Automated Quantitative Analyses of Fatigue-Induced Surface Damage by Deep Learning

Abstract

The digitization of materials is the prerequisite for accelerating product development. However, technologically, this is only beneficial when reliability is maintained. This requires comprehension of the microstructure-driven fatigue damage mechanisms across scales. A substantial fraction of the lifetime for high performance materials is attributed to surface damage accumulation at the microstructural scale (e.g., extrusions and micro crack formation). Although, its modeling is impeded by a lack of comprehensive understanding of the related mechanisms. This makes statistical validation at the same scale by micromechanical experimentation a fundamental requirement. Hence, a large quantity of processed experimental data, which can only be acquired by automated experiments and data analyses, is crucial. Surface damage evolution is often accessed by imaging and subsequent image post-processing. In this work, we evaluated deep learning (DL) methodologies for semantic segmentation and different image processing approaches for quantitative slip trace characterization. Due to limited annotated data, a U-Net architecture was utilized. Three data sets of damage locations observed in scanning electron microscope (SEM) images of ferritic steel, martensitic steel and copper specimens were prepared. In order to allow the developed models to cope with material-specific damage morphology and imaging-induced variance, a customized augmentation pipeline for the input images was developed. Material domain generalizability of ferritic steel and conjunct material trained models were tested successfully. Multiple image processing routines to detect slip trace orientation (STO) from the DL segmented extrusion areas were implemented and assessed. In conclusion, generalization to multiple materials has been achieved for the DL methodology, suggesting that extending it well beyond fatigue damage is feasible.

A Deep Learning Approach for Complex Microstructure Inference

“…Nonetheless, the fact that even simple targeted optimizations of low-variance training data can cause such improvements, implies that dedicated data augmentation pipelines can presumably render models robust against a large range of perturbations in the specimen preparation or imaging. For instance, in our prior study 18 a substantial improvement was achieved by augmentation of our high-variance data. When training images are acquired from different instruments or at different institutions, such regularization methods become increasingly relevant.…”

“…This means, the closer a pixel is to a target pixel, the more it influences the target pixels' predicted class 44 . This represents a CNN-based inductive bias appropriate for many scientific segmentation challenges, such as for fatigue damage localization where image features are dense 18 . However, for phase segmentation tasks, where long-range features (parallelism of distant carbide islands) are relevant, Attention-based networks 45 could improve segmentation performance.…”

“…Deep learning segmentation approach 1 -Vanilla U-Net. A vanilla U-Net model with an architecture implemented in the PyTorch framework 26 that included few adjustments with respect to 18 was trained from scratch. Only two class channels in the output were used since the work at hand covers a binary segmentation problem.…”

“…Different data augmentation techniques from the Albumentations package 29 were applied to investigate their impact on the performance. In contrast to our prior study 18 , a more systematic approach to data augmentation was taken by applying grid, and random search for optimization of relevant hyperparameters in Tune 30 . The set of optimized augmentation parameters for both image modalities is outlined in the Supplemental.…”

“…In contrast, the applied DL methods are directly based on input and target output image pairs (representation learning). Their application to microstructure recognition demonstrated the potential for quantitative microstructure analysis 15,16 , steel type classification 15 , crack path prediction 17 , and micromechanical damage segmentation 18 . A CNN architecture referred to as U-Net 2 has proven its merit in the latter work and represents a common starting point due to its numerous implementations in different DL frameworks and image processing tools 19,20 .…”