Matej Gašperin

Bearing fault prognostics based on signal complexity and Gaussian process models

Standard bearing fault detection features are shown to be ineffective for estimating bearings remaining useful life (RUL). Addressing this issue, in this paper we propose an approach for bearing fault prognostics based on features describing the statistical complexity of the envelope of the generated vibrations and a set of Gaussian process (GP) models. The proposed feature set exhibits continuous trend which can be directly related to the deterioration of bearing condition. Gaussian process models are non-parametric black-box models which differ from most other frequently used black-box identification approaches as they search for the relationships among measured data rather than trying to approximate the modeled system by fitting the parameters of the selected basis functions. Their output is normal distribution, expressed in terms of mean and variance, which can be interpreted as a confidence in prediction. In this paper the GP models are used for filtering noisy features and estimating the RUL based on filtered features. The proposed approach was evaluated on the data set provided for the IEEE PHM 2012 Prognostic Challenge.

Application of Unscented Transformation in Nonlinear System Identification

Abstract This article addresses the problem of system identification of nonlinear dynamical state-space models from input and output data. The problem is tackled by using Expectation Maximization (EM) algorithm for calculating the Maximum Likelihood (ML) estimates of the model parameters. The novelty of the presented algorithm is related to an efficient employment of the Unscented Transformation (UT) in the expectation step, which lowers the number of computations required. This property enables the algorithm to cope with high-dimensional models without a significant increase in computational load. The overall performance of the algorithm is demonstrated using both numerical and real data examples.

Prediction of the remaining useful life: An integrated framework for model estimation and failure prognostics

Machine failure prognostic is concerned with the generation of long term predictions and the estimation of the probability density function of the remaining useful life. Nowadays, a commonly used approach for this task is to make the prediction using a dynamical state-space model of the fault evolution. However, the main limitation of this approach is that it requires the values of the model parameters to be known. This work aims to alleviate the need for extensive prior efforts related to finding the exact model. For this we propose a framework for data-driven prediction of RUL with on-line model estimation. This is achieved by combining the state estimation algorithm with Maximum-Likelihood parameter estimation in the form of the Expectation-Maximization algorithm. We show that the proposed algorithm can be used with different classes of both black-box and grey-box models. First, a detailed solution for linear black-box models with the Kalman filter is presented followed by the extension to nonlinear models using either the Unscented Kalman filter or the particle filter. The performance of the algorithms is demonstrated using the experimental data from a single stage gearbox.

Adaptive Importance Sampling for Bayesian Inference in Gaussian Process models

Abstract Gaussian process (GP) models are nowadays considered among the standard tools in modern control system engineering. They are routinely used for model-based control, time- series prediction, modelling and estimation in engineering applications. While the underlying theory is completely in line with the principles of Bayesian inference, in practice this property is lost due to approximation steps in the GP inference. In this paper we propose a novel inference algorithm for GP models, which relies on adaptive importance sampling strategy to numerically evaluate the intractable marginalization over the hyperparameters. This is required in the case of broad-peaked or multi-modal posterior distribution of the hyperparameters where the point approximations turn out to be insufficient. The benefits of the algorithm are that is retains the Bayesian nature of the inference, has sufficient convergence properties, relatively low computational load and does not require heavy prior knowledge due to its adaptive nature. All the key advantages are demonstrated in practice using numerical examples.

An assessment of water conditions in a PEM fuel cell stack using Electrochemical Impedance Spectroscopy

The success of the fuel-cells technology penetration into commercial applications depends on the fuel-cells-based systems (FCS) durability, reliability and cost competitiveness. One way to improve the FCS durability is usage of advanced diagnostic methods in the context of holistic control. Electrochemical Impedance Spectroscopy (EIS) is for this purpose one of the more promising diagnostic methods because of its non-invasivity and ability to distinguish between the flooding and drying out of the Proton Exchange Membrane (PEM), a key part of each fuel-cell in a stack. Namely, a proper wetness of the stack membranes has an essential impact on the overall fuel cells system performance. This paper deals with the employment of EIS to commercially available PEM FCS. It presents the measuring system, describes EIS methodology, measurements, data processing, calculation procedures and final results.

Condition prognosis of mechanical drives based on nonlinear dynamical models

Forecasting the condition of the equipment is becoming an important ingredient of the advanced maintenance and asset management systems. In this paper a probabilistic approach to the prognosis of damage progression in gearboxes is presented. It is based on a stochastic nonlinear grey-box model of the underlying wear phenomena. Model parameters are estimated from the available vibration records by using an iterative Maximum Likelihood procedure. The procedure relies on the Expectation-Maximization algorithm and the Unscented Kalman filter for estimation of hidden system states. The algorithm has been used to predict the normal operating horizon of a single-stage gearbox system. Several test runs of the system have been preformed to validate the algorithm.

Signal Complexity and Gaussian Process Models Approach for Bearing Remaining Useful Life Estimation

Standard bearing fault detection features are shown to be ineffective for estimating bearings remaining useful life (RUL). In this paper we propose a new approach estimating bearing RUL based on features describing the statistical complexity of the envelope of the generated vibrations and a set of Gaussian process (GP) models. The proposed approach is shown to be sensitive to incipient condition deterioration which allows timely and sufficiently accurate estimates of the RUL. The proposed approach was evaluated on the data set comprising of 17 bearing runs with natural fault evolution.

Decision support system for polymerization production plant using pPIs

The synthesis of plant-wide control structures is recognized as one of the most important production- management design problems in the process industries. This article proposes a decision support system, utilizing key performance indicators (pPIs) as a possible solution to simplify this problem. pPIs represent the translation of operating objectives, such as the minimization of production costs, to a set of measurable variables that can then be used to help production manager to select the operating regime of the production line. The idea of decision support system at the production- management level using pPIs as referenced controlled variables was implemented on the procedural model of a production process for a polymerization plant. Preliminary results show the usefulness of the proposed methodology.