A method for estimating the remaining useful life of a component of an operational wind turbine

Abstract

A method and system for estimating the remaining component life of an operational wind turbine from actual wind turbine operating conditions after commissioning, using a data acquisition module configured to measure real historical data of the operational wind turbine, and an additional state detection unit configured to identify historical operating states. The method includes extracting historical data from the data acquisition module at time intervals, identifying the operating state of the wind turbine, verifying the identified operating state and identifying unmatched uncertain data. Then, a turbine model is simulated for each identified operating state and its wind condition at each time interval, and a fatigue equivalent load is calculated for each operating state and wind condition.

Description

Technical Field

[0001] The invention belongs to the field of health assessment and prognosis for wind turbines and allows estimating the remaining useful life of wind turbine components. In particular, it relates to a method for modeling and simulating actual loads on a wind turbine after the turbine has been commissioned (post-design). The invention provides a very good understanding of the forces and moments actually occurring on the blades, main shaft, tower and other components. The invention describes a method for identifying and distinguishing operating conditions and their respective fatigue loads with respect to the history of the turbine. The fatigue loads can then be used to determine the reliability of individual components as well as the turbine as a whole. Background Art

[0002] Wind turbine components are designed to have a service life of approximately twenty to twenty-five years. Fluctuations in environmental factors can cause changes in loading patterns that can lead to premature component failures. These premature failures result in unplanned service, which is often a major portion of operating costs for wind turbine farm operators.

[0003] Unplanned service schedules and repairs reduce productivity and increase maintenance costs. Maintenance costs for wind turbines increase over time. Some methods for estimating component lifetime are known in the art.

[0004] IEC 61400-1 (Design Requirements) is the traditional standard for estimating turbine driven loads today. The main purpose of the IEC standard is to certify a turbine for a service life of 20 years. The remaining service life RUL is estimated by the wind characteristics of the site through SCADA or meteorological mast measurements or simulated wind models (as described in the approved UL 4143 standard or the DNVGL-ST-0262 Life Extension LTE standard and the IEC61400-28LTE standard being defined). In these cases, the wind trends are characterized and the life is calculated based on wind trend estimates at the wind site level. These wind trends are not the actual wind conditions and operating conditions of a specific wind turbine, and are not accurate and precise estimates of the actual loads that have occurred, and therefore are not estimates of actual damage and future loads.

[0005] For example, document US2016 / 0010628 discloses a method comprising receiving installation data, operating data, environmental data, and historical data of a wind turbine. Multiple correction values ​​corresponding to multiple damage values ​​are determined based on the operating data and the historical data. The multiple correction values ​​include corrections due to occasional failure modes and corrections to the progression of continuous failure modes determined based on data obtained during a maintenance plan. This document discloses a probabilistic method for comparing damage values ​​and models that compare installation data, historical data, and updated operating data and predict continuous failure modes or occasional failure modes. The actual wind loads occurring in the wind turbine from the date of commissioning are not modeled, nor is an accurate model that explicitly considers the actual damage caused by shear and bending moments of each component.

[0006] Document EP 2290597 discloses a method including a performance simulator that uses site conditions of a wind farm, turbine configuration design data and historical field data to simulate the performance of a wind turbine. However, it does not take into account the operating state of a specific wind turbine and how many operating states and operating hours the turbine has experienced in each state in each specific state. This is essential for classifying the historical operating conditions and corresponding loads of the wind turbine, and is also important for accurately estimating the remaining life of the components. In addition, the disclosed data acquisition method is not accurate enough, and a lot of inaccurate and erroneous data will be obtained during the simulation process, resulting in errors in damage and life estimates. Summary of the invention

[0007] A method for estimating the remaining component life of a wind turbine is disclosed herein, by which it has been found that the above-mentioned disadvantages associated with prior art solutions are at least mitigated. The method accurately quantifies the damage of each component from the commissioning date. It accurately estimates the remaining life of each component of a specific operational wind turbine, its own actual operating conditions including its unique location (i.e. wind resource and turbulence intensity and design configuration).

