Gas sensor arrays often lack discrimination power to different analytes and robustness to interferants, limiting their success outside of research laboratories. This is primarily due to the widely sensitive (thus weakly-selective) nature of the constituent sensors. Here, the effect of orthogonality on array accuracy and precision by selective sensor design is investigated. Therefore, arrays of (2-5) selective and non-selective sensors are formed by systematically altering array size and composition. Their performance is evaluated with 60 random combinations of ammonia, acetone and ethanol at ppb to low ppm concentrations. Best analyte predictions with high coefficients of determination (R2) of 0.96 for ammonia, 0.99 for acetone and 0.88 for ethanol are obtained with an array featuring high degree of orthogonality. This is achieved by using distinctly selective sensors (Si:MoO3 for ammonia and Si:WO3 for acetone together with Si:SnO2) that improve discrimination power and stability of the regression coefficients. On the other hand, arrays with collinear sensors (Pd:SnO2, Pt:SnO2 and Si:SnO2) hardly improve gas predictions having R2 of 0.01, 0.86 and 0.28 for ammonia, acetone and ethanol, respectively. Sometimes they even exhibited lower coefficient of determination than single sensors as a Si:MoO3 sensor alone predicts ammonia better with a R2 of 0.68. Graphical abstract Conventional arrays (red) with weakly-selective sensors span a significantly smaller volume in the analyte space than arrays containing distinctly-selective sensors (orthogonal array, green). Orthogonal arrays feature better accuracy and precision than conventional arrays in mixtures of ammonia, acetone and ethanol.
Keywords: Acetone; Ammonia; Electronic nose; Ethanol; Flame spray pyrolysis; Gas sensor; MoO3; SnO2; WO3.