Speaker
Description
Data-driven virtual sensing methods enable cost efficient monitoring of critical system components in predictive maintenance applications. The parameterization of these methods mainly requires the collection of representative usage datasets for the system of interest. In most applications, very long times to failure are a significant challenge for the validation of these approaches. This contribution therefore provides an experimental example dataset, where notched specimens are subjected to individual service loads until failure, using a fatigue test bench.
The dataset is based on acceleration and strain measurements from an instrumented e-bike [1], which are randomly resampled and transformed into new service loads. The experimental setup consists of two servo hydraulic cylinders, which control both force and displacement of the steel specimens. These cylinders are also equipped with acceleration sensor. Separate service loads ensure independence of parameterization and validation datasets and a component SN-curve is obtained for subsequent fatigue analysis.
Similar to predictive maintenance applications in vehicle monitoring, the virtual sensing task is to predict the fatigue damage accumulation in the specimens from the acceleration measurements. This is achieved using two different strategies. In the first method [2], the force signal of the specimens it directly predicted from the acceleration measurements. Following the nominal stress concept, fatigue damage sums are computed using cycle counting and the elementary Miner rule. The second virtual sensing approach [3] instead characterizes the acceleration data using the wavelet Scattering transform and principal component analysis, which is subsequently used for a direct fatigue damage regression. Both approaches are evaluated and compared using an independent validation dataset.
References:
[1] Heindel, Leonhard, Hantschke, Peter und Kästner, Markus. „eBike measurements for fatigue monitoring and maneuver identification tasks“. OpARA TU Dresden, 2022
[2] Heindel, Leonhard, Hantschke, Peter und Kästner, Markus. „ Assessment of hybrid machine learning models for non-linear system identification of fatigue test rigs“. Franklin Open, 2024
[3] Heindel, Leonhard, Hantschke, Peter und Kästner, Markus. „Fatigue monitoring and maneuver identification for vehicle fleets using a virtual sensing approach“. International Journal of Fatigue, 2023
