The trend toward large-scale development of domestic power generation equipment is evident, and un-planned downtime caused by the failure of its supporting rotating mechanical equipment will cause serious economic losses and major safety issues. Rotor imbalance runs through the entire life cycle of rotating mechanical equipment, and diagnosing the condition of an in-service rotor is particularly important. A rotor imbalance fault diagnosis model based on multi-source domain data extraction and machine learning algorithms is proposed to address the problems of large rotating machinery with multiple vibration measurement points and non-stationary vibration signals. Based on multi-source vibration monitoring data, a vibration signal with rich fault information is first extracted based on cross-correlation coefficients, and a high-dimensional mixed feature space is constructed by fusing multi-domain features such as time domain, frequency domain and time-frequency domain. Secondly, a random neighbourhood embedding method based on t-distribution is used to reveal the feature information of the high-dimensional space, which is reflected as a visualised three-dimensional space. Finally, the nearest node algorithm is used for fault clas-sification to determine the unbalanced mass and phase of the rotor. This proposed model uses the cross-correlation coefficients to characterize the richness of fault information in multi-source data, and the combination of machine learning methods to determine the type of rotor unbalance. The effectiveness of the model was verified by designing unbalance state experiments on rotors with different additional masses at multiple speeds, solving the problems of online diagnosis and on-site dynamic balancing of rotors.
维普
万方数据