The AFMS effectively monitors joint and finger motions and, when paired with machine learning (XGBoost), accurately classifies subtle throat motions with accuracy above 85%, demonstrating potential for AI-enabled wearable healthcare and early symptom monitoring. The findings highlight RAPSF-based all-fiber sensors as promising platforms for flexible electronics and AI healthcare, with potential for further development toward broader motion categories and continuous monitoring in real-world scenarios.