Prognostics in switching systems: Evidential markovian classification of real-time neuro-fuzzy predictions

Abstract

Condition-based maintenance is nowadays considered as a key-process in maintenance strategies and prognostics appears to be a very promising activity as it should permit to not engage inopportune spending. Various approaches have been developed and data-driven methods are increasingly applied. The training step of these methods generally requires huge datasets since a lot of methods rely on probability theory and/or on artificial neural networks. This step is thus time-consuming and generally made in batch mode which can be restrictive in practical application when few data are available. A method for prognostics is proposed to face up this problem of lack of information and missing prior knowledge. The approach is based on the integration of three complementary modules and aims at predicting the failure mode early while the system can switch between several functioning modes. The three modules are: 1) observation selection based on information theory and Choquet Integral, 2) prediction relying on an evolving real-time neuro-fuzzy system and 3) classification into one of the possible functioning modes using an evidential Markovian classifier based on Dempster-Shafer theory. Experiments concern the prediction of an engine health based on more than twenty observations.

Features Selection Procedure for Prognostics: An Approach Based on Predictability

“…From the dataset, among 26 available features, 8 were pre-selected in a previous work thanks to information theory and Choquet Integral (Ramasso and Gouriveau (2010)). These 8 features (F1 to F8) are used as a starting point to illustrate the proposed approach of "predictability-based" feature selection (section 3.1).…”

Connexionist-Systems-Based Long Term Prediction Approaches for Prognostics

“…From the dataset, among 26 available features, 8 were preselected in a previous work thanks to information theory, and Choquet Integral [43]. As for the number of prediction inputs, two regressors have been used in addition to a third input, which is the time index as suggested in [43].…”

“…As for the number of prediction inputs, two regressors have been used in addition to a third input, which is the time index as suggested in [43]. The learning data set was composed of 40 runs from previous degraded systems.…”

Prognostic by Classification of Predictions Combining Similarity-Based Estimation and Belief Functions

“…Moreover, both rely on the BBAs in the training dataset L . Thus, the fusion may be performed using the cautious rule [13]: (6) from which a decison concerning the state at time t + h can be made and the result is the estimation of a sequence of statesω t+h . Note that the neutral element is not always the vacuous BBA [13], except for separable BBAs as the ones used in evidential KNN exploited in CPS.…”

“…Continuous states generally represent the value of a signal (an observation or a feature) and their prediction can be made by Kalman-like procedures or by neural networks [4,5], Discrete states generally depict functioning modes reflecting the current degradation and its prediction can be performed by state machines such as Hidden Markov Models [6]. In both cases, data-driven prognostics generally involves a training procedure where statistical models of the degradation are built.…”