Prognostic Algorithm for Prediction of Remaining Useful Life of Metallic Structures

Abstract

Prediction of remaining useful life of metallic structures is very important for maintenance managers to plan the repair/replacement activity. Early detection of failure will help in avoiding harsh circumstances. Accurate prediction is possible through usage of the effective prognostic algorithms, based on historical trends and underlying suitable statistical degradation models. In the proposed research work, the prognostic algorithms are developed for two different case studies taken from two different industries. The first case study is from aerospace industry in which the prognostic algorithms are applied for degradation prediction of aircraft wing structure. The second case study is from electric power distribution industry in which prognostic algorithms are applied on Aerial Bundled Cables (ABCs) for future health state prediction. The developed prognostics algorithms are based on data-driven approach and the model-based approach. In data-driven approach, Particle Filter (PF) based prognostic algorithms are used to predict the future health state. In PF framework, the selection of suitable probability density function as appropriate state transition model for future health prediction plays a vital role in prediction accuracy. Therefore, two approaches for PF implementation are proposed in this research work. In the first approach, a classical PF in which an exponential density as state transition density is used. The main characteristic of an exponential density in the PF framework is that it allows constant degradation growth rate. The rate of degradation is varying with respect to age. Different prediction densities represent varying rates of degradation during the lifecycle. The use of a single prediction density at every prediction stage throughout the life cycle is not considered suitable. Therefore, dynamic prediction vi density based PF framework is proposed in the second approach. The suitable candidate for dynamic prediction density is Weibull density as it is equivalent to different densities by varying the shape parameter. The shape parameter is estimated by applying the Maximum Likelihood Estimation (MLE) technique on the available database of measurements at each life stage. The proposed PF frameworks (Classical and Dynamic) are applied on actual aerospace historical data of an in-service Airbus A310 aircraft. The degradation growth around the countersunk (CSK) rivet-hole are predicted w.r.t. length and depth at different angles. Hence, a 3-Dimensional flaw is predicted using the proposed approach. Further, the developed frameworks are also applied for future health state prediction of ABCs. ABC is a combination of insulated phase conductors bundled tightly together. Insulation degradation is a very common problem with ABCs especially in coastal environments. Actual Non-Destructive Testing (NDT) data from the installed in-service cable is acquired at different time instants to study the phenomena of insulation degradation w.r.t. time. In the reported research work, a novel thermal degradation parameter (TDP) is extracted from the acquired data to characterize the health state of the cable. The proposed prognosis techniques are also compared to showcase the efficacy for prognostics of aero structures and ABCs. Finally, a reliability model is also proposed in this research work which is derived from historical failure data of particular type of ABCs coupled with the life acceleration models. The models are based upon the actual environmental conditions experienced by the cables under study. vii The actual loading data as well as environmental data of two sites of varying distance from Seashore are used to develop the respective reliability models. The reliability prediction from proposed reliability model is then validated using time to failure computation through comparison of historical infrared thermography based Non-destructive testing (NDT) data available in literature. The validation indicates the efficacy of the proposed reliability model.