Reducing Dimensionality of Multi-regime Data for Failure Prognostics

Abstract

Over the last decade, the prognostics and health management literature has introduced many conceptual frameworks for remaining useful life predictions. However, estimating the future behavior of critical machinery systems is a challenging task due to the uncertainties and complexity involved in the multi-dimensional condition monitoring data. Even though many studies have reported promising methods in data processing and dimensionality reduction, the prognostics applications require integration of these methods with remaining useful life estimations. This paper describes a multiple linear regression process that reduces the number of data regimes under consideration by obtaining a set of principal degradation variables. The process also extracts health indicators and useful features. Finally, a state-space model based on frequencydomain data is used to estimate remaining useful life. The presented approach is assessed with a case study on turbofan engine degradation simulation dataset, and the prediction performance is validated by error-based prognostic metrics.

Reconstructing secondary test database from PHM08 challenge data set

“…In general, the performance of a data-driven prognostic method depends on measured condition monitoring data [7], which allows an understanding of system degradation [8]. The features of the PHM08 data challenge include the following characteristics that make it both convenient and suitable for the development of prognostic algorithms on multistep-ahead remaining useful life calculations [2], [4].Each subset contains multiple multivariate time series describing sensor magnitudes over time and three different operational settings that indicate variations of regimes.The sensors are contaminated with noise that simulate the variability of sensor measurements during operations.Each trajectory has a specific initial wear level and manufacturing variation.…”

A neural network filtering approach for similarity-based remaining useful life estimation

“…The common conventional methods include regression models [4,5], Wiener processes [6], wavelets [7,8], hidden Markov model [9,10], Kalman filters [11,12], particle filters [13,14] and the Gaussian process [15]. A widespread range of conventional applications dealing with different domains can be found in the literature, such as fatigue degradation evolution in materials [16], ageing of batteries, [17], fault detection and isolation of mechatronic systems [18][19][20][21][22], and failure of electronic components [23].…”

Achieving a representation of asset data conducive to machine learning driven predictive maintenance

“…Raw asset sensor data has, to date, rarely been usable by data-driven methods, including ML algorithms, due to characteristics such as variations in operating conditions and difficulties choosing a representative sample from datasets containing large number of sensor measurements. [14,19,[31][32][33].…”

“…Poorly trained ML-based PdM models, from improper asset data representation, can leave some ML-based PdM tools vulnerable to high rates of inaccurate fault identification and/or time-of-failure prediction results [17,18,31]. This could potentially lead to value loss during an asset's life cycle.…”

“…1. Asset data is often high-dimensional due to large number of sensor channels perceiving different properties of an asset's life cycle, and ML algorithms generally perform poorly on high-dimensional data [14,31].…”