Probabilistic learning from real-world observations of systems with unknown inputs for model-form UQ and digital twinning

Abstract

In engineering systems, a digital twin serves as a digital replica encompassing both physical assets and their associated processes, such as manufacturing and certification. The implementation of digital twins offers substantial potential for various applications, including improved design, enhanced collaboration, effective energy management, risk mitigation, lifecycle management, and predictive maintenance. However, existing definitions of a “twin” are often ambiguous and lack a structured approach for developing digital twins, particularly for systems with unknown inputs. This paper addresses these shortcomings by proposing a clear definition and a robust methodology for building digital twins. Our methodology integrates projection-based model order reduction, a rapid approach for identifying unknown inputs, and a non-parametric probabilistic method for modeling and quantifying model-form uncertainty. Additionally, it incorporates a probabilistic learning approach for performing stochastic model updating. The effectiveness of this digital twinning methodology is illustrated through a case study involving an elevated truss footbridge located at the Autodesk Research facility at Pier 9 in San Francisco with unknown inputs. This case study underscores the importance of accurately modeling uncertainty to enhance the performance and reliability of digital twins in real-world engineering applications.