This paper presents a hybrid dynamic data-driven approach to achieve simple and multiple drift like fault detection of pitch system sensors. This approach considers the system evolving in nonstationary environments and switching between several control modes. This switching is entailed by changes in the system environments. In each control mode, the system has a different dynamical behavior. The latter is described in a feature space sensitive to normal operating conditions in the corresponding control mode. These operating conditions are represented by restricted zones in the feature space called classes. The latter are characterized by a set of parameters representing their statistical properties, e.g. gravity center and variance-covariance matrix. The occurrence of an incipient fault entails a drift in the system operating conditions until the failure takes over completely. This drift manifests as a progressive change in the classes parameters in each control mode over time. The proposed approach monitors normal classes parameters in order to detect a drift in their characteristics. This drift detection allows achieving the fault in its early stages. It uses two drift indicators. The first indicator detects the drift and the second indicator confirms it. Both indicators are based on the observation of changes in the normal operating conditions characteristics over time. A wind turbine simulator is used to validate the performance of the proposed approach.
The search for alternative clean energy is undoubtedly becoming more and more important in modern societies. The growing interest in wind energy production has led to the design of sophisticated wind turbines (WTs). Like every other complex and heterogeneous system, WTs are faced to the occurrence of faults that can impact their performance as well as their security. Therefore, it is crucial to design a reliable automated diagnostic system in order to achieve fault detection and isolation in early stage.
Fault diagnosis of WTs is a challenging task because of the high variability of the wind speed and the confusion between faults and noises as well as outliers. However, the fault diagnosis of pitch system is particularly a challenging task because of (i) the occurrence of pitch system faults in power optimization zone in which the fault consequences are hidden and (ii) the actions of the control feedback which compensate the fault effects. The role of the pitch system is to adjust the pitch of a blade by rotating it depending on the pitch angle position reference provided by the controller. The latter decides the pitch angle position reference according to the wind speed in order to allow an optimum energy production.
In the literature, there are several methods
[
The major advantages of these methods are their ability to detect both the abrupt and progressive failures via trend analysis, and they give a precise decision or isolation of a failure. However, they suffer from the necessity to depth information about system behavior and failures which is hard to obtain for complex and strong non-stationary systems as wind turbines.
An alternative to overcome this problem is the external
methods [
Few approaches have been proposed to achieve early fault
diagnosis of WTs, in particular pitch sensors. This is due
to the fact that modeling component degradation in strong
non-linear and complex non-stationary environments is very
hard task. Examples of these methods, we can cite genetic
algorithm [
This paper presents a new data-driven based approach in order to achieve a reliable drift monitoring and diagnosis of simple and multiple drift-like faults that can affect wind turbine pitch sensors. This approach takes into account the different dynamical behaviors of WTs according to the wind speed. The goal is to detect a drift from normal operating conditions using only the recent and useful data. Initial offline modeling allows constructing initial classes based on the historical data set. These classes characterize the operating conditions of the pitch system (normal/faulty) and are represented by restricted zones in the feature space. The latter is formed by sensitive features to pitch sensor operating conditions in order to distinguish any drift from normal to fault operating conditions. The modeling tool is an algorithm called AuDyC (Auto-Adaptive Dynamical Clustering) used to initialize the classes that will be dynamically updated.
In this work, two-dimensional feature space is constructed, for the sensor faults. The faulty classes, representing the failure operating conditions of pitch sensor, are considered to be a priori unknown. There is one known class in advance.The class represents the pitch sensor normal operating conditions. It considers gradual degradations in pitch sensor operating condition as a drift in the characteristics of normal class over time. Detecting and following this drift can help to predict the occurrence of pitch sensor failure.
The drift-like fault is monitored using two drift indicators: one to detect a drift and the second one to confirm it. When the drift is detected by the first indicator, a warning is emitted to human operators. Then, the second drift indicator confirms this drift in order to inform human operators of the necessity to react by taking the adequate correction actions.
