Model-Driven Engineering for Machine Learning Code Generation using SysML

Abstract

The complexity of engineering products increases due to more functions, components, and the number of involved disciplines. In this respect, Data-Driven Engineering (DDE) aims to integrate machine learning to support product development and help manage the increasing complexity of engineered systems. Still, the potential and opportunities of DDE are not entirely reflected in practice, which among others originate from the rarely available machine learning experts on the market and the effort for the implementation in practice. In this respect, this work depicts an approach based on model-driven engineering, allowing to automatically derive executable machine learning code based on machine learning task formalization using the general-purpose modeling language SysML. The main focus of the approach is on the generality of the model transformation using templates so that extensions and changes to the code generation can be integrated without requiring profound modifications to the code generator. The approach is evaluated in a use case in the domain of Cyber-Physical Systems, i.e., weather forecast prediction based on data from a Cyber-Physical weather system. The derived executable code promises to reduce the time for the implementation and supports the standardization of machine learning implementations within a company due to templates.