AI-Based Multi-Domain Sensor Fusion for Needle Condition and Process Parameter Monitoring in Flat Knitting Machines

This work explores an AI-driven multi-domain sensor fusion approach for monitoring process parameters and detecting needle faults in flat knitting machines. Acoustic and thread tension sensors were used to capture machine states under varying polishing depths and controlled failure conditions. Convolutional neural network (CNN) classifiers were trained independently on both acoustic and tension data to identify operational and faulty states. Beyond single-modality classification, a late fusion strategy combining both sensor domains via weighted averaging and additional fusion layers – jointly trained for enhanced integration – yielded improved classification performance. The results demonstrate that acoustic and tension sensing enable reliable monitoring of operational and faulty machine states, and that multi-domain fusion further improves robustness and generalization.