[0008] More specifically, according to a first aspect of the invention, the invention describes a method for estimating remaining useful life of components of an operational wind turbine model, performed by a computing element, wherein the computing element is communicatively coupled to the operational wind turbine, the operational wind turbine comprising a data acquisition module configured to measure real historical data of the operational wind turbine, and additionally a state detection unit configured to identify historical states of operation, wherein the method comprises:

[0009] - extracting historical data from the data acquisition module at selected time intervals, the data comprising at least wind speed, blade pitch position and rotor speed;

[0010] - identifying, by a state detection unit, an operating state of the wind turbine for each selected time interval, the operating state comprising at least running, idle and transition;

[0011] - Verify the identified operating status using the data extracted by the data acquisition module at each time interval and identify the unmatched uncertain data;

[0012] -- identifying a dominant operating state, the dominant operating state comprising the state in which the wind turbine has spent the longest amount of time;

[0013] -- identifying wind conditions, said wind conditions comprising at least average wind speed and turbulence intensity at said selected time interval;

[0014] - Obtaining actually occurring loads of the wind turbine model corresponding to wind conditions, at least the prevailing state and its duration, based on the simulation model.

[0015] Preferably, the method further comprises a step for identifying a number of transient events including the number of times said wind turbine has changed state, and a step for obtaining the actually occurring loads for at least one of the transient events and its wind conditions at each time interval based on the simulation model. Preferably, all transient events are taken into account in order to estimate very accurately all fatigue loads during the transient events to which each component of the wind turbine is subjected.

[0016] Transient events may represent short periods of time within a time interval but may have a large impact on loads and fatigue damage. Therefore, one of the advantages of the present invention is to accurately detect these transient events and simulate actual loads for a specific operating wind turbine model and its specific wind condition variations.

[0017] Preferably, the transition operating state is subdivided into three further operating states: start, normal stop and emergency stop. However, the invention is not limited to these operating states. Finer subclassifications, such as different types of emergency stops, can also be used. However, the invention suggests using the mentioned operating states as a minimum.

[0018] As mentioned previously, the state identification unit can be implemented by combining information from the state detector unit with operational metrics from the data acquisition module. This will prevent misleading state identification and allow multiple state changes to be handled.

[0019] Preferably, the state detector unit is part of a wind turbine internal control system and is presented as a logical variable for each operating state.

[0020] Preferably, the data collection module further includes a generator power change at each time interval.

[0021] Alternatively, the state detector unit may be a code executed by a computing element that identifies the operating state from the data acquisition module. For example, the code may be an algorithm that combines data delivered by independent sensors from the data acquisition unit, such as blade pitch, generator power, rotor speed and wind speed, and can verify and identify uncertain, erroneous and / or non-corresponding data by repeatedly comparing instantaneous data.

[0022] Preferably, the wind turbine simulation can be an aeroelastic simulation, a dynamic multibody analysis, a finite element simulation, an aeroservoelastic model or any model that fully represents the wind turbine to obtain reliable fatigue loads. The model of the specific wind turbine should be run in a design of experiments for each operating state. The model should be representative enough to capture the interaction between all intervening parts of the turbine and the wind and / or any other excitation source, such as the earth wake or ocean waves. These interactions can be inertial, elastic, aerodynamic forces generated when the turbine is exposed to the excitation source.

[0023] Preferably, in addition to the previously mentioned interactions, the control mechanism of the turbine should also be considered. For example, if aeroelastic simulation is chosen, the control mechanism should replicate the real control behavior of the specific turbine with sufficient accuracy in order to capture all relevant dynamic responses of the turbine.

[0024] The experimental design for each operating state should include all relevant wind conditions. The present invention is not limited to any particular set of wind conditions, but it is recommended to use at least the average wind speed, turbulence intensity, vertical inflow direction (also known as upflow), horizontal inflow direction (also known as yaw error), wind shear and air density. The units for all conditions need to be consistent. The average wind speed is usually the average wind speed at the height of the turbine hub over a sample period, such as 10 minutes. Turbulence intensity can be calculated as the standard deviation of the wind speed divided by the average wind speed over the same sample period. Other methods of calculating turbulence intensity are also applicable. Wind shear, also known as wind gradient, refers to the difference in average wind speed in the horizontal direction. Wind shear can be expressed by a power law equation. The exponent of the equation can be used as the value of the experimental design. An aeroelastic model is simulated for each experiment. The duration of the simulation should be equal to the time interval (e.g. 10 minutes), plus any transition time required for the model to converge from its initial condition (e.g. static) to the operating state of a particular experiment should be excluded from the simulation because it cannot represent the actual behavior of the turbine under a given experiment (wind condition).

