This paper presents an interesting approach to continuous lifetime monitoring of wind turbine towers and main bearings using strain gauge calibration and virtual load sensors over nearly 10 years of data. The methodology is comprehensive and the long-term dataset is valuable. However, several clarifications and corrections are needed before publication, as outlined below.

• The paper is quite long as another reviewer has also noted. Consider shortening by consolidating or removing redundant material. Some sections could be more concise.
• The paper switches inconsistently between "nearly 10 years," "9 years," and "7.5 years" when describing different analyses. A clear timeline figure would help readers follow which period is being discussed in each section.
• The gearbox mounting stiffness of 20×10^6 N/m is from literature for different turbines. The sensitivity analysis shows this can cause 60% error in P_eq. How representative is this value for your specific drivetrain? How could this be validated in practice?
• The hyperparameter tuning used only 5 hours of data compared to 160 hours for training. Why do the authors use such a small amount? This seems inconsistent.
• You trained on 160 hours but validated on 160 hours from 2019. How was validation set selected? Same k-means or random?
• There's a fundamental clarity issue with the lifetime calculations in section 5.2.1 and Figure 7. The accumulated damage values shown in the figure don't align with the stated lifetimes of 166 years for the front main bearing and 333 years for the rear main bearing. The same issue appears for the tower bottom lifetime. Please verify and clarify these numbers.
• Figure 12b shows the correction factor varies across years for different inputs. Using average value could introduce bias depending on which year dominates. Did the authors check if this correlates with annual wind statistics?
• Figure 11 shows all virtual sensors systematically underpredict tower damage. The proposed correction method needs 6 months to 1 year of real measurements for calibration. This limits the benefit and you still need substantial measurements. What's minimum period needed for reliable correction?
• LSTM with strain inputs (no accelerometer) performed worse than simpler FNN. Needs more explanation than just "LSTM cannot attenuate 1P, 2P, 3P contributions." Why would temporal modeling perform worse with these periodic signals?
• Low turbulence increasing bearing loads at rated wind speed is interesting but not sufficiently explained. The authors cite HIPERWIND D5.4 but no intuition provided. Could this be because control system switches between modes frequently with turbulent inflow, leading to peak thrust? (see https://doi.org/10.1115/1.4041996)