Mirko Knaak

Geometrically Constrained Independent Component Analysis

Acoustical signals are often corrupted by other speeches, sources, and background noise. This makes it necessary to use some form of preprocessing so that signal processing systems such as a speech recognizer or machine diagnosis can be effectively employed. In this contribution, we introduce and evaluate a new algorithm that uses independent component analysis (ICA) with a geometrical constraint [constrained ICA (CICA)]. It is based on the fundamental similarity between an adaptive beamformer and blind source separation with ICA, and does not suffer the permutation problem of ICA-algorithms. Unlike conventional ICA algorithms, CICA needs prior knowledge about the rough direction of the target signal. However, it is more robust against an erroneous estimation of the target direction than adaptive beamformers: CICA converges to the right solution as long as its look direction is closer to the target signal than to the jammer signal. A high degree of robustness is very important since the geometrical prior of an adaptive beamformer is always roughly estimated in a reverberant environment, even when the look direction is precise. The effectiveness and robustness of the new algorithms is proven theoretically, and shown experimentally for three sources and three microphones with several sets of real-world data

A new image segmentation method based on human brightness perception and foveal adaptation

We propose an algorithm for adaptive image segmentation based on human psychovisual phenomena: visual perception-based segmentation. The new method can reliably segment poor quality images with low contrast and low SNRs. Due to its adaptability, it can be applied to a wide range of low quality images with different object sizes. In successful tests with ultrasound and flow field images that are normally difficult to segment, this new method outperforms a conventional texture-based segmentation method as a result of its biological source.

Geometrically constraint ICA for convolutive mixtures of sound

The goal of this contribution is a new algorithm using independent component analysis with a geometrical constraint. The new algorithm solves the permutation problem of blind source separation of acoustic mixtures, and it is significantly less sensitive to the precision of the geometrical constraint than an adaptive beamformer. A high degree of robustness is very important since the steering vector is always roughly estimated in the reverberant environment, even when the look direction is precise. The new algorithm is based on FastICA and constrained optimization. It is theoretically and experimentally analyzed with respect to the roughness of the steering vector estimation by using impulse responses of real room. The effectiveness of the algorithms for real-world mixtures is also shown in the case of three sources and three microphones.

Blind source separation for acoustical machine diagnosis

Acoustical machine diagnosis is frequently made difficult by noisy environments at a production site. This paper evaluates whether blind source separation (BSS) algorithms can be used to enhance machine signals as they enhance speech signals. Unfortunately, the SNR is not significant, since as an energy based number it ignores distortions of the machine signal. In comparison to speech processing where a small distortion does not reduce the intelligibility, it reduces the classification rate in machine diagnosis significantly. Therefore, an assessment for BSS algorithms with respect to machine diagnosis is proposed and used to verify the applicability of a new BSS algorithm.

Fast measurement by integrating model-based methods and advanced test-cell automation

Modern calibration engine control unit requires a huge number of measurements to fill their maps. Steady state measurements at the engine test bed are still very important to find the optimal settings of all control parameters for an engine operating point. Model-based methods - like design of experiments - are more and more used to reduce the needed numbers of experiments for an engine calibration. Nevertheless, since the number of controllable engine parameters arises, new methods for accelerating each measurement are needed. This paper presents methods to shorten the measurement time by using transitions between measurement points and avoidance of stabilization

Determining adjustment ranges for model-based approaches using support vector machines

Model-based methods, such as design of experiments (DoE), have become more and more established in recent years in optimizing control maps in engine ECUs from the aspects of ride comfort, fuel economy and emissions. As a result of the rising number of control parameters and ever shorter development times, the aim in this context is to develop automated intelligent setting strategies for test design. In doing so, it is imperative to find a range of settings at which the engine works safely (adjustment range). This paper presents a method for determining adjustment range limits in engine measurement for high dimensional parameter spaces using the support vector machines (SVM). SVMs are a relatively new method in machine learning and are applied to learn a hull that models the unknown, actual test space on the basis of measurement points

Mehrkanalige Signalschätzung für die technische Diagnose (Multi-channel Signal Estimation for Machine Diagnosis)

Abstract Die akustische Maschinendiagnose hat die Aufgabe, aus dem möglicherweise verrauschten und durch andere Signale gestörten Schallsignal einer Maschine auf deren Fehlerzustand zu schließen. Da dieses Mustererkennungsproblem bereits bei ungestörten Signalen schwierig ist, wird vorgeschlagen, die Störungen mit einer mehrkanaligen Signalschätzung zu eliminieren. Der vorgestellte Algorithmus beruht auf den Methoden der ICA (Independent Component Analysis), integriert vorhandene, jedoch ungenaue geometrische Informationen und erzielt somit sehr gute Schätzraten bei hoher Robustheit. Er ist nicht wie die derzeitigen Algorithmen der akustischen blinden Quellentrennung auf zwei Signale begrenzt und erzielt somit sehr gute Schätzraten bei höherer Robustheit als adaptive Beamformer.

Acoustic Source Number Estimation and Its Application to Machine Monitoring

Abstract The goal of this contribution ia a sound based fault detection in a noiay production hall. For this purpose, a new algorithm based on source number estimation allows to classify the sound in periods when the machine sound ia dominant. This avoids costly insulation measures which are troublesome in the production process. Based on the knowledge that the eigenvalue number of the spatial correlation matrix reflects the number of sources, a simple Mahalanobis classifier ia sufficient to find quiet periods. Employing the data from quiet periods, a feature based classifier ia used for reliable fault detection. The method was verified with real data of washing machines obtained from a production site which have been classified with a considerably good rate. The simplicity of the preprocessing makes an online classification poesible.

Semi-blinde Geräuschrekonstruktion für die technische Diagnose (Semi-blind Sound Reconstruction for Machine Diagnosis)

Wenn der sehr aussagekräftige Luftschall zur objektiven akustischen Güteprüfung benutzt wird, stellt sich in einer Fabrik o.Ä. das Problem starker Nebengeräusche. Dieser Beitrag schlägt ein semi-blindes Verfahren zur Rekonstruktion des Maschinengeräuschs vor. Die blinde Quellentrennung ist eine statistische Methode, die mit geringen Annahmen eine Rekonstruktion von Signalen aus deren Mischung ermöglicht. Zur Lösung ihrer systematischen Probleme für akustische Mischungen werden hier spezifische Eigenschaften rotierender Maschinen ausgenutzt. In diesem Beitrag wird der Algorithmus vorgestellt und mit Labordaten und Originaldaten einer realen Motorenproduktion getestet.