Open-Circuit Faults Diagnosis in Direct-Drive PMSG Wind Turbine Converter

The condition monitoring and fault diagnosis have been identified as the key to achieving higher availabilities of wind turbines. Numerous studies show that the open-circuit fault is a significant contributor to the failures of wind turbine converter. However, the multiple faults combinations and the influence of wind speed changes abruptly, grid voltage sags and noise interference have brought great challenges to fault diagnosis. Accordingly, concerning the open-circuit fault of converters in direct-driven PMSG wind turbine, a diagnostic method for multiple open-circuit faults is proposed in this paper, which is divided into two tasks: The first one is the fault detection and the second one is the fault localization. The detection method is based on the relative current residuals after exponential transformation and on an adaptive threshold, and the localization method is based on the average values of fault phase currents. The scheduled diagnosis method is available to both the generator-side converter and the grid-side converter, allowing to detect and locate single and double open-circuit faults. For validating this, robustness test and multiple open-circuit faults diagnosis are presented in a 2-MW direct-driven PMSG wind turbine system, the results validate the reliability and effectiveness of the proposed method.


Introduction
Power converter is a hub connecting generator and power grid, which is mainly composed of two identical converters: The generator-side converter (GESC) representing a three-phase rectifier that converts the AC voltage in generator-side to a DC voltage across the DC coupling capacitor; the grid-side converter (GRSC) representing the inverter that converts the DC voltage to an AC voltage with the grid's voltage and frequency [1]. Therefore, as the key component of wind turbines (WTs), the reliability level of converters directly determine whether WTs can connect to power grid in a safe state. The insulated gate bipolar transistor (IGBT) is the core component of the converter. The statistical data presented in [2,3] show that failures in converters are a significant contributor to the overall failure rate of modern WTs, the failure rate of converters is higher compared with the other components, where annual failure rate becomes to more than 15%, and 34% of failures in converters result from IGBTs. With the power generation capacity of WTs increases gradually, the demand for reliability and safety of power converters is increasing. Thus, fault detection and localization of IGBTs has always become a research hotspot.
There have been a large quantity of relevant research works that mainly focus on two aspects. The first one is IGBT short-circuit (SC) faults diagnosis research in WT converter. SC faults lead to an abnormal over current that can cause permanent damage to IGBTs, the time between the fault initiation and equipment damage is very short [4]. Therefore, SC faults detection and protection methods are mostly based on hardware circuits in order to deal with the faults rapidly, and the technologies have been extremely mature [5]. The IGBT gate driver with the model of 1IPSE1S45-100, produced by InPower, can detect SC faults by detecting the collector current change rate (di/dt) of high-power IGBT within 2.16 μs after the fault initiation, and then the driver turns off IGBT immediately [6]. The second one is IGBT opencircuit (OC) faults diagnosis research in WT converter. OC faults are relatively mild and do not lead to serious damage compared with SC faults. But this fault easily degrades the WTs performance, affects the power quality and makes the output phase current distortion. In addition, the DC currents generate unequal stress in the upper and lower transistors. These effects may cause secondary faults in the converter, motor, or load [4,7]. The use of the current Park's vector shape monitoring approach for detecting the switch devices of three-parallel power converters in a wind-turbine system is presented in [8]. However, the technique generally has the disadvantage of slow detection. In order to mitigate these problems, the Park's vector normalized currents and errors of their average absolute values were utilized to improve the immunity against the issue of false alarms and processing speed in [9]. The proposed method can allow for the diagnosis of single and a part of multiple OC faults. For the fixed threshold, the lack of adaptability to the system makes it prone to produce a negative impact on the reliability of the diagnosis results. A fault diagnosis scheme based on three phase output currents is proposed for single and multiple transistors open circuit in PWM VSI fed vector control drives in [10]. Evaluated the capability and robustness for accurate fault diagnosis under noisy variable wind conditions, a new fault diagnosis method for permanent magnet synchronous generator (PMSG) WTs converter OC faults is proposed in [11], but this method is only available for diagnosing