Recognition of fault process conditions based on spinline tension in melt spinning

Proper setting of process conditions in the melt spinning setup is one way to yield uniform quality of spinline tension. The optimum setting to give uniform spinline tension is determined using experiment plans in the Taguchi method and significant process parameters are also identified. When the setting shifts from the optimum, the spinline tension becomes non-uniform and downgrades product quality. This study aims to diagnose single or double fault conditions of those significant process parameters based on the spinline tension signal. The critical procedures of fault diagnosis are feature extraction and classification. The tension signal is decomposed into a wavelet packet tree of four resolution levels. Four entropies from the best-basis wavelet packet tree and the lowest entropy at level four are selected as features. The back-propagation neural network acts as a classifier. The experimental results demonstrate that the features and classifier actually work well to identify the single and double fault conditions with high accuracy in melt spinning.