Electronic nose responds at 40Hz

University of Hertfordshire researchers have created an electronic nose that can identify smells in tens of milliseconds using standard metal oxide sensors in an novel way.

The micro-machined sensors are ScioSense’s CCS801 volatile organic compound detector, and SGX Sensortech’s triple detector MiCs-6814 for carbon monoxide, nitrogen oxides and ammonia.

These are duplicated for a total of eight metal oxide (‘MOX’) sensors.

The electronic nose. For scale, the perforated metal packaged MiCs-6814 triple detectors measure 5 x 7mm

MOX sensors change resistance based on on their surface coating, gas concentration and temperature – the latter of which is controlled by individual on-board heaters and temperature sensors.

In many cases, they are ambiguous, with different chemicals producing the same resistance change in the same sensor – the collection on the Heartfordshire board also respond to molecules including ethanol, hydrogen, methane, propane and iso-butane , for example.

The novelty in the Hertfordshire case comes from changing the set temperature of the sensors very fast – from 150°C to 400°C and back to 150°C in 50ms, for example – the result of the tiny size of the detectors – under 0.5mm across – and custom microcontroller-in-the-loop electronics developed for the project.

A Raspberry Pi Pico, just visible under the board, implemented eight PID control loops for the sensor heaters

This high rate of temperature change allowed some compounds that would otherwise cause the same static resistance change in a MOX sensor to be differentiated – providing they have different diffusion rates in and out of the sensitive layer, or reaction speeds inside the layer.

“A key part of the challenge has been temperature control,” said Hertfordshire researcher Nik Dennler (right). “The sensor reaction depends strongly on heat, which is difficult to control in a turbulent environment. We have assembled electronics and devised algorithms that rapidly control the operating temperature. This allows the sensors to work with incredible precision.”

Dennler is a dual graduate, of Hertfordshire and Western Sydney University – which was also involved in the project, as was University College London and the Francis Crick Institute.

Sensor outputs were fed into a set of artificial intelligence algorithms that the researchers trained on one second ‘pulses’ of smells – from synthetic versions of compounds that give characteristic smells to pineapple (ethyl butyrate), banana (isoamyl acetate), eucalyptus (cineol) and cheese (2-heptanone).

Once trained, the e-nose was tested against rapidly changing and gradually changing combinations of the four smells.

“When compared to the performance of mice in distinguishing rapidly switching odours, the electronic nose outperformed them,” according to Hertfordshire, which added: “It has already been proven that mice can reliably distinguish between different temporal structures of odour pairs up to frequencies of 40Hz. These timescales have not hitherto been matched in robotic systems.”

The aim is to develop sensors suitable for use on drones and un-manned ground vehicles, both indoors and outside.

“Many odour-based monitoring scenarios could benefit from a mobile platform, such as a drone or a robot,” said Dennler. “Think for example about localising gas leaks, monitoring wildfires, or finding explosive materials, all based on the odour.”

Beyond this, a switch to neuromorphic computing – essentially analogue neural networks – is being considered to reduce power consumption and otentially to reduce latency.

For more information, including much more on the algorithms used, is available in the clearly-written Science Advances paper ‘High-speed odor sensing using miniaturized electronic nose‘, which can be read in full without payment.

As an aside: The project used smell modulating equipment developed for earlier research that is capable of swapping smells at above 60Hz.