Method and device for monitoring the condition of a machine in a machine system, in particular a wind turbine

Abstract

Method for monitoring the condition of a machine of a machine system (200), in particular a wind turbine, comprising the steps: - Provision of a time series of measured natural vibration spectra of the machine system (200) (S1), - Recording of a deformation parameter in at least one monitoring time interval (S2), wherein the deformation parameter is characteristic for a deviation of the measured natural frequency spectra from a reference natural frequency spectrum of at least one reference machine system, - Acquisition of a noise parameter for at least one monitoring time interval (S2), wherein the noise parameter is characteristic of the noise of the measured eigenvibration spectra, and - Determination of the machine state from the deformation parameter and the noise parameter (S4).

Description

[0001] The invention relates to a method and a monitoring device for monitoring the condition of a machine system, in particular a wind turbine, a vehicle, or a structure, based on measured vibration data of the machine system. Applications of the invention are particularly evident in the monitoring and detection of faulty operating states of wind turbines or other machine systems.

[0002] It is well known that the condition of machinery can be recorded using vibration measurements (see, for example, DE 103 00 947 A1 or DE 10 2018 119 679 A1). In machinery, particularly in wind turbines, both machine operating vibrations and structural vibrations occur. Machine operating vibrations are resonantly excited by periodically moving, particularly rotating, components of the machinery. The frequency range of machine operating vibrations (typically in the Hz to kHz range) is determined by the operating frequencies of the machinery and their higher harmonics. The structural vibrations are the natural frequencies (or eigenvibrations) of the machinery, which include all resonant vibrations except for the vibrations caused by dynamic excitation during machine operation and are typically below the operating frequencies of the machinery in the range of mHz to kHz.For example, the natural vibrations of the tower bending of a wind turbine fall in the range of 0.1 Hz to 10 Hz and the torsional vibrations of a ship's drive shaft fall in the range of 1 Hz to 100 Hz.

[0003] A wind turbine typically comprises a tower with a nacelle that supports a rotor and absorbs a generator for converting the rotor's mechanical rotational energy into electrical energy. Slow structural vibrations of a wind turbine include, for example, tower bending vibrations, which occur analogously to the two mechanical bending vibrations of a rod, and nacelle vibrations, which in particular include pitch vibrations, rolling vibrations laterally around the rotor axis, and torsional vibrations, of which, in addition to the fundamental vibrations, the first excited natural vibrations can be measured. Structural vibrations are excited by all forces acting on the wind turbine, so that stochastic force effects (stochastic excitations), such asthe wind load on the tower, generate random resonance curves characterized by random fluctuations of resonances below the envelope and whose mean amplitudes as a function of frequency are usually Breit-Wigner curves (Lorentz curves), with current realizations being subject to strong random local fluctuations.

[0004] Mechanical changes to the structure of a machine plant affect structural vibrations, particularly by changing the mean amplitudes of natural frequencies and / or deforming the spectrum of structural vibrations. For example, a bolt securing the nacelle to the tower of a wind turbine can crack, altering the nacelle vibrations. Crack growth results in a time-dependent deformation of the spectrum of structural vibrations.

[0005] There is therefore an interest in recording structural vibrations in order to detect changes in the structure of a machine system at an early stage and, if necessary, to carry out maintenance or repair measures in a timely manner. While it is known to analyze structural vibrations in simple structures, such as antenna masts, additional measured environmental data must be taken into account (see DE 10 2018 119 679 A1). Complex machine systems, such as wind turbines, are characterized by highly noisy structural vibration spectra, which are also overlaid by systematic, but speed-dependent excitations of the machine dynamics, as exemplified in Fig. 3A is recognizable.

[0006] Such highly noisy spectra can only be evaluated using conventional methods with severely limited reliability, limited significance, with a restriction to particularly pronounced changes, and / or with the application of additional statistical expertise, so that the need for maintenance or repair is either not detected or not detected in a timely manner. This limited significance means, for example, that a statement about the type of damage or at least the cause of the damage (material aging, broken bolts, or the like) is largely impossible. A further problem with conventional methods arises from the fact that each spectrum only comprises a short-term measurement with relatively constant environmental parameters, and long-term changes to a machine system can only be detected to a limited extent.

[0007] The problems mentioned occur not only in the monitoring of wind turbines, but also in other machinery, such as watercraft or bridges.

[0008] The object of the invention is to provide an improved method and an improved monitoring device for monitoring the condition of a machine system, such as a wind turbine, based on measured vibration data of the machine system, which avoids limitations of conventional techniques. The monitoring method and the monitoring device are intended to be characterized in particular by increased reliability, improved sensitivity, and / or increased informative value, enable simplified output of monitoring results, and / or enable new or expanded applications of vibration-based monitoring.

[0009] This object is achieved by a method and / or a monitoring device for monitoring a machine condition of a machine system, which has the features of the independent claims. Preferred embodiments and applications of the invention are set out in the dependent claims.

[0010] According to a first general aspect of the invention, the above object is achieved by a method for monitoring a machine state of a machine installation, in particular a wind turbine, comprising the steps of providing a time series of measured natural vibration spectra of the machine installation, detecting a deformation parameter in at least one monitoring time interval, wherein the deformation parameter is characteristic of a deviation of the measured natural vibration spectra from a reference natural vibration spectrum of at least one reference machine installation, detecting a noise parameter for the at least one monitoring time interval, wherein the noise parameter is characteristic of noise in the measured natural vibration spectra, and determining the machine state from the deformation parameter and the noise parameter.

