A two-stage deep-learning-based balancing method for rotating machinery

Abstract

Abstract Purpose—balancing is essential to all rotating machinery. To make the balancing process convenient and inexpensive, new balancing technologies are needed. In this work, a two-stage deep-learning-based balancing method is proposed and validated. Design/methodology/approach—the architecture of the method is described. The whole balancing method has two stages. The first stage identifies the unbalanced force vector and the second stage identifies the correlation masses and phases from the unbalanced force series. Deep-learning-based modules can be trained using one-run response data and labeled support force data only. Findings—both numerical and experimental balancing performances are reasonable and comparative. The performances indicate that the proposed method is validated and robust. Originality—the proposed method combines deep learning technology with rotor dynamics knowledge. The proposed method achieves good performance without a weight trail process and provides a competitive approach for rotor balancing technology.