single OC faults. A fault diagnosis method for OC faults based on real-time current amplitude is proposed in [12], which compares the current amplitude in the fault state with the set current amplitude threshold eliminates the misdiagnosis caused by errors. The diagnosis approach proposed in [13] is based on the average absolute value of the normalized stator currents, which is independent of the amplitude changes. However, the methods in [12,13] are only applied for diagnosing single-phase OC faults. Through the installation of voltage sensors, rapid fault detection could be achieved using voltage signal in [14][15][16], nevertheless, they require additional voltage sensors, which lead to a rise in the system cost, therefore they have clear limitations in application. The application of intelligent algorithm in fault diagnosis is not rare recently, machine learning diagnosis methods was adopted in [17][18][19], where Zhang et al. [17] extracts the original voltage signals and proposes a strategy containing wavelet transform, feature analysis, judgment and BPNN classification, Tan et al. [18] extracts the constraint relationship between the three-phase circuits and adopts the ANN fault detection method using correlation features, Zhang et al. [19] presents a strategy containing FP-Growth, WPT, K-means, LSTM, analysis and weight to conduct the reliability assessment of PMSG wind power converter. These series of methods whereas the complex computations increase the diagnosis time, and there is a huge amount of training parameters. Although the methods do not depend on the system model, the general approach to select and set of fault characteristics is not fully investigated at present. Furthermore, the fault diagnosis techniques based on observer have got extensive attentions of scholars. A diagnosis method for OC faults in converters based on interval sliding mode observer was proposed in [20], the Mixed Logic Dynamic model of the converter was established by using the current flow direction of the switch under normal working and fault conditions. A novel diagnosis method for OC fault of voltage sensor less inverter based on second-order sliding mode observer and hybrid logic dynamic model is proposed in [21], but the chattering phenomenon of sliding mode observer is serious. An adaptive diagnosis method for OC faults based on the bilinear observer is proposed in the converter of doubly-fed WTs in [22]. According to the voltage distortion observer, a diagnosis method for single-phase OC faults of generator-side converter is proposed in [23].
A review of literature shows that the existing research works regarding the development of OC faults diagnostic algorithms entirely dedicated to converters in direct-driven PMSG wind turbine (D-PMSGWT) have obvious drawbacks related to the few type of diagnosable fault and lack of adaptability and portability, which cannot be used both for the GESC and the GRSC. Affected by the converter structure and control strategy in D-PMSGWT, the combination of faulty IGBTs is diverse, which leads to the corresponding phase currents distinct degree of distortion. The stable operation of converter is interfered with conditions such as random wind speed and voltage sags. Apart from these, the current sensors signals are often mixed with noise signals, which increase the total harmonics distortion (THD) of currents. Obviously, all of these bring great challenges to OC faults diagnosis. In this context, this paper designs an effective diagnosis method for OC faults of converters in D-PMSGWT. The considered technique requires three PMSG phase currents and three grid phase currents only, allowing to achieve single and double IGBT OC faults detection and localization in both the GESC and the GRSC.
The remainder of the paper is organized as follows. In Section 2, the focused topology and the current equations of direct-driven PMSG wind power converter system are addressed and analyzed. In Section 3, the diagnostic principle and method are described, and the fault states are classified. Robustness test and multiple open-circuit faults diagnosis are presented to verify the reliability and effectiveness of the proposed fault diagnostic method in Section 4. The conclusions of this paper are drawn in Section 5.

Model of Wind Power Converter System
The extensively applied topology of direct-driven PMSG wind power converter system, as depicted in Fig. 1, comprises PMSG and full power converter. Full power converter connects PMSG and power grid, it adopts double-PWM structure, which is indirect AC to AC converter (AC-DC-AC), including the GESC, DC capacitor and the GRSC. Each converter is composed of six IGBTs and their respective antiparallel diodes. Typically, stator flux oriented vector control is applied in the GESC to control PMSG, whereas power system voltage oriented vector control is applied in the GRSC to keep DC-link voltage stable; furthermore, the space vector pulse width modulation (SVPWM) strategies are applied for both converters (GESC and GRSC) to output PWM signals.
