The global metalworking and manufacturing industry serves as a cornerstone of the economy, responsible for producing a wide range of essential consumer products. Yield rate stands as a critical indicator within manufacturing, exerting direct influence over production costs and customer contracts. High-end products frequently comply with stringent zero-defect standards, where even a single defective item incurs significant costs and damages company reputation. Advances in manufacturing quality inspection notwithstanding, mature metalworking processes typically achieve high yield rates, resulting in a minimal quantity of defective products and presenting challenges in defect detection and classification, commonly referred to as the limited data problem. Anomaly detection emerges as an effective method for overcoming obstacles posed by limited data, concentrating on the identification and differentiation of data points that diverge from established patterns, representing potential anomalies or defects. The infrequency of defects necessitates precise detection within a large dataset dominated by normal samples. The paper introduces an Anomaly Detection Mechanism for Manufacturing Defects, leveraging the capability of the GANomaly model to discern subtle distinctions within similar data. By employing extensive training on majority class datasets, this model demonstrates effectiveness in detecting manufacturing defects, with the objective of minimizing return costs and preventing losses in reputation attributable to undetected defective items.