Method and apparatus for diagnosing a fault condition in an electric machine

The disclosure relates to a method and fault diagnosis apparatus for diagnosing a fault condition in an electric machine.

Electric machines, such as Wound Field Brushless Synchronous Generators (WFBLSGs), are used in a variety of safety and mission critical applications, and, thus, their reliability is of utmost importance. Synchronous machines such as WFBLSGs are particularly susceptible to stator faults. Examples of stator faults include winding faults such as turn-to-turn short circuit faults (i.e. inter-turn short circuit faults), phase-to-ground faults and phase-to-phase faults. Turn-to-turn short circuit faults are initiated by winding insulation degradation. Although turn-to-turn short circuit faults are in themselves a low-impact form of fault, they can lead to phase-to-ground faults and phase-to-phase faults, both of which are catastrophic. Although condition monitoring techniques for detecting and diagnosing stator winding faults do exist, they are deficient in a number of respects.

It is therefore desirable to provide an improved method and fault diagnosis apparatus for diagnosing a fault condition in an electric machine.

According to an aspect of the disclosure, there is provided a method for diagnosing a fault condition in an electric machine. The method comprises: measuring at least one physical parameter generated during operation of the electric machine; analyzing the or each measured parameter in a frequency domain; and determining whether the electric machine has a stator or rotor winding fault based on a comparison of an amplitude of the or each analyzed parameter at a first predetermined frequency and a first threshold amplitude for the first frequency. The at least one physical parameter comprises a sound generated by the electric machine.

The first threshold amplitude may be an upper threshold amplitude. The electric machine may be determined to have a stator or rotor winding fault if the amplitude of the analyzed sound at the first frequency is above the upper threshold amplitude.

The method may further comprise determining whether the electric machine is unbalanced based on a comparison of an amplitude of the or each analyzed parameter at a second predetermined frequency and a second threshold amplitude for the second frequency.

The step of determining whether the electric machine has a stator or rotor winding fault may be carried out only if it is determined that the electric machine is not unbalanced.

The second threshold amplitude may be a lower threshold amplitude. The electric machine may be determined to be unbalanced if the amplitude of the or each analyzed parameter at the second frequency is below the lower threshold amplitude.

The at least one physical parameter may further comprise an electrical parameter and/or a mechanical parameter of the electric machine.

Determining whether the electric machine has a stator or rotor winding fault and/or is unbalanced may be based on an average of the difference between the amplitudes of the analyzed parameters and the threshold amplitudes.

The electric machine may be determined to be healthy if one or fewer of the parameters have amplitudes which when compared with the respective threshold amplitude satisfy a criteria which suggests that the electric machine has a stator or rotor winding fault or is unbalanced.

The method may further comprise determining the type of electric machine and/or the loading the electric machine is subjected to. The or each predetermined frequency and its respective threshold amplitude may be determined based on the type of electric machine and/or its loading.

The loading the electric machine is subjected to may comprise a level of loading and/or whether the loading is linear or non-linear.

The or each predetermined frequency and its respective threshold amplitude may be obtained from a database.

The step of analyzing the or each measured parameter in a frequency domain may comprise generating a spectrogram from the measured parameter. The amplitude of the or each analyzed parameter may be determined from the spectrogram.

The sound generated by the electric machine may be measured using an acoustic sensor spaced from the electric machine.

The electric machine may be an electric generator.

The electric generator may be a Wound Field Brushless Synchronous Generator (WFBLSG).

The stator or rotor winding fault may be a turn-to-turn short circuit stator or rotor winding fault.

According to another aspect of the disclosure, there is provided a fault diagnosis apparatus for diagnosing a fault condition in an electric machine. The fault diagnosis apparatus comprises an acoustic sensor and a processor. The acoustic sensor is configured to measure a sound generated by the electric machine. The processor is configured to: analyze the measured sound in a frequency domain; and determine whether the electric machine has a stator or rotor winding fault based on an amplitude of the analyzed sound at a first frequency and a first condition.

The invention may comprise any combination of the features and/or limitations referred to herein, except combinations of such features that are mutually exclusive.

