Batch active learning for reliability-based model validation

Lämmle S, Stecher J, Hodrius T, Haselhoff A, Roos D (2026)

Publication Type: Journal article

Publication year: 2026

Journal

Reliability Engineering & System Safety Elsevier

Book Volume: 272

Article Number: 112631

DOI: 10.1016/j.ress.2026.112631

Abstract

In engineering, quantifying the deviation between model prediction and reality is crucial, particularly when simulation or empirical models are used in safety-critical applications or for product release. Model validation aims to quantify this discrepancy by comparing model predictions to experimental results. However, conducting physical experiments for validation is often costly and time-consuming. Active learning strategies can mitigate this problem by reducing the data needed to assess model accuracy. In practice, these strategies must account for experimental uncertainties, such as material variability. This work proposes an active learning framework for model validation focused on efficiently identifying the limit state between valid and invalid model regions under such conditions. Therefore, we introduce a decision-theoretical basis from which we derive two novel misclassification-based acquisition functions for mean-based and quantile-based validation. Furthermore, the framework is extended to the batch setting, allowing multiple points to be proposed for evaluation while considering nonlinear constraints of the input space. The proposed method is compared to alternative approaches for several analytical and engineering problems. Finally, the validation of a temperature model for an electric motor demonstrates the efficiency of the method in a complex scenario.

Involved external institutions

ZF Friedrichshafen AG DE Germany (DE) Hochschule Ruhr West – University of Applied Sciences DE Germany (DE) Hochschule Niederrhein University of Applied Sciences DE Germany (DE)

How to cite

APA:

Lämmle, S., Stecher, J., Hodrius, T., Haselhoff, A., & Roos, D. (2026). Batch active learning for reliability-based model validation. Reliability Engineering & System Safety, 272. https://doi.org/10.1016/j.ress.2026.112631

MLA:

Lämmle, Sven, et al. "Batch active learning for reliability-based model validation." Reliability Engineering & System Safety 272 (2026).

BibTeX: Download