Reliability Prediction for Offshore Renewable Energy: Data Driven Insights
Accurately quantifying and assessing the reliability of Offshore Renewable Energy (ORE) devices is critical for the successful commercialisation of the industry. At present, due to the nascent stage of the industry and commercial sensitivities there is very little available reliability field data. This presents an issue: how can the reliability of ORE’s be accurately assessed and predicted with a lack of specific reliability data? ORE devices largely rely on the assessment of surrogate data sources for their reliability assessment. To date there are very few published studies that empirically assess the failure rates of offshore renewable energy devices [1]. The applicability of surrogate data sources to the ORE environment is critical and needs to be more thoroughly evaluated for a robust ORE device reliability assessment. This paper tests two commonly held assumptions used in the reliability assessment of ORE devices. Firstly, the constant failure rate assumption that underpins ORE component failure rate estimations is addressed. Secondly, a model that is often used to assess the reliability of onshore wind components, the Non-Homogeneous Poisson Power Law Process (PLP) model is empirically assessed and trend tested to determine its suitability for use in ORE reliability prediction. This paper suggests that pitch systems, generators and frequency converters cannot be considered to have constant failure rates when analysed via nonrepairable methods. Thus, when performing a reliability assessment of an ORE device using non-repairable surrogate data it cannot always be assumed that these components will exhibit random failures. Secondly, this paper suggests when using repairable system methods, the PLP model is not always accurate at describing the failure behaviour of onshore wind pitch systems, generators and frequency converters whether they are assessed as groups of turbines or individually. Thus, when performing a reliability assessment of an ORE device using repairable surrogate data both model choice and assumptions should be carefully considered.