The proposed data-driven approach is composed of five main steps: processing and data analysis, clustering and classification, drift monitoring, updating and interpretation steps.
The wind turbine model under study is composed of five principal parts: the blades, the drive train, the generator with the converter, and the controller (see Figure 1). It can be seen that the blades are fixed to the main axis, which in turn is connected to the generator through the drive train. The generator is electrically connected to the converter, which in turn is connected to a transformer. The blades are pitched by the pitch actuators.
The controller operates in four zones (see Figure 2). Zone 1 is the start-up of the turbines, zone 2 is power optimization, zone 3 is constant power production and zone 4 is no power production due to a too high wind speed.
In order to handle transitions between the control modes,
The satisfaction of this condition generates a discrete event, E23, allowing the switching from control mode 1 to control mode 2. The goal to obtain Pg equal to Pr. This condition is satisfied when the wind speed is greater than predefined threshold for zone 2 (12.5 m/s in Figure 2). Likewise, the control mode should switch from control mode 2 to control mode 1 if the following condition is satisfied: E32 : !g(t) < !nom !
Where !nom is the nominal generator speed and ! is a
small offset subtracted from the nominal generator speed to
introduce some hysteresis in the switching scheme, thereby
avoiding that the control modes are switching all the time
[
The focus of this benchmark model is on the operation of WT in zones 2 and 3. Two control strategies are applied to optimize the energy production and keep it constant at its optimal value: the converter torque control in zone 2 and the blades angle control in zone 3 (see Figure 4). In zone 2, the WT is controlled so that it produces as much energy as possible. To do so, the blades angle is maintained equal to 0 and the tip speed ratio is kept constant at its optimal value. The latter is regulated by the rotating speed control by tuning the converter torque. Once the optimal power production is achieved, the blades angle control maintains the converter torque constant and adjusts the rotating speed by controlling the blades angle. The latter modifies the transfer of the aerodynamic power of the wind on the blades. In this work, the controller modes are modeled by a finite state automaton containing two states (see Figure 4). In the following, zones 2 and 3, respectively, correspond to control modes 1 and 2: Control Mode 1 In this control mode, the power optimum value is achieved by setting the pitch reference to zero [t] = 0 and the reference torque to the converter g;r as follows: !g [t] Ng
2 g;r = Kopt (1) Ng is the gear ratio and n is the sampling time.
Where
Kopt = 21 AR3 CPo3mpatx (2) with the air density, A the area swept by the turbine blades, CPmax the maximum value of power coefficient, and opt the optimal value of is found as the optimum point in the power coefficient CP mapping of the WT. The power coefficient mapping characterizes the efficiency of energy and it depend on and .
Control Mode 2 In this mode, the major control actions are handled by the pitch system using a Proportional Integral (PI) controller trying to keep !g[t] at !g.