Since the current path of converter is formed by IGBTs and their respective antiparallel diodes, the fault phase current will lose its normal waveform, and the healthy phase current will be offset with the occurrence of IGBT OC fault of the converter [24]. According to this characteristic, the current signals can be  implemented for faults diagnosis. The current equations of PMSG and grid-side converter in the dq synchronous rotating frame can be represented as (1) and (2), respectively: where u sd and u sq are PMSG dq-axes voltage components; i sd and i sq are PMSG dq-axes current components; Ψ f is PMSG magnets flux linkage; ω s is PMSG electrical angular speed; R s is PMSG stator resistance; L sd and L sq are PMSG dq-axes stator inductance components and L sd is equal to L sq .
where u gd and u gq are grid-side converter dq-axes voltage components; e gd and e gq are power grid dq-axes voltage components; i gd and i gq are grid-side dq-axes current components; ω g is grid rotational angular speed; R g is grid-side equivalent resistance; L gd and L gq are grid-side dq-axes equivalent inductance components and L gd is equal to L gq .

Fault Detection
In order to detect the phase with OC fault, the detection method mainly includes three steps, designed and illustrated in the block diagram of Fig. 2. The first step is establishing Luenberger observer to obtain three-phase observed currents. The second step is calculating current relative residuals by measured and observed currents to design the detection variables. In the third step, two adaptive thresholds are designed for both converters based on the corresponding detection variables respectively. The detection purposes are achieved through the comparison of detection variables and adaptive threshold. The detailed design process is performed as follows: Step 1: Design of the Luenberger Observer Three-phase observed currents of both the GESC and the GRSC can be obtained by means of a Luenberger observer, which is based on the state space model. Therefore, according to (1) and (2), u sd ,

Detection result
Step 1 Step 2 Step 3 Current Observer The currents i sk , i gk relative residual generation Eq. u sq , u gd and u gq are selected as input variables, concurrently, i sd , i sq , i gd and i gq as state variables, the state space model of direct-driven PMSG wind power converter system is established: where x, u and y represent state variable, input variable and output variable, respectively; _ x is derivative of x; A, B and C represent state matrix, input matrix and output matrix, respectively. In addition, Based on the state space model, the Luenberger observer is designed, which is represented by: wherex is observed value of x andx ¼î sdîsqîgdîgq À Á T ;ŷ is observed value of y andŷ ¼x ; K is the observer gain matrix and K= The gain matrix K is considered as the key parameter of observer design. Therefore, matrix K is determined by pole-placement method adopted in [25,26], assuring the stable convergence of the observer. The structure of the Luenberger observer for direct-driven PMSG wind power converter system in this section is constructed as shown in Fig. 3. Since the observed currents of observer outputs are in the dq synchronous rotating frame, it is necessary to obtain three PMSG phase observed currents and three grid phase observed currents through coordinate transformation.
Step 2: Design of detection variables For the generator(grid) phase k, the current residual is the difference between phase measured current i sk (i gk ) and its observed currentî sk ðî gk Þ, then the relative current residual is calculated by the absolute current residual divided by the absolute measured current value and defined as r sk (r gk ), where k = a, b, c.
Theoretically, the current residual and the relative current residual of each phase are approximately equal to 0 in normal operating condition. But affected by the non-ideal of power IGBT module and measurement noise in practice, the current residual exists even in normal operating condition, leading to the problem that the waveform of r sk (r gk ) appears spicule sign. In order to obtain the detection variables with nearly constant values in normal operating condition and which decrease below a certain threshold value when OC faults occur, the necessary mathematical considerations for r sk and r gk are designed as shown in Fig. 4.