[0025] Preferably, the method further requires calculating a fatigue equivalent load (also called a damage equivalent load) for each operating state and wind condition. This is done by simulating a plurality of turbines at a plurality of time intervals to estimate the fatigue equivalent load at each operating state and each wind condition calculated for each time interval for at least one component in a specific location of the operational wind turbine.

[0026] These load values ​​should cause the same damage to the wind turbine components over a time interval (e.g. 10 minutes) as the original recorded time-varying loads. The present invention does not specify the method for calculating the fatigue equivalent load values. However, it is recommended to use the rain flow counting method for structural components and the load-revolution distribution for all bearings and gear components. The loads are the result of each experiment. Once the turbine model has been run for the complete experimental design, the fatigue equivalent loads for each experiment are collected and associated with their respective wind conditions.

[0027] Preferably, a further step is to model a hyperdimensional response model HDRM representation for each operating state, wind condition and a load type, such as bending moment, torque, thrust, shear force or fatigue equivalent load at a specific location of a determined component of the wind turbine. Preferably, the correlation should then be represented by a hyperdimensional response surface model. Hyperdimensional response surface models explore the relationship between multiple explanatory variables and one or more response variables. In this case, the response surface model explores the fatigue equivalent load under wind conditions. There are a variety of techniques to establish response surface models based on a given set of performed experiments, such as Taylor series, radial basis functions and Kriging. The present invention does not explicitly point out the technology to be used, but suggests that various techniques should be tested to identify the best representation of the correlation. For the same or similar wind conditions, the process of creating a response surface model should be repeated for all operating states (running, idle, transition). This is a useful method for developing hypersurface response models for different turbine models, and a simple method for representing fatigue equivalent load estimates for any wind speed and turbulence intensity under specific operating conditions. This can be done to estimate or correlate another equivalent turbine of the same model in the same wind farm or a different wind farm. Note that the number of transient events varies depending on the wind resource, but for the same wind turbine model, the hyperdimensional response model for fatigue damage to a specific component under a specific operating state and wind condition should be equivalent.

[0028] Which response surface model should be used depends on the operating state determined for each time interval early in the process.

[0029] Preferably, any of the previous steps may be used for a reliability analysis or a remaining useful life of at least one component of the wind turbine using historical operating conditions and fatigue equivalent loads.

[0030] The method may further include a time series of loads experienced by the turbine from the commissioning date to today. Note that this is performed by multiple turbine simulations at multiple time intervals from the commissioning date to accurately estimate the historical time series fatigue equivalent loads of at least one component of the operational wind turbine from the commissioning date. This may be computationally intensive and time intensive, so the hyper response surface model may be an alternative accurate and faster estimate for a given turbine model once it has been calculated for an equivalent wind turbine model or more precisely for the specific wind turbine for which it is modeled.

[0031] Preferably, the method further comprises performing a reliability analysis or a remaining useful life model for each component of the wind turbine with the aid of the actual occurring loads from the commissioning date. More specifically, the load time series can then be used for reliability analysis such as stress cycle curves combined with a linear damage progression model or a remaining useful life calculation (e.g. a particle filter) for each wind turbine component on components such as blades, main shaft, bearings, tower, etc.

[0032] Clear differentiation of actual occurring loads and load durations for each turbine and changes in operating status allows for more refined asset-specific load analysis, demonstrating significantly improved accuracy in performance and life calculations. Using real operating data + leveraging the hypersurface model allows for much higher resolution than assessments based on trended wind estimates and modeling of wind turbines in certified operating conditions.

[0033] Furthermore, if combined with appropriate loading for life analysis such as the model of the present invention, knowing the true loading allows the OEM or third party to provide better updated designs for turbines already in production and provide operators with better failure risk assessments.

[0034] Accurate asset-specific loads are a sine qua non for new reliability analyses of turbines already in production, focusing on long-term planning of asset actions rather than short-term traditional O&M.

[0035] Preferably, the method takes as input a time series of measured wind and operating conditions. The time series can have any time interval, but preferably a time interval of at least 10 minutes (1.67e-3 Hz) or less is used. In addition, preferably, the input time series should include statistical values ​​for each time interval, such as mean, minimum, maximum and standard deviation.

