An improvement in data interpretation to estimate residual stresses and mechanical properties using instrumented indentation: A comparison between machine learning and Kriging model. (September 2022)

• Record Type: Journal Article
• Authors: Salmani Ghanbari, S.; Mahmoudi, A.H.
• Is Part Of: Engineering applications of artificial intelligence. Volume 114(2022)
• Publication Date: 2022-09
• Subjects: FNN Fuzzy Neural Networks -- kNN k-Nearest Neighbor -- MAE Mean Absolute Error -- RMSE Root Mean Square Error; Indentation -- Residual stresses -- Mechanical properties -- Machine learning -- Kriging model -- kNN algorithm; Engineering -- Data processing -- Periodicals; Artificial intelligence -- Periodicals; Expert systems (Computer science) -- Periodicals; Ingénierie -- Informatique -- Périodiques; Intelligence artificielle -- Périodiques; Systèmes experts (Informatique) -- Périodiques; Artificial intelligence; Engineering -- Data processing; Expert systems (Computer science); Periodicals; 620.00285
• DOI: 10.1016/j.engappai.2022.105186
• Languages: English
• ISSNs: 0952-1976

Abstract

Abstract: The instrumented indentation method has been introduced as an effective means of estimating surface residual stresses. This technique has also been widely utilized to measure mechanical properties of the materials. Most studies in recent years have been conducted for the nano and micro-indentation. However, macro-indentation technique can result in interesting results in terms of residual stresses as well as mechanical properties. The instrumented indentation is a rapid and non-destructive method with acceptable precision which is desirable for a variety of different applications. This technique is applicable not only for large samples or those under service but also for small material volumes. In the author's previous work, a portable indentation apparatus was examined for extracting the p–h curve. The fuzzy neural networks were then employed to determine the residual stresses as well as the plastic mechanical properties of materials. The current research uses different data interpretation methods to extract equi-biaxial residual stresses and mechanical properties. A comparative study between two methods, supervised machine learning and Kriging model, was conducted to estimate the unknown parameters. The performance accuracies of machine learning and Kriging were then compared. The results indicated that supervised machine learning using kNN algorithms can perform slightly better than the universal Kriging model. The findings were later compared with the experimental results for an aluminum plate.