Strain-engineered active site migration and machine learning-assisted NO₂ recognition in UV-activated Cs-doped WO_3-x nanosheets with operando mechanistic insights

Pristine metal oxide gas sensors often suffer from poor selectivity and limited polarizability. The objective of this work is to overcome these bottlenecks by developing an advanced material tailored specifically for highly sensitive and intelligent NO₂ sensing applications. Consequently, oxygen vacancy-rich 2D ultrathin Cs-doped WO_3-x nanosheets were synthesized via a solvothermal method. KPFM, UPS, and TAS confirmed that optimal Cs doping (0.08 mol%) optimizes the surface work function and drastically prolongs the photo-generated carrier lifetime (τ₃ = 40.45 ns), maximizing UV-excited charge separation. Under 365 nm UV illumination at 175℃, the sensor exhibited an ultra-high response of 118.4 at 10 ppm NO₂, a 2.41 ppb limit of detection, ultra-fast kinetics (10 s/7 s response/recovery), superior tolerance to moisture interference (stable across a 30–90% RH gradient), and robust 30-day long-term stability. To resolve cross-selectivity limitations, a machine learning model integrating Random Forest (RF) feature extraction and Linear Discriminant Analysis (LDA) was deployed, achieving 98.8% training-set classification accuracy in multi-component gas mixtures. Mechanistically, in situ/operando Raman Two-Dimensional Correlation Spectroscopy (2D-COS) and Density Functional Theory (DFT) revealed that steric hindrance from large Cs⁺ ions induces localized W-O lattice strain while effectively eliminating surface moisture side reactions. This strain creates distorted sites with ultra-high adsorption energy (-1.3538 eV), promoting a preferential shift of the dominant NO₂ adsorption sites from intrinsic W-O-W lattice sites toward Cs-adjacent distorted O-W-O sites. By coupling strain-engineered active site migration with artificial intelligence, this work provides a powerful blueprint for the next generation of highly selective, intelligent gas sensors.