From the summary of previous literatures, it can be seen that the research on the *WDE* life of turbine blades is quite few. At the same time, previous researchers mostly used the incubation period (*Ip*), which is also called the initial crack life, as the useful life of blades. However, in practical engineering applications, in order to save the cost, the blade is allowed to be used even if there is some *WDE* defect on the blade surface, without affecting the equipment safety seriously. In this case, the initial crack life of the material is no longer of engineering reference value, and the more important parameter is the remaining useful life (*RUL*) of the blade.

In this paper, a method for predicting *RUL* of blade under water droplet erosion (*WDE*) based on image recognition and machine learning is introduced. The general procedure of machinery health prognostics, which consists of four technical process, including data acquisition, health indicator (*HI*) construction, health stage (*HS*) division and *RUL* prediction, is adopted in the present study. Firstly, by using the experimental rig for testing the *WDE* characteristics of materials, the erosion morphology images and mass loss data at several moments in the whole *WDE* process are obtained. Then, according to the data processing method of *ASTM-G73* and the cumulative erosion-time curves, the *WDE* life coefficient (*ζ*) is defined and the three *WDE* periods is quantitatively divided, which correspond to *HI* and *HS* respectively. The life coefficient (*ζ*) could be used to calculate the *RUL* of turbine blades. The dataset consists of 600 sets of erosion images and corresponding life coefficients. After dividing the dataset with the proportion of 7:3 for training and prediction, four machine learning regression models, including KNeighbors Regression (*KNR*), Support Vector Regression (*SVR*), AdaBoost Regression (*ABR*) and *ResNet*, are used to train and predict the dataset. It is found that the method proposed in this paper can achieve the highest prediction accuracy of 0.91.

The advantage of this study is that, the end-to-end relationship between *WDE* images and *RUL* is established. The *RUL* of blades can be calculated directly by surface observation, avoiding a large number of theoretical and numerical calculation. At the same time, the application scope of this method is broader than other research, covering three *WDE* periods: incubation period (*Ip*), deceleration period (*Dp*) and terminal period (*Tp*). The *WDE* situation of blades can be determined accurately, which can provide reference for determining the overhaul period of equipment and the replacement time of blades.


Water Droplet Erosion Life Prediction Method for Steam Turbine Blade Materials Based on Image Recognition and Machine Learning

Division / Sponsor:
International Gas Turbine Institute (IGTI)

Paper Proceedings: https://asmedigitalcollection.asme.org/GT/GT2020/volume/84058

Turbo Expo - Virtual Conference

Supporting the need among researchers worldwide for additional online and multimedia content, ASME now offers the ASME Conference Video Collection. This platform will collect and provide long-term access to conference presentation videos, encompassing all technology areas within ASME's technical conference and events program. About 30,000-40,000 engineers across all mechanical engineering and related disciplines participate in ASME conferences each year. This includes authors of approximately 6,000 to 8,000 technical papers published by ASME, the majority of whom also submit video presentations for our virtual conferences. The video collection will be available as an add-on to existing product packages or as a separate subscription.