A Method for Forecasting the Condition of HPT NGVs by Using Bayesian Belief Networks and a Statistical Approach

Abstract

Jet engine maintenance is a very competitive field in terms of time and costs. To increase planning security and reduce turnaround time (TAT) of the maintenance process it is important to get as much engine data as possible before disassembly. Aero engines are especially subjected to environmental and operational influences. For the high pressure turbine (HPT), the following parameters have been identified to describe the deterioration of its nozzle guide vane (NGV): On-wing cycles, NGV material, airport region, engine wing position, thrust rating, vane repair history and customer business segment. The combined influences of the parameters are non-trivial and it is not possible to acquire them analytically. There are no known mathematical laws connecting the above-mentioned parameters. The linear regression method set limits for processing data in an adequate manner. This is confirmed by the analysis of the arithmetic means and standard deviations. Especially the standard deviation values fit in a broad spectrum due to various reasons. Thus, it is not feasible to make an appropriate forecast with a simple statistical method due to the multidimensional character of the parameters influencing the accuracy. For this reason, advanced methods need to be developed to derive a feasible forecast method. By applying a statistical hypothesis test, a bayesian belief network (BBN) has been designed. It allows the use of imprecise data without suffering a significant loss in forecast accuracy and additionally, the implementation of expert knowledge. The objective of this study is to develop an effective BBN in order to adequately predict the next repair of the first stage HPT NGV of the General Electric CF6-80C2 engine. The reason for selecting the NGV is due to its high susceptibility to different influences, combined with the significant costs and TAT during the maintenance process. Having poor forecasting quality by using a simple statistical method, the evaluation of the BBN provides very satisfactory accuracy of above 80 percent which is equivalent to 19 out of 23 vane segments. Furthermore, the developed BBN emphasises robustness when detecting the expected tendencies while having only a limited amount of input parameters. Further work includes application of this method on other engine components as well as establishing the business value of the developed method. In conclusion, BBN have tremendous potential for forecasting the repair of the entire jet engine.