The steam turbine rotor is still the main power generation equipment. Affected by the impact of new energy on the power grid, the steam turbine needs to participate in peak load regulation, which will make turbine rotor components more prone to failure. The rotor is an important equipment of a steam turbine. Unbalance and misalignment are the normal state of rotor failure. In recent years, more and more attention has been paid to the fault detection method based on deep learning, which takes rotating machinery as the object. However, there is a lack of research on actual steam turbine rotors. In this paper, a method of rotor unbalance and parallel misalignment fault detection based on residual network is proposed, which realizes the end-to-end fault detection of rotor. Meanwhile, the method is evaluated with numerical simulation data, and the multi task detection of rotor unbalance, parallel misalignment, unbalanced parallel misalignment coupling faults (coupling fault called in this paper) is realized. The influence of signal-to-noise ratio and the number of training samples on the detection performance of neural network is discussed. The detection accuracy of unbalanced position is 93.5%, that of parallel misalignment is 99.1%. The detection accuracy for unbalance and parallel misalignment is 89.1% and 99.1%, respectively. The method can realize the direct mapping between the unbalanced, parallel misalignment, coupling fault vibration signals and the fault detection results. The method has the ability to automatically extract fault features. It overcomes the shortcoming of traditional methods that rely on signal processing experience, and has the characteristics of high precision and strong robustness.

This content is only available via PDF.
You do not currently have access to this content.