Casting Discontinuity Detection Method based on Deep Learning Algorithms

This paper concerns the detection of material discontinuity and inclusions in self-excited acoustic aluminium systems through the application of artificial intelligence methods. Classical machine learning techniques and convolutional neural networks were employed to tackle the binary classification problem associated with inclusion detection. The process shown involves the partitioning of base signals into samples for machine learning, followed by spectral analysis. Various feature extraction methods were compared, including Mel-Frequency Cepstral Coefficients (MFCC), short-time Fourier transform, Power Spectral Density (PSD) as well as wavelet transform. The study evaluates the efficacy of each method in capturing essential information for inclusion detection within the acoustic signals. Additionally, Principal Component Analysis (PCA) was employed to reduce dimensionality, and its impact on classifier performance was thoroughly analysed. The results of this research provide insights into the effectiveness of different neural network architectures and feature extraction techniques for the task of inclusion detection in self-excited acoustic cast aluminium systems. The findings contribute to the advancement of techniques for enhancing the quality control for the production of such systems through robust and efficient detection mechanisms.