Method and System for Monitoring Wind Turbine Generator Vibration

Technical field

The present invention relates to a method and system for monitoring vibration and, particularly, to a method and system for monitoring vibrations in a wind turbine generator.

Background art

Condition monitoring and fault diagnosis of wind turbine generators (WTG) are generally focused on transmission systems, and such a transmission system comprises a main shaft, a main gearbox and a generator and so on, with such components being assembled by corresponding bearings, and faults in the mechanical transmission components being mainly caused by inadequate lubrication, contamination, overload or inherent defects. The vibration signals obtained from the transmission system are very useful in determining the condition changes of a wind turbine generator, while the traditional vibration monitoring systems in wind turbine generators produce too many false alarms, the main reason is that the behavior pattern of a wind turbine generator is extremely unstable in comparison with gas/steam turbines, centrifugal compressors, fans and other rotary machines, and therefore it leads to questions of the reliability of the vibration monitoring systems for wind turbine generators currently available.

Since aerodynamic forces are highly nonlinear and unstable, the dynamic characteristics of a wind turbine generator vary significantly under different operation conditions. As for gearboxes in an unstable environment, the rotary speed, the torque and the forces acting on the gears are always changing, particularly in cases of wind

turbulence. A more complicated issue is that even in the case of controlled stable power generation the rotary speed and the torque of the rotor still change significantly. In addition, there are differences in the kinetic

characteristics of the various components of the rotor, the tower and the control system, which can produce significant influences on the transmission system, and these problems further affect the accuracy of the vibration monitoring. In actual operation, the statistical analysis of the vibration monitoring data shows significant unstable characteristics, and when a traditional vibration monitoring system is applied to a wind turbine generator, the above-mentioned unstable characteristics are highly likely to cause false alarms.

In operation, the rotary speed of the wind turbine generator rotor and the load of the transmission system are both unstable, and in terms of the rotary speed of the rotor, the pneumatic torque on the transmission system is related to the blade's tip-speed ratio, blade design, wind speed, blade pitch angle, yaw errors and any additional resistance on the blades, whereas the loads on the traditional system are also affected by many factors, for example random factors such as the stator/rotor currents, the wind turbulence, etc. In addition, some peculiar conditions, such as braking events, can generate random extra-large torques within a short time. In this way, even if the wind turbine generator is operating under the same wind speed and power, its normal vibration values will change when other operation data (such as wind deviation or the generator's stator current) change.

Therefore, the definition of the vibration threshold will still lead to high false alarm rate in vibration monitoring if various possible combinations of operation data are not taken into consideration.

Each false alarm from the vibration monitoring system will bring with it high costs, the wind turbine generator's operation has to be stopped for it to have a thorough inspection or even for it to be disassembled, and the loss in power generation and the costs of inspection during this period are both huge; furthermore, if a vibration monitoring system regularly sends out false alarms, the system will soon be shut down or neglected. These factors make it very

difficult to accurately monitor the vibrations in a wind turbine generator, not to mention to accurately diagnose faults.

The object of the present invention is to provide a method and system for monitoring vibrations in a wind turbine generator, which can determine adaptively the thresholds of the monitored parameters by taking into consideration various possible combinations of operation data, so as to reduce the false alarm rate, and at the same time to improve their operability and interpretability, and reduce the difficulty of maintenance. Another object of the present invention is to provide a vibration monitoring system for realizing the method of the present invention for monitoring vibrations in a wind turbine generator, which system is capable of comprehensively taking the influences of various factors on the wind turbine generator' s vibrations into consideration, so as to reduce the false alarm rate and the maintenance costs.

