A Building Automation and Control micro-service architecture using Physics Inspired Neural Networks

Abstract

In this work, we present a micro-service architecture which defines a Digital Twin (DT) framework for adaptive building automation and control. The DT framework primarily involves the orchestration of several containerized micro-services, promoting the scalability and deployability of the proposed framework within the industrial context. In the proposed framework, containerized microservices facilitate: (i) model-based control strategies; (ii) data-driven learning; (iii) data management; (iv) the inclusion of an internal High-Fidelity Simulator (HFS) to enable bootstrapped learning; and (v) a User Interface/User Experience (UI/UE) micro-service orchestrator. To validate the usefulness of the proposed framework, we implement a Physics Inspired Neural Network (PINN) to adapt the model-based control strategies for plant-model uncertainty and utilize bootstrap sampling against an internal HFS.