[0011] According to a second general aspect of the invention, the above object is achieved by a monitoring device which is configured to monitor a machine state of a machine installation, in particular a wind turbine, and which comprises a measuring device which is configured to provide a time series of measured natural vibration spectra of the machine installation, an analysis device which is configured to detect a deformation parameter which is characteristic of a deviation of the measured natural vibration spectra from at least one reference natural vibration spectrum of at least one reference machine installation in at least one monitoring time interval and to detect a noise parameter which is characteristic of noise in the measured natural vibration spectra in the at least one monitoring time interval, and an evaluation device,which is configured to determine the machine condition from the deformation parameter and the noise parameter. Preferably, the monitoring device or one of its embodiments is configured to carry out the method according to the first general aspect of the invention or one of its embodiments.

[0012] According to preferred embodiments of the invention, the monitoring device may comprise programmable logic controllers, programmable logic controllers, and / or at least one FPGA (field programmable gated array) unit with which the analysis device and the evaluation device are implemented.

[0013] Further independent objects of the invention are a data processing device comprising a computer device configured to carry out the method according to the first general aspect of the invention or one of its embodiments, in particular the detection of the deformation parameter and the noise parameter and the determination of the machine state, and a computer program product comprising instructions which, when the program is executed by a computer, cause the computer to carry out the method according to the first general aspect of the invention or one of its embodiments, in particular the detection of the deformation parameter and the noise parameter and the determination of the machine state.

[0014] The machine installation is preferably a wind turbine or another machine equipped with a periodically operating motor and / or generator, such as a ship, or a structure that can be excited to natural vibrations by a periodically operating machine, such as a bridge, a building, in particular a tower, a dam or a tunnel structure.

[0015] The term “machine condition” generally refers to the condition of the mechanical configuration of the machinery, in particular to mechanical properties of the machinery at the time of monitoring the machinery condition, such as material properties, in particular strength and elastic properties, of the machinery and / or its components and / or structural properties, in particular the integrity of materials and / or connections of components, of the machinery, and / or the temporal development of at least one of these properties. The machine condition can, for example, comprise the property of at least one material and / or at least one component of the machinery to be free from embrittlement or cracks. Quantitative measures of the machine condition are provided by the natural vibration spectra, by spectral features of the natural vibration spectra, such as, for example,The position of maxima, the amplitudes of maxima, the spectral width of maxima, and / or the spectral shape, and / or by stochastic properties of the natural vibration spectra, and / or their temporal development. The determination of the machine condition of the machine system is based on the acquisition of the deformation parameter and the noise parameter from at least one of these quantitative measures. Advantageously, the machine condition of the machine system can be acquired independently of its operating state, which characterizes the current machine operation, such as running under partial or full load.

[0016] Determining the machine condition of the machine system may include outputting status information, e.g., in text form and / or graphic form and / or as an optical and / or visual signal, and / or diagnosing faults, particularly material and / or structural defects in the machine system. A fault-free and error-free machine condition is also referred to as a regular or normal machine condition. Deviations from the fault-free and error-free machine condition are also referred to as an irregular or abnormal machine condition. Monitoring the machine system, and in particular determining the machine condition, can be carried out intermittently, at selected times, or continuously, during the operation of the machine system.

[0017] The provision of the time series of measured natural vibration spectra of the machine system can preferably comprise a vibration measurement of raw vibration data with at least one vibration sensor, in particular for detecting surface vibrations, particularly preferably a plurality of vibration sensors, such as at least one strain sensor and / or acceleration sensor, a conversion of the raw vibration data into amplitude spectra and a separation of the natural vibration spectra from a mostly noise-induced background and / or contributions of the typically speed-dependent machine dynamics.

[0018] Natural vibration spectra can encompass all frequencies or at least a sub-frequency range of the mechanical vibrations of the machine system, except for the dynamically excited machine operating vibrations and their higher harmonics. The machine operating vibrations and their higher harmonics can be determined using specified operating parameters of the machine system or by measuring the machine operating vibrations (speed measurement) during operation of the machine system. The optional speed measurement can be performed continuously or at specific sampling times.

[0019] Providing the time series of measured natural vibration spectra involves separating the natural vibrations from the raw data. The time series of measured natural vibration spectra can be provided locally directly at the machine system and / or via a data connection to a monitoring device located remotely from the machine system.

[0020] The deformation parameter recorded according to the invention is a quantitative deformation measure of the deviation of the natural vibration spectra from a reference natural vibration spectrum (standard spectrum) of one or more (e.g., up to 10 or more) reference machine(s) in at least one monitoring time interval. The monitoring time interval is a time range (time window) from which recorded natural vibration spectra are taken into account and which includes the measurement times of the associated natural vibration spectra and, optionally, measurement pauses between the measurements. The measurement time of an individual natural vibration spectrum can be selected such that, for example, a frequency range of 10 -2 Hz to about 10 4Hz and / or a time interval in the range of a few seconds up to typically 10 minutes is covered. When monitoring tower vibrations in the frequency interval 0.1 Hz to 10 Hz, measurement times of, for example, 5 minutes to 10 minutes with a clock rate of the individual vibration measurements of 0.01 seconds are useful for acceptable resolution. The duration of the measurement break can, for example, be up to 3 hours each. A number of natural vibration spectra are measured in the monitoring time interval. The number of natural vibration spectra per monitoring time interval is selected such that the deformation parameter can be recorded with sufficient statistical certainty. Preferably, 10 to 100, e.g. 40 to 60, natural vibration spectra are measured per monitoring time interval. The duration of the monitoring time interval is selected, for example, in the range of 3 hours to 2 days.With continuous monitoring, the monitoring time interval is shifted along the time series of measured natural vibration spectra as a continuous monitoring window, resulting in sequences of time-varying spectra. The duration of the continuous monitoring can be, for example, months or years.