In order to achieve the above objects, the present invention proposes a method for monitoring vibrations in a wind turbine generator, which method comprises:

establishing a rule base for the range of vibration characteristic values;

acquiring real-time operation data of various operation parameters of the wind turbine generator;

predicting the range of the vibration characteristic values of the wind turbine generator according to the measured real-time operation data;

calculating the threshold of a corresponding vibration characteristic value according to the predicted range of the vibration characteristic values;

comparing said real-time operation data of said

corresponding vibration characteristic value with said threshold of the vibration characteristic value; and

sending out an alarm signal when said real-time operation data of said corresponding vibration characteristic value is larger than said threshold of the vibration characteristic value;

wherein said rule base for the range of the vibration characteristic values is established by a method as follows:

acquiring historical data of the various operation parameters and the vibration characteristic values of the wind turbine generator under normal operation, and storing the same in a database;

discretizing the data stored in the database; carrying out data mining to said discretized data based on a rough set, so as to constitute a knowledge system; since there are a large number of wind turbine generator parameters, and each set of data has uncertainty and

inaccuracy to some extent, a rough set can solve the problems well; and

extracting rules for the range of the vibration characteristic values of the wind turbine generator under normal operation so as to establish the rule base for the range of the vibration characteristic values. According to another method for monitoring vibrations in a wind turbine generator, in this case the above historical data of the various operation parameters include the

supervisory control and data acquisition (SCADA) data of the wind turbine generator and the operation data with the control variables under known normal conditions.

According to another method of the present invention for monitoring vibrations in a wind turbine generator, in this case the method for carrying out data mining based on the rough set is applied to carry out attribute reduction to a attribute set constituted by the various operation parameters of the wind turbine generator, so as to determine the minimum attribute set, thus simplifying the knowledge presentation and increasing the system processing efficiency and

facilitating a user' s decisions. The present invention also proposes a system for using the method of the present invention for monitoring vibrations in a wind turbine generator, which system comprises:

a rule base for the range of the vibration characteristic values comprising rules for the range of the vibration characteristic values;

a real-time data acquiring unit for acquiring the SCADA data and control variables of the wind turbine generator in operation;

a vibration characteristic value range predicting unit for predicting a range of the vibration characteristic values according to the rules in the rule base for the range of the vibration characteristic values and the data acquired in real-time operation;

a vibration characteristic threshold calculating unit for calculating the threshold of the vibration characteristic values of the wind turbine generator according to the predicted range of the vibration characteristic values;

a vibration characteristic value monitoring unit for monitoring the corresponding vibration characteristic value of the wind turbine generator and acquiring the real-time data of the vibration characteristic value;

a comparison unit for comparing the threshold of said vibration characteristic value and the real-time data of said vibration characteristic value; and

an alarm unit for sending out a corresponding alarm signal when said real-time data of said vibration

characteristic value are larger than said threshold of said vibration characteristic value.

In the method and system of the present invention for monitoring vibrations in a wind turbine generator, on one hand, since the threshold of the vibration characteristic values of the wind turbine generator is calculated on the basis of the mutual relationships of the automatically selected operation data, which can greatly reduce the false alarms under various normal operation conditions. On the other hand, since in the present invention the rule base for the range of the vibration characteristic values is

established on the basis of the historical data of the wind turbine generator by way of data mining based on the rough set, these rules are represented as a compact set of the automatically selected operation data, instead of the

operation data of all types. By means of the rule base and taking various combinations of the operation data into consideration, the threshold of the vibration characteristic values is calculated under various normal operation

conditions of the wind turbine generator, thus greatly reducing the false alarm rate.

Furthermore, in the method and system of the present invention for monitoring vibrations in a wind turbine generator, the rule base for the range of the vibration characteristic values is represented as explicit and

interpretable multiple rules and rules derived from tests, standards or maintenance experiences of other similar type wind turbine generators, which can be easily added into an existing rule base by maintenance personnel. Furthermore, by way of an automatic and periodical database knowledge discovery and upgrading mechanism, it enables "learning" continuously and automatically from the experiences in monitoring the wind turbine generator, and in this way, the accuracy of the system for monitoring vibrations is further increased and the false alarm rate further reduced over time. Description of the accompanying drawings

The following accompanying drawings are only intended for the exemplary illustrations and explanations of the present invention and are not to limit the scope of the present invention. In the drawings,

Fig. 1 is a flow chart for establishing a rule base of a wind turbine generator under normal operation conditions in the present invention;

Fig. 2 is a flow chart of the method of the present invention for monitoring vibrations in a wind turbine generator; and

Fig. 3 is a block diagram of the system of the present invention for monitoring vibrations in a wind turbine generator.

