About this Abstract |
| Meeting |
MS&T25: Materials Science & Technology
|
| Symposium
|
Materials Processing and Fundamental Understanding Based on Machine Learning and Data Informatics
|
| Presentation Title |
Preprocessing of Inconsistent Creep Data Collected from a Literature Survey to Provide Reliable and Consistent Creep Life Prediction |
| Author(s) |
Taejoo Lee, Chang-Seok Oh, Yoon Suk Choi |
| On-Site Speaker (Planned) |
Taejoo Lee |
| Abstract Scope |
Since the machine learning-based creep life prediction is heavily influenced by the quality and integrity of the collected creep data, data preprocessing is required to eliminate outliers. In this study, a machine learning-based data screening methodology was developed to detect and eliminate outliers. The methodology consisted of selecting appropriate machine learning models for the collected creep data through an assessment of their validity in creep physics, evaluating the prediction accuracy and variability of collected datapoints through bagging, and identifying inconsistent datapoints by ranking their residuals and prediction variabilities. The proposed methodology was successfully applied to the multi-source collected creep data of a Ni-base single crystal superalloy CMSX-4 and led to improved accuracy and consistency in creep life prediction. In addition, the proposed methodology was validated by predicting the creep life of a newly generated dataset that was independent from the training data. |