[0021] The reference machine can preferably be another machine with the same or a similar mechanical configuration as the monitored machine. A similar mechanical configuration is given, for example, if the reference machine, in its components, predominantly has the same design and / or construction as the monitored machine. The reference machine or one of several reference machines can, in particular, be the monitored machine itself in a regular machine state in the past. The reference machine is characterized by a fault- and error-free machine state, in particular by vibration characteristics during intended operation, without stochastic external forces, such as gusts of wind.

[0022] Alternatively, the (or one of the) reference machinery(s) may comprise an idealized, fault-free machinery. Accordingly, the deformation parameter may represent a deviation of an arithmetically averaged resonance curve of the monitored machinery from a standard shape (e.g., Breit-Wigner shape).

[0023] The deformation parameter can be determined as the deviation of the natural vibration spectra from a reference natural vibration spectrum by arithmetic subtraction of the spectra. The reference natural vibration spectrum (standard spectrum) of several reference machines can be calculated by arithmetic averaging or, in the case of reference machines characterized by minor technical differences, by weighted averaging of the reference natural vibration spectra of the reference machines. Alternatively, the spectra of the reference machines can be considered as fixed points in a function space (Hilbert space), so that the change in the distances of a vibration spectrum of the machine under investigation from these spectra can be used as a measure of the machine condition of the machine in question and thus as a deformation parameter.

[0024] A reference natural vibration spectrum can be based, in particular, on historical data from the other and / or monitored machine system, assuming that these were recorded in a regular machine condition. This assumption can be based, for example, on a user-based decision or on a numerical analysis of reference natural vibration spectra from the past and the selection of a set of identical reference natural vibration spectra.

[0025] The deformation parameter quantitatively represents how strongly the natural vibration spectrum deviates from the reference natural vibration spectrum and thus represents a distance measure to the reference systems or historical states of the system under consideration.

[0026] The reference natural vibration spectrum is characterized by lower noise than the natural vibration spectra. The noise parameter (disorder parameter) measured according to the invention is a quantitative disorder measure that characterizes the stability and statistical reliability of the temporal evolution of the natural vibration spectra. The noise parameter is generally smaller in regular operation than in irregular operation, i.e., the noise parameter increases when natural vibrations are excited more strongly than in regular operation, e.g., due to mechanical instabilities, and in particular, the spectral width of the maxima at the natural frequencies increases.

[0027] Advantageously, the inventor has found a combination of two mathematically meaningful indicators, the deformation parameter and the noise parameter, which together allow a reliable and reproducible, optionally automated, quantitative evaluation of the natural vibration spectra. For each monitoring time interval, the noise parameter provides information about the noise in the natural vibration spectrum, e.g., caused by stochastic external influences. The noise parameter improves the evaluation of the deformation parameter in such a way that changes in the natural vibration spectra can be more reliably attributed to non-critical stochastic effects or to the desired structural changes in the machine system. In other words, a deformation of the natural vibration spectra caused by non-critical noise not only leads to the identification of an irregular machine condition.The deformation parameter and the noise parameter respond to small changes in natural vibration spectra, so that irregular operating conditions can be measured with increased sensitivity and reliability compared to conventional methods.

[0028] A further advantage is that the deformation parameter and the noise parameter are generally valid indicators, which are determined primarily on a functional theory or functional analysis (mathematical) basis and essentially do not depend on learning data or expert knowledge. The indicators react to deviations from the typical appearance of the natural vibration spectra and to increased measurement inaccuracies.

[0029] Other machine system faults, such as sensor or machine defects, can also be detected and, if necessary, trigger an alarm. The deformation parameter and the noise parameter allow for simplified output of monitoring results.

[0030] According to a preferred embodiment of the invention, the deformation parameter may comprise a distension of an amplitude-time function of the measured natural vibration spectra in at least one interval of vibration frequencies with respect to a reference amplitude-time function of the reference natural vibration spectra of the at least one reference machine installation in the at least one interval of vibration frequencies.

[0031] According to a further preferred embodiment of the invention, the noise parameter may comprise a distension of an amplitude-time function of the measured natural vibration spectra in at least one interval of vibration frequencies with respect to a smoothed amplitude-time function of the measured natural vibration spectra in the at least one interval of vibration frequencies.

[0032] The interval of vibration frequencies considered when calculating the distension to determine the deformation or noise parameter is chosen so that it contains a maximum (peak) of the natural vibration spectrum. If the natural vibration spectrum has several maxima, i.e. several characteristic natural vibrations, the distension is determined accordingly in several intervals of vibration frequencies, each containing one of the maxima. The calculation in a single interval of vibration frequencies with a single maximum of the natural vibration spectrum can be provided, for example, if only a single natural vibration is excited and / or is sufficient for assessing the machine condition of the machinery (e.g. if only one natural vibration of the nacelle is to be monitored) and / or if a greater risk of error can be tolerated when assessing the machine condition of the machinery.