Exemplary embodiments

In order to understand the technical features, objects and effects of the present invention more clearly, particular embodiments of the present invention are described with reference to the accompanying drawings.

Fig. 1 shows the process of establishing a rule base for the range of vibration characteristic values. Before

monitoring vibrations, step SlO is carried out within a particular normal operation period of a wind turbine

generator (generally at least for six months or a year) to collect various relevant data and store the same in a database and then discretize them. Such data include the data of normal vibration characteristic values, the historical data of supervisory control and data acquisition (SCADA), known normal operation conditions and control variables of the wind turbine generator and so on, and such data are historical data for data mining of the operating conditions of the wind turbine generator.

The data of vibration characteristic values can be statistical parameters of vibration signals in the time domain, frequency domain or other domains, such as the virtual values of the vibration speed and time signals (10- 1000Hz) acquired at the high-speed end of the main gearbox. Generally speaking, the more historical data for data mining there are, the lower the possibility of false alarms in the established vibration monitoring system. Table 1 shows a set of attributes of the collected historical data, and of course this attribute set is also the set of attributes of the conditional data to be mentioned below. Table 2 shows a set of decision attributes of the range of the vibration

characteristic values for data mining, and this attribute set includes the data of the normal operating conditions of the wind turbine generator. As decision attributes, the ranges of the vibration characteristic values can be determined by current standards (such as German Standard VDI 3834) or by the operation experience of the wind turbine generator, for example, each normal vibration speed range in the standard VDI 3834 can be equally divided into 5 sections, and each section can be defined as a decision attribute in Table 2.

Table 1 Set of conditional attributes

Table 2 Set of decision attributes

At step S12, data mining based on a rough set is carried out on the data collected at step SlO in the data mining unit in Fig. 1 to construct a knowledge system. From the wind turbine generator' s operation data and the vibration

characteristic values measured during normal operation of the wind turbine generator, the knowledge system is defined as follows, i.e. the knowledge system is the decision table, which can be used for rule extraction:

DT=<U,CvD,V,f> (i) in which, DT is the knowledge system;

U is a finite set of N objects, e.g. the sample data obtained for N times from the wind turbine generator;

C is a conditional attribute set, e.g. the various wind turbine generator operation data as shown in Table 1;

D is a decision attribute set, e.g. the ranges of the vibration characteristic values as shown in Table 2;

V is defined as follows:

(2) wherein V _q is the domain set of the attribute g, ge ( CU D ) ;

f is defined as follows:

wherein f represents all of the decision functions, and for each g e ( C U D) and x e U,

f(*><l)eV, ₍₄₎

There are many data types in the operation parameters of a wind turbine generator, and each type of data has

inaccuracy and uncertainty to some extent, so the rough set method can solve the problem quite well.

Part of the attributes in a knowledge system can be redundant and reducible, and the reduction of the redundant attributes shall not lose necessary information. As one of the key contents of the rough set theory, attribute reduction can determine a smaller attribute set for a knowledge system, and the reduced attribute set contains the knowledge which is the same as or similar to the original attribute set.

Accordingly, a reduced set (RED(A)) of an original attribute set A can be defined as:

E = RED(A) <z> (E a AJND(E) = IND(A)) in which, E is the reduced attribute set of A;

IND ( ) represents the irresolvable relationship generated by the attribute set.

In the knowledge system DT (referring to equation (I)), if the reduced attribute set E can distinguish all objects distinguishable in the original attribute set A and E cannot be further reduced, then E is called the minimum attribute set. Hence it can be seen that after the attribute reduction, the minimum attribute set as the key part of the knowledge system has simplified the knowledge representation while it can still distinguish all the objects distinguishable in the original attribute set, so that the system processing efficiency is increased and user decision facilitated. On the basis of the data in Tables 1 and 2, a decision table for monitoring vibrations in a wind turbine generator can be established by data mining, as shown in Table 3.

