Microcontroller Based Edge Computing for Pipe Leakage Detection

In the embedded system field a correct resource management is crucial, especially in systems that use Machine Learning (ML) algorithms. The resources in that case are in terms of memory, footprint and time used to compute the tasks. The system should be able to be both accurate and compact although the precision is directly proportional to the memory used to storage data. In this paper we describe a comparison between three ML models implemented in a microcontroller, with an application scenario devoted to monitor a Water Distribution Network by using vibrations input and trying to investigate the computational complexity of each tested solution.