Reliability of remaining useful life predictions using physics-informed models and digital twins in predictive maintenance
Physics-informed machine learning is widely presented as a means of making Remaining Useful Life (RUL) prediction for rotating machinery more accurate, data-efficient, and trustworthy, often under the label of a digital twin. This review tests these claims against data splits, physics calibration, baselines, and validation conditions. Journal articles issued from 2022 to 2026 were retrieved from the OpenAIRE Graph and Crossref with a Boolean string requiring physics-informed, prognostic, and asset or digital twin terms. The design combined a bibliometric map of 176 articles, an abstract-level evidence map of 49 core studies on bearings, aero-engines, gears, and motors, technical appraisal of five full texts, and a structured synthesis of matched contrasts. Output rose from 9 articles in 2022 and 2023 combined to 122 in the partial year 2026, spread across 122 sources. Only 21% of the 103 abstracts that used the digital twin label described model updating or synchronization, and only three core abstracts framed their model as a twin. Among core abstracts, 23% mentioned unseen or changing operating conditions, 6% calibration or coverage, and none code availability. Four matched contrasts from three studies favored physics, with log error ratios from -0.19 to -0.30 on held-out units and -2.41 in a field case whose reference RUL came from the physics-weighted model itself, but only one contrast rested on physics demonstrably calibrated without test data. No pooled estimate was defensible. Physics helps most when it constrains extrapolation, yet the literature rarely isolates that benefit, and twin claims outrun twin evidence.
Authors
- Hanif Farhan
Publication Details
- Journal
- Journal of Engineering and High Technology
- Published
- 2026-11-15
- DOI
- https://doi.org/10.55942/jeht.v2i2.2367
- Primary Topic
- Machine Fault Diagnosis Techniques
- Type
- article
- Field-Weighted Citation Impact
- 0.00