An Incremental Self-Learning Interval Type-2 Fuzzy Logic Based Explainable Approach to Predictive Maintenance

Predictive maintenance, powered by real-time tracking to anticipate and avert failures, is reshaping industrial maintenance, with the Industrial Internet of Things (IIoT) at its core, providing instant data for condition monitoring. However, the adoption in the water pumping industry is slow due to the perceived complexity of AI systems used in predictive maintenance. To counter this, our research presents a self-learning, explainable AI system that combines the clarity of rule-based systems with optimized Interval Type-2 Fuzzy Logic Systems. This innovation, inspired by natural learning and forgetting processes, continually adapts, learning from past experiences to enhance performance. It aims to foster trust and understanding of AI decisions among engineers, offering clear explanations for its predictions, thus enabling informed maintenance decisions. Developed using data from 30 devices at UK water pumping sites, the system constantly evolves, outperforming existing solutions in predicting four distinct faults. It integrates rule-based systems with self-learning capabilities, autonomously optimizing its parameters, reducing human intervention. In tests, it showed an 8.99% higher average accuracy compared to its T1FLSs counterpart and a 529.21% increase over decision trees. A survey revealed 80.3% of industry experts agreed substantially with the AI's explanations, indicating high trust in its decisions. The XAI approach promises significant operational efficiencies, preventing unnecessary site visits and ensuring engineers are better prepared when visits are essential, thereby reducing downtime and conserving resources. In conclusion, our innovative AI system offers not only superior accuracy but also the transparency and adaptability needed to earn industry professionals' trust, paving the way for optimized, reliable predictive maintenance in the water pumping sector.