At step S14, the rules for the range of the vibration characteristic values of the wind turbine generator under normal conditions are extracted from the knowledge system according to Table 3, and, with reference to Table 4, these rules provide conditions for monitoring vibrations in a wind turbine generator.

Table 4 Rules for the range of vibration characteristic values

No. Rule description

IF C3 [1118.03, 1118.65)) AND C5 [223.81, 226.06) ) AND C6 ( [272.08, 272.69) ) THEN D4

IF C3 ( [1338.90, 1357.26) ) AND C5 ( [412.03, 427.51) ) AND C6 ( [26858, 271.10) I THEN DlO

IF C3 ( [1519.88, 1523.31) ) AND C5 ( [602.81, 613.37) ) AND C6 ( [272.69, 273.98) I THEN Dl 6

IF C3 ( [1439.89, 1459.66) ) AND C5 ( [505.78, 534.61) ) AND C6 ([271.10, 272.08) I THEN Dl 6

IF C3 ( [1534.49, 1543.31) ) AND C5 ( [625.34, 629.58) ) AND C6 ( [263.91, 265.38) ) THEN Dl 6

IF C3 ([1793.51, 1795.51)) AND C4 ([-2.80, -2.69)) THEN D28

IF C3 ([1786.84, 1793.51)) AND C4 ([-0.79, -0.69)) THEN D22

IF C3 ([1757.24, 1760.11)) AND C4 ([4.94, 4.96)) THEN D16

IF C3 ([1765.58, 1772.01)) AND C4 ([-2.20, -2.19)) THEN D16

10 IF C3 ([1772.01, 1775.25)) AND C4 ([-2.68, -2.60)) THEN D22

11 IF C3 ([1779.25, 1784.55)) AND C4 ([6.22, 6.23)) THEN D22

12 IF C3 ([1763.71, 1765.58)) AND C4 ([4.03, 4.26)) THEN D22

13 IF C3 ([1711.05, 1754.33)) AND C4 ([-7.06, -6.89)) THEN D16

14 IF C3 ([1765.58, 1772.01)) AND C4 ([-0.12, -0.11)) THEN D16

15 IF C3 ([1523.31, 1526.46)) AND C4 ([-7.10, -7.08)) THEN DlO

16 IF C3 ([1393.17, 1398.10)) AND C4 ([-2.69, -2.68)) THEN DlO

17 IF C3 ([1276.83, 1277.75)) AND C4 ([2.07, 2.15] THEN DlO

IF C3 ([1142.49, 1147.01)) AND C4 ([6.51, 6.53] THEN DlO

19 IF C3 ([1079.18, 1080.29)) AND C4 ([1.22, 1.33)) THEN D4

As shown in Table 4, after the discretization of the conditional attribute set C, the rule base for the range of the vibration characteristic values of the wind turbine generator can be established by attribute reduction, the rules are represented as a compact set of the automatically selected conditional attributes, and apparently the

automatically selected conditional attributes are generally a part of the original attribute set. The rule base mines the important relationships between the vibration characteristic values and the operation data of the wind turbine generator, and then the range of the vibration characteristic values can be predicted by using the rule base in combination with the real-time operation data of the wind turbine generator.

Once the rule base for the range of the vibration characteristic values of the wind turbine generator has been established, it can be used for monitoring vibrations. Fig. 2 shows a flow chart of the method of the present invention for monitoring vibrations.

As shown in Fig. 2, at step S20, various real-time data representing the performance of the wind turbine generator are acquired, which data mainly include the supervisory control and data acquisition data and the control variables used in the data mining step S12 shown in Fig. 1 ;

at step S22, corresponding rules for the range of the vibration characteristic values are extracted from the rule base for the range of the vibration characteristic values of the wind turbine generator established at step S14 shown in Fig. 1, and according to the rules, real-time operation data are used to predict the range of the vibration characteristic values of the wind turbine generator;

at step S24, the corresponding threshold of the vibration characteristic values is calculated from the predicted range of the vibration characteristic values, and, for example, the alarm threshold can be defined as the predicted virtual value of the vibration speed upper limit multiplied by various weight factors;

at step S25, the corresponding vibration characteristic values of the actual operating wind turbine generator are monitored and the real-time data acquired; at step S26, the threshold of the vibration characteristic values calculated at step S24 is compared with the real-time data of the vibration characteristic values measured at step S25. If the real-time data of the vibration characteristic values are within a normal range, the

monitoring process continues; and if the real-time data of the vibration characteristic values exceed the threshold of the vibration characteristics values, a corresponding fault alarm is sent out at step S28.