Firstly, the amplitude of r sk (r gk ) is adjusted in the range of 0 to 1 through the exponential transformation represented as (6) and the amplitude reconstruction represented as (7):  where R 1 is the amplitude of r sk (r gk ) after exponential transformation; R 2 is the amplitude of R 1 after amplitude reconstruction. Then, the detection variable of the generator (grid) phase k is obtained through a low-pass filter, which is defined as N sk (N gk ).
The relative current residual r sk (r gk ) is approximately equal to 0, and the detection variable N sk (N gk ) is approximately equal to 1 in normal operating condition. Once OC faults occur in the generator(grid) phase k, there will be severe distortion in the current waveform, which results in r sk (r gk ) increasing and the corresponding N sk (N gk ) decreasing, whereas the detection variables of healthy phase keeping invariant. Therefore, it is essential to compare each detection variable with a threshold.
Step 3: Design of adaptive threshold for detection variables Generally, OC faults diagnosis technique depends on comparing the detection variables with the threshold to achieve the detection purposes. The adaptive threshold can be automatic adjusted when the operating conditions change, which helps to improve the reliability and reliability of the diagnosis technique. Thus, two adaptive thresholds D s (for the GESC) and D g (for the GRSC) are designed based on N sk and N gk , respectively: where d s1 and d s2 (d g1 and d g2 ) are the maximum and median value of N sa , N sb and N sc (N ga , N gb and N gc ), respectively. Thus, d s1 , d s2 and D s (d g1 , d g2 and D g ) will be approximately equal to 1 in normal operating condition. Once OC faults occur, there will be an error between N sk (N gk ) and D s (D g ).
For the convenience of describing fault detection results, the detection sign of the generator(grid) phase k is defined as F sk (F gk ): where ɛ s and ɛ g are constant values. If the absolute value of error between N sk (N gk ) and D s (D g ) is less than ɛ s (ɛ g ), F sk (F gk ) is equal to 0, it represents that there is no OC fault in the generator(grid) phase k. On the contrary, the absolute value of error between N sk (N gk ) and D s (D g ) is more than ɛ s (ɛ g ), F sk (F gk ) is equal to 1, it represents that there are OC faults in the generator(grid) phase k.

Fault Localization
Full power converter is three-phase six-switch structure, so the same variety occur to the detection sign of each phase. However, each phase consists of upper and lower bridge arm IGBTs, just using the detection variable would bring difficulties to the following analysis. It is essential to design the fault located variables. In this paper the average value of fault phase current is used to solve this problem, and the located variable of the generator (grid) phase k is defined as L sk (L gk ):, where k = a, b, c; △t is the initial time; T is one current period. It should be noted that the integral interval here is a moving time-window with the length of one current period, which is shown in Fig. 5.
In normal operating condition, the generator (grid) phase k current is sine wave, and L sk (L gk ) is 0 during an integral interval, as illustrated in Fig. 5. When different types of OC faults occur in phase k, respectively, the current waveforms have distinct degree of distortion, L sk (L gk ) also changes with phase k current distortion. In this case, the fixed thresholds M s and M g are respectively introduced for L sk and L gk and the threshold intervals are designed. The localization purposes are achieved through the comparison of L sk (L gk ) and the corresponding threshold interval.

Process of Faults Diagnosis
In practice, OC faults are almost entirely caused by the open-circuit of single and double IGBTs. Both the GESC and the GRSC can be simplified into three-phase voltage-type PWM topology. According to the structural characteristics of three-phase six-arm converter, unilateral converter with OC faults can be classified as follows:   6 shows the whole process of OC faults diagnosis of converters in D-PMSGWT with the proposed method, including the fault detection and the fault localization. As the basis of fault diagnosis, three PMSG phase currents and three grid phase currents are collected. Afterward, the observed current is obtained by designed observer, and the average value of each phase current is calculated at the same time. Then, fault detection algorithm is executed to detect fault phase. If no fault occurs, it will continue to monitor the system operation; otherwise, according to fault localization algorithm, the faulty IGBTs are located through the comparison of located variables and the corresponding threshold intervals finally.