[0036] The source of the time series is not relevant to this approach, but input data can ideally be obtained from the Supervisory Control and Data Acquisition (SCADA) of the specific wind turbine and, if available, supplemented by a nearby meteorological tower to include wind conditions such as wind shear and turn, wind updrafts, turbulence intensity, and air density.

[0037] Preferably, two variables of interest should be determined that provide information about the state of the turbine during each 10 minute data point. The first is the number of times the turbine changes state; the example runs paused, then ready and returns to running, i.e. three state changes. The second is a code indicating the maximum time the turbine has spent in a given state. When operating continuously, this feature, together with the operating sensor, again ensures that the turbine is operating in a specific state. However, there may be multiple state changes within a time interval.

[0038] In addition to identifying whether the turbine is generating or idling, the present invention also seeks to identify uncertain data, i.e. conflicting information from the state detection unit and SCADA is marked or identified as "unidentified". In addition, missing or invalid values, such as "None", "Null" or physically impossible values, are marked as "junk".

[0039] Each data point has an associated state in the data set (e.g., Run, Idle, Garbage, Unrecognized). In addition, the temporal nature of the states allows identification of stop events. Emergency stops are identified in stops based on the state variables or in the operating information. The remaining identified stops are labeled as "Normal Stop". The transition from the "Idle" state to the "Run" state is labeled as "Start".

[0040] According to a second aspect of the invention, the invention describes a system for estimating the remaining useful life of components of a wind turbine model, comprising at least the wind turbine model, the wind turbine model comprising a data acquisition module configured to measure real historical data of the operational wind turbine, a state detection unit configured to identify historical states of operation, wherein the system comprises a computing element communicatively coupled thereto:

[0041] - extracting historical data from the data acquisition module at selected time intervals, the data comprising at least wind speed, blade pitch position and rotor speed;

[0042] - identifying, by a state detection unit, an operating state of the wind turbine for each selected time interval, the operating state comprising at least running, idle and transition;

[0043] - verifying the identified operating state using the data extracted from the data acquisition module at each time interval and identifying unmatched uncertain data;

[0044] - identifying a dominant operating state, said dominant operating state comprising the state in which said wind turbine has spent the longest amount of time;

[0045] - identifying wind conditions, said wind conditions comprising at least average wind speed and turbulence intensity at said selected time interval;

[0046] - obtaining actually occurring loads of said wind turbine model corresponding to wind conditions, at least the prevailing state and its duration, based on the simulation model.

[0047] Preferably, the computing element is capable of performing any one of the steps of the method disclosed above in the first aspect of the present invention.

[0048] According to a third aspect of the present invention, the present invention discloses a computer program adapted to execute the above method and any one of the steps.

[0049] This method and an implementation program adapted to perform said method enable accurate simulation of the real damage of each component of a specific wind turbine subjected to determined specific wind conditions, transient events and corresponding real fatigue loads. As a result, an accurate improved updated design of components for a specific wind turbine in operation can be accurately done. Furthermore, appropriate component changes and the timing of said changes can be performed. Furthermore, knowing the real loads allows providing a better failure risk assessment if combined with appropriate loads such as a life analysis as described in the present invention.

[0050] Accurate asset-specific loads are a sine qua non for new reliability analyses of turbines already in production, focusing on long-term planning of asset actions rather than short-term traditional O&M. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] To supplement the description made and to facilitate a better understanding of the characteristics of the invention, a set of drawings is attached as an integral part of said description, according to a preferred example of its practical embodiment, in which, with an illustrative and non-limiting character, the following are listed:

[0052] Figure 1 A block diagram of a preferred embodiment of a method for estimating the lifetime of a wind turbine component is shown.

[0053] Figure 2 Exemplary graphs of data acquired by a data acquisition module and exemplary identification of transition events by the data acquisition module are shown.

