The Transformers are modules that map data (textual or image) into embeddings conventionally used in natural language processing. These are currently examined for utility to industrial work situations that generate huge amount of visual data during processing or quality inspections. The present research adopts a novel method based on Visual Transforms (ViTs) for surface defect detection in Aluminium die castings as an alternate approach to Convolution Neural Networks (CNNs). This technique overcomes limitations in local feature extraction of CNN method through ‘self-attention methods’ to capture global dependencies in the fault pattern. The work highlights the adept nature of ViTs in detecting complex anomalies thereby introducing better robustness and accuracy. The comparative values redeem intrinsic benefits of ViTs and associated long term relationships with defective images captured. The Visual transforms are promising to revolutionize fault detection through deeper insights by improved scalability and adaptability that form vital requirement in the current production scenario. The results reveal huge potential for non-invasive fault detection of aluminium die casting as a route to practice sophisticated quality policy using the proposed technique.