D2ai: A Rami 4.0–aligned Architecture For Industrial Analytics and Real-Time Visualization

The IIoT, big data analytics, and artificial intelligence are increasingly relied upon in Industry 4.0 to support data-driven decision-making. Nevertheless, secure, end-to-end, on-premises data-science workflows are challenging to implement, especially when sensitive production data must remain within the factory. This paper describes D2AI, an operational architecture of secure industrial analytics, model operationalization, and real-time dashboard visualization based on RAMI 4.0. Its architecture is structured into presentation, logic, and data layers, which allow regulated data ingestion, anonymization, segregation between production and clean data space, and prompt delivery of model outputs to end users. By doing so, data scientists will be able to operate with analytics-ready data, without necessarily accessing production databases. To test the proposal, three end-to-end experiments were performed using a dataset of simulated CNC temperature stream under normal, drift, and fault-injection conditions. The findings indicated a stable throughput of about 1 event/s, low event loss during fault conditions, and low inference latency (median 6 ms, p95 < 10 ms), demonstrating the feasibility of secure near-real-time industrial analytics and decision support within the factory shop floor.