Robust Structural Health Monitoring of CFRP Structures under Varying Loads using NSFD and Machine Learning with Combined Usage of Acceleration and Strain Data

Structural health monitoring (SHM) of objects in situ often requires a suitable combination of different methods and measurement techniques. In the field of aerospace applications, the previous research project "Combined acoustic and modal structure monitoring" dealt with the development of a robust SHM system for damage detection in carbon-fibre-reinforced polymer (CFRP) structures under varying realistic loads. The methods combined within the project were based on guided waves, acoustic emission, different vibration monitoring techniques and strain measurement. The present paper deals with the application of the nullspace-based fault detection algorithm (NSFD) combined with a machine learning approach based on strain measurements to classify different load conditions of the structure. Due to the high sensitivity to changes in the statistical properties of the measurement data, the NSFD algorithm reacts very sensitively to structural changes, but also to changes in the environmental conditions, such as changes in the external loads and the resulting stress state of the structure. In the paper, the impact of varying external loads on the NSFD damage indicator is analysed and shown. To prevent the NSFD algorithm from causing false alarms due to changes in the load conditions, but still being sensitive to small structural changes due to impact damages, the strain sensor data is used to predict the external loads on the structure. For this, machine learning is used to create clusters of similar stress states. For every cluster, a baseline for the NSFD algorithm must be made. The resulting approach is robust against changes in external loads and sensitive to small changes in the structure due to damages at the same time. The theory and the algorithms are successfully tested with measured data sets from a CFRP-airplane structure.