A system was developed for predicting the length of cracks generated in the first-stage nozzles of 1100 °C class gas turbines. The system consists of three programs, namely, crack input, crack display, and crack predicting programs, and a database containing data on 210,000 cracks generated in first-stage nozzles gathered from existing maintenance records maintained at three major power plants in Japan. The crack database also contains numerous crack data collected during inspections, operating hours and number of starts of gas turbine up to the time of inspection. The distinctive features of the system are 1) the crack data can be input as an image over the drawing of first-stage nozzles with a mouse, 2) by accumulating crack data, the sections in which most cracks occur in the first-stage nozzles can be clarified, 3) the correlation of crack length to operating time and number of starts can be analyzed simply, and 4) the length of cracks growing in any direction of any section of first-stage nozzles can be predicted.
Crack data collected from three power plants were analyzed. It was found that the cracks that were comparatively long and grew in proportion to an increase in operating time and number of starts were only 11 crack patterns as compared to the entire first-stage nozzles. The length of these 11 crack patterns was predicted for new first-stage nozzles and compared with the measured values obtained from inspection. As a result, it was verified that the crack length could be predicted.