RESEARCH TOPICS

The Signal Processing group comprises several researchers focused on advanced signal processing methods, which are later used in several areas of our department activities. Our field of research are:

  • vibration signal processing
  • varying operational conditions
  • automatic signal validation
  • instantaneous frequency estimation
  • cyclostationarity
  • wavelet demodulation
  • Triple Corelation

We pay special attention to the practical implementation of our research. We cooperate with several companies, which develop market products based on our know-how. Our algorithms were used in aerospace, power generation, oil & gas and many other industries.

The research is coordinated by prof. Wiesław Staszewski and prof. Tomasz Barszcz.

Integration of modulating components over a selected carrier frequency band

The figure illustrates integration of modulating components over a selected carrier frequency band, which offers attractive alternative to classical envelope analysis.

 

 

Vibration Signal Processing

Rotating machinery is a vital part of modern economics. These machines are basis for many industries, like power generation (both conventional and renewable), chemical industry, oil & gas, transportation, aerospace, and many others. Our research is focused on methods and algorithms for early detection of failures, which – if unaddressed – might result in catastrophic failures, loss of production, and increased costs of maintenance. With our methods, users can detect and identify malfunctions such as:

  • imbalance
  • misalignment
  • looseness
  • bearing failures
  • gear failures
  • machine wear

Some of our methods were submitted for a patent, like e.g. ICPCP method. The method enables detection of failures of planetary gearboxes, especially used in wind turbines or mining machinery.

Instantaneous energy generated during meshing of ring gear with planet gear of a wind turbine.

The figure illustrates instantaneous energy generated during meshing of ring gear with planet gear of a wind turbine.

 

 

Varying Operational Conditions

The classical fault detection methods assumed stationarity of operating conditions and used the frequency analysis in order to detect a change in the signal. The real world is much more complex and – strictly speaking – no machine can be assumed to work in stationary condition. Even it is operates in ideally stationary conditions, the machine needs to start and stop, though it is a trivial case. Apart from this, there are numerous groups of machinery for which non-stationarity is inherent. To name only a few, there are wind turbines, mining machines, jet engines, car engines, etc.. The fundamental problem in analysis of the nonstationary signals is understanding how varying operational parameters influence the vibrations signal. Then, it is possible to detect a real failure despite much larger influence from e.g. varying load.

The figure illustrates how the vibration signal changes its characteristics form random stationary (up tp 3rd second) to cyclostationary over increasing speed, which calls for different analysis methods.

The figure illustrates how the vibration signal changes its characteristics form random stationary (up tp 3rd second) to cyclostationary over increasing speed, which calls for different analysis methods.

 

 

 

The similar approach is applied for identification process of non-stationary systems. Variable operating conditions make the use of conventional identification techniques difficult (or impossible) to apply. In that cases the solution can be application of suitable signal processing method. For non-stationary systems the concept of adaptive wavelet filtering was developed. The technique enables the separation of the individual frequency components contained in the signal response.

Vibration signal of the turbo generator foundation during its rundown and the time-frequency representation of the signal.

The figure illustrates the vibration signal of the turbo generator foundation during its rundown and the time-frequency representation of the signal.

 

The concept of adaptive wavelet filtering applied to damping identification process

The figure presents the concept of adaptive wavelet filtering applied to damping identification process.

 

Automatic Signal Validation
Latest development of distributed condition monitoring for heavy duty machinery has revealed a major challenge concerning reliable data acquisition. Due to frequent failures of diagnostic algorithms caused by incorrectly registered data, validation of vibration signals became a very attractive field of research. Classical data acquisition procedures might be enhanced by implementation of signal validation methods based on both machine process parameters signals and machine vibration signals. Validation covers analysis of process parameters, data stability constrains and also independent and comparative validation of vibration signals. From diagnostic analysis point of view, classification of useless vibration signals as ‘‘invalid’’ is sufficient, without a preclassification between ‘‘correct’’ and ‘‘incorrect’’ signals. However, the latter differentiation carries important information from a system maintenance point-of-view, since it might for instance inform about vibration sensor failure. The figure shows exemplary vibration signals classified according to developed rules.

By prefiltering of inevitable incorrect and invalid signals during acquisition, implementation of validation rules results in significantly lower rate of false alerts caused by misinterpretation of high values of diagnostic indicators.

By prefiltering of inevitable incorrect and invalid signals during acquisition, implementation of validation rules results in significantly lower rate of false alerts caused by misinterpretation of high values of diagnostic indicators.

 

Instantaneous frequency estimation

Most advanced methods of signal processing require information about the rotational speed of the machine. It requires an additional sensor and in some cases is not possible to apply. There are numerical methods, which can help to estimate the instantaneous rotational speed (also known as instantaneous frequency estimation) using the raw vibration signal only.
Problem of the Instantaneous Frequency (IF) estimation is one of fundamental issues in signal processing. As a generalization of the definition of frequency, IF is defined as the rate of change of the phase angle at time t of the analytic version of the signal.

Cyclostationarity

Cyclostationarity is a new direction of research, which uses the cyclic changes in random signals in order to find a hidden information, e.g. a fault in a machine. This approach was first used in the field of telecommunication in order to model the process of signal modulation. It allows to build communication systems, which are much more immune to eavesdropping. Science then, it has become popularized especially in the field of communication and financial analysis.
The application of cyclostationarity to signal processing allows obtaining many useful information, including: detection of random components, classification of multiple received signals present in noisy data set according to their modulation types, estimation of signal parameters, prediction of a future behavior of the random signal etc.

Wavelet demodulation
In nonlinear acoustics test the phenomena which occur as a result of acoustic wave interaction with damage are very often used. One of them is signal modulation. Due to modulation the number of sidebands can be observed around high frequency component. Sidebands analysis can provide information on the size of the damage and even its location. Using demodulation process the best parameters for nonlinear acoustics test can be selected.

Algorithm of wavelet demodulation applied for selection of frequency value of acoustic wave.

Figure illustrates the algorithm of wavelet demodulation applied for selection of frequency value of acoustic wave.

 

Triple Correlation
The triple correlation is somehow less popular than the standard (doubled) correlation and is mainly used when multiple observations – embedded in additive noise and corrupted by trends – are present in analysed signals. The method is applied to compare the sets of signals acquired during vibro-acoustic modulation test. Thereby correlation between successive higher harmonics can be determined

Triple correlation coefficients for two specimen (damaged and undamaged).

The figures presents examples of Triple correlation coefficients for two specimen (damaged and undamaged).