Typed Telemetry, Advisory Analytics: An EPICS–Spark Architecture for Dental Robotic Arms

Robot-assisted dentistry is entering routine clinical practice, intensifying the need for deterministic control, typed telemetry, and a provable separation between motion and analytics near sensitive anatomy. We present a partitioned architecture for safety-critical dental robotic arms that couples Experimental Physics and Industrial Control System (EPICS 7) with Apache Spark Structured Streaming. EPICS—via pvAccess, QSRV, and Normative Types—publishes coherent, self-describing supervisory data (pose, depth, forces, images) with delta-efficient subscriptions and deterministic alarms [1]–[3]. Long-term, audit-grade histories are retained by the EPICS Archiver Appliance [7]. Selected process variables are mirrored to Apache Kafka using an EPICS→Kafka bridge, enabling Spark to compute advisory analytics with exactly-once semantics and operational latencies commonly in the sub-second regime, including ~100–250 ms p99 micro-batch performance and an optional real-time trigger reaching tens of milliseconds for suitable stateless workloads [4]–[6], [8]. The design formalizes safety partitions and latency budgets to ensure all motion commands remain within the robot controller and safety programmable logic controller (PLC), with operator-gated actions exposed only through supervised EPICS channels—never direct actuation from analytics [1]–[3], [8]. We map the architecture to robotics and medical frameworks, including ISO 10218 (2025) for robot and system integration safety, ISO/TS 15066 for collaborative force/pressure limits, ISO 13485 for quality management systems, IEC 60601-1 for medical electrical safety and essential performance, and IEC 61508 for functional safety lifecycle [9]–[13]. A verification and validation plan demonstrates EPICS coherence, Spark resilience (checkpointed recovery, watermarks), collaborative safety compliance, electrical safety, and cybersecurity hardening; essential performance is preserved even under analytics failure. Clinical context and adverse-event reports motivate fast deviation detection and auditable operator gating in patient-adjacent workflows [14]. Contributions include: (1) a publishable reference design, (2) quantitative latency and safety partitions, (3) standards mapping, and (4) a practical V&V checklist for deployment.