50 % Seminar: Data-Informed Method Selection for Predictive Maintenance and Degradation-Aware Battery Operation in Power Systems

Predictive maintenance in power systems increasingly relies on machine-learning methods, but the choice of method should depend on both the maintenance objective and the information available in the data. A model that performs well in retrospective evaluation may still be unsuitable for deployment if the available measurements do not adequately reflect the degradation or failure mechanism of interest, or if operational constraints limit its practical use.

This seminar presents the work completed so far on GUIDE–PdM, a five-phase decision-support framework for predictive-maintenance method selection. The framework covers problem formulation, data assessment, method selection, model development, and deployment with feedback. It explicitly considers whether the available data are sufficiently informative for the intended objective, whether improving the data is justified, and whether a data-driven method is preferable to simpler alternatives. The seminar will also discuss the development of the framework, including expert consultation, and its application to four case studies.

The final part presents ongoing work on battery degradation and operation. This includes capacity forecasting across different operating conditions and degradation-aware battery dispatch for voltage control in distribution networks with high photovoltaic penetration. The seminar concludes with an outlook on the remaining work toward the PhD thesis.