Copyright © 2026 Authors retain the copyright of this article. This article is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

@article{200561,
        author = {BIDIPTA BISWAS and Asis Marceline V},
        title = {VIGIL AI: A Multimodal Industrial Anomaly Detection Platform with Autonomous Closed-Loop Maintenance Orchestration},
        journal = {International Journal of Innovative Research in Technology},
        year = {2026},
        volume = {12},
        number = {12},
        pages = {1447-1454},
        issn = {2349-6002},
        url = {https://ijirt.org/article?manuscript=200561},
        abstract = {Industrial predictive maintenance platforms have historically suffered from three structural limitations: unimodal signal analysis that ignores complementary evidence across sensor, visual, and historical modalities; detection-only architectures that leave downstream maintenance workflows unintegrated; and static health status that persists without autonomous re-evaluation as conditions evolve. This paper presents VIGIL AI, a cloud-native multimodal industrial anomaly detection and maintenance orchestration platform that addresses all three limitations simultaneously. VIGIL AI fuses three independently computed health signals — a rolling-window z-score sensor anomaly score computed client-side using TensorFlow.js, a structured visual defect severity score produced by Google Gemini 2.5 Flash via tool/function calling, and an LLM-driven historical failure-pattern match confidence score — into a single composite health index called the VIGIL Score, defined as a weighted linear combi-nation with modality weights derived from the relative evidential directness of each channel. The platform implements a fully closed-loop maintenance workflow wherein detected anomalies automatically generate structured alerts with AI-generated root cause hypotheses, alerts escalate directly into scheduled maintenance work orders, completed maintenance execution triggers autonomous anomaly resolution and health score adjustment, and returning-to-normal sensor readings automatically restore machine status without manual intervention. Experimental evaluation across a twelve-machine simulated fleet demonstrates a macro-averaged health status classification F1-Score of 0.945, VIGIL Score calibration validated against empirical near-term failure rates increasing monotonically from 0.5% in the Normal zone to 72.1% in the Critical zone, and an 89.5% reduction in alert volume relative to naive per-reading alerting while maintaining high true positive coverage. These results confirm the analytical validity of the multimodal fusion approach and the operational effectiveness of the closed-loop maintenance architecture.},
        keywords = {predictive maintenance, multimodal anomaly detection, industrial IoT, health index fusion, large vision models, LLM diagnostic reasoning, closed-loop maintenance, real-time monitoring},
        month = {May},
        }