Evidence fusion method for robust state monitoring of heterogeneous textile circular knitting machine control panels

Real-time monitoring of circular knitting machines is important for production continuity and fabric quality. In textile workshops, specular reflection, low illumination, and chromatic similarity between active light-emitting diodes and passive panel components complicate visual state recognition. A lightweight monitoring framework for heterogeneous circular knitting machine control panels is presented, based on subjective logic and conflict-aware evidence fusion. Hue saliency, nonlinear red-green-blue ratios, topological coherence, luminance contrast, and local entropy are converted to opinions comprising belief, disbelief, and uncertainty. Entropy-guided vacuity discounts unreliable observations, while panel profiles specify region-of-interest geometry and operating priors for configuration-driven deployment. For the field-collected dataset, the method achieved a count-weighted accuracy of 94.7% and an F1 score of 0.922. The framework provides interpretable state estimates at a computational cost suitable for online monitoring.