Arrangements will now be described, by way of example, with reference to the accompanying drawings, in which:

FIG. 1 is a plan schematic view of an example fault diagnosis apparatus for detecting and diagnosing a fault in an electric machine;

FIG. 2 is a schematic view of a first example of a system architecture used in the fault diagnosis apparatus;

FIG. 3 is a flow diagram of a first example method performed by the fault diagnosis apparatus using the architecture of FIG. 2 ;

FIG. 4 is a time-frequency spectrogram for a WFBLSG under linear load at a loading level of 40%;

FIG. 5 is a time-frequency spectrogram for a WFBLSG under linear load at a loading level of 60%;

FIG. 6 is a time-frequency spectrogram for a WFBLSG under non-linear load at a loading level of 40%;

FIG. 7 is a flow diagram of a second example method performed by the fault diagnosis apparatus using the architecture of FIG. 2 ;

FIG. 8 is a schematic view of a second example of a system architecture used in the fault diagnosis apparatus;

FIG. 9 is a flow diagram of a third example method performed by the fault diagnosis apparatus using the architecture of FIG. 8 ;

FIG. 10 is a flow diagram of a fourth example method performed by the fault diagnosis apparatus using the architecture of FIG. 8 ;

FIG. 1 shows a fault diagnosis apparatus 2 for diagnosing a fault condition in an electric machine 4 . The electric machine 4 is a wound field brushless synchronous generator (WFBLSG), and comprises a rotor and a stator (not shown) housed within a housing 20 . The stator comprises a plurality of stator windings, each comprising a plurality of insulated coils. The stator windings are susceptible to stator winding faults such as turn-to-turn short circuit faults (i.e. inter-turn short circuit faults), phase-to-ground faults and phase-to-phase faults. The stator comprises three terminals, each terminal being connected to grid or a separate load, for example a lighting load, a pump load or a motor load. Ideally, the magnitudes of the loads connected to each terminal are the same. If loads having different magnitudes are connected to one or more of the terminals, the electric machine 4 is considered to have a load unbalance condition.

The fault diagnosis apparatus 2 generally comprises a sensor 6 , a processor 8 and a display 7 . The fault diagnosis apparatus 2 may form part of a mobile phone or a tablet computer. Alternatively, the fault diagnosis apparatus 2 may be a dedicated device. In use, the sensor 6 is positioned externally with respect to the electric machine 4 , and is separated from the electric machine 4 by an air gap 21 . The sensor 6 is an acoustic sensor (such as a microphone) configured to measure sound (i.e. acoustic waves) produced by the electric machine 4 , and output a signal to the processor 8 that corresponds to the measured sound. The sensor 6 is configured to measure sound in the frequency range of a few hertz to 20 kilohertz. The fault diagnosis apparatus 2 , in particular the processor 8 , is configured to diagnose a fault condition based on characteristics in the sound generated by the electric machine 4 . The characteristics may indicate the presence or absence of stator winding faults, in particular stator winding faults such as turn-to-turn short circuit faults, phase-ground faults and phase-phase faults, or a load unbalance fault. The diagnosis made by the processor 8 may be displayed on a display 7 , thereby allowing a user to take an appropriate course of action based on the diagnosis.

FIG. 2 shows a first example architecture used in the fault diagnosis apparatus 2 . As shown, the fault diagnosis apparatus 2 comprises a data acquisition unit 26 , a signal processing unit 28 , an information collection unit 30 and a fault detection and diagnosis unit 32 . The sensor 6 forms part of the data acquisition unit 26 . The signal processing unit 28 , information collection unit 30 and fault detection and diagnosis unit 32 form part of the processor 8 .

The fault diagnosis apparatus 2 is described with reference to FIG. 3 , which shows a flow diagram of a first example method performed by the fault diagnosis apparatus 2 using the first example architecture. Upon initiation of the fault diagnosis apparatus 2 , step S 2 is carried out by the data acquisition unit 26 . During step S 2 , power is supplied to the electric machine 4 such that the rotor moves relative to the stator. This movement generates sound, which is transmitted through the air gap 21 to the sensor 6 . The sensor 6 is thus able to measure the sound generated by the electric machine 4 over a period of time, and subsequently output a signal in the form of an electrical signal to the signal processing unit 28 that is representative of the sound generated by the electric machine 4 .

Step S 4 is carried out by the signal processing unit 28 . The signal processing unit 28 comprises a spectrograph. The spectrograph receives the signal from the data acquisition unit 26 and performs frequency spectrum analysis on it so as to generate a spectrogram based on the signal.

FIG. 4 shows an example of a spectrogram. As shown, the spectrogram is a frequency-time plot, with both the frequency and time axes being represented linearly. The amplitude of the signal received from the data acquisition unit 26 at a specified frequency and at a specified time is represented on the spectrogram by the color of the spectrogram at a position corresponding to that particular frequency and time. The spectrogram may show up to 20th order harmonics.

