R.A. Potyrailo, S. Shan, T. Wang
GE Vernova Advanced Research Center,
United States
Keywords: AI-enabled, gas sensors, high-accuracy, robust, economic impact, societal impact
Summary:
AI impacts our lives in many ways, e.g. when AI processes data from sensors to drive a car or to diagnose medical conditions without human input. However, in this talk we will discuss where and how AI can create a new generation of sensors. But why do we need new sensors? Many sensors are already good at providing data with desired accuracy and affordability – e.g. microphones, accelerometers, gyroscopes in mobile devices are annually shipped at billions of units. We will discuss how AI can replace “not good enough” sensors with desired good sensors. Today, gas sensors are acknowledged as “really not good enough”. They are based on last-century designs to have a simple single output (e.g., resistance, current) to detect relatively large concentrations of gases for safety. Unfortunately, by design, single-output sensors have poor gas selectivity. “Simple sensors perform best when pollution levels are high and when the compound of interest swamps others” because “the biggest headaches are caused by interfering chemicals” – highlighted in Nature (2016). The negative impact of “not good enough” sensors is substantial. For example, US EPA points out (2024, 2013) that such “sensors have inherent limitations that are critical to understand before collecting and interpreting the data”. As a result, “data of poor or unknown quality is less useful than no data since it can lead to wrong decisions”. We create AI-enabled new generation multiparameter (multivariable) gas sensors to disrupt century-old approaches in air monitoring. Our first goal is to replace traditional analytical instruments based on last-century designs to detect volatiles in complex samples (urban air, exhaled breath, etc.). These traditional analytical instruments are power-hungry, dishwasher- or microwave-oven-size, require substantial maintenance, but are an unavoidable alternative to “not good enough” gas sensors. Our second goal is to replace sniffer dogs introduced in 1888 to detect suspicious odors. Today, ~100,000 dogs in law enforcement in USA alone are an unavoidable alternative to “not good enough” gas sensors to detect explosives, drugs, and illegal agricultural products. We apply AI to design new generation high-accuracy robust gas sensors to be on par or better performers than traditional analytical instruments, yet to consume much less materials and energy to manufacture and energy to operate, providing green sustainable gas-monitoring solutions. These high-accuracy sensors will also be on par or better than sniffer dogs that require costly training and manual handling. Our AI-enabled next generation gas sensors are built across the electromagnetic spectrum from radio and microwave to optical frequencies. These sensors implement two distinct design components: (1) physical transducer designed to operate with independent variables and excitation conditions, and (2) sensing material designed with different response mechanisms to multiple gases. We use AI to design transducers and sensing materials and to boost data analytics capabilities. We will discuss achieved expected benefits –multi-gas detection– based on our design concepts. We will also discuss unexpected benefits –increased signal-to-noise, reduced/eliminated temperature effects, reduced/eliminated baseline drift. These next-generation gas sensors will be impactful in energy- and materials-efficient robust solutions for diverse gas-sensing scenarios.