Temporal responses of chemically diverse sensor arrays for machine olfaction using artificial intelligence

Abstract

Machine sensing has progressed through innovative breakthroughs. Perhaps the most significant advancements have been in the field of machine vision through the advent of the charge-coupled-device (CCD) chip and the development of artificial neural networks. This is evident in the range of sophisticated behaviors that have been replicated including facial recognition and object tracking. Due to parallel processing of large amounts of temporally evolving data, these systems can rapidly learn, adapt and predict in changing environments. While there have been significant advances in chemically diverse sensor arrays that mimic the mammalian olfactory system under well-defined conditions, relatively little work has been done to understand the temporal response of these systems and to develop neural architectures that will enable machines to become chemically aware of their surroundings. In this work, the temporal response behavior for sensors is investigated, demonstrating that both the response evolution and steady-state provide useful information about the chemical environment under dynamic conditions and that neural networks are ideally suited to combine this information in unsupervised learning to perceive odors in a dynamic environment.