No open-access location is known for this record yet.
No open-access location is known for this record yet.
Abstract When using Monte Carlo simulation involving repeated finite element analysis (FEA) to perform fatigue reliability evaluation for an impeller, a variety of uncertainties should be considered to ensure the comprehensiveness of fatigue predictions. These uncertainties include the aleatory uncertainty from the geometric, material and load condition, and epistemic uncertainty from the parameters of the physics‐of‐failure (PoF) model to yield fatigue prediction. However, the latter uncertainty is often ignored in fatigue reliability analysis. And the reliability assessment will become computationally unaffordable and inefficient when there are many random variables involved, as an enormous amount of FEAs are demanded. To address this problem, a Whale Optimization Algorithm‐extreme gradient boosting (WOA‐XGBoost) surrogate model is developed, based on relatively few FEA results obtained using a Latin hypercube sampling (LHS). Its strengths lie in the interpretability of the design variables and effective determination of fine‐tuned hyperparameters. A case study on an impeller is conducted considering uncertainties from 11 input variables, where an efficient XGBoost model with an R2 greater than 0.93 on test set is established using 400 samples from practical FEAs. In addition, the importance analysis indicates that elasticity modulus and density play the greatest impact on the maximum strain, showing a combined importance of 82.3%. Furthermore, the reliability assessment results under fatigue parameter derived from the Median method tend to be more conservative compared to those obtained from the Seeger method.
Loading…
Stored only when both papers are in the corpus — 8 citations exist upstream.
Papers that cite the same references — the relational fallback while embeddings are still backfilling.
Imported from OpenAlex (CC0) — nothing here is generated.