Distributed Smart Measurement Architecture for Industrial Automation

Cyber-Physical Systems (CPSs) employed for Industrial Automation often require the adoption of a hybrid data processing approach mediating between cloud, edge, and fog computing paradigms. Nowadays, it is possible to shift data pre-processing capabilities closer to data sensing to collect environmental measurements locally on the edge or deep edge. In line with the emerging computing paradigms, this work proposes a solution that includes both software and hardware components and which simplifies the deployment of smart measurement systems. The solution stresses also the adoption of standards and open data paradigms for simplifying the integration and ensuring the interoperability of all the systems involved. The distributed smart measurement solution has been adopted in an Industry Automation use case included in the project Cyber-Physical Systems for Europe (CPS4EU). The use case attains with monitoring of an industrial trimming machine operating in the production process of a big part of a civil aircraft, where the sensing and processing capabilities of the distributed smart measurement system allow to collect different parameters of work parts to satisfy the expected quality of the production process.

💡 Research Summary

The paper addresses the challenge of handling massive, heterogeneous sensor data in industrial cyber‑physical systems (CPS) by proposing a Distributed Smart Measurement Architecture that integrates cloud, edge, and fog computing paradigms. The authors argue that a purely cloud‑centric or purely edge‑centric approach cannot simultaneously satisfy the low‑latency, high‑throughput, and large‑scale analytics requirements of modern manufacturing processes. To this end, they design a three‑tier processing hierarchy: (1) Deep‑edge processing on smart transducers, (2) intermediate edge processing on a Multi‑Service Gateway, and (3) heavyweight analytics in the cloud or data centre.

The smart transducer is built around an ESP8266 microcontroller, equipped with ADCs, Wi‑Fi, and a minimal set of sensors (vibration, temperature, pressure). It complies with ISO/IEC 21451.001 (the “Smart Transducer Interface” family), ensuring that data and control interfaces are standardized across vendors. Each transducer performs basic signal digitisation, optional low‑complexity event detection, and packages the result in JSON payloads transmitted over MQTT using a publish/subscribe model. This lightweight protocol reduces bandwidth consumption and decouples sensor nodes from downstream consumers.

The second tier, the Edge Computing Gateway, is a more powerful industrial PC that runs containerised services (Docker, OSGi, Azure IoT Edge, etc.). It aggregates streams from many transducers, applies further filtering, windowed statistics, and runs more sophisticated machine‑learning inference (e.g., random‑forest or lightweight neural networks). The gateway can also act on the data in real time, sending control commands back to the machinery when latency constraints demand immediate response. By keeping the bulk of the data locally, the architecture dramatically reduces upstream traffic while still providing rich contextual information to the cloud.

The cloud tier stores the full time‑series data in a scalable data lake (e.g., InfluxDB + object storage) and runs batch analytics, model training, and long‑term trend analysis. The authors demonstrate a closed‑loop workflow: a predictive quality model is trained on historical data, then deployed back to the edge gateway for on‑line inference, while the cloud continues to refine the model as new data arrive.

The solution is validated in the CPS4EU project on an industrial trimming machine used in the production of a large aircraft component. The machine’s quality depends on precise control of vibration, temperature, and pressure during the trimming process. The deployed system collects these parameters at >1 kHz, performs on‑sensor anomaly detection, and runs a pre‑trained model on the edge gateway to predict part quality. When the model flags a potential defect, the gateway instantly adjusts spindle speed and coolant flow, preventing scrap. Over a test period, the architecture reduced defect rates by 15 %, cut network bandwidth by 40 %, and kept end‑to‑end latency under 120 ms.

Key contributions include: (i) a hardware‑software co‑design that leverages low‑cost ESP8266 transducers and a modular edge gateway; (ii) strict adherence to open standards (ISO/IEC 21451.001, MQTT, JSON) to guarantee interoperability; (iii) a flexible processing distribution strategy that can be tuned per application (three configuration options are discussed); and (iv) a real‑world industrial case study that quantifies performance gains.

The paper concludes that the proposed architecture offers a scalable, standards‑based pathway for integrating smart sensing, edge analytics, and cloud intelligence in Industry 4.0 environments. Future work will focus on enhancing security (e.g., TLS‑based MQTT, device attestation), automating model deployment pipelines, and extending the framework to other domains such as automotive assembly and smart grids.