Industrial defect detection has traditionally been done using human eyesight, which has always been costly, subjective, and often inaccurate. This underscores the importance of automated techniques like computer vision for effective quality control. We propose to incorporate Deep Convolutional Generative Adversarial Network (DCGAN) as a data approach for the augmentation of other computer vision methodologies, specifically Convolutional Neural Networks (CNNs), in categorizing defective pistons images. Our approach deals with issues associated with limited and unequal distributions by creating artificially-induced defects in the training set. We categorized our piston images into three groups: Normal, Defected 1 (broken, shaped out, and fallen), and Defected 2 (rust, oil, and grease marks). The results indicated that employing synthetic data generated by DCGAN improves CNN’s F1-score tremendously from the classical dataset’s 0.7750 to 0.9386. The F1-score for the combined datasets, consisting of synthetic and classical datasets, was equal to 0.8959. These findings demonstrate that DCGAN enhances overall accuracy for defect detection in the manufacturing industry, thereby providing them with reliable solutions that reduce the need for manual labeling processes.