The recent rise of MEMS and Microsensors have posed a new and interesting problem to designers. Before MEMS, you barely had one temperature sensor, maybe one pressure sensor in your system. Then you would put some code around … Et Voila ! your device was capable of measuring Temperature and Pressure.
Now with the advent of MEMS, nothing prevents you to have a lot of sensors, of different types, at different positions in your system. The issue has become “How do I combine all these together to produce something useful” ? This what Sensor Fusion is about.
![sensor fusion](http://www.em-company.fr/blog/wp-content/uploads/2017/11/sensor-fusion-300x139.jpg)
Your smartphone has typically several temperatures sensors, several light sensors, several accelerometers. Modern accelerometers often implement smart functions like “Significant Motion detection”, “No Motion detection”, etc .. . This is even an OS requirement (see here, for Android).
My own definition: A Sensor Fusion Controller is a piece of Software and/or Hardware that implements some smart algorithms out of a variety of inputs (the sensors), essentially to derive a richer information than the sum of all the inputs.
The following describes an implementation of an off-arm detection based on sensor fusion. The Wearable Device I was working on was an arm-worn type of device. Is it very important to understand, in those kind of devices, whether it is actually worn by the User, or simply left aside on a table or within a bag. All kind of Power Saving scenarios can then be implemented.
So in this specific case, it looked simple ! Put a Proximity Sensor and you got it. Those components are available, one example here. But he problem was that, the manufacturer wanted to shave off the 1 dollar of component …. so Software had to find a solution. Easy 🙁
Problem description
The basic question was:
“Are we able to infer that the device is worn or not, based on all those sensors ?”.
And of course, the corollary design constraint:
“We shall NOT infer it is worn when it is not …”.
We very rapidly agreed that, we’d better concentrate on a few “solid” cases, like: the device is laying on a table, facing up, since more that 5 minutes. For all the difficult cases (for example, Worn but not moving ), we can just not do anything.
In the next articles, I will detail the rationale in using a Fuzzy Logic approach for solving this problem, the implementation of the whole thing … and some results / possible improvements.
References
- Original paper from Pr. Lofti Zadeh, 1965
- Introduction to Fuzzy Logic , by Franck Dernoncourt (MIT)