Fault diagnosis systems are used to improve the productivity and reduce the costs of the manufacturing process. However, the feature variables in existing systems are extracted based on the classification performance of the final model, thereby limiting their applications to models with different conditions. This paper proposes an algorithm to improve the characteristics of feature variables by considering the cutting conditions. Regardless of the frequency band, the noise of the measurement data was reduced through an oversampling method, setting a window length through a cutter sampling frequency, and improving its sensitivity to shock signal. An experiment was subsequently performed to confirm the performance of the model. Using normal and wear tools on AI7075 and SM45C, the diagnosis accuracies were 97.1% and 95.6%, respectively, with a reduction of 85% and 83%, respectively, in the time required to develop a diagnosis model. Therefore, the proposed algorithm reduced the model computation time and developed a model with high accuracy by enhancing the characteristics of the feature variable. The results of this study can contribute significantly to the establishment of a high-precision monitoring system for various processing processes.

Roll-to-roll manufacturing systems have been widely adopted for their cost-effectiveness, eco-friendliness, and mass-production capabilities, utilizing thin and flexible substrates. However, in these systems, defects in the rotating components such as the rollers and bearings can result in severe defects in the functional layers. Therefore, the development of an intelligent diagnostic model is crucial for effectively identifying these rotating component defects. In this study, a quantitative feature-selection method, feature partial density, to develop high-efficiency diagnostic models was proposed. The feature combinations extracted from the measured signals were evaluated based on the partial density, which is the density of the remaining data excluding the highest class in overlapping regions and the Mahalanobis distance by class to assess the classification performance of the models. The validity of the proposed algorithm was verified through the construction of ranked model groups and comparison with existing feature-selection methods. The high-ranking group selected by the algorithm outperformed the other groups in terms of training time, accuracy, and positive predictive value. Moreover, the top feature combination demonstrated superior performance across all indicators compared to existing methods.