r (t) = r (t 1) + kp:e (t) + (ki:Ts:kp) :e (t When e(t) = !r(t) !nom. In this case the converter reference is used to suppress fast disturbances:
As we said before, the benchmark model allows simulating the WT behavior in two power zones: 1) zone 2 (power optimization) where g is controlled and r is equal to zero and; 2) zone 3 (optimal energy production) where g is kept r constant and is controlled. In this paper, we focus on pitch sensor faults as it is discussed in subsection 2. 3
The considered WT is horizontal-axis based with three blades. Each blade is equipped with an actuator. The role of the pitch actuator is to adjust the pitch of a blade by rotating it; Each actuator is provided by the same pitch angle reference r. The pitch angle of a blade is measured on the cylinder of the pitch actuator, each pitch position (angle) mi where i 2 f1; 2; 3g is measured with two sensors where index mi represents the ith sensor of the corresponding variable (see Figure 5). The pitch system feedback f is an internal variable used to model the pitch position error caused by sensor faults:
1 f = r
( k;m1 + k;m2)
2
The controller is fed by the mean value of the readings of the
two sensors. Hence, this sensor fault is modeled as a change
(7)
in the pitch references, meaning that a sensor fault resulting
in changed mean value should also change the pitch
reference accordingly [
The hydraulic pitch system is modeled in the benchmark as
a closed loop of dynamic system. The state representation
of the nominal pitch system dynamics is defined as follows
[
The pitch system represent a hybrid dynamic system and especially it belongs to the class of Discretely Controlled Jumping Systems (DCJS), In these systems, the continuous state variables change discontinuously under the influence of an external action (e.g., a command) as the case for electromagnetic systems with pulse inputs [?]. The pitch system h : iT state variable xp = k k changes discontinuously under the influence of an external action defined by Equation 5 and 6. 5
In this paper the types of fault which are considered in this work are simple and multiple drift-like fault in pitch sensors. The following subsections detail the generation of several scenarios representing drift-like faults with three different (8) speeds in pitch sensor m1 and pitch sensor both pitch sensors m1 and m2. 5.1
Each blade is equipped with an actuator. Each actuator is provided by the same pitch angle reference r. In addition, each pitch position, (angle) mi is measured with two sensors where index i represents the ith sensor of the corresponding variable. The fault scenarios related to simple drift-like fault in pitch sensor n 1 and sensor n 2 and multiple drift-like fault in both pitch position sensor n 1 and sensor n 2 in blade n 3 are summarized respectively in Table 1, Table 2 and Table 3. The state representation of the pitch system after the integration of a fault in sensor mi, i 2 f1; 2g is defined as follow: : xp = Axp + Bu yp = Cxp + f (t) f (t) = i: (tb te) (9)
Therefore the parameter i, i 2 f1; 2g is used in the simulation to generate a fault in sensor mi during the time period (tb te) where tb is the start time and te is the end time of sensor drift-like fault.
In this paper the simple drift-like fault scenarios in pitch sensor 1 ( m1) scenarios are modeled as a gradual change in the coefficient 1 of pitch sensor n 1 in blade n 3 where tb is the beginning of the drift and te is the end of the drift. Nine scenarios for simple sensor drift-like fault are generated in order to simulate slow, moderate and high degradation speeds represented by slow, moderate and high drift speeds (see Figure 6). Each drift speed scenario is generated at three different time instances. Thus, parameter 1 is changed linearly from 1N to 1F in a period of 30s, 60s and 90s, corresponding respectively to high, moderate and slow drift speeds. Then, the fault remains active for 200s. Finally the parameter 1 decreases again to return to its initial value 1N (see Figure 6 for the case of high drift speed in sensor 1 ( m1)).
The simple drift-like fault scenarios in pitch sensor 2 ( m2) scenarios are modeled as a gradual change in the coefficient 2500s -2730s 2500s -2760s 2500s -2790s 2600s -2830s 2600s -2830s 2600s -2890s 2700s -2930s 2700s -2960s 2700s -2990s Period 2800s -3030s 2800s 3060s 2800s-3090s 2900s -3130s 2900s -3130s 2900s -3190s 3000s -3230s 3000s -3260s 3000s -3290s 60s 90s 30s 60s 90s 1N ! 1N ! 1N ! 1N ! 1N ! 1N ! 1N ! 1N ! 1N !
