The design of a gas turbine compressor vane carrier (CVC) should meet mechanical integrity requirements on, among others, low-cycle fatigue (LCF). The number of cycles to the LCF failure is the result of cyclic mechanical and thermal strain effects caused by operating conditions on the components. The conventional LCF assessment is usually based on the assumption on standard operating cycles — supplemented by the consideration of predefined extreme operations and safety factors to compensate a potential underestimate on the LCF damage caused by multiple reasons such as non-standard operating cycles. However, real operating cycles can vary significantly from those standard ones considered in the conventional methods. The conventional prediction of LCF life can be very different from real cases, due to the included safety margins.
This work presents a probabilistic method to estimate the distributions of the LCF life under varying operating conditions using operational fleet data. Finite element analysis (FEA) results indicate that the first ramp-up loading in each cycle and the turning time before hot-restart cycles are two predominant contributors to the LCF damage. A surrogate model of LCF damage has been built with regard to these two features to reduce the computational cost of FEA. Miner’s rule is applied to calculate the accumulated LCF damage on the component and then obtain the LCF life. The proposed LCF assessment approach has two special points. First, a new data processing technique inspired by the cumulative sum (CUSUM) control chart is proposed to identify the first ramp-up period of each cycle from noised operational data. Second, the probability mass function of the LCF life for a CVC is estimated using the sequential convolution of the single-cycle damage distribution obtained from operational data. The result from the proposed method shows that the mean value of the LCF life at a critical location of the CVC is significantly larger than the calculated result from the deterministic assessment, and the LCF lives for different gas turbines of the same class are also very different. Finally, to avoid high computational cost of sequential convolution, a quick approximation approach for the probability mass function of the LCF life is given. With the capability of dealing with varying operating conditions and noises in the operational data, the enhanced LCF assessment approach proposed in this work provides a probabilistic reference both for reliability analysis in CVC design, and for predictive maintenance in after-sales service.