Leveraging Generative AI for Essential Predictive Maintenance in Industrial Consumer Electronics

Centrifugal compressors are pivotal in the manufacturing and cooling systems of consumer electronics, critical for ensuring operational efficiency and product quality. As Industry 5.0 emphasizes personalized production, human-centric automation, and sustainability, optimizing these systems is crucial for enhanced production throughput and reduced downtime. This study presents a methodology for fault trend prediction in centrifugal compressors, combining tensor decomposition-based data imputation, VAE-GAN for health indicator (HI) generation, and LSTM networks for time-series forecasting. Missing data is imputed using Tucker decomposition to ensure data integrity. The VAE-GAN model then generates high-fidelity HIs, capturing complex degradation patterns, which are used as inputs for the LSTM network to provide precise forecasts of future health trends. Experimental results show significant improvements in accuracy and reliability, effectively addressing data incompleteness, non-linear data relationships, and long-term dependency modeling. This approach not only enhances predictive maintenance strategies, increasing efficiency and reducing downtime but also supports sustainable manufacturing practices crucial in the consumer electronics industry under Industry 5.0.