An Anthropomimetic Olfactory System for the Digitalization of In-Vehicle Air Quality

The digitization of odors is essential for precise control, particularly in managing in-vehicle air quality. This study presents an anthropomimetic olfactory system that mimics professional sniffers' evaluations of automotive materials. Specifically, the system consists of multiple gas sensors functioning like olfactory receptors, a circuit board for signal processing, and an algorithm to interpret odor characteristics. To enhance odor detection, the sensor array with broad sensitivity is selected to identify key automotive odors, such as 4-methyl-2-pentanone and ethyl acetate from materials such as PVC fabric and steering wheels. To furtherly improve the measurement accuracy, first-order hysteresis and least mean squares filters are applied to reduce background noise and sensor drift. As the gradient boosting regression tree (GBRT) algorithm excels in highly nonlinear processing, it successfully fulfills the complex requirements of in-vehicle odor detection. The algorithm was subsequently trained and validated through a novel dataset with 5 sniffers' label. As a result, this anthropomimetic olfactory system succeeds in replicating sniffers' subjective evaluation results of odor intensity and comfort, with accuracy rates of 99.0% and 98.6%, respectively. This research represents a significant step toward digitalizing in-vehicle odor assessment, paving the way for automated odor control in large-scale manufacturing.