DEEP LEARNING-BASED ANOMALY DETECTION FOR RELIABLE INDUSTRIAL IOT MAINTENANCE

Abstract

This paper introduces an AI-based anomaly detection system in Industrial IoT (IIoT) networks that can help improve predictive maintenance in intelligent manufacturing settings. The suggested model combines LSTM- autoencoders with attention systems to analyze time-series sensor data of industrial machinery, such as CNC machines, robotic arms, and PLC systems. The findings show that the accuracy of detection is very high (97 percent) and the F1-score is high (95 percent), which is much higher than the traditional rule-based systems (76 percent). The structure reduces the level of false positive (18 to 7) which increases reliability and minimizes the number of unnecessary maintenance measures.

The AI-centered model which was introduced leads to reduced downtimes by 30 percent in comparison to preventive maintenance strategies which usually only lead to improvement by 15 percent. Also, the system saves about 25 percent in costs through optimization of maintenance schedules and minimization of unforeseen failures. The edge-based processing achieves a 40 percent better real-time response through a reduction of 150 ms in the detection latency. The framework also has a high level of scalability and can support a performance efficiency of more than 89% when implemented in large scale industrial environments and 200 connected devices.

All in all, the research indicates that AI-based anomaly detection can help to enhance the efficiency of operations, the reliability of the system, and the cost-effectiveness of IIoT networks. The findings support the idea of using smart, data-driven maintenance strategies in Industry 4.0 and the need of the expanded computational efficiency and interpretability of the model in the more general industrial implementation.