Surface asperities play a leading role in determining the fatigue life of as-built Ti-6Al-4 V components manufactured by electron beam powder bed fusion (PBF-EB). Several roughness parameters are available to characterize the surface asperities This study focuses on identifying the surface roughness parameter that correlates best with fatigue life. To this end, several fatigue test specimens were manufactured using the PBF-EB process and utilizing different contour melting strategies, thus producing as-built surfaces with varying roughness. The focus variation microscopy technique was employed to obtain surface roughness parameters for the as-built surfaces. Selected specimens were characterized using x-ray computed tomography (XCT). Tomography can detect surface-connected features obscured by other parts of the surface that are not visible through optical microscopy. The fatigue life of all specimens was determined using four-point bend testing. Through regression model analysis, maximum pit height (Sv) was identified as the statistically significant roughness parameter with the best fit affecting fatigue life. The fracture zone was closely inspected based on the data collected through XCT prior to fatigue tests. This led to another estimate of the worst-case value for the statistically significant roughness parameter Sv. The Sv parameter values obtained from optical microscopy and XCT were used as the initial crack size in a crack growth model to predict fatigue life. It is observed that life estimates based solely on optical measurements of Sv can be overly optimistic, a situation that must be avoided in predictive design calculations.

Within quality critical industries, e.g. aerospace, quality control with non-destructive evaluation (NDE) is essential. The surface quality is often important and e.g. visual inspection is often applied. Part of the inspection is the data interpretation, not easily made automatic for critical products. Recent studies on the automatization have indicated promising results utilizing deep-learning-based artificial intelligence. However, many such algorithms are known to be overconfident when subjected to unexpected input (e.g. new/rare material defects) far from the training dataset, so-called out-of-distribution (OOD) data. We claim that safe computer-based interpretation of NDE data within quality critical applications, must respond sensible also to OOD data.

A sensible response could be that the algorithms identify such OOD data and forward it to a human for further analysis. Such an OOD detector could facilitate a human-machine collaboration in a NDE 4.0 vision. In this work we have explored if a recently proposed (for industrial x-ray images) auto-encoder-based approach can be utilized as OOD detector (one-class classifier) for visual inspection data. The model is trained in an unsupervised manner on accepted input to reconstruct it at high precision. Simultaneously it is trained to remove synthetically added defect indications to generate a clean image patch, similar to denoising-auto-enoders. The difference between the input and reconstructed input is analyzed for OOD detection. We train and test the algorithm on a publicly available visual inspection dataset with surface defects. We achieve true positive rates at 0.90 with true negative rates at 0.99 and demonstrate detection of OOD data.

X-ray inspection is often an essential part of quality control within quality critical manufacturing industries. Within such industries, X-ray image interpretation is resource intensive and typically conducted by humans. An increased level of automatization would be preferable, and recent advances in artificial intelligence (e.g., deep learning) have been proposed as solutions. However, typically, such solutions are over confident when subjected to new data far from the training data, so-called out-of-distribution (OOD) data; we claim that safe automatic interpretation of industrial X-ray images, as part of quality control of critical products, requires a robust confidence estimation with respect to OOD data. We explored if such a confidence estimation, an OOD detector, can be achieved by explicit modeling of the training data distribution, and the accepted images. For this, we derived an auto encoder model trained unsupervised on a public dataset with X-ray images of metal fusion welds and synthetic data. We explicitly demonstrate the dangers with a conventional supervised learning-based approach and compare it to the OOD detector. We achieve true positiverates of around 90% at false positive rates of around 0.1% on samples similar to the training data and correctly detect some example OOD data.

X-ray computed tomography (XCT) is increasingly utilized industrially at material- and process development as well as in non-destructive quality control; XCT is important to many emerging manufacturing technologies, for example metal additive manufacturing. These trends lead to increased needs of safe automatic or semi-automatic data interpretation, considered an open research question for many critical high value industrial products such as within the aerospace industry. By safe, we mean that the interpretation is not allowed to unawarely or unexpectedly fail; specifically the algorithms must react sensibly to inputs dissimilar to the training data, so called out-of-distribution (OOD) inputs. In this work we explore data interpretation with deep neural networks to address: robust safe data interpretation which includes a confidence estimate with respect to OOD data, an OOD detector; generation of realistic synthetic material aw indications for the material science and nondestructive evaluation community. We have focused on industrial XCT related challenges, addressing difficulties with spatially correlated X-ray quantum noise. Results are reported on training auto-encoders (AE) and generative adversarial networks (GAN), on a publicly available XCT dataset of additively manufactured metal. We demonstrate that adding modeled X-ray noise during training reduces artefacts in the generated imperfection indications as well as improves the OOD detector performance. In addition, we show that the OOD detector can detect real and synthetic OOD data and still model the accepted in-distribution data down to the X-ray noise levels.