Fig. 3 shows a block diagram of a system of the present invention for monitoring vibration in a wind turbine

generator, and the vibration monitoring system comprises: a real-time data acquiring unit 30 for acquiring the SCADA data and control variables of the wind turbine

generator in operation;

a rule base of the vibration characteristic value ranges 31, which rule base 31 is the rule base for the range of the vibration characteristic values of the wind turbine generator established in the vibration monitoring method of the present invention as shown in Fig. 1, formed by rules for the vibration characteristic value ranges;

a vibration characteristic value range predicting unit 32 for predicting the range of the vibration characteristic values;

a vibration characteristic value threshold calculating unit 34 for calculating the threshold of the vibration characteristic values;

a vibration characteristic value monitoring unit 35 for monitoring the corresponding vibration characteristic values of the wind turbine generator and acquiring the real-time data of the vibration characteristic values;

a comparison unit 36 for comparing the threshold of the vibration characteristic values and the real-time data of the vibration characteristic value; and

an alarm unit 38 for sending out a corresponding fault alarm when the particular data of the vibration characteristic values are larger than the threshold of the vibration characteristic values.

As shown in Fig. 3, the system for monitoring vibration in the wind turbine generator acquires various real-time data of the wind turbine generator by the real-time data acquiring unit 30, including the SCADA data and the control variables in Table 1. The system extracts the corresponding rules for the range of the vibration characteristic values from the vibration characteristic value range rule base 31, and the vibration characteristic value range predicting unit 32 predicts the range of the vibration characteristic values of the wind turbine generator according to the extracted rules and the real-time operation data. The vibration

characteristic value threshold calculating unit 34 calculates the threshold of the vibration characteristic values of the wind turbine generator from the predicted range of the vibration characteristic values. The vibration characteristic value monitoring unit 35 monitors the corresponding vibration characteristic values of the wind turbine generator and acquires the real-time data of the vibration characteristic values. When the real-time data of the wind turbine

generator is acquired, the comparison unit 36 compares the threshold of the vibration characteristic values and the real-time data of the vibration characteristic values. The alarm unit 38 sends out a corresponding fault alarm in a certain form if the real-time data of the vibration

characteristic value exceed the threshold of the vibration characteristic values.

In the method and system of the present invention for monitoring vibrations in a wind turbine generator, the rule base for the range of the vibration characteristic values is established by way of the data mining method (such as rough set calculation) on the basis of a large quantity of

historical data, and the rules are represented as a compact set of the automatically selected operation data. Then the threshold of the vibration characteristic values under any operation status can be calculated by means of the rule base mentioned above by taking into consideration various

combinations of the operation data. Since the threshold is calculated on the basis of the relationships between the automatically selected operation data, it can significantly reduce the false alarms in the system. At the same time, the rule base for the range of the vibration characteristic values is represented in the form of explicit and explainable multiple rules, which is easily understandable and convenient for automatic or manual maintenance, and the rules obtained from tests, standards or maintenance experiences of other similar wind turbine generators can be easily added into the existing rule base by the maintenance personnel. Furthermore, by way of an

automatic and periodical database knowledge acquiring and upgrading mechanism, the system is able to "learn"

continuously and automatically from the wind turbine

generator monitoring experiences, so as to increase its own prediction accuracy continuously. In this way, the accuracy of the system for monitoring vibrations will be further improved and the false alarm rate further reduced over time.

The above descriptions are only exemplary particular embodiments of the present invention and are not to limit the scope of the present invention. Equivalent changes,

modifications or combinations made by any persons skilled in the art without departing from the spirit and principle of the present invention shall belong within the protective scope of the present invention.