Prognostics and Statistical Predictive Maintenance, 6+3 hp

Course responsible: Erik Frisk (erik.frisk@liu.se)

This is a static information page about the course. For up-to-date information about current lectures and course material, visit the public page https://gitlab.liu.se/vehsys/phd_prognostics.

Course objectives, content, and organization

This is a third-cycle (PhD) course in prognostics and predictive maintenance. The course aim to cover model-based and statistical, data-driven, techniques. The main focus of the course is statistical modeling and data-driven techniques for survival modeling and Remaining Useful Life (RUL) regression models. The objective is to give theoretical foundations but also practical experience in Python with industrially relevant datasets. Since the course is limited to 6 credits, the course will aim for some general coverage, but methods will be oriented towards topics relevant for research at the department.

• Introduction to prognostics and maintenance policies
• Statistical modeling of component life-time
• Model-based prognostics
• Data-driven techniques
  • Classical statistics methods: Kaplan-Meier and Cox Proportional Hazard models
  • Machine learning models for prognostics-Random Survival Forest and RUL-regression
  • Deep nets for prognostics - GNN, Mixture Density Networks, …
• Optimal maintenance polices
• Experimenting on research and industrially relevant datasets in Python

Examination: The course is examined through active participation in the lectures and handing in solved exercises/code.
Course literature: Book chapters and research papers.
Pre-requisites: Basic knowledge in statistics. Some programming experience in Python and Pytorch beneficial but not strictly necessary.

The basic course is planned for 6 credits. For those interested, small projects can be formulated for 3 extra credits.

Preliminary lecture plan

This is a preliminary lecture plan that might change during the course.