Prognostics by classifying degradation stage on Lambda architecture

To enhance the reliability and availability of an asset in its life, predicting the remaining useful life of an asset is strongly encouraged by assessing the extent of deviation or degradation of the asset's monitored parameters from its expected normal operating conditions. Although intelligent fault prognostic techniques such as machine learning and artificial neural networks have been applied in modern industries, application in actual industrial conditions requires that the forecasting process is revealed and more descriptive. To investigate the issue and increase the accuracy, this paper proposes an additional technique that can be further applied to any recent intelligent prognostic methods. The proposed method consists of two steps. First, the entire training set is divided into several degradation stages before regression using a heuristic approach and then the regression results are synthesized for each stage. The proposed method will increase the monotonicity of the predictive parameters, thus helping improve the predictive model's accuracy. To demonstrate the hypothesis, real condition monitoring data of high-pressure LNG pump and acceleration experimental data of a rotating machine is used for an experiment. Moreover, a system in which the proposed method can be appropriately executed is introduced with Lambda architecture. Finally, by demonstrating that the proposed method is capable of parallel computing, it is proven suitable for use in the proposed large-scale distributed processing system.