[0054] Figure 3 - An exemplary hyper-dimensional response surface model is shown that relates actual historical wind conditions (wind speed and turbulence intensity) to the fatigue equivalent angular moment of a low-speed main shaft. DETAILED DESCRIPTION

[0055] Figure 1A block diagram of a preferred embodiment of the method disclosed herein is shown. In the described preferred embodiment, the method for estimating the remaining component life of an operational wind turbine comprises a data acquisition module and an additional state detection unit, the data acquisition module comprising a SCADA conventional monitoring system configured to measure real historical data of the operational wind turbine, the state detection unit being used to identify historical operating states, wherein the method comprises extracting historical data from the data acquisition module at 10 minute intervals, the data comprising wind speed, power, pitch position and hub angular velocity. Figure 2 An exemplary representation of data collected by the data acquisition module is shown. Figure 1 The subsequent step of the method is to identify the operating state of the wind turbine by the state detection unit for each 10-minute time interval, the operating state comprising at least running, idle and transition, wherein the transition is further subdivided into start-up, normal stop and emergency stop.

[0056] In the above preferred embodiments, the state detection unit is a logical variable given by the internal control system of the operational wind turbine.

[0057] Furthermore, each operation status detected at each interval is verified using data extracted from SCADA at every 10-minute time interval, and uncertain or erroneous data is discarded.

[0058] The number of times a wind turbine changes state and the maximum amount of time the wind turbine spends in a given operating state during each time interval are detected and marked for each time interval.

[0059] After performing these steps, collecting actual and pure data for each time interval, and taking into account the time of the operating state and the transition events, the aeroelastic simulation model for each identified operating state and its wind condition should be executed for a specific turbine model to identify the fatigue load for each operating state, wind condition (speed and turbulence intensity). In a preferred embodiment, multiple hyper-surface response models are executed for each wind condition and each operating state of the wind turbine model. Therefore, the actual occurring loads and their duration and transition events are accurately identified based on the simulation and / or hyper-response surface model and the method for collecting data disclosed herein.

[0060] This process is repeated for each time interval of the wind turbine starting from the commissioning date.

[0061] like Figure 1 As shown, in a preferred embodiment, the method further includes a plurality of hyper-dimensional response surface models that relate wind and operating conditions to fatigue equivalent loads for each operating state calculated in the aeroelastic simulation. For example, Figure 3An exemplary representation of the hyper-dimensional response surface model is shown, which relates wind speed and turbulence intensity to the fatigue equivalent angular moment at the low-speed main shaft for the operating state RUN.

[0062] As described above, in a preferred embodiment, the method comprises calculating a fatigue equivalent load for each operating state and each time interval and for each component of the wind turbine. The rain flow counting method is used for structural components and the load revolution distribution method is used for all bearings and gear components. The load is the result of each experiment at each time interval and it is the equivalent load that should produce the same damage as the dynamic load at each time interval.

[0063] Therefore, a historical time series consisting of the mean values ​​of wind conditions at each time interval and the corresponding fatigue equivalent loads starting from the commissioning date is modeled.

[0064] Furthermore, reliability analysis or remaining useful life analysis is performed for each component of the wind turbine using the historical time series of operating conditions and the fatigue equivalent load at each interval.

[0065] Hyperdimensional response surface models explore the relationship between multiple explanatory variables and one or more response variables. In this case, the response surface model explores fatigue equivalent load under wind conditions. There are multiple techniques to build response surface models based on a given set of execution experiments, such as Taylor series, radial basis functions, and Kriging. The present invention does not specify which technique to use, but suggests that various techniques should be tested to identify the best representation of the correlation. The process of creating a response surface model should be repeated for all operating states (running, idle, transition) for the same or similar wind conditions.

Claims

1. A method for estimating remaining useful life of a component of an operational wind turbine model, performed by a computing element, wherein the computing element is communicatively coupled to the operational wind turbine, the operational wind turbine comprising a data acquisition module configured to measure real historical data of the operational wind turbine, and an additional state detection unit configured to identify historical states of operation, wherein the method comprises: - extracting historical data from the data acquisition module at selected time intervals, the data comprising at least wind speed, blade pitch position and rotor speed; - identifying, by the state detection unit, an operating state of the wind turbine for each selected time interval, the operating state comprising at least running, idle and transition states, the transition comprising starting, normal stop and emergency stop; - verifying the identified operating state using the data extracted from the data acquisition module at each time interval, identifying and discarding unmatched uncertain data; - identifying a number of transient events including the number of times the wind turbine has changed state; - identifying a dominant operating state, the dominant operating state comprising the state in which the wind turbine has spent the longest amount of time over its selected time interval; - identifying wind conditions, said wind conditions comprising at least average wind speed and turbulence intensity at said selected time interval; - Repeat the previous steps for multiple time intervals, wherein the method further comprises performing a plurality of simulations for the selected plurality of time intervals to estimate fatigue equivalent loads, the simulations comprising obtaining loads for the wind turbine model corresponding to the following wind conditions; - the dominant operating state identified at each time interval and its duration, and - Transient events identified at each of their time intervals.