[0033] Distension is a quantitative measure of the distortion of the amplitude-time function, in particular including scaling (enlargement or reduction), stretching (or compression) and / or inflation of the amplitude-time function. Determining distension can involve calculating a distance, such as a Euclidean distance, a Hamming distance or another distance measure, between the amplitude-time function and the reference amplitude-time function or the smoothed amplitude-time function at the predetermined oscillation frequencies for each measured eigenfrequency spectrum of the time series and summing the distances over the time series. Distension can be determined in particular as described, for example, in “Control Theory in Physics and other Fields of Science” by Michael Schulz (Springer, 2005) or in “Computational Intelligence” by Rudolf Kruse et al. (Vieweg + Teubner, 2011).

[0034] The acquisition of the distension of the amplitude-time function of the measured natural vibration spectra to determine the deformation parameter and / or the noise parameter has the advantage that the distension provides the deformation parameter and / or the noise parameter with particularly high sensitivity and reliability.

[0035] Advantageously, the acquisition of the deformation parameter and the noise parameter and the determination of the machine condition of the machine system can be repeated continuously at new monitoring time intervals, with the provision of each currently measured natural vibration spectrum of the time series of measured natural vibration spectra. This embodiment of the invention can have particular advantages for the continuous monitoring of a machine system, in particular a wind turbine.

[0036] Preferably, the analysis device of the monitoring device can be configured to detect the distension for determining the deformation parameter and / or the noise parameter, e.g. continuously.

[0037] According to a further advantageous variant of the invention, the provision of each natural vibration spectrum of the time series of the measured natural vibration spectra can comprise the steps of measuring raw vibration data with a plurality of vibration sensors arranged on the machine system for vibration measurement, converting the raw vibration data into vibration spectra of the machine system, wherein the vibration spectra contain dynamically excited machine vibrations and natural vibrations of the machine system, and filtering the vibration spectra to eliminate the dynamically excited machine vibrations, whereby the measured natural vibration spectra are obtained.

[0038] The vibration sensors are part of the measuring system of the monitoring device. A conversion device of the monitoring device can be configured to convert the raw vibration data, and a filter device of the monitoring device can be configured to filter the vibration spectrum.

[0039] Optionally, smoothing of the filtered vibration spectrum can also be provided. Smoothing can be achieved by repeatedly measuring the natural vibration spectra, for example, over a period of up to four hours in the case of a wind turbine, and averaging the natural vibration spectra. This smoothing advantageously reduces the effects of wind on the natural vibration spectra.

[0040] The number of vibration sensors can be selected depending on the specific application conditions and the monitoring task. In the case of a wind turbine, at least 5, preferably 8 to 15 vibration sensors are used for vibration measurement. These sensors are located, for example, on the main bearing, the generator, the tower, and in the gear stages of the generator gearbox.

[0041] Particularly preferably, the filtering of the vibration spectrum can comprise applying a Kalman filter to the vibration spectra. The Kalman filter is a time series filter applied to amplitude-time functions at given frequencies of consecutive natural vibration spectra. The Kalman filter is constructed, for example, as described in "Control Theory in Physics and other Fields of Science - Concepts, Tools, and Applications" by Michael Schulz (Springer, 2005) or in "Optimal Control" by Arturo Locatelli (Birkhäuser, 2001). The Kalman filter, which can be a temporally discrete or continuous filter, distinguishes between measurement-related observation errors and deviations in the machine and structural dynamics resulting from disturbances or noise influences and is therefore particularly effective at removing random influences from the vibration spectra.Advantageously, the Kalman filter eliminates both noise contributions and residual machine dynamics (machine operating vibrations, i.e. speed, and higher harmonics) from the natural vibration spectra with high efficiency.

[0042] Alternatively, the provision of each natural vibration spectrum of the time series of the measured natural vibration spectra can occur while no machine vibrations are periodically excited by the machine of the machine system. In this embodiment of the invention, the provision of each natural vibration spectrum can thus comprise a measurement of raw vibration data with a plurality of vibration sensors arranged on the machine system for vibration measurement while the machine system is in a state without dynamically excited machine vibrations, and a conversion of the raw vibration data into vibration spectra of the machine system, wherein the vibration spectra form the measured natural vibration spectra. The state without dynamically excited machine vibrations is present, for example, when a machine system is switched off, e.g.a wind turbine is at a standstill or in a spin mode, or a machine system is being monitored without an exciting machine.

[0043] Even in the case of the non-dynamically periodically excited machine system, smoothing and / or filtering of the vibration spectra can optionally be provided, which includes the application of the Kalman filter to the vibration spectrum, whereby an additional smoothing of the natural vibration spectra is advantageously achieved.

[0044] According to a particularly preferred embodiment of the invention, a multidimensional evaluation parameter is determined from the deformation parameter and the noise parameter, whereby the machine condition is determined from the multidimensional evaluation parameter. The determination of the multidimensional evaluation parameter advantageously facilitates the determination of the machine condition of the machine system. The multidimensional evaluation parameter allows a direct assessment of whether a machine condition is regular or irregular, or whether a critical intermediate condition exists.

[0045] Preferably, the monitoring device may include an evaluation device configured to determine the multidimensional evaluation parameter from the deformation parameter and the noise parameter and to determine the machine state from the evaluation parameter.

