Load calculations in wind turbine power trains - design, maintenance & reliability
Due to the COVID-19 crisis measures the PhD defence of Damian Rommel will take place online without the presence of an audience.
The PhD defence can be followed by a live stream.
Damian Rommel is a PhD student in the research group Mechanics of Solids, Surfaces and Systems (MS3). His supervisor is prof.dr.ir. T. Tinga from the faculty of Engineering Technology (ET).
Due to high failure rates and long downtimes, extensive research has been done on wind turbines during the last decades. A particular focus has been given to power train components like gearboxes and generators. However, significant reductions of the failure rates could not be achieved despite of the enhancements in gearbox designs and the development of direct driven wind turbines. As a consequence of this, the wind industry has started to focus on failure prediction in order to reduce and optimize cost and thus, has initiated research to develop methods for effective failure and maintenance prediction.
Failures of wind turbine systems are commonly predicted by data-driven and statistics-based approaches. However, statistics-based methods have the drawback to be less precise than model-based ones, especially when large variations in design, loads or operating conditions occur, as is typical for wind turbines. As these variations cannot be adequately incorporated in statistics-based methods, large uncertainties in life predictions and associated conservative maintenance intervals are found. On the other hand, model-based approaches using physical equations and calculations aim to incorporate these variations and consequently, to improve the life time prediction.
Therefore, in this thesis, global loads will be the input for the load-based predictions as they can be of different sources and can be aggregation of various external loads. The term global loads is used here because different sources of global loads (root causes) can lead to various external loads, i.e. forces and moments. These external loads generate internal loads (e.g. stresses), which govern the degradation and failure of components. For example, global loads are: the reaction loads associated to the power transmission (torque and speed), component motions or misalignments.
The usage of (global) load-based prediction approaches provides the advantage that the failure or remaining useful life can already be predicted when the system is still “healthy”, i.e. at a stage where the real system still runs at or close to normal conditions and system degradations cannot be measured yet. Further, the proposed load-based prediction policy can be combined with measurements determining the actual system loads. This has the potential that the system life time can be evaluated more accurately than during the system design procedure. During the system design (global) loads are often assumed. Estimating the system loads based on measurements allows a more detailed specification of the load profiles during operation. This means that the load-based prediction approaches also provide feedback for improving the system design.
However, to apply load-based prediction, physics-based models used for global load calculations must be developed. As the global loads depend on the system design, operation and environment, these models should be i) usable for design purposes and ii) able to handle measurements of operational and environmental responses (conditions). Further, the physics-based models should allow detailed analyses of system loads and behavior to evaluate the most dominant influences of system design, operation and environment on the system life time. To comply with these requirements, the load-based prediction approaches are, in addition to physics-based, also analytical-based in this thesis because finite element based methods are very time consuming.
It is demonstrated that (global) load-based prediction approaches cannot only be used for system design and maintenance but also for system reliability. Application of these approaches in practice yields three major challenges, namely: i) the lack of information of component dimensions and materials, ii) the measurement of operational conditions and iii) the conversion of calculated loads into life time. By using design standards the third problem can be solved. However, design standards require design details from the Original Equipment Manufacturer (OEM) and thus, lead to the first problem, lack of information. As a system operator can probably realize the measurements of operational (and environmental) conditions, the remaining challenge is the lack of information, which can be reduced by either the cooperation with the system OEM or by engineering expertise.
In this thesis, analytical models are developed for a number of power train components. Chapter 2 provides a method to calculate the loads (e.g. exerted on bearings) caused by the wind turbine rotor. It also shows how these loads are influenced by different effects of asymmetric wind flow. Chapter 3 develops a method to calculate the reaction loads of flexible connecting couplings with metal disc packs. The load calculations considers both joint kinematics and disc pack deformations due to misalignment between gearbox and generator. Further, a load sensitivity analysis is executed to clarify the effect of misalignment on gearbox and generator bearings. Chapter 4 proposes methods to evaluate i) transformer core and winding losses based on rms input and output currents and voltages and ii) the hot spot temperature in transformer windings and cores based on a virtual twin derived from transformer rating information. An accurate detection of the transformer hot spot temperature is important for the transformer life prediction.
This thesis also shows that the maintenance intensity and reliability of any system (here wind turbine power trains) are directly related to their design. This means that the observation of high maintenance activities and low system reliability indicates that system loads and system design are created from different assumptions. Further, the design of any system should be chosen such that the mechanical and electrical subsystem as well as the different components are decoupled from each other so that the effects of transient behaviors and dynamic loads of subsystems and components are not transmitted, i.e. they are minimized or even avoided. This will provide the lowest maintenance costs and the highest system reliability.
This concept is again demonstrated for several components. Chapter 5 considers the drive train reliability from a physics-based design perspective. Methods are developed to improve the drive train reliability during early concept and design stages. Principles of system load reduction are applied to achieve a higher reliability, i.e.: minimization of load magnitude, prevention of load superposition, specification of system excitation and evaluation of transfer functions. Chapter 6 discusses the system transfer behavior for reliability improvements. The transfer functions are derived for shaft-inertia and gear-gear assemblies. To calculate the transfer function based on fundamental dimensions a simple gear mesh stiffness approximation is provided. Furthermore, the transfer behavior of a direct drive and geared wind turbine are compared. Chapter 7 derives requirements for the stability of renewable energy grids. It is discussed whether a grid connected generator or electrical power converter fulfills these requirements. Based on the discussion of the results an alternative power train concept is presented which seeks to stabilize energy grids based on renewable sources.
Finally, the (global) load-based prediction methods provided in this thesis are not only applicable to wind turbine, industrial or automotive (e.g. electrical car) power trains, the renewable energy grid or any other energy system. Rather, as global loads are a generic concept and any system is determined and exposed to them, global load-based prediction methods (for system design, maintenance and reliability) can be developed for and applied to any mechanical and electrical system. In this sense the methods and calculations presented in this thesis are just a sample of the potential and possibilities which provide global load-based prediction approaches.
