The periodic maintenance, repair, and overhaul (MRO) of turbine blades in thermal power plants are essential to maintain a stable power supply. During MRO, older and less-efficient power plants are put into operation, which results in wastage of additional fuels. Such a situation forces thermal power plants to work under off-design conditions. Moreover, such an operation accelerates blade deterioration, which may lead to sudden failure. Therefore, a method for avoiding unexpected failures needs to be developed. To detect the signs of machinery failures, the analysis of time-series data is required. However, data for various blade conditions must be collected from actual operating steam turbines. Further, obtaining abnormal or failure data is difficult. Thus, this paper proposes a classification approach to analyze big time-series data alternatively collected from numerical results. The time-series data from various normal and abnormal cases of actual intermediate-pressure steam-turbine operation were obtained through numerical simulation. Thereafter, useful features were extracted and classified using K-means clustering to judge whether the turbine is operating normally or abnormally. The experimental results indicate that the status of the blade can be appropriately classified. By checking data from real turbine blades using our classification results, the status of these blades can be estimated. Thus, this approach can help decide on the appropriate timing for MRO.