Defect-dipole coupling and anisotropic 2D crystallization in Bi/Mn co-doped BaTiO3 for flexible pressure sensors with integrated AI-based motion classification

Abstract

The development of lead-free piezoelectrics with high sensitivity and real-time signal intelligence is critical for advancing wearable electronics, motion-tracking systems, and self-powered biomedical devices. Barium titanate (BaTiO3) is a promising alternative, but intrinsic charge screening and limited dipole alignment restrict its performance. Here, a dual-site defect–dipole coupling strategy addresses these issues through Bi3+/Mn4+co-doping at the A- and B-sites of BaTiO3. Bi3+, with a stereochemically active 6s2lone pair and smaller ionic radius than Ba2+, induces strong off-centre displacement and lattice tetragonality, while Mn4+acts as a redox-stable trap for oxygen-vacancy electrons, suppressing internal charge screening. A microwave-assisted sol–gel process with PEG-mediated crystallization enables anisotropic 2D BaTiO3 microsheets in a single-step, low-temperature synthesis—unlike conventional multi-step hydrothermal methods. The optimized pellet-like composite film (C3), comprising 25 wt% Ba0.9Bi0.1Ti0.9Mn0.1O3 in a PDMS matrix, shows high dielectric constant (ε′ ≈ 138), ultra-low loss (tan δ ≈ 0.0052), and strong piezoelectric response (d33 ≈ 88 pC N−1, g33 ≈ 0.072 V m N−1). Under dynamic loading, the C3-based sensor delivers ∼97 V peak-to-peak output, 5.31 V kPa−1sensitivity, and a detection limit of 0.58 kPa, enabling stable signal capture during motions like running, squatting, and hand–object interaction. To extend functionality, a lightweight AI model is integrated for on-device biomechanical signal classification. The DrCIF model achieves the highest accuracy (≈89.97%), outperforming CNNs and ensemble methods. This framework, which combines defect engineering, anisotropic crystallization, and AI-assisted interpretation, offers a scalable pathway to intelligent, lead-free piezoelectric sensors for sports analytics, soft robotics, and wearable healthcare. Code available at:https://github.com/Zahid672/Pressure_Sensor_Classifcation_Via_DrCIF. © 2026 Elsevier Ltd and Techna Group S.r.l. All rights are reserved, including those for text and data mining, AI training, and similar technologies.