Supervisors: Professor Mark Mueller, University of Edinburgh
Sponsoring Company: Gaia Wind
PhD Student: JenHao Wu
Reliability assessment plays an important role in different phases of a product from design, manufacturing, installation, to maintenance. In the perspective of this research, a product's life time is separated into three periods, the past, the future and the present. These periods contain specific information from different phases. Three Bayesian Hierarchical models, where a Bayesian Hierarchical model is a statistics belief structure, are created to describe the reliability performance of a wind turbine in these three periods.
The capabilities and features of these models in different period are listed below.
Historical time to failure (TTF) data in a warranty record can be analysed. Explanatory variables can be evaluated, such as turbulence intensity (TI), average wind speed (AWS), repair effectiveness. The result is a jointed posterior probability distribution function (p.d.f) which contains all the parameters interested. The TTF distribution of a turbine can be predicted as well as the hazard function, the reliability curve etc.
Conventionally, the uncertainties in introducing a new design is hard to be quantified. And this can make the comparison between alternative ideas impossible. In the proposed model, new design ideas can be described as new prior distributions which can be imported in the Bayesian Hierarchical model. The result is a updated TTF distribution which comprises the information from the historical data and new design ideas. Therefore, alternative design ideas can be compared with a greater confidence.
Condition monitoring (CM) technique is well known for providing the on-line status of a product. However, unsurprisingly, the judgements are hard to be made due to the insignificant changes of the feedback signal. The proposed model can monitor the same feedback signal while knowing the historical behaviour of the turbine which enrich the information used for making judgements.