Deep learning regression-based stratified probabilistic combined cycle fatigue damage evaluation for turbine bladed disks

• Published in: International journal of fatigue 2022-06, Vol.159, p.106812, Article 106812
• Main Authors: Li, Xue-Qin, Song, Lu-Kai, Bai, Guang-Chen
• Format: Article
• Language: English
• Subjects: Combined cycle; Combined cycle fatigue; Correlation; Cumulative damage; Damage assessment; Damage evaluation; Deep learning; Disks; Fatigue failure; Fatigue life; High cycle fatigue; Low cycle fatigue; Materials fatigue; Nickel base alloys; Nonlinearity; Probabilistic analysis; Regression models; S N diagrams; Statistical analysis; Superalloys; Turbines
• ISSN: 0142-1123; 1879-3452
• DOI: 10.1016/j.ijfatigue.2022.106812

•A deep learning regression-stratified strategy (DLR-SS) is proposed for the first time.•Deep learning regression model is built by synchronous mapping and error control.•The DLR-SS can address high nonlinearity and correlated relationship issues.•A stratified framework for probabilistic CCF damage evaluation is first constructed.•The DLR-SS is validated to hold high accuracy and efficiency in probabilistic CCF damage evaluation. Probabilistic combined cycle fatigue (CCF) damage evaluation involves complex large-scale simulations of low cycle fatigue (LCF) damage, high cycle fatigue (HCF) damage and cumulative damage. Due to the high nonlinearity of performance function and correlated relationship of LCF/HCF damages, low simulation efficiency will be incurred if the traditional direct evaluation methods are employed, and low computing accuracy will also have appeared if the separate evaluation methods are applied. In response to this problem, a deep learning regression-stratified strategy (DLR-SS) is proposed, which transforms the complex evaluation problem into the stratified sub-evaluation problems: constitutive response sub-evaluation (stress/strain) and life/damage sub-evaluation; in constitutive response sub-evaluation, the synchronous mapping-based deep learning regression (DLR) model is developed to deal with the correlated relationships between constitutive responses; in damage evaluation sub-evaluation, the fatigue life models (Coffin-Manson model, S-N curve, miner cumulative model) are adopted to assess the LCF/HCF/CCF damages. With the dual-level collaborative analysis of DLR-SS, the nonlinearity degree in each level is reduced and the correlated relationships between LCF/HCF are well-considered. By selecting a typical turbine bladed disk with nickel-base alloy GH4133 material as an engineering case, the feasibility and effectiveness of the proposed method are verified. The current efforts of this study will shed a light on high-fidelity probabilistic CCF evaluation.