Multimodal Perception through Synergistic Interface Fusion and Machine-Learning Decoupling in Vertically Integrated Flexible Sensors

Abstract

Abstract Flexible integrated sensors have been constrained by signal crosstalk from planar layouts and material coupling, restricting the capability for high-precision multiparameter sensing. Here, we develop a biomimetic vertically integrated flexible sensor (BVIS) based on a poly(vinyl alcohol)-cellulose nanofiber-reduced graphene oxide multiwalled carbon nanotube (PVA-CNF-RGO-MCNT, PCRM) material system. Serpentine temperature-sensitive, humidity-sensitive, and square pressure-sensitive arrays are assembled into an ultrathin six-layer architecture through a water-induced interfacial fusion mechanism. The molecular-level interpenetrating network formed by the PCRM composite enhances the interfacial adhesion and suppresses multimodal interference. The humidity units exhibit sensitivities of 2.8%/%RH (70–80% RH) and 8.57%/%RH (80–95% RH), respectively, while the tangential interfacial and T-peel strengths reach 2.592 and 0.285 MPa. By combining structural isolation with a random forest algorithm for signal decoupling, the platform achieves recognition accuracies of 96.79% and 99.8% for temperature-position and humidity-position mapping, respectively. The study establishes a novel vertical-layered multimodal sensing paradigm for highly integrated and anticrosstalk flexible intelligent sensing systems.