The number of aging concrete infrastructures has dramatically increased with the past urban development. Deterioration is primarily caused by cracking, particularly internal cracks in concrete, which are not visible through visual inspection on the surface. In existing structures, internal cracks can be formed in various orientations, such as diagonal cracks, which present additional challenges in damage evaluation. Crack propagation is a major concern, as it can significantly compromise structural integrity, durability, and overall performance of concrete structure. Non-Destructive Testing (NDT) techniques, such as the Impact-Echo (IE) method, are commonly used to evaluate damage and estimate repair conditions. In this study, internal crack was artificially introduced using starch-type polysaccharide sheet to simulate interior damage in a real situation. Typically, internal conditions are assessed based on waveform characteristics obtained from elastic wave propagation in applying IE method. For interpretation of waveform data, it is commonly converted to the frequency domain using Fast Fourier Transform (FFT) analysis. While horizontal internal cracks typically result in a clear peak corresponding to crack depth, diagonal cracks often produce multiple dominant frequencies, which require more complex data interpretation. To address this, AI-based analysis utilizing supervised deep learning models with frequency domain data is proposed to evaluate and estimate the geometrical information of crack inside concrete. However, labelling datasets for training supervised models is challenging in the context of aging infrastructures, as these structures often lack specific details due to their unique and varied characteristics. To overcome this limitation for preparing datasets, this study also proposes the application of unsupervised deep learning models as a potential solution. The unsupervised approach is validated through comprehensive evaluation, demonstrating its effectiveness in detecting and characterizing internal defects in concrete specimens.