Paper Type
short
Description
The availability of datasets for analytical solution development is a common bottleneck in data-driven predictive maintenance. Therefore, novel solutions are mostly based on synthetic benchmarking examples, such as NASA’s C-MAPSS datasets, where researchers from various disciplines like artificial intelligence and statistics apply and test their methodical approaches. The majority of studies, however, only evaluate the overall solution against a final prediction score, where we argue that a more fine-grained consideration is required distinguishing between detailed method components to measure their particular impact along the prognostic development process. To address this issue, we first conduct a literature review resulting in more than one hundred studies using the C-MAPSS datasets. Subsequently, we apply a taxonomy approach to receive dimensions and characteristics that decompose complex analytical solutions into more manageable components. The result is a first draft of a systematic benchmarking framework as a more comparable basis for future development and evaluation purposes.
Recommended Citation
Zschech, Patrick; Bernien, Jonas; and Heinrich, Kai, "Towards a Taxonomic Benchmarking Framework for Predictive Maintenance: The Case of NASA’s Turbofan Degradation" (2019). ICIS 2019 Proceedings. 4.
https://aisel.aisnet.org/icis2019/data_science/data_science/4
Towards a Taxonomic Benchmarking Framework for Predictive Maintenance: The Case of NASA’s Turbofan Degradation
The availability of datasets for analytical solution development is a common bottleneck in data-driven predictive maintenance. Therefore, novel solutions are mostly based on synthetic benchmarking examples, such as NASA’s C-MAPSS datasets, where researchers from various disciplines like artificial intelligence and statistics apply and test their methodical approaches. The majority of studies, however, only evaluate the overall solution against a final prediction score, where we argue that a more fine-grained consideration is required distinguishing between detailed method components to measure their particular impact along the prognostic development process. To address this issue, we first conduct a literature review resulting in more than one hundred studies using the C-MAPSS datasets. Subsequently, we apply a taxonomy approach to receive dimensions and characteristics that decompose complex analytical solutions into more manageable components. The result is a first draft of a systematic benchmarking framework as a more comparable basis for future development and evaluation purposes.