Measurement Uncertainty and Remaining Useful Life Prediction: A Case Study Using Testing Data From Shape Memory Alloy Wires

In a globalized and interconnected world, the importance of reliability, maintenance, and quality continues to grow.At the same time, the prediction of the remaining useful life (RUL) is essential for maintenance planning, facing Prognostics and Health Management of technically complex products.Given the importance of maintenance measures for a reliable product life cycle, it is all the more important to achieve the most accurate prediction results possible.A relevant influencing factor here is the measurement uncertainty of the database used.This paper presents a case study on the impact of measurement uncertainty on the RUL prediction.The study employs data from tests of cyclically stressed shape memory alloy wires.Real data from long-term life tests conducted in a test rig are given.Firstly, the RUL of the shape memory alloy wires is predicted using linear regression models.The training data is fitted with a Gaussian least-squared regression model, and the RUL is estimated using forecasts generated by this model and an adaptive y-target value for the failure time.The second step is a comprehensive measurement uncertainty analysis, which determines and quantifies all relevant uncertainty components of the measurement process.The extended measurement uncertainty is determined according to ISO 22514-7 and VDA 5. Thirdly, Monte Carlo simulations are conducted based on the original time series and the determined measurement uncertainty.Representative time series are generated, for which each the RUL is predicted.Subsequently, descriptive statistics are applied to the obtained set of simulated RULs, with the results compared to the original time series and the true RUL.The paper concludes with a discussion of the results and an outlook on future work.