The spectrogram shown in FIG. 4 is generated by the signal processing unit 28 in step S 4 for a first example electric machine, operated over a period of 80 seconds. The measured frequency range is from 0 to 800 Hertz. The amplitude is represented in decibels. The 0 decibel reference value is selected such that all amplitudes are displayed as having a negative decibel amplitude.

Returning to FIG. 3 , step S 6 is carried out by the information collection unit 30 . The information collection unit 30 comprises a knowledge base (i.e. a database), which contains signature data. The signature data can be determined experimentally or mathematically. The signature data comprises load unbalance condition signature data and stator winding fault signature data. In the present arrangement, the stator winding fault signature data relates to turn-to-turn short circuit faults (i.e. inter-turn short circuit faults). However, in other arrangements, the stator winding fault signature data may relate to other stator winding faults such as phase-to-ground faults and phase-to-phase faults. A signature data set comprising load unbalance condition signature data and stator winding fault signature data exists for a variety of different types of electric machine 4 (for example WFBLSGs or brushed DC motors) operating under a variety of different loading conditions. The loading conditions relate to the type of load that the electric machine 4 is subjected to (e.g. linear or non-linear load) and the level of loading that the electric machine 4 is subjected to (e.g. 40% loading or 60% loading).

The load unbalance condition signature data for each type of electric machine 4 under each type of loading condition comprises a frequency value (hereinafter referred to as the unbalance signature frequency) and an associated threshold condition (hereinafter referred to as the unbalance signature threshold condition). Likewise, the stator winding fault condition signature data for each type of electric machine 4 under each type of loading condition comprises a frequency value (hereinafter referred to as the stator signature frequency) and an associated threshold condition (hereinafter referred to as the stator signature threshold condition). The unbalance signature threshold condition and the stator signature threshold condition each comprise an amplitude value and an indication as to whether the amplitude value is an upper threshold or a lower threshold. Step S 6 comprises identifying what type of electric machine 4 is being tested, what type of loading it is being subjected to, and retrieving the associated load unbalance condition signature data and stator winding fault condition signature data from the knowledge base.

Step S 8 is also carried out by the information collection unit 30 . In step S 8 , an amplitude (hereinafter referred to as the extracted unbalance amplitude) at the unbalance signature frequency at a current or recently elapsed time is extracted (i.e. measured or determined) from the spectrogram. Likewise, an amplitude (hereinafter referred to as the extracted stator amplitude) at the stator signature frequency at the current or recently elapsed time is also extracted (i.e. measured or determined) from the spectrogram. The extracted unbalance amplitude and the extracted stator amplitude may represent the amplitudes of the spectrogram at a single, discrete point in time. Alternatively, they may represent the average amplitude of the spectrogram over a period of time, for example 1 second. The extracted unbalance amplitude and the extracted stator amplitude, along with their associated unbalance signature threshold condition and stator signature threshold condition, are outputted to the fault detection and diagnosis unit 32 , whereupon step S 10 takes place.

Step S 10 is carried out by the fault detection and diagnosis unit 32 . In step S 10 , a preliminary check is carried out in order to ensure that the electric machine 4 does not have a load unbalance condition. Specifically, the extracted unbalance amplitude is assessed as to whether it meets the conditions prescribed by the unbalance signature threshold condition. If the extracted unbalance amplitude does meet the conditions prescribed by the unbalance signature threshold condition, the electric machine 4 is determined to have a load unbalance condition, whereupon step S 2 is repeated. This error condition may be displayed on the display 7 . If the extracted unbalance amplitude does not meet the conditions prescribed by the unbalance signature threshold condition, the electric machine 4 is determined not to have a load unbalance condition, whereupon step S 12 is carried out.

Step S 12 is also carried out by the fault detection and diagnosis unit 32 . In step S 12 , a main check is carried out in order to determine whether the electric machine 4 has a stator winding fault, in particular a turn-to-turn short circuit fault (i.e. an inter-turn short circuit fault). Specifically, the extracted stator amplitude is assessed as to whether it meets the conditions prescribed by the stator signature threshold condition. If the extracted stator amplitude does meet the conditions prescribed by the stator signature threshold condition, the electric machine 4 is determined to have a stator winding fault. This error condition may be displayed on the display 7 . If the extracted stator amplitude does not meet the conditions prescribed by the stator signature threshold condition, the electric machine 4 is determined not to have a stator winding fault and to be healthy, whereupon step S 2 is repeated. The healthy condition of the electric machine 4 may be displayed on the display 7 , until such time as the electric machine 4 is determined to have an error condition.

