Dynamic Observability for Automated ML Pipelines in Predictive Maintenance for Industrial IoT
Journal of Circuits, Systems and Computers · 15.11.2026 · 35 (18)
DOI: 10.1142/s0218126625420022 ↗Kurz gefasst
Fachartikel in der Zeitschrift Journal of Circuits, Systems and Computers, online erschienen am 29.10.2025. Verfasst von M. Thanjaivadivel und 5 weiteren Autor:innen. Die Arbeit wurde bisher 1-mal zitiert (Crossref) und nennt 27 Literaturangaben.
Abstract (Originalsprache)
The widespread uptake of Industrial Internet of Things (IIoT) technologies has revolutionized predictive maintenance through real-time monitoring and fault prediction. Increasing complexity in automated machine learning (ML) pipelines, however, requires greater transparency, traceability and resilience. This work presents a new framework for dynamic observability of predictive maintenance pipelines for IIoT systems. The solution of choice combines modern observability features like Grafana, Prometheus, and mltrace with deep neural networks like LSTM and transformers for real-time monitoring, tracing and optimizing all pipeline elements. Employing the NASA C-MAPSS data set, the system efficiently eliminates noise and performs dynamic feature extraction using moving averages and time derivatives. The main innovations are approximation-based metric assessments, streaming data ingest through reservoir sampling and count-min sketch, and end-to-end metric consolidation through ML metrics fused with operational KPIs. The observability framework facilitates anomaly detection, diagnosis of performance degradations and recovery on auto-pilot, improving model accuracy, lessening feature staleness and pipeline latency as well as throughput. Case studies illustrate its power in the identification of data drifts, fix of feature corruption, and optimization of model output towards business objectives. Empirical findings indicate an 8.2% increase in accuracy, 11.5% precision and 25% reduction in latency in preprocessing, confirming the performance of the framework. Feature staleness and corruption scores also decreased significantly, confirming improved data integrity. Visual analyses also confirm classifier stability and feature freshness improvements. In general, the dynamic observability framework provides a scalable and adaptive platform for predictive maintenance in IIoT, enhancing operational efficiency, reliability and self-healing capabilities.
Zitieren
APA
Thanjaivadivel, M., Jayaprakasam, B. S., Gupta, M., Mythrayee, D., Abbas, Q., & Muniyandy, E. (2026). Dynamic Observability for Automated ML Pipelines in Predictive Maintenance for Industrial IoT. Journal of Circuits, Systems and Computers, 35(18). https://doi.org/10.1142/s0218126625420022
Vancouver
Thanjaivadivel M, Jayaprakasam BS, Gupta M, Mythrayee D, Abbas Q, Muniyandy E. Dynamic Observability for Automated ML Pipelines in Predictive Maintenance for Industrial IoT. Journal of Circuits, Systems and Computers. 2026;35(18). doi:10.1142/s0218126625420022
BibTeX
@article{Thanjaivadivel2026,
author = {Thanjaivadivel, M. and Jayaprakasam, Bhagath Singh and Gupta, Manish and Mythrayee, D. and Abbas, Qaisar and Muniyandy, Elangovan},
title = {Dynamic Observability for Automated ML Pipelines in Predictive Maintenance for Industrial IoT},
journal = {Journal of Circuits, Systems and Computers},
year = {2026},
volume = {35},
number = {18},
doi = {10.1142/s0218126625420022}
}
EUVIMED