Synthetic data generation in manufacturing: a review of methods, domains, and emerging trends

Data scarcity remains a major barrier to the effective deployment of AI in manufacturing, where labeled data is often limited, costly, or difficult to obtain. This review investigates how synthetic data generation techniques are being applied to address this challenge in manufacturing AI applications. Eighteen recent papers (Jan 2024- May 2025) were analyzed and categorized based on generation methods, application domains,
and data modalities. Techniques covered include GAN (Generative Adversarial Networks), VAEs (Variational Autoencoders), diffusion models, simulation-based approaches, SMOTE (Synthetic Minority Oversampling Technique), and hybrid combinations. Their use spans tasks such as defect detection, predictive maintenance, process modeling, material design, and human–robot collaboration. The review highlights emerging
trends, methodological trade-offs, and practical challenges shaping the future of synthetic data in intelligent manufacturing systems. In addition to consolidating recent work, the review identifies underexplored research gaps and methodological challenges that shape future directions in synthetic data use for manufacturing AI.