Automated Generation of Consistent and Diverse Graph Execution Trace Models

While graphs provide key abstractions when engineering critical software-intensive cyber-physical systems, the availability of real graph models is often limited or insufficient.

Hence synthetic graph generators are increasingly important for the training, benchmarking or quality assurance of various graph-based applications.

However, existing graph generators can only derive a set of individual graphs representing a snapshot of the system, but cannot derive a sequence (or path) of consistent graph snapshots, which are called graph (execution) traces or filmstrip models.

The TRACIFY project will provide novel scalable techniques and open-source software tools to automatically generate a diverse set of consistent domain-specific filmstrip models.

Filmstrip model generation will innovatively fuse refinement-based partial graph generation with rule-based design space exploration and machine-learning based heuristics, while developing efficient novel techniques for partial model differencing, and incremental evaluation of temporal constraints.

As a key practical application, automatically generated filmstrip models are planned to be used as test cases during system-level assurance of autonomous vehicles.