Interpretable and Reusable Reduced Order Models for Digital Twins in Manufactory as a Service

                                
==Abstract==

The increasing interest on Digital Twins (DTs) solutions in Industry 4.0 (I4.0) is transforming industrial processes towards a more profitable and sustainable production. In an industrial environment DTs enable the creation of virtual replica of industrial products, services, and processes, allowing a more effective management. The DIGITbrain project [1] aims for the development of an integrated digital platform to provide Small and Medium-sized Enterprises (SMEs) access to DT technology. Within this context several use cases have been created using different types of models. We have developed models using CAELIA, an authoring tool developed at ITAINNOVA for the generation and the management of Reduced Order Modelling (ROM) -based models [2]. CAELIAROM models are obtained through the Twinkle library (which can work both on dense and sparse data, and it is especially designed for unstructured data), and are based on self-adapted Tensor Rank Decomposition (TRD). Moreover, we have developed a real-time co-simulation structure linking Gazebo robot environment and controllers in Matlab using delay compensators [3]. The use case is a quarter car, as a sample for the foreseen vehicles to be integrated into a set of automated robots’ fleet, by means of co-simulation. Such a system of robotic vehicles uses RabbitMQ for node-master communications, enabling remote control and autonomous movements. All use cases developed at ITAINNOVA were conceived within the DIGITbrain environment, where all applications must be entirely reusable. A recombination of use case parts were proven to be reused in different scenarios, such as ROM, co-simulation, and Machine Learning (MLS)