Masking the Problem

Face masks can help to protect people from certain types of air pollution and infectious diseases. But a mask is only as good as the fit allows it to be. An out of place or loose mask making poor contact with the skin allows infectious agents and air pollutants to pass freely around the barrier that was designed to stop them. It has been found that younger people, women, and men with beards in particular have difficulty finding a mask that offers a proper fit. Even for those that a mask is likely to fit well, studies have shown that most people have a difficult time assessing fit quality.

The guesswork surrounding mask fit may soon be a thing of the past. A group of engineers at MIT have developed a flexible array of sensors that can be fitted to virtually any face mask. The paper-thin strip of sensors line the inner edges of the mask. 17 capacitive sensors surround the entire device, and ​​temperature, humidity, and air pressure sensors, in addition to an accelerometer are also included. These sensors can be used to determine positioning of the mask on the face with the help of a machine learning classifier, and can also assess environmental variables. Additionally, they make it possible to detect activities like speaking and coughing for more advanced functions.

The sensors are embedded in a thin polyimide film designed to protect them and keep them in the right positions. The film is attached to the mask and face using an adhesive that was inspired by the sticky feet of geckos. The sensors are connected to a pair of microcontrollers on a small, external circuit board enclosed in a plastic case that attaches to a corner on the outside portion of the instrumented mask. The microcontrollers are used to acquire data from the sensors, run the machine learning algorithm that assesses fit quality, and for communicating wirelessly via an onboard Bluetooth Low Energy radio.

The seventeen capacitive sensors on the device’s periphery can detect if the mask is in contact with the skin, while the accelerometer can provide information about motion and orientation. A k-means clustering algorithm was trained to interpret these measurements and determine if the mask is correctly positioned on the face. An overall average classification accuracy of 92.8% was observed in a validation study, with women and bearded men being more difficult to assess mask positioning for. The researchers believe that if women wore smaller face masks, the classification accuracy would increase further.

This device represents a much more accessible way to assess mask fit — the current standard is to use a machine called a mask fit tester that is rarely available outside of hospitals and a few other specialized institutions. Mask fit testers assess differences in air particle concentrations inside and outside of a mask to determine how well it is sealed. But by using a clever combination of sensors and machine learning, the team at MIT has effectively demonstrated that much simpler, and less expensive, approaches also exist.

Aside from helping people to get a better fit from existing masks, the researchers hope that their work inspires manufacturers to produce masks that fit a wider variety of face sizes and shapes. They wonder why different sizes of shoes are being made, but not different sizes of masks. That is a good question, and perhaps we will see some new developments come from this research.