Adaptive Multidimensional Measurements with Validated Uncertainty: A Machine-Learning-Enabled Framework

This paper describes the objectives, methodology, and expected impact of a project launched last year to advance smart, multidimensional measurement. It presents the project’s methodological framework and a pre-registered validation protocol; experimental results are planned and will be reported in follow-up publications. The protocol specifies datasets, metrics, and acceptance criteria to ensure reproducible evaluation. As measurement tasks grow in complexity, fixed measurement plans become inefficient, motivating automated strategies that decide where and how to measure next while adapting to local conditions and uncertainty. The project targets three building blocks, automated, adaptive, and uncertainty-aware, by developing methods for efficient reconstruction and completion of high-dimensional data from sparse or irregular sampling, and for robust, validated uncertainty evaluation suitable for closed-loop workflows. Generative machine learning is combined with compressed sensing to enable data-efficient acquisition and accurate reconstruction from limited measurements. In parallel, Bayesian inference is integrated with the GUM framework to deliver uncertainty evaluation that is traceable, interpretable, and compatible with automation. Case studies spanning different measurement modalities are used to illustrate validity, generalizability, and practical deployment considerations. By reducing measurement effort while improving data quality through AI-driven automation and digitalization, the project aims to accelerate testing and production, strengthen industrial metrology and quality assurance, and contribute to sustainability via lower resource use and waste.