Design and prototyping of self-learning methods for semi-automated model creation as a reference behavior in distributed embedded system

In recent years, a paradigm shift from traditional centralized systems to a more distributed architecture has been observed. This adoption of Distributed Embedded Systems (DES) has been triggered by the ever-declining costs and increasing functionality of the embedded networking devices. DES systems are becoming more sophisticated and complex. Growing complexity of these systems stems from the complex interactions between the networked devices, present at the nodes. The absence of a central controlling entity, allows the addition of new nodes, causing minimum to null disturbance to the original system. Validation of such scalable systems becomes a tedious task, when using DES components from different manufacturers and inclusion of third-party services. Validation, usually done by building models at varying levels of abstraction, requires a more flexible approach for accommodating this heterogeneity. The idea of reverse engineering models by observing the system operation, in the form of traces or event log, has recently gained momentum. However, the vast amount of tracing data from a large and complex DES system, can sometimes be overwhelming for the system engineer, manually constructing a behavioral model of the system. This thesis proposes a machine learning approach in deriving meaningful insights from the incremental tracing data and iteratively constructing a behavioral state machine model of the system. This executable UML stereotyped state machine can be used for validation and anomaly detection, by on-line trace verification of a DES industrial plant assembly setup