Local monitoring of embedded applications and devices using artificial neural networks


Reliability, security, and safety become even more challenging in times of the Internet of Things (IoT). Devices operate jointly in large distributed networks and may affect each other’s functionality due to failures or attacks. Identifying abnormal system behavior is therefore the solution to protect the device itself and other network participants to ensure service availability and system integrity. We propose a monitor concept based on long short-term memory recurrent neural networks which adapts to new devices by learning the nominal behavior automatically. No fault model is needed to identify erroneous behavior. The monitor can operate locally on the device, so our approach addresses the limited bandwidth and connectivity of IoT devices. Experiments evaluate our approach for a simulated controller under varying runtime conditions.

Görschwin Fey

Entwurfsautomatisierung, digitale Schaltungen und Systeme, Fehlertoleranz, Robustheit