2 of pitch sensor n 2 in blade n 3 where tb is the beginning of the drift and te is the end of the drift. As for the case of simple drift-like fault in pitch sensor m1 scenarios, nine scenarios for simple sensor drift-like fault are generated in order to simulate slow, moderate and high degradation speeds represented by slow, moderate and high drift speeds (see Figure 7). Each drift speed scenario is generated at three different time instances. Thus, parameter 2 is changed linearly from 2N to 2F in a period of 30s, 60s and 90s, corresponding respectively to high, moderate and slow drift speeds. Then, the fault remains active for 200s. Finally the parameter 2 decreases again to return to its initial value 2N (see Figure 7 for the case of high drift speed in sensor 2, ( m2)). In this chapter the generated scenarios of the multiple driftlike fault in pitch sensor 1 ( m1) and sensor 2 ( m2) are modeled as a gradual change at the same time in the drift coefficient ( 1 and 2) of both pitch sensors n 1 and pitch sensors n 2 in blade n 3. As for the case of simple driftlike fault in pitch sensor scenarios, nine scenarios for multiple sensor drift-like fault are generated in order to simulate slow, moderate and high degradation speeds representing by slow, moderate and high drift speeds (see Table 3). Each drift speed scenario is generated at three different time instances. Thus, parameters 1 and 2 are changed linearly from 1N and 2N to 1F and 2F in a period of 30s, 60s and 90s, corresponding respectively to high, moderate and slow drift speeds. Then, the fault remains active for 200s. Finally the parameter decreases again to return to their initial values (see Figure 8 for the case of high drift (degradation) speed in both sensor 1 ( m1) and sensor 2 ( m2)). In this section, hybrid dynamic data-driven approach is developed in order to achieve condition monitoring and drift like fault detection of pitch sensor. It performs predictive diagnosis by detecting a drift of the system operating conditions from normal to faulty modes. The proposed approach is based on 5 steps developed in the following subsections (see Figure 9). 6.1
This step aims at finding the features that are sensitive to the system operating conditions in order to construct the feature space. A feature space representing the operating conditions of each assembly of WT is defined, this feature space will be responsible of the detection and isolation of faults impacting this components. The research of sensitive features is based on the signals provided by the pitch sensors as well as the prior knowledge about the system dynamics. These features are chosen in order to maximize the discrimination between operating conditions in the feature space. In this paper, twodimension feature space is constructed for the sensor fault. The goal of the feature space use, at the level of component, is to facilitate the drift-like fault isolation and to enhance the diagnosis robustness.
The position of the pitch actuators is measured by two redundant sensors for each of the three pitch positions k;mi, k = 1; 2; 3, i = 1; 2, with the same reference angle r provided to each of them. In order to enhance the robustness against noise, the measurements are filtered by a first order filter using time constant = 0:06.
For the drift like fault detection and isolation of the sensor faults, we propose to explore the physical redundancy in order to generate residuals as follows: s1 = j r + f s2 = j r + f m1j m2j (10) (11)
To do so, the residual sn, n = 1; 2, is generated by the comparison between the pitch angle measurement mi, i = 1; 2, m = 1; 2; 3 and the command computed by the sum of the desired value of the pitch angle r and the feedback pitch system f (see Figure 5). The residual is computed within a time window which is tuned to be several times the actuator time response.
The evolution of these residuals with respect to each of the two sensors is considered as meaningful features. Indeed, the residual s1 respectively s2, is equal to zero when the corresponding sensor m1 respectively m2, is in normal operating conditions. When, the sensor m1 respectively m2, is in faulty operating conditions, the residual s1, s2 will be different of zero because this sensor will not measure the new value of command ( r + f ) (see Figure 5). Indeed, the command ( r + f ) will change in order to compensate the difference between the two sensors due to the fault of sensor m1 respectively m2. 6.2
The clustering looks to determine the number of classes contained in the learning set and to initialize their parameters. The classification aims at designing a classifier able to assign a new pattern to one of the learnt classes in the feature space. A new pattern characterizes the actual operating conditions (normal or faulty in response to the occurrence of a certain fault) of the system. Examples of these approaches are present in [5] as well as in the references of this paper.
Auto-adaptive Dynamical Clustering Algorithm
(AuDyC) [
In the sensor feature space, four classes are considered: the fault of sensor 1, m1 , the fault of sensor 2, m2, the fault of both sensor 1, m1 and sensor 2 m2, and the normal functioning. Figure 10 shows the classes representing normal and failure operating conditions of pitch sensor in the feature space constituted by the two residuals defined by Equation 10 and 11. In zone 2, the effects of this fault are hidden because the actuators are not operated. Moreover, it is strongly difficult to distinguish the fault occurrence to the noise in the case of small angles. Therefore an overlapping region is created between the normal and failure classes (see Figure 10 and Figure 15).