[0046] Advantageously, various options are available for recording the multidimensional evaluation parameter. According to a first variant, the evaluation parameter can comprise a position of the deformation parameter and the noise parameter, which are recorded at a common monitoring time interval, in an at least two-dimensional evaluation field. The evaluation field is, for example, a flat map with two coordinates, which are represented by the deformation parameter and the noise parameter. For example, a Cartesian coordinate system can be provided, the axes of which represent the deformation parameter and the noise parameter. Alternatively, other coordinate systems can be used and / or by incorporating other current status parameters of the machine system, such as SCADA data (control data) collected for monitoring energy systems and / or information from the system control and / or available environmental data or the like, such asthe oil temperature of gear oil, a reactance of the generator, a rotational speed of the generator, the position of rotor blades and / or weather data, a higher-dimensional evaluation field (higher-dimensional coordinate system) can be realized.

[0047] The evaluation parameter provides a current position and, preferably with continuous monitoring, a trajectory within the evaluation field. Different areas within the evaluation field can be assigned, for example, to a regular machine state or an irregular machine state, or optionally, to a critical intermediate state. The machine state can be determined by recording the position or, optionally, the trajectory within one of the areas within the evaluation field.

[0048] Furthermore, various sub-areas within the evaluation field, representing an irregular machine condition or, optionally, a critical intermediate condition, can be assigned to specific malfunctions of the machine system or its components, so that the current position and, preferably, the trajectory within the evaluation field also enables a diagnosis of which components of the machine system exhibit an irregularity. This diagnosis can be based, for example, on reference measurements on comparable systems and / or empirical values.

[0049] Alternatively or additionally, according to a further variant, the evaluation parameters can comprise at least a two-dimensional functional of the deformation parameter and the noise parameter, which are recorded at a common monitoring time. Advantageously, the functional allows the immediate classification of the machine state as regular or irregular, or optionally as a critical intermediate state.

[0050] According to another particularly preferred embodiment of the invention, the determination of the machine status can include a classification of the evaluation parameter, e.g., using a classification device of the monitoring device, by comparing it with predetermined parameter ranges, e.g., the ranges in the evaluation field or predefined value range intervals of the functional, and an output of the machine status, e.g., using an output device of the monitoring device, depending on the result of the classification. The output of the machine status can be, e.g., in text form or in graphic form or as an optical and / or visual signal, in particular an alarm, and / or can include a diagnosis of errors in the machine system.

[0051] Advantageously, the classification allows the set of machine system states to be split into at least two, e.g., two, three, or more subsets, such as Normal, Critical, and Alarm. For example, all known or detectable anomalies can be verified as critical and / or alarm-worthy, and all operating states that appear normal can be assigned to the Normal class.

[0052] Further details and advantages of the invention are described below with reference to the accompanying drawings. The drawings schematically show: Fig. 1: Features of methods for monitoring a machine state of a machine system according to embodiments of the invention; Fig. 2: Features of devices for monitoring a machine state of a machine system according to embodiments of the invention; Fig. 3: Examples of practical applications of the invention; and Fig. 4: a visualization of an evaluation parameter in a two-dimensional evaluation field.

[0053] Features of embodiments of the invention are described below with exemplary reference to the monitoring of a wind turbine 200, which is shown schematically in Fig. 2 and comprises a tower 210, a rotor 220 and a nacelle 230. The wind turbine 200 is provided with a plurality of vibration sensors 11, each of which monitors a measuring channel of the measuring device 10 of the monitoring device 100 according to Fig. 2. The application of the invention is not limited to the monitoring of a wind turbine, but can be applied to other machinery. Properties of wind turbines, their operation, especially control, and vibration measurement on wind turbines are not explained, as these are known per se from conventional wind turbines.

[0054] Fig. 1 shows a flowchart with steps of the method for monitoring the machine condition of the wind turbine 200. According to various embodiments of the invention, all steps in combination or a portion of the steps (e.g., the method shown without step S3) or additional steps (e.g., for reliability assessment) may be provided, as explained below. The method is preferably executed continuously during the operation of the wind turbine 200, i.e., raw vibration data collected over years of service life can be evaluated and used for monitoring.

[0055] The procedure according to Fig. 1 is carried out for at least one of the vibration sensors 11. A single vibration sensor 11 can be used for monitoring, or several or all vibration sensors 11 can each be used separately for monitoring. Each of the vibration sensors 11 can be arranged to detect a different structural vibration of the wind turbine 200. An irregular machine state of the wind turbine 200 can be detected if the evaluation of the vibration data from at least one of the vibration sensors 11 results in the detection of a fault or if the evaluation of the vibration data from several or all of the vibration sensors 11 results in the detection of faults. Raw vibration data is measured with the vibration sensors 11 of the measuring device 10. The vibration sensors 11 are sampled at a rate of approximately 10 3 Hz, with approximately 10 7 Measuring points are recorded.

[0056] The procedure according to Fig. 1 is preferably used with the Fig. 2, which comprises the measuring device 10 with the vibration sensors 11, a conversion device 20, a filter device 30, an analysis device 40, an evaluation device 50 with a classification device 60, and an alarm device 70. The conversion device 20, the filter device 30, the analysis device 40, and the evaluation device 50 are preferably provided by one or more computer units, such as one or more FPGAs. For each measuring channel, i.e. for each vibration sensor 11, a separate conversion device 20, a filter device 30, an analysis device 40, and an evaluation device 50 can be provided. Alternatively, the conversion device 20, the filter device 30, the analysis device 40, and the evaluation device 50 can each contain separate data processing channels for the measuring channels.The alarm device 70 comprises, for example, at least one visual alarm and / or at least one audio alarm. Different alarm signals can be provided for different faults, e.g., in different components of the wind turbine 200, and / or different measurement channels.

