Adaptive AI Models for Predictive Maintenance in Smart Factories

作者：Rohan U A, S. P, Praveen Kumar G D, S. S, Srinivas R, T. V · 发表于：2024 International Conference on Augmented Reality, Intelligent Systems, and Industrial Automation (ARIIA) · 年份：2024 · DOI：10.1109/ARIIA63345.2024.11051773 · 被引用次数：2

This paper proposes an adaptive AI-based predictive maintenance model particularly suitable for smart factories in view of dynamic operational conditions, non-stationary environments, and real-time decision-making. Traditional predictive maintenance systems use static models on historical data and provide poor adaptability to changing equipment behaviors and variable conditions, leading to suboptimal maintenance schedules and unexpected failures of equipment. It brings into light a hybrid architecture that embeds deep learning for feature extraction, reinforcement learning for dynamic decisions, and transfer learning for adapting knowledge across domains. The proposed model self-adjusts predictions as incoming sensor data streams in, continuously learning and adapting to new conditions without large-scale retraining. It is further supported through the experimental results carried out on the simulated environment of the smart factory that demonstrates a great attainment in predictive accuracy-94%, 30% less unexpected downtime, and low false alarm rate compared to a conventional static model. Experimental results confirm that the proposed adaptive AI model is highly effective in improving operational efficiency; hence, it has the potential to transform the predictive maintenance strategy for the next-generation industrial environment.