A Novel method for online prognostics performance evaluation

Prognostics performance assessment technologies are necessary for building and qualifying a real and mature prognostics and health management (PHM) system. The further refinements in concepts and definitions for prognostics performance are made in this paper. The limitations for the application of offline prognostics performance evaluation are discussed. Two novel metrics: Relative Accuracy (RA) and Relative Precision (RP) are proposed for online prognostics performance evaluation according to the random uncertainty existing in both the prediction values and the actual values. A novel method for impartially evaluating the prognostics performance is presented. It synthesizes the evaluations of the predictions at different times before and close to the End-of-Useful-Predictions time tEOUP to achieve the evaluation of the minimum effective prediction horizon (MEPH) prognostics performance. A BP neural network prognostics method for hydraulic pump is evaluated for the demonstration intents. The results show the online evaluation method can work well without the failure data of the product.