PSI - Issue 38

Frédéric Kihm et al. / Procedia Structural Integrity 38 (2022) 12–29

29

18

Kihm, Miu, Bonato / Structural Integrity Procedia 00 (2021) 000 – 000

6. Conclusions This paper discusses the creation of predictive models using placeholder signals, such as temperature, pressure or acceleration – instead of strain – to estimate damage. It was shown that accurate damage estimation is sometimes possible even with a small subset of sensor data and that increasing the number of signals brings diminishing returns. It is worth noting that the accuracy of the prediction will strongly depend on the quantity and quality of input data available. In this paper, we proposed a general framework for exploring data, which is inspired from the field of data science. It was shown how engineers can extend their understanding of the problem domain thanks to the use of basic data science concepts. Two case studies were presented to illustrate the use of the methodology to construct a predictive model from measured data. Both use cases show that an engineer who is using a priori physics knowledge but no data science might have under or over designed a model. The first case study shows that the engineer would have probably missed parameters that actually turn out to contribute to the output damage. The second case study will show that the engineer would have probably over complicated the model by considering extra signals that actually don’t nec essarily contribute to the output damage. This paper advocates the fact that both statistical techniques and physics-based approaches need to be combined to construct sensible models and maximize the confidence in its estimates. References J. A. Bannantine, J. J. Comer and J. L. Handrock, Fundamentals of Metal Fatigue Analysis, Prentice Hall, 1990. A. Halfpenny, "A practical introduction to fatigue," in New technology 2001, Warwickshire, UK, 2001. M. Matsuishi and T. Endo, "Fatigue of Metals Subjected to Varying Stress," in Japan Society of Mechanical Engineers, Jukvoka, Japan, 1968. James, Gareth, et al. An introduction to statistical learning. Vol. 112. New York: springer, 2013. Friedman, Jerome, Trevor Hastie, and Robert Tibshirani. The elements of statistical learning. Vol. 1. No. 10. New York: Springer series in statistics, 2001. Montgomery, Douglas C. Design and analysis of experiments. John wiley & sons, 2017. Bracke, S., et al. "Reliability engineering based on operating data and monitoring systems within technical products: Challenges, requirements and approaches." (2018): 1069-1076. J. Bendat and A. G. Piersol, Random Data: Probability, random variables, New York: 3rd ed. John Wiley & Sons, 2000 S, S., SV, H., Mendez, A., and Dodds, C., "Development of a Specific Durability Test Cycle for a Commercial Vehicle Based on Real Customer Usage," SAE Technical Paper 2013-26-0137, 2013, https://doi.org/10.4271/2013-26-0137.

UNRESTRICTED