In this paper, the authors describe a methodology to detect malfunctioning or inoperable wind turbine drivetrain sensor readings and then correct for their effect on other diagnostic indicators. Two examples of “masking” of these faulted sensors are given, and the accuracy of the historical fault diagnoses are recalculated. The article addresses an important real-life problem and the first few Sections are well written. However, not being well-versed in autoencoders I had difficulty interpreting the Figures shown in the Results section and related text, which is also relatively brief. My biggest desire in revision would be to expand the Results section in this respect. I also offer additional comments in the attachment.

The paper proposes a fault diagnosis and prediction method based on SCADA data from wind turbines. From the results, it is evident that training models with a sufficient amount of data can enable the prediction of certain faults, demonstrating practical significance. The authors are advised to consider the following issues:

1. The proposed model’s performance is analyzed in terms of reconstruction errors under masked sensor errors and nominal conditions, as well as its ability to distinguish between healthy and unhealthy data. To better demonstrate the superiority of the model, it is recommended to include comparisons with other models, particularly in the prediction aspect.
2. The authors’ work primarily relies on SCADA data for analysis, and the fault information is mainly based on faults that can be reported in wind farms. Therefore, the authors should further emphasize the practical significance of the proposed method in key sections.