In order to answer the challenges inherent to the system operation, the normal and failure classes are split into five classes and the pitch actuator dynamics are represented by two different control modes. The first one corresponds to the case of zone 2 low wind speed; while the second control mode represents the case of zone 3 high wind speed (see Figure 16). Class 1 is the ambiguity class. It gathers the patterns representing pitch sensor normal or faulty operating conditions. This class represents the control mode 1. Class 2 represents the normal operating conditions class in control mode 2. Class 3 represents failure class caused by simple drift-like fault in pitch sensor 1, m1 in control mode 2, class 4 represents failure class caused by simple drift-like fault in pitch sensor 2, m2 in control mode 2 and class 5 represents failure class caused by multiple drift-like fault in pitch sensor 1, m1 and sensor 2, m2 in control mode 2.
The updating step aims at reacting to the changes in classes characteristics in the feature space. AuDyC continuously updates the classes parameters by using the recursive adaptation Rules 12 and 13. In such a way, its validity and where xnew and xold are respectively, the newest and the oldest arrived pattern in the time window Nwin .
Initial off-line modeling allows the construction of initial classes that characterize knowledge from historical data. The historical data are usually sensor data that are saved. AuDyC is used to initialize the parameters of classes that will be dynamically updated. Knowledge of failure modes given from (labeled) historical data can help building a classification scheme for fault diagnosis. However, in reality, these data are hard to obtain.
In this work, we suppose that only data corresponding to normal operating conditions (normal classes) are known in advance. The training of the process by applying AuDyC is made based on features that are extracted from historical sensor data once finished; the class corresponding to normal operating conditions is retained. We denote this class by CN = ( N ; N ).
In on-line functioning, the parameters of CN are dynamically updated by AuDyC for each new pattern arrived in control mode 2. This yields changes in the class parameters which continuously reflect the distribution of the newest arriving patterns. We denote by Ce = ( e; e) the evolving classes in feature space. We have Ce (t = 0) = ( e; e) = CN .
In control mode 1 of pitch system, pitch sensor normal and faulty behaviors cannot be distinguished. Thus, in the proposed approach, the decisions about the status (normal/faulty) of patterns located in this region are delayed. Therefore in this case, the classifier will not be updated in order to avoid integrating in the drift time window useless patterns. In order to detect the drift as soon as possible, AuDyC updates the classes parameters by using a window that contains only the patterns belonging to control mode 2. AuDyC is dynamic by nature in the sense that it continuously updates the parameters of the classes as new patterns arrive. 6.3
When a new pattern is classified in the ambiguity class (A), in sensor feature space, assigning it to normal or failure operating conditions is a risky decision since normal and failure classes are overlapped in this region of the feature space. In order to reduce this risk, the decision about the status (normal or faulty) of any pattern classified in this region is delayed by assigning the label (A) (ambiguity decision). Then, this ambiguity can be removed by analyzing the past and future decisions of this pattern. The analysis of the pattern decision sequence is achieved by using a set of decision rules allowing assigning to ambiguity patterns label (N) or label (F) (normal or faulty) as follows. Let us suppose that XA = fxt; xt+1; : : : ; xt+ng is a set of patterns associated with decision (A). Let xt 1 be the previous pattern arrived just before xt. Let D (xt 1) 2 fA; N; Fig be the decision of this pattern. Let xt+n+1 the pattern arrived just after xt+n. Let D (xt+n+1) 2 fA; N; Fig be the decision for this pattern. Then, the decision can be updated as follows:
Rule 16 signifies that the fault has occurred somewhere in control mode 1 where its consequences on the pitch system dynamical behavior can be observed. Rule 17 indicates that the failure has disappeared in the control mode 1 either because of maintenance actions or because the fault is intermittent. The key problem of drift monitoring is to distinguish between variations due to stochastic perturbations and variations caused by unexpected changes in a system’s state. If the sequence of observations is noisy, it may contain some inconsistent observations or measurements errors (outliers) that are random and may never appear again. Therefore, it is reasonable to monitor a system and to process observations within time windows in order to average and reduce the noise influence. Moreover, the information about possible structural changes within time windows can be interpreted and processed more easily. As a result, a more reliable classifier update can be achieved by monitoring within time windows. The latter must include enough of patterns representing the drift.