The abovementioned method ensures early detection of stator winding faults, which, if left undetected, may develop into catastrophic faults. Early fault detection allows a system to move from scheduled maintenance to predictive maintenance, which increases the availability of the electric machines 4 . Furthermore, acoustic based winding fault detection is simple, non-intrusive and easy to implement. The method does not require current and voltage sensors, which must be rated equally to the power rating of the electric machine 4 . The use of an acoustic sensor 6 is non-intrusive. Specifically, electric machine 4 need not be modified in order for the acoustic sensor 6 to accurately measure the sound generated by the electric machine 4 . Further, the acoustic sensor 6 need not even be in contact with the electric machine 4 , and can be spaced therefrom by an air gap 21 . Accordingly, the use of mountings for the sensor 6 on the electric machine 4 is obviated. Further, since the acoustic sensor 6 is not integral with the electric machine 4 , it does not affect the operation of the electric machine 4 , and, as such, the measurements taken by the sensor 6 . Consequently, the accuracy of the sound produced by the electric machine 4 and sensed by the sensor 6 is improved, thus improving accuracy of fault detection.

The first example method will now be illustrated with reference to an actual electric machine 4 under a known loading condition. In the following example, the electric machine 4 is a WFBLSG, the type of load is a linear load, and the level of loading is 40%. The spectrogram generated by the signal processing unit 28 in step S 4 for this type of electric machine 4 under these loading conditions is shown in FIG. 4 (referred to previously).

In step S 6 , the type of electric machine 4 being tested is identified as a WFBLSG under linear load at 40% loading. An unbalance signature frequency, an unbalance signature threshold condition, a stator signature frequency and a stator signature threshold condition for a WFBLSG under linear load at 40% loading are retrieved from the knowledge base. In the present example, the unbalance signature frequency is 500 Hertz, the unbalance signature threshold condition is less than −30 decibels, the stator signature frequency is 100 Hertz and the stator signature threshold condition is greater than −15 decibels.

In step S 8 , the amplitude at the unbalance signature frequency of 500 Hertz (i.e. the extracted unbalance amplitude) is extracted from the spectrogram and determined to be −25 decibels. Likewise, the amplitude at the stator signature frequency of 100 Hertz (i.e. the extracted stator amplitude) is extracted from the spectrogram and determined to be −20 decibels.

In step S 10 , the extracted unbalance amplitude of −25 decibels is determined not to meet the requirements of the unbalance signature threshold condition of being less than −30 decibels. Accordingly, the electric machine 4 is determined not to have a load unbalance condition, whereupon step S 12 is carried out.

In step S 12 , the extracted stator amplitude of −20 decibels is determined not to meet the requirements of the stator signature threshold condition of being greater than −15 decibels. Accordingly, the electric machine 4 is determined not to have a stator winding fault condition and to be healthy, whereupon step S 2 is carried out.

This process repeats, until approximately 50 seconds has elapsed. After 50 seconds has elapsed, in step S 8 , the extracted unbalance amplitude is again determined to be −25 decibels. In contrast, extracted stator amplitude is now determined to be −10 decibels.

In step S 10 , the extracted unbalance amplitude of −25 decibels is again determined not to meet the requirements of the unbalance signature threshold condition of being less than −30 decibels. Accordingly, again, the electric machine 4 is determined not to have a load unbalance condition, whereupon step S 12 is carried out.

In step S 12 , the extracted stator amplitude of −10 decibels is determined to meet the requirements of the stator signature threshold condition of being greater than −15 decibels. Accordingly, the electric machine 4 is determined to have a stator winding fault condition.

The first example method will now be illustrated with reference to another actual electric machine 4 under known loading conditions. In the following example, the electric machine 4 is again a WFBLSG under linear load, however the level of loading is 60%. The spectrogram generated by the signal processing unit 28 in step S 4 for this type of electric machine 4 under these loading conditions is shown in FIG. 5 .

