SimIMA: a virtual Simulink intelligent modeling assistantSimulink intelligent modeling assistance through machine learning and model clones

Intelligent virtual model assistance is a key challenge in cultivating model-driven engineering proliferation and growth. Such assistance will help improve the quality of software models, support education for students learning modeling, and lower the entry barriers to new modelers. We present SimIMA, an intelligent modeling assistant for Simulink, which is an extremely popular modeling language in both industry and academia. SimIMA provides modelers with two different forms of data-driven guidance using a knowledge base of configurable repositories and sources. The first form of guidance, SimGESTION, suggests to modelers single-step operations they can perform on their models as they edit them in their modeling environment. These suggestions are based on the machine learning technique of ensemble learning through association rule mining and frequency classification. The second form of guidance, SimXAMPLE, presents modelers with similar/related Simulink systems for modelers to either insert directly into their environments or to view for inspiration. SimXAMPLE accomplishes this through model clone detection. To validate SimIMA, we conduct experiments using an established, open, and curated large set of Simulink models coming from a variety of application domains. Our results show that both of SimIMA’s forms of guidance are inferring the appropriate model and element suggestions given SimIMA’s knowledge base and that SimIMA is both scalable and efficient. Through our evaluation, SimIMA demonstrates a prediction accuracy of 78.86% for block-level suggestions and 82.04% for full system suggestions.