DYNA JOURNAL ENGINEERING

Key words:

Edge AI, machine learning, deep neural networks, cyber-physical systems, industrial valves, PLC, embedded inference, anomaly detection, real-time monitoring, regression models, classification accuracy.

Article type:

ARTICULO DE INVESTIGACION / RESEARCH ARTICLE

Section:

RESEARCH ARTICLES

In industrial process installations, the improper operation or misconfiguration of safety-critical components, such as manually operated ball valves, can seriously compromise both process performance and plant safety. This work proposes a sensorless Edge AI method to estimate hand-operated ball valves states without the use of physical position sensors. Using multivariate time-series data collected from a PLC-based pilot plant, a benchmark evaluation is conducted comparing four Deep Learning (DL) and four classical Machine Learning (ML) models for classification and regression tasks. The models are deployed on an embedded platform, enabling real-time inference at the edge with a minimum latency of 500ms.
Results show Decision Tree (DT) and Random Forest (RF)
achieve high regression accuracy (R2 >0.98, MAE < 0.5), while
all eight model reach high classification accuracy. Additionally,
the computational efficiency metric that combines model
accuracy, latency, and size, confirming DT as the most efficient
model (1.83/(ms.KB) for edge deployment. This work contributes
a cost-effective and scalable monitoring strategy, particularly
suitable for complex industrial environments where physical
sensing and visual inspection are limited, offering a viable path
toward early anomaly detection and intelligent supervision
within cyber-physical systems.
• Key Words: Edge AI, machine learning, deep neural networks,
cyber-physical systems, industrial valves, PLC, embedded
inference.