Development of fault detection modules for various industrial processes based on measurement data
The basic task is
- to enhance currently available techniques with new data-driven learning algorithms for process models (with improved predictive quality and flexibility), automatically identified from the system based on collected measurement data
- an advanced generation and analysis of fault indicators based on the identified models in order to detect various type of faults as early as possible and furthermore to significantly improve the quality of machines and products.
The models will be developed for a range of processes from industry, where measurement data with various characteristics is recorded and supervised, ranging from stationary to dynamic data (time delays), including low-dimensional to high-dimensional variable spaces as well as containing un-labeled and labeled data (fault classes).