For instance, F sa F sb F sc (F ga F gb F gc ) is 100, which means that OC faults occur in generator (grid) phase a. Similarly, F sa F sb F sc (F ga F gb F gc ) is 110, which means that OC faults occur in generator(grid) phase a and phase b. If L sa (L ga ) is in the range of ±M s (±M g ) and L sb (L gb ) is less than -M s (-M g ), IGBT S1 and S3

Analysis of Examples
The OC faults of converters diagnosis method was tested in a 2-MW D-PMSGWT as depicted in Fig. 7. As shown, the wind turbine system comprises blades, a 2-MW PMSG, full power converter, filter, control module and fault diagnosis module.  The main parameters of this study are introduced in Tab. 2. To be noted, all mentioned OC faults are applied by removing their corresponding gate trigger pulse signals. Furthermore, the gain matrix K of the observer is determined by pole-placement method (k 1 = k 2 = 9000, k 3 = k 4 = 20000). Based on this, robustness test for the fault diagnosis method is carried out, and then different types of OC faults are diagnosed and analyzed.

Robustness Test
In practice, OC faults of converters diagnosis in D-PMSGWT is inevitably compromised by various factors. Among them, random wind speed and grid voltage sags are common conditions during normal operation of converter; regarding the current sensors, they are deployed in harsh environments with high magnetic fields, which leads to the actual collected current signals mixed with noise signals. The robustness test of the proposed diagnosis method is addressed in this section.
Firstly, Fig. 8 shows the waveforms of the generator-side currents together with the detection variables, under wind speed changes abruptly. From t = 0.4 s, wind speed gradually increases from 8 m/s to 11.5 m/s, which is its ratings. At the instant t = 0.7 s, it exceeds 11.5 m/s and becomes constant after reaching 13.5 m/s. After t = 1.1 s, it decreased and stabilized at its ratings. Next, Fig. 9 shows the waveforms of grid voltages, the grid-side currents and detection variables, under phase a grid voltage sags. At t = 0.4 s phase a voltage sags to 60% of the grid voltage ratings. It is obvious from Figs. 8 and 9 that only the current amplitudes change, while the current waveforms are still sine waves and the absolute value of error between N sk (N gk ) and D s (D g ) is still in a safe range during the above conditions. Therefore, the robustness of the fault diagnosis method is verified for wind speed changes abruptly and grid voltage sags.
In addition, 20 dB white Gaussian noise is respectively added into the generator-side and grid-side currents through the awgn function, and the waveforms of corresponding currents, detection variables and THD of currents under noise interference are shown in Figs. 10 and 11, respectively. THD of the generator-side currents increases from 0.87% to 3.28% and THD of grid-side current increases from 0.95% to 3.35% after the noise addition at t = 0.4 s. Although both are less than 5% [27], the increased harmonics will cause power loss, leads to the distortion of grid voltage, and affects power grid security [24]. Despite this, the detection variables are still in a safe range. It is obvious that noise interference will not cause the fault false alarm during the diagnosis process.

