Machine learning for Non-Destructive Evaluation (NDE) has the potential to bring significant improvements in defect characterization accuracy due to its effectiveness in pattern recognition problems. However, the application of modern machine learning methods to NDE has been obstructed by the scarcity of real defect data to train on. This paper demonstrates how an efficient, hybrid finite element and ray-based simulation can be used to train a Convolutional Neural Network (CNN) to characterize real defects. To demonstrate this methodology, an inline-pipe inspection application is considered. This uses four plane wave images from two arrays, and is applied to the characterization of cracks of length 1-5 mm and inclined at angles of up to 20° from the vertical. A standard image-based sizing technique, the 6 dB drop method, is used as a comparison point. For the 6 dB drop method the average absolute error in length and angle prediction is ±1.1 mm, ±8.6° while the CNN is almost four times more accurate at ±0.29 mm, ±2.9°. To demonstrate the adaptability of the deep-learning approach, an error in sound speed estimation is included in the training and test set. With a maximum error of 10% in shear and longitudinal sound speed the 6 dB drop method has an average error of ±1.5 mm, ±12° while the CNN has ±0.45 mm, ±3.0°. This demonstrates far superior crack characterization accuracy by using deep learning rather than traditional image-based sizing.
An investigation was performed into the effects of operating an absolute eddy-current testing (ECT) probe at frequencies close to its electrical resonance. A previously undocumented defect signal enhancement phenomenon, resulting from associated shifts in electrical resonant frequency, was observed and characterized. Experimental validation was performed on three notch defects on a typical aerospace superalloy, Titanium 6Al–4V. A conventional absolute ECT probe was operated by sweeping through a frequency range about the electrical resonance of the system (1−5MHz). The phenomenon results in signal-to-noise ratio (SNR) peak enhancements by a factor of up to 3.7, at frequencies approaching resonance, compared to those measured at 1MHz. The defect signal enhancement peaks are shown to be a result of resonant frequency shifts of the system due to the presence of defects within the material. A simple, operational approach for raising the sensitivity of conventional industrial eddy-current testing is proposed, based on the principles of the observed near electrical resonance signal enhancement (NERSE) phenomenon. The simple procedural change of operating within the NERSE frequency band does not require complex probe design, data analysis or, necessarily, identical coils. Therefore, it is a valuable technique for improving sensitivity, which complements other ECT methods
1 September 2018 in Composites Part B: Engineering
Deep learning for nondestructive evaluation (NDE) has received a lot of attention in recent years for its potential ability to provide human level data analysis. However, little research into quantifying the uncertainty of its predictions has been done. Uncertainty quantification (UQ) is essential for qualifying NDE inspections and building trust in their predictions. Therefore, this article aims to demonstrate how UQ can best be achieved for deep learning in the context of crack sizing for inline pipe inspection. A convolutional neural network architecture is used to size surface breaking defects from plane wave imaging (PWI) images with two modern UQ methods: deep ensembles and Monte Carlo dropout. The network is trained using PWI images of surface breaking defects simulated with a hybrid finite element / ray-based model. Successful UQ is judged by calibration and anomaly detection, which refer to whether in-domain model error is proportional to uncertainty and if out of training domain data is assigned high uncertainty. Calibration is tested using simulated and experimental images of surface breaking cracks, while anomaly detection is tested using experimental side-drilled holes and simulated embedded cracks. Monte Carlo dropout demonstrates poor uncertainty quantification with little separation between in and out-of-distribution data and a weak linear fit ( $R=0.84$ ) between experimental root-mean-square-error and uncertainty. Deep ensembles improve upon Monte Carlo dropout in both calibration ( $R=0.95$ ) and anomaly detection. Adding spectral normalization and residual connections to deep ensembles slightly improves calibration ( $R=0.98$ ) and significantly improves the reliability of assigning high uncertainty to out-of-distribution samples.
This paper reports a novel, negligible-cost and open-source process for the rapid prototyping of complex microfluidic devices in polydimethylsiloxane (PDMS) using 3D-printed interconnecting microchannel scaffolds. These single-extrusion scaffolds are designed with interconnecting ends and used to quickly configure complex microfluidic systems before being embedded in PDMS to produce an imprint of the microfluidic configuration. The scaffolds are printed using common Material Extrusion (MEX) 3D printers and the limits, cost & reliability of the process are evaluated. The limits of standard MEX 3D-printing with off-the-shelf printer modifications is shown to achieve a minimum channel cross-section of 100×100 μm. The paper also lays out a protocol for the rapid fabrication of low-cost microfluidic channel moulds from the thermoplastic 3D-printed scaffolds, allowing the manufacture of customisable microfluidic systems without specialist equipment. The morphology of the resulting PDMS microchannels fabricated with the method are characterised and, when applied directly to glass, without plasma surface treatment, are shown to efficiently operate within the typical working pressures of commercial microfluidic devices. The technique is further validated through the demonstration of 2 common microfluidic devices; a fluid-mixer demonstrating the effective interconnecting scaffold design, and a microsphere droplet generator. The minimal cost of manufacture means that a 5000-piece physical library of mix-and-match channel scaffolds (100 μm scale) can be printed for ~$0.50 and made available to researchers and educators who lack access to appropriate technology. This simple yet innovative approach dramatically lowers the threshold for research and education into microfluidics and will make possible the rapid prototyping of point-of-care lab-on-a-chip diagnostic technology that is truly affordable the world over.