05-02-2021 | | By Robin Mitchell
Recently, Tufts University engineers developed flexible thread-based sensors that can be used to read data from the neck as direction, angle, and displacement. How does the engineers' sensor work, and how is AI becoming more important in sensor systems?
Recently, Tufts University engineers developed flexible sensors that can be either directly attached to the skin or woven into the material to determine displacement, rotation, and direction. The sensors' key factor is not the thread material, but the coating which provides electrical conductivity. The coating used is a carbon-based ink, and as the threads are deformed, stretched, and strained, they conduct electricity differently.
To make the sensors detect a wide range of different readings, two threads were placed in an X shape that can either be connected to the skin directly using a patch or woven into clothing. However, simply measuring the sensors' electrical conductivity is not enough to determine the current activity (rotation, displacement, and direction).
Instead, an AI reads the results from the sensors, and past data and training allow the AI to make many determinations about what the sensor is detecting. The AI's accuracy rate is currently around 93%, and this includes more than simple activity monitoring (i.e. rotation angle and amount of displacement is inferred by the AI).
Normally when an announcement is released to the public, the immediate content is what is of interest. However, in the case of the sensors produced by the engineers from Tufts University, the use of AI is what makes this announcement of particular interest.
Historically, sensors have been designed to react to some environmental stimuli, and produce either a simple binary value or proportional reading to that stimuli. For example, a temperature sensors resistance may linearly (approximately), increase as the temperature increases, and thus a binary number can be generated from this data to display the temperature.
However, the introduction of AI is now creating a new generation of sensor technology. Creating a cross-shaped conductor placed on the back of the neck can demonstrate resistance changes, but to try and determine what is going on by simply looking at the sensor's raw resistance values can be nearly impossible.
Instead of determining relationships between sensor readings and activities through basic equations, AI allows for a sensor to be connected to an AI system and then trained to compare the sensors' readings to what the environment is currently doing. As such, sensors that appear to produce incoherent data can be used to provide reliable data on their surroundings.
The use of AI in sensors is currently in its infancy, but there have been other AI examples in sensors. For example, BOSCH recently released its BHI260AP, which is a self-learning 3-axis gyroscopic sensor. As it operates, it can learn new repetitive behaviours that allow the sensor to recognise new tasks. Instead of being hardcoded into either software or hardware, the self-learning system can look at data that appears incoherent and make sense from it.
AI-driven sensors will provide engineers with a wide range of capabilities not currently possible with sensor technology. To start, the use of AI in sensors allows for adaptable sensors that can be retrained to look for different activities or measurements.
For example, a strain sensor such as the one developed by Tufts University engineers could be used to identify the movement in different joints (arm, leg, foot etc.). Gas sensors that typically look for absolute levels could be used to provide early warning systems of a potentially failing sensor (which also carries potential into safety systems such as fire alarms).
AI-driven sensors also provide the possibility to create complex sensor systems from basic sensor technology. In the case of the engineers' sensors from Tufts University, simple conductive threads could provide data on rotation and movement. Thus, other basic sensors (such as photodetectors and thermistors) could also provide complex data from simplistic sensors in different configurations. It may even be possible to detect sound from vibrating conductive threads that could in turn be used to determine noise levels.
The use of AI in sensors provides many opportunities, and the ability for AI to adapt and learn creates a strong foundation for the future sensors.