A study of supervised combined neural-network-based ultrasonic method for reconstruction of the spatial distribution of material properties

Abstract

This paper examines the performance of the commonly used neural-network-based classifiers for investigating such structural noise in metals as grain size estimation. It is extremely difficult to determine the grain size of objects only by the internal structure features of the object. When the structured data is obtained, a proposed feature extraction method is used to extract the feature of the object. Afterwards, the extracted features are used as the inputs for the classifiers. This research study is focused on using basic ultrasonic sensors to obtain object’s structural grain size. The performance for the used neural-network-based classifier is evaluated based on recognition accuracy for an individual object. Furthermore, traditional neural networks, namely, convolutional and fully connected dense networks are shown as a result of the grain size estimation model. To evaluate the robustness property of neural networks, the original samples data are mixed for three types of grain sizes. Experimental results show that combined convolutional and fully connected dense neural networks with classifiers outperform the other single neural networks with original samples with high signal-to-noise ratio data. The dense neural network as itself demonstrated the best robustness property when the object samples did not differ from trained datasets.