Quality Assessment of Metal Additive Manufactured Parts by a Multiscale Convolutional Fuzzy Neural Network Using Ultrasound Images as Input Data

The successful production of metallic workpieces through selective laser melting requires a quality assurance process that can effectively and nondestructively assess internal defects. Ultrasound testing is a nondestructive testing modality that can be used to identify defects and characterize the microstructure and properties of a material. Moreover, developments in computer vision techniques have led to the increased use of neural networks (NNs) in quality assurance processes. Therefore, a novel multiscale convolutional fuzzy NN (MCFNN) that uses ultrasound images as input data was developed in this study to automatically evaluate Inconel 718 workpieces fabricated through selective laser melting under various combinations of process parameters. The average accuracy, precision, recall, F1 score, and number of required parameters for the developed MCFNN were comparable to those of state-of-the-art models. Moreover, several concatenation methods, multiscale fusion strategies, and fuzzy inference systems were implemented in the developed MCFNN for performance comparison. Subsequently, the workpieces were examined through microcomputed tomography to verify the results obtained using ultrasound images. The experimental results indicated that among the compared models, the MCFNN achieved the highest average accuracy (91.44% ± 4.73%), precision (92.74% ± 3.79%), recall (91.44% ± 4.73%), and F1 score (91.35% ± 4.82%) and required the fewest parameters (107,138). The experimental results demonstrate that the developed MCFNN has high potential for implementation in the embedded devices of portable ultrasound scanning systems.