Multi-sensor data-driven intelligent product interaction fault self-diagnosis system

Abstract

Intelligent fault diagnosis is crucial for the safety and reliability of dynamic smart manufacturing systems. Existing single-sensor approaches do not model complex, multi-dimensional interactions between robotic components during dynamic interactions. This paper introduces a Kronecker Fused Variational-Evolutionary Reconstruction Framework for Adaptive Intelligent Diagnostics (KROVER-AID), which enables real-time, cross-sensor, and across-sensor fault diagnosis. The Kronecker fusion method captures the high-order cross-dependencies among vibration, current and torque signals to generate unified and discriminative feature embeddings. UNVAE (Unified Nested Variational Autoencoder) serves as a nested encoder-decoder to uncover latent fault features and accurately reconstruct fused representations, thus exposing subtle faults. UNVAE is trained by ERM (Evolutionary Reconstruction Model) using adaptive evolutionary optimization to fine-tune the parameters of the decoder, thus reducing the reconstruction error and improving the adaptability to unseen fault cases. Experimental validation on industrial-scale electric motor datasets with variable load and speed conditions shows that KROVER-AID outperforms the benchmark approaches by 6.5% in fault detection accuracy and 5.8% in F1-score. The proposed framework has demonstrated significant robustness, cross-domain generalization, and autonomous adaptability; therefore, it is a scalable and intelligent approach to predictive maintenance and fault management in next-generation smart manufacturing systems.