[0057] According to Fig. 1, the monitoring of the wind turbine 200 initially comprises a step S1 of providing a time series of measured natural vibration spectra of the wind turbine 200. The time range covered by the acquired time series of measured natural vibration spectra is also referred to as the monitoring time interval. The subsequent acquisition of the deformation parameter and the noise parameter in step S2 refers to the monitoring time interval, which is selected, for example, in the range of 3 hours to 2 days. During ongoing monitoring, the monitoring time interval is shifted by a predetermined increment, e.g., by 3 to 4 hours, after each run (cycle) of the method (see below, step S8).

[0058] In step S1, the conversion device 20 converts the raw vibration data into vibration spectra of the wind turbine 200. The conversion comprises a frequency-amplitude analysis of the raw vibration data and the creation of data sets, each representing a vibration spectrum. Each vibration spectrum initially contains dynamically excited machine vibrations, i.e., the periodic excitations due to the generator's rotational speed, and the natural vibrations of the wind turbine 200.

[0059] In step S1, the filter device 30 extracts the natural vibrations of the wind turbine 200 from the vibration spectra data sets, since the inventive monitoring of the wind turbine 200 is based on the qualitative assessment of the natural vibrations. The dynamically excited vibrations and their higher harmonics are eliminated using the filter device 30. The filtering of the vibration spectra is based, for example, on a known FFT (Fast Fourier Transformation) application and a subtraction of the dynamically excited machine vibrations and the higher harmonics in the frequency domain. The frequency of the dynamically excited machine vibrations is generally known from the operating conditions of the machine system and, in particular, from a speed measurement on the generator of the wind turbine 200.

[0060] As a result of filtering out the dynamically excited machine vibrations, a time series of measured natural vibration spectra is provided, as exemplified in Fig. 3A, where f represents a linear frequency axis with frequencies, for example, in the range from 0 Hz to 1 kHz, A represents a linear amplitude axis of the vibration amplitudes occurring at the frequencies (relative units), and t represents a linear time axis over a period of, for example, 250 days (duration of continuous monitoring). The monitoring time interval is, for example, 1 day. Fig. Figure 3A illustrates, for example, that the obtained temporal evolution of measured eigenvibration spectra is initially characterized by noise, which would make it difficult to derive irregularities directly from the eigenvibration spectra.

[0061] First, unsmoothed amplitude-time functions are obtained from the natural vibration spectra at a variety of specified frequencies. Each amplitude-time function comprises amplitudes of the natural vibration spectra at the respective specified frequency within the monitoring time interval. For example, eight frequencies are selected in the recorded frequency range, for each of which a data set is determined that represents the corresponding amplitude-time function.

[0062] To reduce noise, the natural oscillation spectra, in particular the amplitude-time functions of the natural oscillation spectra, are preferably smoothed with the filter device 30 by applying a Kalman filter to the amplitude-time functions of the oscillation spectra. The Kalman filter is applied as a time series filter to each amplitude-time function, removing noise and determining a smoothed signal with the highest probability as a smoothed amplitude-time function.

[0063] After applying the Kalman filter, the natural oscillation spectra can be recovered from the smoothed amplitude-time functions with reduced noise. As a result of filtering with the Kalman filter, a time series of measured natural oscillation spectra with significantly reduced noise is provided, as exemplified in Fig. 3B or Fig. 3C is shown.

[0064] The Fig. 3B and Fig. 3C illustrate the beneficial effect of smoothing the eigenmode spectra with the Kalman filter. In contrast to the highly noisy image of the Fig. 3A, natural oscillations and their temporal development are clearly visible. According to Fig. 3B, the natural oscillation has a substantially constant frequency and relatively little change in amplitude, which allows the detection of a regular machine state of the wind turbine 200. According to Fig. 3C splits the natural oscillation into two partial oscillations, whereby the frequency of the new split natural oscillation changes over time, which allows the detection of a change in the mechanical configuration and thus an irregular machine state of the wind turbine 200.

[0065] Subsequently, in step S2, the analysis device 40 determines the deformation parameter of the natural vibration spectra as a distension of the smoothed amplitude-time functions of the natural vibration spectra in an interval of vibration frequencies around one or more maxima of the natural vibration spectra with respect to a reference amplitude-time function of a reference natural vibration spectrum of at least one reference wind turbine in the interval of vibration frequencies.

[0066] Furthermore, in step S2, the analysis device 40 determines the noise parameter of the natural vibration spectra as a distension of the unsmoothed amplitude-time functions of the natural vibration spectra in the interval of vibration frequencies around the one or more maxima of the natural vibration spectra with respect to the previously calculated smoothed amplitude-time function of the measured natural vibration spectra in the interval of vibration frequencies.

[0067] To calculate the distension, the deviations (distances) of the smoothed amplitude-time functions from the reference amplitude-time function or the deviations of the unsmoothed amplitude-time functions from the smoothed amplitude-time function are summed.

