Predicting damage location information based on single actuator sensor path information using regularised multilinear regression

In structural health monitoring with guided ultrasonic waves, probability reconstruction algorithms are a method to locate a damage. They work by calculating the probability for each actuator sensor path on whether there is a damage on the path or not. By superposition of each path and its damage probability, a damage localisation is done. The disadvantage of this is, that the damage localisation resolution is limited by the number of paths crossing each other. To overcome this, the hypothesis of this investigation is that the information of a path can not only be used to determine whether a damage is present, but that additional information about the location within the path can be calculated as well. This way a localisation resolution can be higher than by only relying on the path density. To verify this assumption, an experimental setup was chosen in which the path lengths always remain the same while the distance between damage and the direct path varies. This is implemented by a moving ultrasonic microphone simulating the sensor. The varying distance is the local information, which is determined in this study using the information of a single path. For this purpose, a prediction is calculated using regularised multilinear regression. The input features are characteristic values of five sections of the sensor signal in the time domain. The sections are manually chosen based on arriving wave events. The result confirms the hypothesis. Therefore, it is plausible to increase the detection resolution of probability reconstruction algorithms by calculating damage location estimations for each path.