Analysis of Faults Diagnosis Results
The diagnosis results of 42 OC faults states produced from the proposed method is addressed in this section, where the diagnostic variables include detection variables and located variables. In Fig. 12 double OC faults in IGBT S1 and S2 of the GESC is considered. In normal operating condition, the generator-side current waveforms are standard sine waves, N sa , N sb , N sc and D s are approximately equal to 1, the detection sign F sa F sb F sc is 000, and L sa , L sb and L sc are equal to 0. S1 fault happens at t = 0.525 s, the current waveforms with important fault info have distortion, especially for phase a, where positive current blocking occurs within a short period of time before the antiparallel diode of S2 turning on. Concerning behaviors of the diagnostic variables, after the fault initiation N sa becomes Figure 9: Waveforms of grid voltages, grid-side currents and detection variables, under phase a grid voltage sags Figure 10: Waveforms of generator-side currents, detection variables and THD of generator-side currents, under noise interference less than D s , whereas N sb and N sc remain close to D s . At t = 0.554 s, the absolute value of error between N sa and D s reaches 0.047, which is more than ɛ s , so F sa F sb F sc becomes 100, which means that OC faults occur in generator phase a. The average value of phase a current L sa is equal to -193.1, which is less than -M s . Combining all the information to compare with Tab. 1, faulty IGBT is correctly located to S1, accounting for 36.2% of the current fundamental period (0.08 s). Then S2 fault happens at t = 0.915 s, phase a current has worse distortion, as a result N sa continues to decrease, whereas N sb and N sc still remain close to D s . At t = 0.953 s, F sa F sb F sc becomes 100, L sa is equal to 3.157, which is in the range of ±M s , the faulty IGBTs are correctly located to S1 and S2, accounting for 47.5% of the current fundamental period.  In Fig. 13 double OC faults in IGBT S1 and S3 of the GESC is considered. S3 fault happens at t = 0.915 s. At t = 0.939 s, F sa F sb F sc becomes 110, L sa is equal to 62.30, which is in the range of ±M s , and L sb is equal to -249.6, which is less than -M s . Combining all the information, the faulty IGBTs are correctly located to S1 and S3, accounting for 30% of the current fundamental period.
In Fig. 14 double OC faults in IGBT S1 and S6 of the GESC is considered. S6 fault happens at t = 0.915 s. At t = 0.931 s, F sa F sb F sc becomes 101, L sa is equal to -330.4, which is less than -M s , and L sb is equal to 352.7, which is more than M s . Combining all the information, the faulty IGBTs are correctly located to S1 and S6, accounting for 20% of the current fundamental period.
To further evaluate the diagnosis performance under faults in the GRSC, double OC faults in IGBT G1 and G2 of the GESC is considered in Fig. 15. The behaviors in normal operating conditions are the same as those in the GESC. At the instant t = 0.361 s G1 fault happens, phase a current only takes negative values, because of no positive current flowing through G1 and the antiparallel diode of G2. In Figure 12: Waveforms of generator-side currents and diagnostic variables, for OC faults in IGBT S1 and S2 Figure 13: Waveforms of generator-side currents and diagnostic variables, for OC faults in IGBT S1 and S3 addition, after the fault initiation N ga becomes less than D g , whereas N gb and N gc remain close to D g . At t = 0.368 s, the absolute value of error between N ga and D g reaches 0.102, which is more than ɛ g , so F ga F gb F gc becomes 100, which means that OC faults occur in grid phase a. And at t = 0.368 s, L ga is equal to -617.8, which is less than -M g , indicating the existence of fault in G1, which takes a time period equivalent to approximately 60% of the current fundamental period (0.02 s). Then, G2 fault happens at t = 0.445 s, and F ga F gb F gc becomes 100 at t = 0.451 s. It can be seen that L ga takes value in the range of ±M g , the faulty IGBTs are correctly located to G1 and G2, accounting for 30% of the current fundamental period.
Next, double OC faults in IGBT G1 and G3 of the GRSC is introduced at t = 0.361 s and t = 0.445 s respectively in Fig. 16. At t = 0.453 s, F ga F gb F gc becomes 110, at t = 0.453 s, L ga is equal to -538.3, which is in the range of ±M g , and L gb is equal to -616.7, which is less than -M g . Combining all the information, the faulty IGBTs are correctly located to G1 and G3, the fault diagnosis process takes a time Figure 14: Waveforms of generator-side currents and diagnostic variables, for OC faults in IGBT S1 and S6 Figure 15: Waveforms of grid-side currents and diagnostic variables, for OC faults in IGBT G1 and G2 period equivalent to 65% of the current fundamental period. Moreover, double OC faults in IGBT G1 and G6 of the GRSC is introduced at t = 0.361 s and t = 0.445 s respectively in Fig. 17. At t = 0.44 9 s, F ga F gb F gc becomes 101, L ga is equal to -975.5, which is less than -M g , and L gc is equal to 699.6, which is more than M g . The faulty IGBTs are correctly located to G1 and G6, which takes a time period equivalent to approximately 20% of the current fundamental period.
