Alexander Ypma

Blind separation of rotating machine sources: bilinear forms and convolutive mixtures

Abstract We propose the use of blind source separation (BSS) for separation of a machine signature from distorted measurements. Based on an analysis of the mixing processes relevant for machine source separation, we indicate that instantaneous mixing may hold in acoustic monitoring. We then present a bilinear forms-based approach to instantaneous source separation. For simulated acoustic mixing, we show that this method may give rise to a more robust separation. For vibrational monitoring, a convolutive mixture model is more appropriate. The demixing algorithm by Nguyen Thi–Jutten allows for separation of the contributions of two coupled machines, both in an experimental setup and in a real-world situation. We conclude that BSS is a feasible approach for blind separation of distorted rotating machine sources.

Support objects for domain approximation

We propose a novel algorithm for extracting samples from a data set supporting the extremal points in the set. Since the density of the data set is not taken into account, the method could enable adaptation to novel (e.g. machine wear) data. Knowledge about the clustering structure of the data can aid in determination of the complexity of the solution. The algorithm is evaluated on its computational feasibility and performance with progressively more dissimilar data.

Robust machine fault detection with independent component analysis and support vector data description

We propose an approach to fault detection in rotating mechanical machines: fusion of multichannel measurements of machine vibration using independent component analysis (ICA), followed by a description of the admissible domain (part of the feature space indicative of normal machine operation) with a support vector domain description (SVDD) method. The SVDD method enables the determination of an arbitrary shaped region that comprises a target class of a dataset. In this particular application, it provides a way to quantify the compactness of the admissible class in relation to data preprocessing. Application to monitoring of a submersible pump indicates that combination of measurement channels with ICA gives improved results in fault detection, without requiring detailed prior knowledge on origin and type of the failure.

A method for detection of Alzheimer's disease using ICA-enhanced EEG measurements

OBJECTIVE Many researchers have studied automatic EEG classification and recently a lot of work has been done on artefact-removal from EEG data using independent component analyses (ICA). However, demonstrating that a ICA-processed multichannel EEG measurement becomes more interpretable compared to the raw data (as is usually done in work on ICA-processing of EEG data) does not yet prove that detection of (incipient) anomalies is also better possible after ICA-processing. The objective of this study is to show that ICA-preprocessing is useful when constructing a detection system for Alzheimers disease. METHODS AND MATERIAL The paper describes a method for detection of EEG patterns indicative of Alzheimers disease using automatic pattern recognition techniques. Our method incorporates an artefact removal stage based on ICA prior to automatic classification. The method is evaluated on measurements of a length of 8s from two groups of patients, where one group is in an initial stage of the disease (28 patients), whereas the other group is in a more progressed stage (15 patients). Both setups include a control group that should be classified as normal (10 and 21, respectively). RESULTS Our final classification results for the group with severe Alzheimers disease are comparable to the best results from literature. We show that ICA-based reduction of artefacts improves classification results for patients in an initial stage. CONCLUSION We conclude that a more robust detection of Alzheimers disease related EEG patterns may be obtained by employing ICA as ICA based pre-processing of EEG data can improve classification results for patients in an initial stage of Alzheimers disease.

Pump Failure Detection Using Support Vector Data Descriptions

For good classification preprocessing is a key step. Good pre-processing reduces the noise in the data and retains most information needed for classification. Poor preprocessing on the other hand can make classification almost impossible. In this paper we evaluate several feature extraction methods in a special type of outlier detection problem, machine fault detection. We will consider measurements on water pumps under both normal and abnormal conditions. We use a novel data description method, called the Support Vector Data Description, to get an indication of the complexity of the normal class in this data set and how well it is expected to be distinguishable from the abnormal data.

Improved unscented kalman smoothing for stock volatility estimation

We introduce a novel approximate inference algorithm for nonlinear dynamical systems. The algorithm is based upon expectation propagation and Gaussian quadrature. The first forward pass is strongly related to the unscented Kalman filter. It improves upon unscented Kalman filtering by only making Gaussian approximations in the latent and not in the observation space. Smoothed estimates can be found without inverting latent space dynamics and can be improved by iteration. Multiple forward and backward passes make it possible to improve local approximations and make them as consistent as possible. We demonstrate the validity of the approach with an interesting inference problem in stochastic stock volatility models. The traditional unscented Kalman filter is ill suited for this problem: it can be proven that the traditional filter effectively never updates prior beliefs. The novel algorithm gives good results and improves with iteration

Novel approximations for inference in nonlinear dynamical systems using expectation propagation

We formulate the problem of inference in nonlinear dynamical systems in the framework of expectation propagation, and propose two novel algorithms. The first algorithm is based on Laplace approximation and allows for iterated forward and backward passes. The second is based on repeated application of the unscented transform. It leads to an unscented Kalman smoother for which the dynamics need not be inverted explicitly. In experiments with a one-dimensional nonlinear dynamical system we show that for relatively low observation noise levels, the Laplace algorithm allows for the best estimates of the state means. The unscented algorithm however is more robust to high observation noise and always outperforms the conventional inference methods against which it was benchmarked.

Deterministic and Stochastic Gaussian Particle Smoothing

In this article we study inference problems in non-linear dynamical systems. In particular we are concerned with assumed density approaches to filtering and smoothing. In models with uncorrelated (but dependent) state and observation, the extended Kalman filter and the unscented Kalman filter break down. We show that the Gaussian particle filter and the one-step unscented Kalman filter make less assumptions and potentially form useful filters for this class of models. We construct a symmetric smoothing pass for both filters that does not require the dynamics to be invertible. We investigate the characteristics of the methods in an interesting problem from mathematical finance. Among others we find that smoothing helps, in particular for the deterministic one-step unscented Kalman filter.

Health Monitoring with Learning Methods

In this paper we provide a framework for the design of a practical monitoring method with learning methods. We demonstrate that three medical and industrial monitoring problems involve subproblems that can be tackled with our approach. Application of interference removal, novelty detection and learning of a signature leads to a feasible monitoring method in these cases.

The role of subclasses in machine diagnostics

In machine diagnostics it is difficult to collect for learning all possible operating modes of machine functioning. Some of the operating modes are often missing. In these circumstances, it is important to know which modes (subclasses) are the most valuable for successful machine diagnosis. It is also of interest to investigate the usefulness of noise injection to cover the missing operating modes in the data. In this paper, we study the importance of selecting different operating modes of a water-pump and using them for learning in both 2-class and 4-class problems. We show that the operating modes representing different running speeds are more valuable than those representing machine loads. We also demonstrate that the 2-nearest neighbours directed noise injection is useful when filing in missing operating modes in the data.