To distinguish the useful patterns, the pitch sensor dynamics are represented by two different control modes. In the control mode 2, the degradation consequences of pitch sensor can be observed. Therefore, all patterns in this mode are useful to be analyzed and to be included in the drift time window. In the control mode 1, the degradation consequences are masked. Patterns representing normal operating conditions cannot be distinguished from patterns representing pitch sensor degradations. Therefore in this case, no decision (normal/drift) will be taken in order to avoid integrating in the drift time window useless patterns.
The proposed scheme makes use of classes parameters (Mean, Variance-covariance matrix) which are dynamically updated at each time but only with the patterns belonging to control mode 2. Drift indicators are defined based on these parameters and the detection of faults inception will be made based on their values. We define two drift indicators Ih1 (x) ; Ih2 (x) as follows:
Ih1 (x) = dMah (CN ; e)
Ih2 (x) = dE ( N ; e) Where dMah and dE are, respectively, the Mahalanobis and Euclidean metrics.
Euclidean metric computes the distance between the center n of the normal class CN and the center e of evolving class Ce; on the other side Mahalanobis metric computes the distance between the normal class CN and the evolving class center e. Therefore, these two distances are calculated as follows: dMah (CN ; e) = q
( N q e) e)
Ih2 (x) > thd ) drift is confirmed The selection of thd is motivated statically by taking three (standard deviations) of the data in the normal operating conditions.
In the case of pitch sensor faults, three scenarios may appear in the sensor feature space: fault impacting sensor 1 ( m1), fault impacting sensor 2 ( m2) or fault impacting both sensors ( m1 and m2) at the same time. The direction of the evolving class in the sensor feature space depends on which of these scenarios happened. Therefore, for sensor fault isolation, we use a drift direction indicator in order to monitor the direction of the evolving class. This will allow to determine which of these three scenarios happened and hence to isolate the abnormal drift source. When drift occurs, the evolving class will migrate from normal operating condition to failure. The direction indicator Dr and direction isolation DI are used to isolate the sensor which caused the drift-like fault. The idea is to consider the angle 1 respectively 2, between the vector e relating the center of the evolving class and the origin of the feature space, and the vector e1 respectively e2 relating the origin with the projection of the center of the evolving class according to feature 1 respectively feature 2, of the feature space. These angles define the movement direction of the evolving class.
In order to calculate 1 and 2, the scalar products between !e1 and !e and between !e2 and !e are calculated as follows: !e (x) !e1 (x) = k e(x)k k e1(x)k cos 1 (24) (18) (19) e)T (20) (21) (22) (23) !e (x) !e2 (x) = k e(x)k k e2(x)k cos 2
If the drift is detected and confirmed by the two drift indicators Ih1 (x) and Ih2 (x), then the drift isolation (to determine if sensor 1 or sensor 2 or both is the source of this drift) is achieved as follows:
If Dr = 1
2 > tha and 1 > 2 ) DI = 1 : fault in sensor 1 ( m1)(26) If Dr = 1
2 > tha and 2 < 1 ) DI = 2 : fault in sensor 2 ( m2) (27) If Dr = 1
2 < tha ) DI = 3 : fault in both sensors ( m1and m2) (28) where tha is the angle threshold. tha is defined according to the variation of patterns within the normal class CN . Therefore, tha is determined experimentally using the patterns belonging to CN .
