Abstract: Digital Twins integrate computational models with monitoring data to represent structural behaviour and support the assessment of civil infrastructure. Achieving sustained synchronisation between physical structures and their digital counterparts requires these systems to adapt to changing structural conditions and evolving data availability. A Self-Generating Digital Twin (SGDT) is conceived as a full realisation of a true Digital Twin: one that continuously updates, learns and evolves through autonomous data assimilation and model regeneration. Within this framework, incomplete monitoring data present a critical challenge, potentially compromising the information required for reliable structural-response estimation and model updating. This seminar presents research undertaken through the I-X Global Early-Career AI Fellowship at Imperial College London, investigating AI-enabled reconstruction of missing monitoring data to address this challenge. Using experimental earthquake-response measurements from a three-storey cross-laminated timber rocking-wall structure, the study evaluates reconstruction under random data loss and contiguous sensor outages. Conventional approaches, including linear interpolation and Gaussian process regression, are compared with generative methods, including TimeGAN and conditional diffusion, across different missing-data patterns and severities.  Performance is assessed through both the fidelity of reconstructed response histories and their influence on subsequent Digital Twin predictions. The study evaluates whether accurate reconstruction of missing measurements preserves the reliability of structural response estimates by integrating the recovered data into a fixed data-assimilative Digital Twin. By connecting reconstruction quality with downstream engineering performance, the study assesses the contribution of data imputation to resilient operation within the SGDT framework. The seminar will discuss methodological capabiliti

Description supplied by the event organiser.