Skip to main content

Signal processing and measurement data analysis

The lectures and laboratories in the Department of Materials Science contain various experiments for the determination of technological or comparable material parameters, which are provided by means of destructive and non-destructive testing methods. In addition to the use of measurement and testing technology and data acquisition, data processing and analysis play a major role in the evaluation of materials and the integrity of components and structures.

In signal processing, pattern recognition in data makes it possible to determine regularities, repetitions, similarities or regularities in data in order to isolate them for further analysis and thus to use them for the interpretation of the recorded signals or values.

Artificial neural networks are often used for this purpose with the aim of determining the probability that an event belongs to one or the other event category (pattern) and assigning it to the category with the highest probability. The basic feature recognition is achieved by means of statistical methods, while superordinate inference methods rely on special knowledge of the application area. For the concrete example of the characterization of the material microstructure, this means that different defects, which are e.g. based on the dislocation structure, microcracks, pores, etc., lead to characteristic measurement signal courses with specific features. Instead of evaluating features in the signal characteristics according to ready-made rules, these are measured as values and combined in a so-called feature vector. A mathematical function can assign a defect category to every conceivable feature vector based on its probability.

The procedure requires a signal calibration and can be "learned", for example, by changing the measurement signal along artificially introduced defects or defined stress conditions. While structural pattern recognition only checks for the presence of different features, superordinate structural procedures, such as Bayesian networks, for example, make it possible to combine individual results, calculate the overall result from this and can thus, for example, reproduce information on the material microstructure, which is characterized by the superposition of different phenomena.

Furthermore, by comparing successive snapshots in identical loading situations, a dynamic system can be developed which enables the development of defects, such as an increase/decrease in dislocation density or crack growth. In the course of the lecture the measurement signals based on temperature (thermocouples) and magnetics will be acquired by means of a myDAQ measurement card and the data will be treated with respect to pattern recognition in order to separate different fatigue mechanisms.

The following steps are required for pattern recognition:

  •  Recording of the measurement signals
  •  Digitization of the measuring signals
  •  Creation of patterns and representation in feature vectors
  •  Data reduction through pre-processing of the data
  •  Extraction of characteristics
  •  Reduction of features to essential features
  •  Assignment in characteristic classes

For example, in systems oscillating freely in resonance, the frequency domains for different fatigue stages can be determined from the Fast Fourier Transformation and analyzed with regard to the higher harmonics (oscillations) that occur. The increase of local defects in the test volume of the material microstructure leads to an increase of the vibration order. The ratio of the amplitudes of the first harmonic to the higher harmonics of the second, third, ..., n-th order describes the so-called harmonic distortion factor, whereby the sole consideration of the amplitudes of odd order or of quotients of individual amplitudes is also possible here. Increasing material damping due to increasing damage leads, for example, to a drop in the amplitudes of the higher harmonics and to a drop in the entire frequency spectrum. Other relevant features include the shift of the Fourier or energy coefficient.

The aim of the project is the practical and timely teaching of the handling of measurement and testing techniques as well as the associated data acquisition and analysis. Therefore, 5 laptops are to be purchased, which, together with the National Instruments myDAQ measurement cards (purchase via application WS 2018), enable a timely and application-oriented measurement data analysis for student use.