Non-Intrusive system behavior tracing using power consumption

Abstract

This thesis presents a novel non-intrusive method of creating a map of black-box systems’ behavior to commands by observing the power consumption of the system. We expand our method in the form of a flexible framework which supports communication abstraction layer to send commands on different channels. The framework also integrates data processing and the opportunity of adding a validation module that framework users can utilize to detect anomalies in the system. The framework furthermore builds on the idea of anomaly detection by providing two operating modes: one for generating training data for machine learning models, and one for run-time validation of the system’s behavior. From the experiments, we confirm the effectiveness of the framework for creating a map of the system’s power consumption against a series of queries. Our experiments also show some of the challenges presented with this method when different commands induce similar behaviour in the system.