Machine Learning is a key technology opening an enormous potential also in the field of NDE and impact-echo as well as ultrasonics specifically. The echolyst A.I. module makes this technology easy to access for NDE practitioners. The module implements artificial intelligence in a way that makes the process not only highly effective but also very transparent and understandable. It addresses practitioners that have no experience with Machine Learning but is also a very handy tool for experienced Machine Learning application engineers.
The module selects the echolyst data sets and offers a variety of preprocessing options targeted at impact-echo and ultrasonic data. The preprocessing includes signal processing in the time as well as the frequency domain, time-frequency spectrogram calculation using the Continuous Wavelet and Short-Time Fourier trans-form, filtering, etc. The data set management includes labeling of the data sets and splitting them into balanced training and test sets. It is handled interactively and supported by graphical user interfaces.
Deep Learning architectures can be defined in user-friendly dialogs and are displayed in technical plots. Multi-Layer Perceptron Neural Networks (MLP) as well as Convolutional Neural Networks (CNN) are offered. Classification as well as regression models can be trained. The resulting learning curves state the basis for further model adjustment.
The software accesses the widely established Tensor-flow library for model generation, thus ensuring con-sistent results and compatibility with Python coding.