2. A method for estimating the remaining useful life of a component of an operational wind turbine model, performed by a computing element, according to claim 1, comprising a plurality of hyper-dimensional response models, each of said response models comprising, for an operational state, all fatigue loads of a load component, including forces and / or moments for a first position of said wind turbine model.

3. A method for estimating remaining useful life of a component of an operational wind turbine model performed by a computing element according to claim 1, comprising a plurality of fatigue load calculations at a plurality of time intervals from a commissioning date to estimate a historical time series fatigue equivalent load for at least one load component, including forces and / or moments for a first position of the wind turbine model.

4. The method for estimating remaining useful life of a component of an operational wind turbine model performed by a computing element according to claim 1, further comprising a reliability analysis or remaining useful life of at least one component of the wind turbine using historical operating conditions and fatigue equivalent loads.

5. The method for estimating remaining useful life of components of an operational wind turbine model performed by a computing element according to claim 1, wherein the simulation model is an aeroelastic simulation.

6. The method for estimating the remaining useful life of a component of an operational wind turbine model, performed by a computing element, according to claim 1, wherein the state detection unit is a logical variable given by an internal control system of the operational wind turbine.

7. The method for estimating remaining useful life of components of an operational wind turbine model performed by a computing element according to claim 1, wherein the data acquisition module includes generator power data at each time interval.

8. The method for estimating the remaining useful life of a component of an operational wind turbine model executed by a computing element according to claim 7, wherein the state detection unit is a separate code executed by a computing element that identifies the operating state from the data acquisition module.

9. The method for estimating remaining useful life of a component of an operational wind turbine model performed by a computing element according to claim 1, wherein the time interval is 10 minutes or less.

10. The method for estimating remaining useful life of components of an operational wind turbine model performed by a computing element as recited in claim 1, wherein said data acquisition module is a conventional Supervisory Control and Data Acquisition (SCADA) system of said wind turbine.

11. The method for estimating remaining useful life of components of an operational wind turbine model performed by a computing element according to claim 1, further comprising measuring wind conditions from a nearby meteorological tower at each time interval.

12. A system for estimating the remaining useful life of a component of an operational wind turbine model, comprising at least the wind turbine model, the wind turbine model comprising a data acquisition module configured to measure real historical data of the operational wind turbine, a state detection unit configured to identify historical states of operation, wherein the system comprises a computing element communicatively coupled thereto: - for extracting historical data from the data acquisition module at selected time intervals, the data comprising at least wind speed, blade pitch position and rotor speed; - for identifying, by means of the state detection unit, an operating state of the wind turbine for each selected time interval, the operating state comprising at least running, idle and a transition state, the transition state comprising starting, normal stop and emergency stop; - for verifying the identified operating state using data extracted from said data acquisition module at each time interval, Identify and discard unmatched uncertain data; - for identifying a number of transient events including the number of times said wind turbine has changed state; - for identifying a dominant operating state, the dominant operating state comprising the state in which the wind turbine has spent the longest amount of time over a selected time interval; - for identifying wind conditions, said wind conditions comprising at least average wind speed and turbulence intensity at said selected time interval; - Repeat the previous steps for multiple time intervals; as well as - performing a plurality of simulations for said plurality of time intervals to estimate fatigue equivalent loads, said simulations comprising obtaining loads for said wind turbine model corresponding to the following wind conditions; - the dominant operating state identified at each time interval and its duration, and - Transient events identified at each of their time intervals.

13. System for estimating remaining useful life of components of an operational wind turbine model according to claim 12, wherein the computing element is capable of executing the method according to any one of claims 1-11.

14. A computer program adapted to perform the method according to any of claims 1-11.

15. Computer program according to claim 14, embodied on a storage medium.

Images