Chemical Sensing for Safer Environments with NevadaNano and Machine Learning

Bob Vigdor8/18/2020

Developers of air-monitoring equipment now have a new core technology to deploy for detecting dangerous conditions. These conditions include methane gas from industrial operations, heavily legislated refrigerant gases, and flammable gas exposure of industrial workers or first responders. NevadaNano has continuously improved and delivered new versions of its Molecular Property Spectrometer (MPS™) MEMS-based chemical sensing platform. MPS is a platform for safety and environmental monitoring that enables next-generation sensor upgrades with the latest firmware on the embedded ARM® Cortex® M4F based MCU from STMicroelectronics. However, much of the design of new features for a product launch involves a tremendous amount of data collection, reduction, and analysis. The data also needs to be stored for future work and periodic quality assurance testing. We have to maintain data used for design and calibration and save data sets for testing. Our data collection and management system are a key to our success, and tools such as Edge Impulse take on the heavy lifting for us.

Card image cap

The MPS sensor currently ships in three versions, a flammable gas sensor, a methane gas sensor, and a version tuned to detect combustible refrigerant gases. In each case, the MPS sensor delivers an accurate account of the composition of the air it is monitoring. Whether it is the safety of first responders who might need an LEL (lower explosive limit) Gas Monitor, or watching over the environment with a methane detector, the MPS sensors deliver accurate chemical decomposition and analysis. When the concentration exceeds a predetermined threshold, the MPS sensor will trigger an alarm or warning, allowing workers, employees, or residents to be alerted to the heightened situation.

To test our sensors in different environments, we use automated data collection software, simplifying the steps and improving the process to build robust, predictive data-science models. We test and calibrate every sensor in-house in all the environments customers need to monitor, giving rise to tremendous dataset sizes, up to 10s of TBs for all the sensors that we ship. Machine learning and traditional digital signal processing allow us to deliver sensors that are best-in-class in terms of accuracy. Unlike catalytic bead sensors, the MPS is highly immune to the poisoning effects that cause severe accuracy degradation over time, and ultimately force expensive replacement cycles. Unlike nondispersive infrared (NDIR) sensors, our MPS can detect hydrogen. With near-zero accuracy drift over time, the MPS can stay in field deployments up to 10 years without requiring any calibration cycles.

What sets the NevadaNano MPS sensor apart is its ability to classify the gases that it is sensing. For example, the current version of the MPS flammable sensor can detect 12 different combustible gases and separate them into six classes. The ability to classify the detected gases gives product developers more information about the air's composition, which in turn opens up more feature-rich industrial gas detectors.

Card image cap

What can machine learning do for us? How can we get more insight from our data and then use the knowledge to move towards more intelligent products faster? We are dealing with things on an impossibly small scale, and all these small sensors inform us about the chemistry of the environment, and the quality and safety of the air. Platforms like Edge Impulse automate the collection, labeling, and storage of our data in a format that is readily available for classical DSP and more advanced machine learning algorithm development. Edge impulse frees us up to focus on the bigger picture while having our valuable and important data collected, stored, and used for product development, characterization, and testing. The interface allows us to understand the data trends and points out anomalies as we move from design to production.

Workers, managers, and government officials can be more informed about the environments where we live, work, and play. With our sensors' production process, we can deliver a core technology with features that provide valuable insight for safety products, such as methane gas sensors used in portable or fixed flammable gas detector systems. Everyone would agree that knowing more about the air we breathe is an excellent outcome.