AI-Driven Universal IoT Architecture for Real-Time Industrial Monitoring and Decision Support

Industrial monitoring systems are frequently assembled from incompatible device firmware, proprietary messaging conventions and bespoke dashboards, so that each new sensor or machine repeats the same integration work. This paper presents a universal Internet of Things (IoT) architecture, implemented as the Universal AIoT Platform, that removes this repetition through three mechanisms: a single canonical device protocol, automated generation of device software, and a real-time decision-support dashboard. The canonical protocol, designated aiot/v1, defines a versioned Message Queuing Telemetry Transport (MQTT) topic namespace and validated message schemas for telemetry, commands, acknowledgements and presence. From a declarative device description, a software development kit (SDK) generator produces ESP32 firmware and runtime code, so that a device joins the platform without hand-written integration code; the SDK is publicly distributed as the npm package aiot-sdk. A virtual device and sensor/actuator simulator, enabled by default, allows the complete protocol to be exercised, tested and demonstrated without physical hardware. Security is treated as an architectural property: each device receives unique login credentials at SDK download, credentials are bound to that device's topics, organisations are isolated from one another, and human users act under role-based authorisation. The platform was evaluated headlessly in a laboratory setting. The end-to-end acceptance flow passed 19 of 19 steps, workspace test suites passed 195 of 195 tests, all 5 workspace builds succeeded, the standalone SDK passed 44 of 44 tests, and generated firmware compiled for ESP32 into a binary of 1,048,432 bytes, while the whole system runs within approximately 512 MB of memory. Evaluation used simulated devices rather than a factory deployment. Within that scope, the results show that a canonical protocol combined with generated device software can make real-time industrial monitoring reproducible, testable and inexpensive to extend.