Condition monitoring of vibrating structures using data-based modeling
During the service life of load-bearing structures (e.g. bridges) structural changes occur which can lead to damage or failure. In order to ensure safe operation and to detect possible sources of damage at an early stage, information about the condition is necessary. Therefore, the objective of this research project is the development of a methodology to make statements about the condition of the monitored object based on measurement data.
Such a procedure falls into the field of Structural Health Monitoring (SHM), which can be divided into different levels of increasing complexity:
Damage detection (Is damage present?)
Damage localization (Where does the damage occur?)
Damage identification (What type of damage occurs?)
Damage quantification (How severe is the damage?)
With this information, prediction models can be used to estimate the remaining life of the structure or to propose maintenance measures. The accuracy of the predictions depends on the quality of the available information and the complexity of the SHM system. However, the question arises how, on the one hand, measurement data with the relevant information can be generated and, on the other hand, how this information can be extracted from the measurement data.
Within the scope of this project, the measured information is supposed to be based on the vibration state of the monitored object. A change of state of the structure affects the wave transmission in the structure and thus its vibration state. The vibration data, measured for example by acceleration sensors, can be scanned for indications of possible damage using suitable algorithms.
Machine learning algorithms are practical for this purpose. These algorithms construct a statistical model, that „learns“ patterns and correlations from empirical data in order to subsequently apply the extracted information to unknown data. In the project context, this means that vibration data (e.g. frequency responses) of the undamaged and the damaged object are generated simulatively and experimentally. The relation of these samples to the different stages of SHM is then established by data-based models. Subsequently, these models will be validated by applying them to real measurement data of the example object in order to detect existing or developing damages.