Encouraging the production of casting products while cutting down on inspection expenses and time commitments is made possible through automated surface fault inspection. As a key element of contemporary modern manufacturing, intelligent systems with image classification capabilities are frequently used in visual inspection. In comparison to traditional machine learning methods, Convolutional Neural Networks have demonstrated notable improvements in image categorization tasks. The purpose of this research is to investigate the use of transfer learning techniques for the recognition and classification of casting defects on the surface using the Resnet152V2-based Convolutional Neural Network architecture. To demonstrate the model's adaptability in heterogeneous casting, a range of image-processing techniques were first used to casting datasets for data enhancement. The lightweight ResNet152V2 model that has already been trained is then modified, and its hyperparameters are adjusted to maximize its performance. The effort produces a scalable, lightweight, and adaptive model that is perfect for edge devices with limited resources. The model has been assessed using different performance matrix. The accuracy of the model is 99.65%, precision achieved is 99.95%, recall is 99%, based on statistical parameters. These findings demonstrate that a CNN framework based on the ResNet150V2 architecture can be effectively used for the identification of casting surface defects of impellers used in pumps in manufacturing sectors, despite its uncomplicated architecture.