Aimed at the difficulty of diagnosing the transmission system of wind turbine under variable working conditions, a novel health condition monitoring method based on common features distribution adaptation is proposed in this article. In the method, envelope analysis is first performed on the collected signals, and then the time-frequency features are extracted to be combined as new input samples. The feature set under the working condition similar to target working condition is selected as the auxiliary sample set in source domain through the evaluation of the transferability. The kernel function is used to map the labeled auxiliary samples and unlabeled target samples to a reproduced kernel Hilbert space, which effectively reduces the data distribution discrepancy between source and target domains. The problem of class imbalance in each domain is taken into account when performing fault recognition, which improves the effect of transfer learning. Finally, the adjusted source domain is used to train the classifier, which is applied to the target domain to get the predicted labels of the test data. Experiment shows that the proposed method has better working performance than traditional fault diagnosis methods.