Smart Factory: A real-time predictive maintenance platform based on microservices architecture and deep learning

Industrial equipment failures lead to production losses and safety risks. This paper presents Smart Factory, a real-time predictive maintenance platform that integrates IIoT ingestion, signal processing, feature extraction, anomaly analysis, and RUL prediction within a scalable microservices architecture. Seven microservices communicate through Apache Kafka, and LSTM, GRU, and TCN models are trained on the NASA C-MAPSS dataset. The main contribution is the software integration of prognostic and diagnostic functions in a unified industrial platform. Because NASA C-MAPSS has no explicit anomaly labels, anomaly analysis is treated as a validation of the pipeline using proxy labels derived from anomaly scores. Experiments on FD001 show that the ensemble RUL model achieves a mean absolute error of approximately 12 cycles, with error reductions of 25.7% and 35.0% over representative LSTM- and CNN-based baselines. The supervised anomaly-classification stage achieves F1-scores between 0.97 and 0.98. The proposed platform supports end-to-end real-time predictive maintenance and maintenance-oriented decision support.