[0068] If the smoothed amplitude-time functions change more significantly with respect to the reference amplitude-time function, a greater distension or a larger deformation parameter is determined than if the smoothed amplitude-time functions change less. Accordingly, the deformation parameter directly provides a measure of the change in the natural vibration spectra during the monitoring time interval. Furthermore, if the unsmoothed amplitude-time functions exhibit greater noise with respect to the smoothed amplitude-time function, a greater distension or a larger noise parameter is determined than if the noise is less. Accordingly, the noise parameter directly provides a measure of the stochastic disorder of the natural vibration spectra during the monitoring time interval.

[0069] In the optionally provided step S3, a multidimensional evaluation parameter is determined from the noise and deformation parameters using the evaluation device 50. An example of the evaluation parameter is the position in a two-dimensional evaluation field 51, as shown schematically in Fig. 4. The dimensions of the evaluation field 51 are the noise and deformation parameters. The evaluation parameter is given by the position of the current noise and deformation parameters in the evaluation field 51, and it can be displayed on a display of the monitoring device 100. Step S3 is not mandatory. The noise and deformation parameters can also be used directly to determine the machine state of the wind turbine 200. Furthermore, another evaluation parameter can be used, such as a functional based on the noise and deformation parameters.

[0070] Subsequently, the current machine state of the wind turbine is determined in step S4 by classifying the evaluation parameter using the classification device 60. For example, it is determined whether the current evaluation parameter falls into one of the three areas "Regular," "Critical," or "Irregular" of the evaluation field 51 (see Fig. 4).

[0071] In addition to the classification, a reliability assessment can be performed in step S4. A reliability measure is determined, which, for example, quantifies the error probability of the evaluation parameter. The deviation of the error distribution of the measured machine system from the error distribution of the reference machine systems, estimated using the Kolmogorov-Smirnov method, can be used as a reliability measure, for example. The reliability measure allows appropriate measures to be taken on the wind turbine 200 in response to the detection of a critical or irregular machine condition. For example, in the event of an alarm with a high error probability, the operation of the wind turbine 200 can initially be continued automatically in order to verify the current monitoring result, or with a low error probability, an emergency shutdown of the wind turbine 200 can be provided.

[0072] If a deviation from the normal state is detected in step S5, i.e., if the evaluation parameter falls, in particular, into the "Critical" or "Irregular" range of the evaluation field 51, step S6 is followed by an output of the result of the machine state determination and an alarm via the alarm device 70. The alarm via the alarm device 70 can contain various signals depending on the detected state. Subsequently, regardless of the alarm, monitoring of the wind turbine 200 can be continued by postponing the monitoring time interval S7 and restarting the method with step S1 using the updated monitoring time interval. Alternatively, the method can be terminated, particularly in the event of an alarm-related termination of operation of the wind turbine 200.

[0073] In response to the alarm, for example, a detailed analysis of the vibration data, a change in the operating conditions of the wind turbine 200 (possibly with a shutdown) and / or maintenance of the wind turbine 200 by maintenance personnel may be provided.

[0074] Alternatively, if no deviation from the normal state is detected in step S5, i.e., if the evaluation parameter falls within the "Regular" range of the evaluation field, no alarm is triggered. Instead, in step S8, the result of determining the machine state is output, the monitoring time interval is shifted, and the process jumps to step S1 with the updated monitoring time interval. Alternatively, the method can be terminated, particularly in the case of a scheduled termination of operation of the wind turbine 200.

[0075] The result of determining the machine status in step S6 or S8 can be output, for example, in text form or preferably in the form of a color signal. The operating history can be displayed, for example, on a display of the monitoring device 100 or a separate control unit of the wind turbine 200 as a continuous strip with a color code, with different colors indicating different operating states.

[0076] The features of the invention disclosed in the above description, the drawings and the claims may be important both individually and in combination or sub-combination for the realization of the invention in its various forms.