In step S 6 , the type of electric machine 4 being tested is identified as a WFBLSG under linear load at 60% loading. An unbalance signature frequency, an unbalance signature threshold condition, a stator signature frequency and a stator signature threshold condition for a WFBLSG under linear load at 60% loading are retrieved from the knowledge base. In the present example, the unbalance signature frequency is 500 Hertz, the unbalance signature threshold condition is less than −25 decibels, the stator signature frequency is 100 Hertz and the stator signature threshold condition is greater than −5 decibels.

In step S 8 , the amplitude at the unbalance signature frequency of 500 Hertz (i.e. the extracted unbalance amplitude) is extracted from the spectrogram and determined to be −20 decibels. Likewise, the amplitude at the stator signature frequency of 100 Hertz (i.e. the extracted stator amplitude) is extracted from the spectrogram and determined to be −10 decibels.

In step S 10 , the extracted unbalance amplitude of −20 decibels is determined not to meet the requirements of the unbalance signature threshold condition of being less than −25 decibels. Accordingly, the electric machine 4 is determined not to have a load unbalance condition, whereupon step S 12 is carried out.

In step S 12 , the extracted stator amplitude of −10 decibels is determined not to meet the requirements of the stator signature threshold condition of being greater than −5 decibels. Accordingly, the electric machine 4 is determined not to have a stator winding fault condition and to be healthy, whereupon step S 2 is carried out.

This process repeats, until approximately 20 seconds has elapsed. After 20 seconds has elapsed, in step S 8 , the extracted stator amplitude is again determined to be −10 decibels. In contrast, the extracted unbalance amplitude is now determined to be −35 decibels.

In step S 10 , the extracted unbalance amplitude of −35 decibels is determined to meet the requirements of the unbalance signature threshold condition of being less than −25 decibels. Accordingly, the electric machine 4 is determined to have a load unbalance condition, whereupon step S 2 is carried out.

This process repeats for the entirety of the remainder of the 80 second period, shown in FIG. 5 .

The first example method will now be illustrated with reference to yet another actual electric machine 4 under known loading conditions. In the following example, the electric machine 4 is a WFBLSG subjected to 40% loading, as per the example shown in FIG. 4 . However, the WFBLSG is under a non-linear load, rather than a linear load. The spectrogram generated by the signal processing unit 28 in step S 4 for this type of electric machine 4 under these loading conditions is shown in FIG. 6 .

In step S 6 , the type of electric machine 4 being tested is identified as a WFBLSG under non-linear load at 40% loading. An unbalance signature frequency, an unbalance signature threshold condition, a first stator signature frequency and a first stator signature threshold condition for a WFBLSG under non-linear load at 40% loading are retrieved from the knowledge base. For the type of electric machine 4 being tested under the specific type of load conditions, there is also a second stator signature frequency and a second signature threshold condition.

In the present example, the unbalance signature frequency is 500 Hertz, the unbalance signature threshold condition is less than −30 decibels, the first stator signature frequency is 400 Hertz, the first stator signature threshold condition is greater than −20 decibels, the second stator signature frequency is 700 Hertz and the second stator signature threshold condition is greater than −25 decibels.

In step S 8 , the amplitude at the unbalance signature frequency of 500 Hertz (i.e. the extracted unbalance amplitude) is extracted from the spectrogram and determined to be −25 decibels. Likewise, the amplitude at the first stator signature frequency of 400 Hertz (i.e. the first extracted stator amplitude) is extracted from the spectrogram and determined to be −30 decibels, and the amplitude at the second stator signature frequency of 700 Hertz (i.e. the second extracted stator amplitude) is extracted from the spectrogram and determined to be −30 decibels.

In step S 10 , the extracted unbalance amplitude of −25 decibels is determined not to meet the requirements of the unbalance signature threshold condition of being less than −30 decibels. Accordingly, the electric machine 4 is determined not to have a load unbalance condition, whereupon step S 12 is carried out.

In step S 12 , the first extracted stator amplitude of −30 decibels is determined not to meet the requirements of the first stator signature threshold condition of being greater than −20 decibels. Likewise, the second extracted stator amplitude of −30 decibels is determined not to meet the requirements of the second stator signature threshold condition of being greater than −25 decibels. Accordingly, the electric machine 4 is determined not to have a stator winding fault condition and to be healthy, whereupon step S 2 is carried out.

This process repeats, until approximately 50 seconds has elapsed. After 50 seconds has elapsed, in step S 8 , the extracted unbalance amplitude is again determined to be −25 decibels and the second extracted stator amplitude is again determined to be −30 decibels. In contrast, the first extracted unbalance amplitude is now determined to be −15 decibels.

