embedded world 2021: a digital nose gas sensor with AI - Embedded.com

embedded world 2021: a digital nose gas sensor with AI

Bosch Sensortec has launched a new version of its gas sensor that adds artificial intelligence capability on-sensor plus a gas scanning mode for volatile sulfur compounds (VSCs). The new BME688 enables customers to train the sensor for their own custom gas sensing applications, for applications like bad breath and fresh food detection.

Bosch Sensortec claims the new BME688 is the first air quality MEMS sensor that combines gas, humidity, temperature and barometric pressure sensing with artificial intelligence (AI) capability – essentially the “world’s smallest” four-in-one air quality measurement solution.

The BME688 hardware is fully backward compatible with the existing BME680 sensor, with both built upon the same platform. But the new sensor features a higher gas resistance range enabled in an application specific integrated circuit (ASIC). It also now features a specific gas scanning mode for VSCs, and its new BME AI-Studio software, which allows data on specific user-environments to be imported, labelled, and then used to create a custom-tailored algorithm for that use case.

The customer categorizes their own data and then applies it in the development of their AI model in the BME AI-Studio, essentially training the BME688 to recognize the tell-tale signs of bacterial growth on the food, as an example. Once the sensor is trained, the final AI code runs on a system microcontroller in the customer’s end product. This AI code is generally lightweight and will run quite easily on the existing microcontroller that handles system control and management tasks. 

Customized applications can include things like checking for bad breath, checking the state of a diaper, or to detect the freshness of food in a fridge. Whether it is food spoilage detection or timely forest fire detection, the sensor enables detection of the gases present, and tracking temperature and humidity changes.

Bosch BME AI-Studio
A customer can input and label data within BME AI-Studio, which then creates code to train the BME688 gas sensor to recognize things like tell-tale signs of bacterial growth on the food, as an example. (Image: Bosch Sensortec)

In conjunction with the company’s new BME AI-Studio software tool, this should make it straightforward for customers to rapidly develop custom solutions for their specific use cases. Bosch Sensortec also provides an Adafruit-compatible development kit to enhance development.

The gas sensor now detects the presence of many gases, including volatile organic compounds (VOCs), volatile sulfur compounds (VSCs) and other gas types such as carbon monoxide and hydrogen, in the parts per billion (ppb) range.

VSCs can indicate spoiled food or bad breath detection example

If a customer wants to develop a sensor-based product that can detect spoiled food, this would be indicated by the VSCs produced by bacteria in the food. Similarly, bad breath or body odor could be detected based on the VSCs they produce.

The optimal approach is to collect indicative real-life data directly in the field. For instance, by sampling gases in the vicinity of both fresh and decaying food, to thereby create different combination models for the VSCs present in the given air sample. The BME688 is capable, by default, of very accurate detection of VSCs, and the BME AI-Studio enables it to be optimized for other gas mixtures and applications.

By sampling gases in the field instead of the lab, the derived algorithms that are utilized by the new detection devices are far more reliable in evaluating actual conditions. Together with gas data, the BME688 simultaneously measures humidity, barometric pressure and temperature as supplementary data inputs for complete AI modeling.

The BME688 has been developed for mobile and connected applications where size and low power consumption play a critical role. The sensor is housed in a compact package, measuring just 3.0 x 3.0 x 0.9 mm3. Current consumption can be configured between 2.1 µA and 11 mA depending on the required data rates and functions and can be optimized using the BME AI-Studio.

Related Contents:

For more Embedded, subscribe to Embedded’s weekly email newsletter.

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.