Related Experiment Video
Updated: Jun 13, 2026
09:17
Surrogate Model Development for Digital Experiments in Welding
Published on: March 28, 2025
Dynamic Ensemble Learning with Transfer Learning for Fatigue Performance Prediction in Ni-Based Superalloys.
Jiaxing Yang1, Fenglou Du1, Haopeng Lv1
1Hebei Short Process Steelmaking Technology Innovation Center, School of Materials Science and Engineering, Hebei University of Science and Technology, Shijiazhuang 050018, China.
Materials (Basel, Switzerland)
|June 12, 2026
Summary
This study introduces a novel machine learning framework for predicting Ni-based superalloy fatigue performance, overcoming data scarcity. The dynamic ensemble and transfer learning approach significantly improves prediction accuracy for fatigue stress and life.
Area of Science:
- Materials Science
- Mechanical Engineering
- Data Science
Background:
- Predicting fatigue performance in Ni-based superalloys is challenging due to limited data and poor generalization of standard machine learning models.
- Existing methods struggle with the scarcity of high-quality fatigue data, hindering reliable material design and performance assessment.
Purpose of the Study:
- To develop an advanced machine learning framework for accurate fatigue performance prediction in Ni-based superalloys, addressing data limitations.
- To enhance the generalization capability of predictive models by integrating dynamic ensemble and transfer learning techniques.
Main Methods:
- A dynamic weighted error feedback ensemble algorithm (DWELA) was developed to optimize base regressor weights in real-time, improving tensile property prediction (R² from 0.90 to 0.95).
- A feature alignment transfer learning (FATL) strategy was employed to transfer knowledge from a tensile dataset to a fatigue dataset, aligning shared features and fine-tuning domain-specific ones.
- The combined framework, ETFPM, was trained on 1025 tensile and 622 fatigue samples.
Main Results:
- The DWELA improved tensile property prediction R² to 0.95, outperforming the best single model.
- The ETFPM model achieved R² of 0.93 for fatigue stress and 0.81 for fatigue life on independent samples, surpassing the best fatigue-trained single model (SVR R² of 0.89 and 0.72).
- Twenty candidate alloys were successfully screened using the predictive model.
Conclusions:
- The proposed framework offers a practical and effective solution for fatigue performance prediction in data-limited scenarios for Ni-based superalloys.
- The novel DWELA and FATL strategies demonstrate significant improvements in prediction accuracy and model generalization.
- Further experimental validation is recommended to confirm the broader applicability of the developed methodology.
Related Concept Videos
01:21Fatigue
Fatigue occurs when materials rupture under repeated or fluctuating loads, even at stress levels far below their static breaking strength. It typically results in brittle failure, even for ductile materials. It is a critical consideration in designing machines and structural components subjected to repetitive or varying loads. The nature of these loadings can range from fluctuating loads like unbalanced pump impellers causing vibrations to repeatedly bending a thin steel rod wire back and forth...
01:22Fatigue Strength of Concrete
Fatigue, in the context of materials science and engineering, refers to the weakening or failure of a material caused by repeatedly applied loads, even if these loads are below the strength limit of the material. Fatigue strength in concrete is a critical property that influences its durability and longevity. Concrete can fail in two ways due to fatigue. Static fatigue or creep rupture occurs under a constant load or one that increases slowly. The other failure mode is due to cyclical or...
01:10Work and Energy for Variable Forces
When an object is acted upon by a variable force, the amount of work done and the change in energy of the object can be more complex to calculate compared to when a constant force is applied. Work is the product of force and displacement, while energy is the capacity of a system to do work. When a constant force is applied to an object, the work done can be calculated as the product of the force and the distance moved in the direction of the force. However, when a variable force is applied, the...