Pop-In Identification in Nanoindentation Curves with Deep Learning Algorithms

Abstract

High–speed nanoindentation rapidly generates large datasets, opening the door for advanced data analysis methods such as the resources available in artificial intelligence. The present study addresses the problem of differentiating load–displacement curves presenting pop-in, slope changes, or instabilities from curves exhibiting a typical loading path in large nanoindentation datasets. Classification of the curves was achieved with a deep learning model, specifically, a convolutional neural network (CNN) model implemented in Python using TensorFlow and Keras libraries. Load–displacement curves (with pop-in and without pop-in) from various materials were input to train and validate the model. The curves were converted into square matrices (50 × 50) and then used as inputs for the CNN model. The model successfully differentiated between pop-in and non-pop-in curves with approximately 93% accuracy in the training and validation datasets, indicating that the risk of overfitting the model was negligible. These results confirmed that artificial intelligence and computer vision models represent a powerful tool for analyzing nanoindentation data.

Multiscale Mechanical Characterization of Polyether-2-ketone (PEKK) for Biomedical Application

“…The experimental curves obtained from nanoindentation tests are depicted in Figure 11 . Given the absence of pop-in features during the loading phase of the indentation curves, it can be assumed that no crack occurred during the tests [ 61 ]. As a matter of fact, the SEM analysis reported below confirms the absence of cracks at the apex of the indentation imprints, which is a typical feature of brittle materials [ 62 ].…”

The Instrumented Indentation Test: An Aiding Tool for Material Science and Industry

“…A number of scientists, users, developers, engineers, and technical experts from the academic, industrial, and research areas have contributed to this first volume of the Special Issue on The Instrumented Indentation Test: An Aiding Tool for Materials Science and Industry. The present state of the art on the subject indicates that, in spite of promising experiences which encourage the transfer of IIT from the laboratory to industry [ 9 , 10 , 11 , 12 , 13 , 14 , 15 , 16 , 17 , 18 , 19 ], other factors currently act as an obstacle to its use (see, for instance [ 14 , 16 ]).…”

“…This has encouraged researchers to develop more efficient numerical methods to manage and post-process such data. For instance, Kossman and Bigerelle [ 19 ] took advantage of artificial intelligence and computer vision models to discriminate the presence of pop-in in a large number of recorded indentation curve data sets. The developed convolutional neural network model achieves an accuracy of 93% for the classification stage while efficiently sorting the P-h curves.…”

Prediction of Steel Nanohardness by Using Graph Neural Networks on Surface Polycrystallinity Maps

“…indentation hardness), our model may also be fine-tuned to accurately forecast indentation-induced strain bursts (i.e. pop-ins) [30] and associated statistical distributions solely based on microstructural inputs.…”