On the influence of physical sensor placement on the virtualization of wind turbine blades testing

Wind turbine (WT) blade manufacturers are asked to test each wind turbine blade individually prior to its installation. Several tests are typically carried out on each individual blade, including static tests, fatigue tests and dynamic tests with different types of excitations and boundary conditions. During these extensive test campaigns, several types of sensors are used to instrument the blade, e.g. accelerometers, strain gauges, inclinometers etc. Indeed, each test requires the use of different sensor types, depending on the system characteristics to be identified. The high cost of these sensors and the time-consuming instrumentation procedure often implies drawbacks in terms of time efficiency and cost, or instead prompts the deployment of a limited number of sensors. In this context, Virtual Sensing (VS) techniques can be adopted for enriching the monitored information via creation of so-called “virtual sensors", i.e., response quantities (e.g. strains) inferred at locations where a physical sensor could not be positioned. This work analyzes the influence of the adopted physical sensing configurations on the virtual predictions delivered via use of Kalman-type filters for joint input-state estimation. The outlined study is carried out using data acquired during an extensive test campaign performed by DTU Wind Energy on a research wind turbine blade