Abstract As nanoindentation-collection becomes increasingly high-throughput, manual anomaly detection and filtration become increasingly prohibitive. In this work, we present the first open data set for anomaly detection on load–displacement curves. This data set consists of 9600 expert-annotated indentation curves over 150 NiFe thin-film electrodeposited samples. We describe a novel taxonomy of anomaly types based on these expert annotations, as well as the potential limitations and bias of a single label set. To that end, we also describe a suite of consistency checks derived from the initial labelling to create a human-tractable set of indents to be reverified by the expert, and describe the resulting second label set for improved ground truth. Both the software for reverification and the dataset are openly available. These tools provide the critical foundation for reproducible machine-learning research for anomaly detection and classification in this space.
Zakariya et al. (Mon,) studied this question.