In step S 10 , the extracted unbalance amplitude of −25 decibels is determined not to meet the requirements of the unbalance signature threshold condition of being less than −30 decibels. Accordingly, the electric machine 4 is determined not to have a load unbalance condition, whereupon step S 12 is carried out.

In step S 12 , the first extracted stator amplitude of −15 decibels is determined to meet the requirements of the first stator signature threshold condition of being greater than −20 decibels. Accordingly, the electric machine 4 is determined to have a stator winding fault condition.

In alternative example methods, both the requirements of the first stator signature threshold condition and the second stator signature threshold condition must be met in order for the electric machine 4 to be determined to have a stator winding fault condition.

As outlined above, in the first example method, the step of determining whether the electric machine 4 has a stator winding fault (i.e. step S 12 ) is only ever carried out if the electric machine 4 is determined not to be unbalanced (in step S 10 ). Accordingly, step S 10 acts as a preliminary check, reducing the number of false alarms of occurrences of stator winding faults. A second example method is shown in FIG. 7 , in which step S 12 is carried out regardless of whether the electric machine 4 is determined to have a stator winding fault. Specifically, in the second example architecture, step S 12 is carried out regardless of the outcome of step S 10 .

FIG. 8 shows a second example architecture used in a fault diagnosis apparatus 102 . As per the fault diagnosis apparatus 2 of the first example architecture, the fault diagnosis apparatus 102 of the second example architecture is configured to diagnose a fault condition present in the electric machine 4 .

The fault diagnosis apparatus 102 comprises a data acquisition unit 126 , a signal processing unit 128 , an information collection unit 130 and a fault detection and diagnosis unit 132 . The data acquisition unit 126 comprises an acoustic sensor 106 a substantially corresponding to the sensor 6 of the first example system architecture 2 . The data acquisition unit 126 further comprises an electrical sensor 106 b and a mechanical sensor 106 c.

The fault diagnosis apparatus 102 is described with reference to FIG. 9 , which shows a flow diagram of a third example method performed by the fault diagnosis apparatus 102 using the second example architecture. Upon initiation of the fault diagnosis apparatus 102 , step S 102 is carried out by the data acquisition unit 126 . Step S 102 substantially corresponds to step S 2 , but additionally comprises measuring and generating a signal corresponding to one or more of terminal voltage, exciter field current, main field current or exciter current using the electrical sensor 106 b , and measuring and generating a signal corresponding to one or more of vibration or temperature using the mechanical sensor 106 c . In other examples, the electrical sensor 106 b may measure and generate a signal corresponding to any type of electrical parameter and the mechanical sensor 106 c may measure and generate a signal corresponding to any type of mechanical parameter. The signals generated by the acoustic sensor 106 a , the electrical sensor 106 b and the mechanical sensor 106 c are outputted in the form of separate electrical signals to the signal processing unit 128 .

Step S 104 is carried out by the signal processing unit 128 . Step S 104 substantially corresponds to step S 4 . Specifically, a spectrograph in the signal processing unit 128 generates a frequency-time spectrogram (hereinafter referred to as an acoustic spectrogram) from the signal produced by the acoustic sensor 106 a . In addition, the signal processing unit 128 generates separate frequency-time spectrograms from the signals produced by the electrical sensor 106 b (hereinafter referred to as the electrical spectrogram) and the mechanical sensor 106 c (hereinafter referred to as the mechanical spectrogram).

Step S 106 is carried out by the information collection unit 130 . Step 106 substantially corresponds to step S 6 . As per step S 6 , step S 106 comprises retrieving load unbalance condition signature data and stator winding fault condition signature data from the knowledge base for the type of electric machine 4 being analyzed under the applicable load conditions. As is done for the first example architecture 2 , load unbalance condition signature data and stator winding fault condition signature data is retrieved for the acoustic spectrogram. In addition, in step 106 , load unbalance condition signature data and stator winding fault condition signature data is retrieved for the electrical spectrogram and the mechanical spectrogram.

Step S 108 is also carried out by the information collection unit 130 . Step S 108 substantially corresponds to step S 8 , however extracted unbalance amplitudes and extracted stator amplitudes are additionally extracted from the electrical spectrogram and the mechanical spectrogram at their respective unbalance signature frequencies and stator signature frequencies. The extracted unbalance amplitudes and the extracted stator amplitudes, along with the unbalance signature threshold conditions and the stator signature threshold conditions for each of the acoustic spectrogram, the electrical spectrogram and the mechanical spectrogram are outputted to the fault detection and diagnosis unit 132 , whereupon step S 110 takes place.

