Real-Time Anomaly Detection in Business Process Analytics Using Self-Supervised Transformers and Digital Twins

In order to overcome the difficulties of tracking intricate and ever-changing business processes, this paper introduces a novel real-time anomaly detection framework that combines digital twin technology with a self-supervised transformer-based model. The suggested method, which makes use of the Vision Transformer Masked Autoencoder (ViT-MAE), can be deployed scalable across domains with little human oversight because it doesn't require labeled data. The framework is appropriate for integration into digital twin systems that require real-time feedback and adaptability due to its strong generalizability and low-latency inference. The approach outperforms state-of-the-art baselines in anomaly detection and localization, according to evaluations performed on the MVTec AD dataset, which is used as a proxy for quality-driven business process outputs. The findings show that the framework can facilitate intelligent, interpretable, and domain-neutral anomaly monitoring in actual business and industrial settings.