Two-day advance prediction of a blade tear on a steam turbine of a coal power plant - Using only automated mathematical data-analysis techniques and no prior knowledge or human experience

摘要

A blade tear on a steam turbine of a coal power plant is a major failure causing large financial losses. Knowing that such an event is going to occur in one or two days time, allows the operational staff to shut down the turbine before the damage is done. The affected blade may then be removed and normal operations resumed. To this end, we must verify two hypotheses: (1) It is possible to predict two days in advance that a blade tear is going to occur and (2) it is possible to determine which blade(s) is affected. Moreover, we wish to do so in an automated way not involving human experience or knowledge so that the prediction is fully objective, can be continuously run and is relatively inexpensive to implement. This paper describes the verification of both hypotheses on an actual example of a steam turbine in a coal power plant in Germany.