Developing an Implicit Creep Model From Open Literature Data

Abstract

As pressure ratios and firing temperatures continue to rise, creep becomes of greater concern everywhere within a gas turbine engine. As a rule of thumb, just a 14°C increase in metal temperature can halve the expected rupture life of a part. In the past, companies might be satisfied with conservative creep estimates based on Larson-Miller-Parameter curves and 1D calculations. Now companies need functional implicit-creep models with finite element analysis for an ever-increasing number of materials. Obtaining adequate test data to create a good creep prediction model is an expensive and time-consuming proposition. Test costs depend on temperature, material, and location, but a single, 10,000hr, rupture test may reasonably be expected to cost > $20,000. Other than large OEMs, small companies and individuals lack the resources to create creep models from their own data. This paper will lead the reader through the creation of a modified theta projection creep model of Haynes 282, a high-temperature, combustion alloy, using only literature data. First, literature data is collected and reviewed. Data consists of very few complete curves, estimated stresses for rupture and 1% strain, and discrete times to individual strains for individual tests. When adequate data exists, individual tests are fit to theta projection model curves. These “local” theta fits of different test conditions are used as input for the global model. Global fits of theta parameters, as a function of stress and temperature, are made from the full data set. As the global creep model is improved, correction factors introduced to account for true stress and strain effects. A statistical analysis is made of actual rupture time versus predicted onset of failure time, theta5=1. A time-based scatter factor is determined to evaluate temperature margin required to ensure reliability. After the creep model was completed, Haynes International, the material inventor, provided specific test conditions (stress and temperature) of 5 tests that had already been run. Creep predictions were generated for these test conditions, before viewing the actual results. The creep model predicted strain curves matched actual tests very well, both in shape and time to rupture. Continued refinement is possible as more data is acquired.