The diagnostic variables have the similar characteristics under the same OC fault types. Extensive diagnosis analyses were performed for other 34 OC faults in the GESC and the GRSC. The obtained results can correctly diagnose the faulty IGBTs, which are similar the ones presented here, so they are unnecessary to go into detail.
Furthermore, considering that the wind power system is actually operating under variable wind speed conditions, we take OC faults in IGBT S1 and S3 of the GESC under constant wind speed as an example. Fig. 18 shows waveforms of generator-side currents and diagnostic variables. Compared with Figure 16: Waveforms of grid-side currents and diagnostic variables, for OC faults in IGBT G1 and G3 Figure 17: Waveforms of grid-side currents and diagnostic variables, for OC faults in IGBT G1 and G6 Fig. 13, generator-side currents are not only distorted, but also changed in amplitude, however, the diagnosis was not affected, faulty IGBTs are correctly located to S1 and S3.
The analyzed results show that the thresholds are adaptive and always changed with the corresponding detection variables at the same time during the different fault states, which guarantees the effectiveness and the robustness of the proposed diagnostic method. It also can be seen that OC faults are diagnosed in a time zone less than 65% of the current fundamental period. On the other hand, the localization time depends on the faulty converter, the fault localization for the GESC can be achieved at the same time as the detection, whereas for the GRSC, it is accomplished in a time zone of about 25% of the grid current period after the detection. The diagnostic performance of the proposed method is justified due to its superior diagnostic capabilities, since it allows identifying 42 distinct and complex faulty states, instead of simple single IGBT OC faults.

Comparison Analysis
In order to evaluate the performance of the proposed method, the detection parameter, research object, effectiveness, diagnosable type, Adaptive threshold and robustness are compared with the existing methods.
Tab. 4 lists the comparison results. It can be seen that the proposed fault diagnosis method has the efficient and comprehensive fault diagnosis performance while the other methods have one or more deficiencies. In addition, the proposed fault diagnosis method effectively takes into account the characteristics of the PMSG system, where the used observer signal already exists in the system. Figure 18: Waveforms of generator-side currents and diagnostic variables, for OC faults in IGBT S1 and S3 under constant wind speed 1. This paper proposes a simple and real-time strategy for detection and localization of single and double IGBT OC faults of converters in D-PMSGWT, which only requires three phase currents, and the parameters used are all existing in the system. Therefore, no additional sensors are added and the original topology of the system will not be damaged. 2. The diagnosis method via adaptive threshold has the ability to ensure the robustness to the transient condition, such as variable wind speed condition, grid voltage sags and noise interference, and realize the diagnosis of OC fault on the premise of preventing misinterpretation under the transient condition. 3. The faults diagnosis results show that the OC faults detection and localization are achieved within 20%-65% of the current fundamental period. Moreover, the proposed diagnostic method is simple and efficient, it allows to achieve 42 distinct and complex faulty states detection and localization of single and double IGBT OC faults in both the GESC and the GRSC, showing higher integrity and superior diagnostic capabilities compared with the existing methods of the same type, which meets the needs of modern industry exceedingly. Conflicts of Interest: The authors declare that they have no conflicts of interest to report regarding the present study. Note: "C" three phase currents; "V" three phase voltages; "GESC" generator-side converter; "GRSC" grid-side converter; "I,II,III,IV, V" five OC fault types; "-"not involved.