Step S 110 is carried out by the fault detection and diagnosis unit 132 . Step S 110 comprises three steps S 110 a , S 110 b , S 110 c that are carried out simultaneously or sequentially. Step S 110 a is similar to step S 10 . As previously mentioned, step S 10 involves determining whether the extracted unbalance amplitude meets the conditions prescribed by the unbalance signature threshold condition. This is a binary determination; the extracted unbalance amplitude is either determined to meet the conditions prescribed by the unbalance signature threshold condition or determined not to meet the conditions prescribed by the unbalance signature threshold condition. In contrast, in step S 110 a , the extent to which the extracted unbalance amplitude meets the conditions prescribed by the unbalance signature threshold condition is determined. In particular, a difference between the extracted unbalance amplitude and the unbalance signature threshold may be calculated. If the extracted unbalance amplitude does not meet the unbalance signature threshold condition, then the difference is negative, whereas if it does, then the difference is positive (or vice versa). A similar determination is carried out for the electrical and mechanical spectrograms in steps S 110 b and S 110 c.

In step S 111 , the difference values generated in steps S 110 a , S 110 b and S 110 c are averaged. It may be necessary for the values to be normalized prior to averaging. The values may also be weighted in the average such that certain measures make a greater contribution. The weighting may be based on the expected accuracy of the measure or some other reason. If the averaged values from the acoustic, electrical and mechanical spectrograms exceed a threshold, the electric machine 4 is determined to have an unbalance load condition, whereupon step S 102 is carried out. The threshold may be set at zero or some other value. If the averaged values from the acoustic, electrical and mechanical spectrograms do not exceed the threshold, the electric machine 4 is determined not to have an unbalance load condition, whereupon step S 112 is carried out.

Step S 112 is also carried out by the fault detection and diagnosis unit 132 . Step S 112 comprises three steps S 112 a , S 112 b , S 112 c that are carried out simultaneously or sequentially. Step S 112 a is similar to step S 12 . As previously mentioned, step S 12 involves determining whether the extracted stator amplitude meets the conditions prescribed by the stator signature threshold condition. In contrast, in step S 112 a , the extent to which the extracted stator amplitude meets the conditions prescribed by the stator signature threshold condition is determined. In particular, a difference between the extracted stator amplitude and the stator signature threshold may be calculated. If the extracted stator amplitude does not meet the stator signature threshold condition, then the difference is negative, whereas if it does, then the difference is positive (or vice versa). A similar determination is carried out for the electrical and mechanical spectrograms in steps S 112 b and S 112 c.

In step S 113 , the difference values generated in steps S 110 a , S 110 b and S 110 c are averaged. It may be necessary for the values to be normalized prior to averaging. The values may also be weighted in the average such that certain measures make a greater contribution. The weighting may be based on the expected accuracy of the measure or some other reason. If the averaged values from the acoustic, electrical and mechanical spectrograms exceed a threshold, the electric machine 4 is determined to have a stator winding fault. The threshold may be set at zero or some other value. If the averaged values from the acoustic, electrical and mechanical spectrograms do not exceed the threshold, the electric machine 4 is determined not to have stator winding fault and to be healthy, whereupon step S 102 is carried out.

It will be appreciated that the method described above allows multiple modalities to be combined to provide a more accurate determination. In particular, a single measurement technique (or even two measurement techniques) may not identify a load unbalance or a stator winding fault, but provided the other measurements are sufficiently conclusive to offset this, then a load unbalance condition or stator winding fault will be identified.

FIG. 10 shows a fourth example method performed by the fault diagnosis apparatus 102 using the second example architecture. The fourth example method substantially corresponds to the third example method. In particular, steps S 202 , S 204 , S 206 and S 208 correspond to steps S 102 , S 104 , S 106 and S 108 , respectively. Step S 210 comprises steps S 210 a , S 210 b and S 210 c , which substantially correspond to steps S 110 a , S 110 b and S 110 c , respectively. However, as per step S 10 , in each of steps S 210 a , 210 b and 210 c , a binary (i.e. yes/no) decision is made as to whether their respective extracted unbalance amplitudes meet the conditions prescribed by their respective unbalance signature threshold conditions.