The interpretation step aims at interpreting the detected changes within the classifier parameters and structure. This interpretation is then used as a prediction about the tendency of the future development of the WT current situation. This prediction is useful to formulate a control or maintenance action. 7
The failures of pitch sensors are caused by a continuous degradation of its performance over time. This degradation can be seen as a continuous drift of the normal operating conditions characteristics (normal class) of the pitch sensor. Detecting and following this drift can help to predict the occurrence of the pitch sensor failures. The two monitoring indicators defined by Equation 18 and Equation 19 are used to detect and to confirm this drift for the twenty-seven scenarios of simple and multiple drift-like fault in pitch sensors are defined in section 2. 7.1
m1 Figure 18 and Figure 19 represent, respectively, first and second residuals used in the pitch sensor feature space in presence of an abnormal drift in pitch sensor 1, m1. We can see in the case of an abnormal drift in pitch sensor 1, m1, that only residual s1 is impacted, while residual s2 has similar behavior as the one without abnormal drift in m1.
Table 4 show the values of the drift indicators Ih1 (x) and Ih2 (x) for the nine defined drift-like fault scenarios. These values represent the required time (starting from the drift beginning) to detect and confirm the drift occurrence. Thus, they can be used as an evaluation criterion to measure the time delay to detect a drift before its end.
Figures 20 and 21 show the obtained results using the two drift detection indicators Ih1 (x) and Ih2 (x), for simple drift-like fault in pitch sensor m1. The degradation is observed when the pitch actuator operate in control mode 2, the drift like fault in pitch sensor is successfully detected by both indicator Ih1 (x) and Ih2 (x), for all drift speeds (see Figure 20 and Figure 21).
The drift-like fault in pitch sensor 1 ( m1), is detected in early stage before the end of this drift (arriving to the failure mode due to drift fault in pitch sensor). As an example, in the case of a drift of slow speed (F6s) (see Table 4), the pitch sensor reaches the failure mode resulting from a driftlike fault in 1 (degradation in 1) after 90 seconds of the beginning of the drift. In the proposed approach, this drift is detected 15.10 seconds and confirmed 29.40 seconds after its beginning. Therefore, the drift like fault in pitch sensor is confirmed 60 seconds before its end. This enables to achieve an early fault diagnosis and therefore helps the human operators of supervision to take efficiently the right actions.
Figure 22 and Figure 23 represent, respectively, evolving class angle and the direction indicator of the pitch sensor fault. These figures show the obtained results in presence of simple drift-like fault in pitch sensor 1, based on Figure 22 and Figure 23 the sensor 1 ( m1), fault is successfully isolated by the direction indicator. Indeed, the direction angle shows that the evolving class exceeds the angle threshold (see Figure 17.a). Based on Equation 26, the drift-like fault in sensor 1 ( m1), is isolated (see Figure 29).
Table 5 show the values of the drift indicators Ih1 (x) and Ih2 (x) for the nine defined drift-like fault scenarios. These values represent the required time (starting from the drift beginning) to detect and confirm the drift occurrence. Thus, they can be used as an evaluation criterion to measure the time delay to detect a drift before its end.
Figures 26 and 27 show the obtained results using the two drift detection indicators Ih1 (x) and Ih2 (x), for simple drift-like fault in pitch sensor 2 ( m2). The degradation is observed when the pitch actuator operate in control mode 2, the drift-like fault in pitch sensor 2 is successfully detected by both indicators Ih1 (x) and Ih2 (x) for all drift speeds (see Figure 26 and Figure 27).
The drift-like fault in pitch sensor 2 ( m2), is detected in early stage before the end of this drift (arriving to the failure mode due to drift fault in pitch sensor). As an example, in the case of a drift of slow speed (F9s) (see Table 5), the pitch sensor reaches the failure mode resulting from a driftlike fault in 2 (degradation in 2) after 90 seconds of the beginning of the drift. In the proposed approach, this drift is detected 14.90 seconds and confirmed 28.10 seconds after its beginning. Therefore, the drift like fault in pitch sensor is confirmed 60 seconds before its end. This enables to achieve an early fault diagnosis and therefore helps the human operators of supervision to take efficiently the right actions.
