A quantitative diagnostic method of feature coordination for machine learning model with massive data from rotary machine

With significant growth in smart manufacturing, recent studies have raised concerns regarding the maintenance of industrial machinery. They show a profound interest in data-driven fault diagnosis methods. The process of forming a steady maintenance system is partially but essentially dependent on the diagnostic model. The development of the model relies on the corresponding machinery's data. However, the data-driven model pro-cesses substantial data, resulting in excessive processing time. The processing time includes the time spent to form a diagnosis model and the time for the model to evaluate the machinery condition with quantitative ac-curacy. This study proposes a method to form a fault diagnosis model and to classify a certain operating con-dition with specific logic. The proposed method, Feature variable Dimensional Coordination (FDC) is a conceptual logic based on feature engineering and distance evaluation of condition indicators. FDC reduces substantial data needed by coordinating the selection of valid feature variables. Furthermore, distance evaluation of coordinated feature variables provides a quantitative indicator called FDC Number to evaluate its current efficiency. The proposed method was verified with three different datasets. Applying the FDC method resulted in the processing time reduction by up to 61% in one of the cases. Numerous other cases showed an identical tendency regarding time. FDC showed improvement in accuracy of diagnosing defects on equivalent verification cases from 54.8% to 93.7%. In conclusion, FDC enhanced the performance of a condition diagnosis model by maintaining the classification accuracy over approximately 90% in most cases and reducing the need for data, resulting in the processing time reduction.