Claims

[1] Method for monitoring a machine condition of a machine installation (200), in particular a wind turbine, comprising the steps: - Provision of a time series of measured natural vibration spectra of the machine system (200) (S1), - Recording a deformation parameter in at least one monitoring time interval (S2), wherein the deformation parameter is characteristic of a deviation of the measured natural vibration spectra from a reference natural vibration spectrum of at least one reference machine system, - detecting a noise parameter for the at least one monitoring time interval (S2), wherein the noise parameter is characteristic of a noise of the measured natural vibration spectra, and - Determination of the machine condition from the deformation parameter and the noise parameter (S4). [2] Method according to claim 1, in which - the deformation parameter comprises a distension of an amplitude-time function of the measured natural vibration spectra in an interval of vibration frequencies with respect to a reference amplitude-time function of a reference natural vibration spectrum of the at least one reference machine in the interval of vibration frequencies. [3] Method according to one of the preceding claims, in which - the noise parameter comprises a distension of an amplitude-time function of the measured natural vibration spectra in an interval of vibration frequencies with respect to a smoothed amplitude-time function of the measured natural vibration spectra in the interval of vibration frequencies. [4] Method according to one of the preceding claims, in which - the recording of the deformation parameter and the noise parameter and the determination of the machine condition are repeated continuously in new monitoring time intervals with the provision of each current measured natural vibration spectrum of the time series of measured natural vibration spectra. [5] Method according to one of the preceding claims, wherein the provision of each natural vibration spectrum of the time series of the measured natural vibration spectra (S1) comprises: - measuring raw vibration data with a plurality of vibration sensors arranged on the machine system (200) for vibration measurement, - converting the raw vibration data into vibration spectra of the machine system (200), wherein the vibration spectra contain dynamically excited machine vibrations and natural vibrations of the machine system (200), and - Filtering of the vibration spectra to eliminate the dynamically excited machine vibrations, whereby the measured natural vibration spectra are preserved. [6] Method according to claim 5, in which - the filtering of the vibration spectra involves applying a Kalman filter to the vibration spectra. [7] Method according to one of claims 1 to 4, wherein the provision of each natural vibration spectrum of the time series of measured natural vibration spectra (S1) comprises: - measuring raw vibration data with a plurality of vibration sensors arranged on the machine system (200) for vibration measurement while the machine system (200) is in a state without dynamically excited machine vibrations, and - Conversion of the raw vibration data into vibration spectra of the machine system (200), whereby the vibration spectra form the measured natural vibration spectra. [8] Method according to claim 7, comprising the step - Filtering of the vibration spectra, which involves applying a Kalman filter to the vibration spectra. [9] Method according to one of the preceding claims, comprising the step - Determination of a multidimensional evaluation parameter from the deformation parameter and the noise parameter, whereby the machine state is determined from the multidimensional evaluation parameter. [10] Method according to claim 9, wherein the multidimensional evaluation parameter comprises: - a position of the deformation parameter and the noise parameter, which are recorded at a common monitoring time interval, in an at least two-dimensional evaluation field (51), and / or - at least a two-dimensional functional of the deformation parameter and the noise parameter recorded at a common monitoring time. [11] Method according to one of claims 9 or 10, wherein the determination of the machine state (S4) comprises - Classification of the evaluation parameter by comparison with predetermined parameter ranges, and - Output of the machine status depending on the classification result. [12] Monitoring device (100) which is designed to monitor a machine state of a machine installation (200), in particular a wind turbine, comprising - a measuring device (10) which is designed to provide a time series of measured natural vibration spectra of the machine system (200), - an analysis device (40) which is configured to detect a deformation parameter which is characteristic of a deviation of the measured natural vibration spectra from at least one reference natural vibration spectrum of at least one reference machine system (200) in at least one monitoring time interval and to detect a noise parameter which is characteristic of noise in the measured natural vibration spectra in the at least one monitoring time interval, and - an evaluation device (50) which is arranged to determine the machine state from the deformation parameter and the noise parameter. [13] Monitoring device according to claim 12, in which - the analysis device (40) is configured to detect, as the deformation parameter, a distension of an amplitude-time function of the measured natural vibration spectra in an interval of vibration frequencies with respect to a reference amplitude-time function of the reference natural vibration spectra of the at least one reference machine system (200) in the interval of vibration frequencies. [14] Monitoring device according to claim 12 or 13, in which - the analysis device (40) is configured to detect, as the noise parameter, a distension of an amplitude-time function of the measured natural oscillation spectra in an interval of oscillation frequencies from a smoothed amplitude-time function of the measured natural oscillation spectra in the interval of oscillation frequencies. [15] Monitoring device according to one of claims 12 to 14, in which - the analysis device (40) is arranged to continuously repeat the detection of the deformation parameter and the noise parameter and the determination of the machine state in new monitoring time intervals with the provision of each current measured natural vibration spectrum of the time series of measured natural vibration spectra. [16] Monitoring device according to one of claims 12 to 15, comprising - a plurality of vibration sensors (11) of the measuring device (10) arranged on the machine system (200) for measuring raw vibration data, - a conversion device (20) which is designed to convert the raw vibration data into a vibration spectrum of the machine system (200), wherein the vibration spectrum contains dynamically excited machine vibrations and natural vibrations of the machine system (200), and - a filter device (30) which is arranged to filter the vibration spectrum in order to eliminate the dynamically excited machine vibrations, whereby the measured natural vibration spectrum is obtained. [17] Monitoring device according to claim 16, in which - the filter device (30) is arranged to filter the vibration spectrum by applying a Kalman filter to the vibration spectrum. [18] Monitoring device according to one of claims 12 to 15, comprising: - a plurality of vibration sensors (11) arranged on the machine system (200) for measuring raw vibration data while the machine system (200) is in a state without dynamically excited machine vibrations, and - a conversion device (20) which is designed to convert the raw vibration data into a vibration spectrum of the machine system (200), wherein the vibration spectrum forms the measured natural vibration spectrum. [19] Monitoring device according to claim 18, comprising - a filter device (30) which is arranged to filter the vibration spectrum by applying a Kalman filter to the vibration spectrum. [20] Monitoring device according to one of claims 12 to 19, comprising - an evaluation device (50) which is arranged to determine a multi-dimensional evaluation parameter from the deformation parameter and the noise parameter and to determine the machine state from the evaluation parameter. [21] Monitoring device according to claim 20, comprising - a classification device (60) which is arranged to classify the evaluation parameter by comparison with predetermined parameter ranges, and - an output device (70) which is arranged to output the machine status as a function of the result of the classification. [22] Monitoring device according to one of claims 12 to 21, comprising - programmable logic controllers, - programmable logic controllers, and / or - an FGPA unit. [23] A data processing apparatus comprising a computer device configured to carry out the method according to any one of claims 1 to 11. [24] A computer program product comprising instructions which, when executed by a computer, cause the computer to carry out the method according to any one of claims 1 to 11.

Images