For the drift isolation, Figure 28 and Figure 29 are used. They represent, respectively, evolving class angle and the direction indicator of the pitch sensor fault. These figures show the obtained results in presence of simple drift-like fault in pitch sensor 2, based on Figure 28 and Figure 29 the sensor 2 ( m2), fault is successfully isolated by the direction indicator. Indeed, the direction angle shows that the evolving class exceeds the angle threshold (see Figure 17.b). Based on Equation 27, the drift-like fault in sensor 2 ( m2), is isolated (see Figure 29).
Figure 30 and Figure 31 represent, respectively, first and second residuals used in the pitch sensor feature space in presence of an abnormal drift in both pitch sensor m1 and m2 at the same time. We can see that both residual s1 and s2 are impacted by the occurrence of the abnormal drift in m1 and m2.
Table 6 show the values of the drift indicators Ih1 (x) and Ih2 (x) for the nine defined drift-like fault scenarios. These values represent the required time (starting from the drift beginning) to detect and confirm the drift occurrence. Thus, they can be used as an evaluation criterion to measure the time delay to detect a drift before its end.
Fault N
F10h F10m F10s F11h F11m F11s F12h F12m F12s fault in pitch sensor is successfully detected by both indicator Ih1 (x) and Ih2 (x) for all drift speeds in both sensors (see Figure 32 and Figure 33).
Figures 32 and 33 show the obtained results using the two drift detection indicators Ih1 (x) and Ih2 (x), for multiple pitch sensor fault. The degradation is observed when the pitch actuator operate in control mode 2. The drift like
The multiple drift-like faults in pitch sensors are detected in early stage before the end of these drifts (arriving to the failure mode due to drift fault in both pitch sensors). As an example, in the case of a drift of slow speed (F12s) (see Table 6), the pitch sensors reache the failure mode resulting from a drift-like fault in 1 and 2 (degradation in 1 and 2) after 90 seconds of the beginning of the drift. In the proposed approach, this drift is detected 14.70 seconds and confirmed 28.25 seconds after its beginning. Therefore, the multiple drift-like fault in pitch sensor is confirmed 60 seconds before its end. This enables to achieve an early fault diagnosis and therefore helps the human operators of supervision to take efficiently the right actions.
For the drift isolation, Figure 34 and Figure 35 are used. They represent, respectively, evolving class angle and the direction indicator of the pitch sensor fault. These figures show the obtained results in presence of a multiple drift-like fault in both pitch sensors m1 and m2, as we can see in Figure 34 and Figure 35 the fault is successfully isolated by the direction indicator. Indeed, the direction angle shows that the evolving class evolve within the axe of the normal class (see Figure 17.c). Based on Equation 27, the multiple drift-like isolation in both pitch sensors is isolated (see Figure 35).
In this paper, an approach of condition monitoring and driftlike fault detection was developed. It is based on the use of a classifier able to achieve a reliable drift monitoring and early diagnosis of simple and multiple pitch sensors faults. This approach considers the system switching between several control modes. This approach based on the monitoring of the drift of the characteristics of classes representing the normal operating conditions of pitch system in each control mode. These characteristics are described by the mean and variance covariance matrix of these classes. They are monitored using two indicators in order to monitor and follow the drift. Both are defined based on the computation of the distance between the class representing normal operating conditions and the evolving class. The first indicator is based on the Mahalanobis distance and is used to detect the drift; while the second indicator is based on Euclidean distance and is used to confirm the drift. The drift indicators have detected successfully all drift scenarios of three speeds in early stage before the end of this drift for the case of simple and multiple drift-like faults in pitch system.
Future work will focus on the drift like fault of other wind turbine critical components as the generator and drive train as well as the use of other indicators to detect drifts of other types or natures.