Planning Under Partial Observability: A Study in High-Precision Manufacturing

S‐N curve parameter determination is a time‐ and cost‐intensive procedure. A standardized method for simultaneously determining all S‐N curve parameters with minimum testing effort is still missing. The Bayesian optimal experimental design (BOED) approach can reduce testing effort and accelerates uncertainty reduction during fatigue testing for S‐N curve parameters. The concept is applicable to all S‐N curve models and is exemplary illustrated for a bilinear S‐N curve model. We demonstrate the fatigue testing workflow for the bilinear S‐N curve in detail while discussing steps and challenges when generalizing to other S‐N curve models. Applying the BOED to the bilinear S‐N curve models, minor errors and uncertainties for all S‐N curve parameters are obtained after only 10 experiments for data scatter values below 1.1. For such, the relative error in fatigue limit estimation was less than 1% after five tests. When S‐N data scatter higher than 1.2 is concerned, 17 tests were required for robust analysis. The BOED methodology should be applied to other S‐N curve models in the future. The high computational effort and the approximation of the posterior distribution with a normal distribution are the limitations of the presented BOED approach.

Section: Related Workmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Reduction of testing effort for fatigue tests: Application of Bayesian optimal experimental design

Frie,

Kolyshkin,

Mordeja

et al. 2023

Analysis of data‐driven models for predicting fatigue strength of steel components with uncertainty quantification

Frie,

Kolyshkin,

Eberl

2023

Material informatics has emerged as a valuable research field in material science, providing solutions to previously unsolvable problems or accelerating deliverables. Fatigue failure, as a complex and non‐deterministic phenomenon, requires a probabilistic approach to assess the uncertainty of the fatigue strength prediction. This study compares various probabilistic data‐driven models for credible fatigue strength predictions for three distinct steel groups. The analysis considers data and model uncertainty, evaluating their impacts on predictive quality from engineering and data science perspectives. Results reveal that deep ensembles outperform other probabilistic models regarding negative log‐likelihood (NLL), while random forest exhibits the lowest root mean square error (RMSE). Notably, the prediction accuracy of case‐hardened steels is negatively affected by insufficient material properties definitions, while stainless steels demonstrate the best performance compared to other steel types.

Exploration of materials fatigue influence factors using interpretable machine learning

Frie,

Riza Durmaz,

Eberl

2024

Data‐driven fatigue strength predictions are gaining popularity. Nevertheless, many machine learning models lack trustworthiness due to their limited decision‐making transparency which often hinders their practical application. In this investigation, we assess the expressiveness of the model‐agnostic explainable AI method known as SHapley Additive exPlanations (SHAP) for data‐driven fatigue strength prediction. Our study demonstrates that the SHAP feature sensitivity analysis underpins known physical relations from materials processing and fatigue theory. This even applies in view of the high‐dimensional, cross‐correlated fatigue feature space and despite data heterogeneity (different steels, component designs, and loads). For instance, SHAP indicates a fatigue strength increase with higher solid solution‐strengthening element concentrations, such as chromium and nickel. SHAP identifies correlations rather than causality. Thus, data science and domain knowledge should be closely linked during the SHAP assessment. If this is satisfied, plausible causal relations can be inferred, and spurious ones arising from confounding variables can be discarded.