In step S 110 , if two or more of the outputs from steps S 210 a , 210 b and 210 c indicate that the extracted unbalance amplitudes have met the conditions prescribed by their respective unbalance signature threshold conditions, the electric machine 4 is determined to have a load unbalance condition, whereupon step S 202 is carried out. If one or no outputs from steps S 210 a , 210 b and 210 c indicate that the extracted unbalance amplitudes have met the conditions prescribed by their respective unbalance signature threshold conditions, the electric machine 4 is determined not to have a load unbalance condition, whereupon step S 212 is carried out.

Step S 212 comprises steps S 212 a , S 212 b and S 212 c , which substantially correspond to steps S 112 a , S 112 b and S 112 c , respectively. However, as per step S 12 , in each of steps S 112 a , S 112 b and S 112 c , a binary (i.e. yes/no) decision is made as to whether their respective extracted stator amplitudes meet the conditions prescribed by their respective stator signature threshold conditions.

In step S 213 , if two or more of the outputs from steps S 212 a , S 212 b and S 212 c indicate that the extracted stator amplitudes have met the conditions prescribed by their respective stator signature threshold conditions, the electric machine 4 is determined to have a stator winding fault. If one or no outputs from steps S 212 a , S 212 b and S 212 c indicate that the extracted stator amplitudes have met the conditions prescribed by their respective stator signature threshold conditions, the electric machine 4 is determined not to have a stator winding fault, whereupon step S 202 is carried out.

By using electrical and mechanical data in addition to acoustic data, the accuracy of fault detection and diagnosis of the third and fourth example methods is improved. In other examples, the acoustic data may be supplemented with only electrical or only mechanical data.

In some arrangements, more than one acoustic sensor 6 may be used, again, in order to improve accuracy.

With reference to the third and fourth example methods, it has been described that the signal processing unit 128 generates separate frequency-time spectrograms from the signals produced by the electrical sensor 106 b and the mechanical sensor 106 c , and that data extracted from these spectrograms (in conjunction with the acoustic spectrogram) is subsequently used to determine whether the electric machine 4 has a load unbalance condition or a stator winding fault condition. However, the signal processing unit 128 can be used to process the signals produced by the electrical sensor 106 b and the mechanical sensor 106 c in any known manner in order to produce an output that can be used in steps S 106 to S 113 and S 206 to S 213 to determine whether the electric machine 4 has a load unbalance condition or a stator winding fault condition.

The electric machine 4 has been described as being a WFBLSG. However, the electric machine 4 may be any type of electric machine, such as an induction machine or a permanent magnet electric machine, such as a permanent magnet synchronous machine. The electric machine 4 may be or act either as an electric generator or an electric motor. The electric machine 4 may use any type of winding.

It has been described that the fault diagnosis apparatus determines whether the electric machine 4 has a turn-to-turn short circuit fault (i.e. an inter-turn short circuit fault). However, the fault diagnosis apparatus may determine whether the electric machine 4 has other types of stator winding faults such as phase-to-ground faults or phase-to-phase faults. The fault diagnosis apparatus may also be used to identify rotor winding faults.

Although it has been described that the determination as to whether the electric machine 4 is unbalanced is carried out prior to the determination as to whether the electric machine 4 has a stator winding fault, this need not be the case. Instead, the determination as to whether the electric machine 4 has a stator winding fault may be carried out prior to the determination as to whether the electric machine 4 is unbalanced.

The example methods have been described as being carried out over a period of time, for example 80 seconds. The methods may be carried out throughout the entire period of operation of the electric machine 4 . If the electric machine 4 is found to have a load unbalance condition or not to have a stator winding fault, the method repeats indefinitely. However, in alternative arrangements, the method may not repeat. Instead, a single iteration of the method may be carried out. Multiple non-repeating iterations of the method may be carried out at regular intervals of time, for example 10 second intervals.

In the example methods, it has been described that a spectrogram is produced from which an extracted unbalance amplitude and an extracted stator amplitude can be obtained. As described previously, a spectrogram is a frequency-time plot, with amplitude at a specific frequency and time represented by way of color, for example. It is not, however, necessary to produce a spectrogram in order to determine extracted unbalance amplitudes and extracted stator amplitudes. Instead, individual amplitude-frequency plots may be produced, one for each cycle of the method.

The electric machine 4 may be used for any application, for example aeronautical applications, marine applications and energy and nuclear applications. The electric machine 4 may be used for on board power generation. The electric machine 4 may, for example, be used as a shaft generator in energy efficient hybrid propulsion systems, such as those used in marine vessels.