JP2005251185A - Electric equipment diagnostic system - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an electric equipment diagnostic system capable of diagnosing and determining a state of electric equipment using statistical pattern recognition technique without requiring vast quantity of data. <P>SOLUTION: In the electric equipment diagnostic system 1, when featured values detected by a sensor 3 in which a sensor which detects current and voltage of the electric equipment 2, a sensor which detects a rotational angle position and rotation speed of a rotator, a torque sensor, a temperature sensor and a vibration sensor, etc. are generically named are inputted in a learning part 5 via an A/D conversion circuit 4, the learning part 5 parameterizes the respective featured value corresponding to a normal state, a failed state and deterioration state of a motor based on a hidden Markov model and stores them. A diagnostic part 6 determines the state of the electric equipment 2 by deciding the parameter in which output probability becomes maximum upon calculating the output probability of patterns of the featured values detected by an operation state of the electric equipment 2 based on the parameters stored in the learning part 5 and displays the state on a display part 7. <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

  The present invention relates to an electrical equipment diagnosis system that diagnoses the state of electrical equipment provided in a factory, a building, or the like using a statistical pattern recognition method.

  Conventionally, electrical equipment diagnostic systems for diagnosing failures and deterioration that occur in electrical equipment such as electric motor systems and power receiving systems in factories and buildings, etc., frequency-analyzed waveforms such as voltage, current, noise, and vibration detected by sensors. In addition, there is one that diagnoses and determines a failure or deterioration of an electrical facility by detecting a specific frequency component that appears in the failure state or deterioration state of the electrical facility (see Non-Patent Document 1).

In the case of the above-described conventional electrical equipment diagnostic system, when detecting the specific frequency component to diagnose and determine the state of the electrical equipment, the same frequency component may be seen even if the electrical equipment has a different failure state. Even if it is possible to determine whether or not there is a failure, it is difficult to specify the failure state. Furthermore, it is difficult to specify a failure state from frequency components in an environment that is susceptible to noise.
In addition, by simply comparing the waveform patterns of features such as voltage, current, noise, vibration, etc. detected by the sensor, when diagnosing electrical equipment failure or deterioration, there are various factors such as normal, deterioration, and abnormality. Since the waveform according to the state is required, the number thereof is enormous and a large storage area is required. Further, when the sensor picks up noise, each waveform is deformed by the influence of the noise, so that the number of prepared waveforms is further increased. At the same time, it is difficult to specify the failure state of the electrical equipment when simply comparing waveforms that vary in value due to the influence of noise.
Proceedings of the Annual Conference of the Institute of Electrical Engineers of Japan (5) Electrical Equipment pp. 501-504 “Recent trends in high-performance induction machine technology: Failure diagnosis technology for induction machines” Authors Kurihara, Nakamura, Miyashita, Ogata

  Therefore, in the present invention, it should be solved to provide an electrical equipment diagnosis system capable of diagnosing and determining the state of electrical equipment from the overall shape of the feature amount pattern using a statistical pattern recognition method. It is to be an issue.

The above problems can be solved by the electrical equipment diagnosis system described in the appended claims.
According to the electrical equipment diagnosis system according to claim 1, the learning means detects a feature amount in each normal state, failure / abnormal state, deterioration state, etc. of the electrical equipment to be diagnosed, Each feature pattern is parameterized based on the hidden Markov model, and each parameter is stored. The diagnostic means calculates the generation probability (likelihood) of the feature amount pattern detected in the operating state of the electrical equipment based on the respective parameters stored in the learning means, and sets the maximum probability. Based on the diagnostic information corresponding to the parameters, the electrical equipment is diagnosed as normal or abnormal, and when it is diagnosed as an abnormal state, the contents are determined.

  It should be noted that fluctuations occur in the pattern of the feature amount generated from each electric equipment even when the electric equipment is in the same state. How to deal with these fluctuations becomes a problem when diagnosing electrical equipment with the shape of the same pattern. Therefore, in this invention, said subject is solved by applying a hidden Markov model. In the Hidden Markov Model, this pattern is expressed using probability parameters such as state transition probability and output probability, so fluctuations in time expansion, fluctuations in amplitude, spectrum, etc., fluctuations due to noise, etc. are modeled statistically. Can do.

  Hidden Markov models do not diagnose using a single point value at a certain time, but diagnose the state of the electrical equipment from the overall shape of the feature pattern, which makes it possible to realize a diagnostic system that is strong against variations in measured values. .

  In addition, the frequency spectrum of the observed feature value changes from time to time due to the influence of observation noise, etc., that is, the size of a specific frequency component also changes from time to time. When determining the failure state of the electrical equipment by the size of the component, there is a possibility of making a wrong diagnosis. However, when the hidden Markov model is applied, fluctuations due to observation noise or the like can also be expressed as probabilities, so that erroneous determination due to variations in feature value values can be reduced.

  In addition, when simply comparing the waveform patterns of the feature quantities generated from the respective electric facilities as in the conventional case, the data that must be prepared in advance is enormous. In addition, when simple waveform pattern comparison is performed under the condition that the feature amount is affected by noise, waveform patterns with various feature amounts that take noise into account are required, and the number of features increases further, and the state of electrical equipment It becomes difficult to diagnose.

  On the other hand, the present invention prepares in advance for diagnosing and determining the state of the electrical equipment by expressing the pattern of the feature quantity stochastically even with the feature quantity including noise as described above. It is possible to reduce the number of data of the characteristic feature pattern. Thereby, the calculation amount until calculating the diagnosis result can also be reduced. In addition, it is possible to solve the difficulty of diagnosis of electrical equipment due to the influence of noise, and to improve the accuracy of diagnosis of failure and deterioration of electrical equipment.

In addition, according to the electrical equipment diagnosis system of the second aspect, the state of the electrical equipment can be roughly determined from the waveform of the specific frequency component by the frequency analysis means. In addition to this, since the learning means and the diagnosis means can diagnose and determine the electrical equipment, the state of the electrical equipment can be diagnosed and judged more accurately.
Furthermore, according to the electrical equipment diagnostic system of the third aspect, since the parameters stored in advance in the learning means can be updated or exchanged, the versatility of the electrical equipment diagnostic system can be enhanced.

Here, the concept of the hidden Markov model will be described with reference to the schematic diagram shown in FIG.
In FIG. 1, a11, a12, a22, a23, a33, a34 indicate state transition probabilities, and aij is the probability of transition from state Si to state Sj. Further, bi (x) indicates the probability of outputting the feature quantity x in the state Si. S1 indicates an initial state, and S4 indicates a final state. Thus, the hidden Markov model has the state transition probability, the output probability, and the initial state probability πi (probability that the initial state is Si) as parameters, and each parameter is a normal state of the electrical equipment, a failure to be recognized, This is stored for each abnormality and deterioration state, and corresponds to the function of the learning means described above.

  According to the present invention, whether or not an electrical facility is normal by detecting a feature amount generated during operation of the electrical facility and applying a hidden Markov model, which is a stochastic statistical pattern recognition method, to the feature amount pattern. It is possible to accurately diagnose and determine whether a failure or an abnormality is present.

Next, an embodiment of the present invention will be described.
FIG. 2 is a block diagram showing the overall configuration of the electrical equipment diagnosis system 1. As shown in FIG. 2, the electrical equipment diagnosis system 1 includes various sensors for detecting a current flowing through the electrical equipment 2, a voltage applied to the electrical equipment 2, sound and vibration generated from the electrical equipment 2, temperature, and the like. 3 is provided. Since the detection signals of the feature amounts detected by these sensors 3 are generally analog signals, each analog signal is converted into a digital signal by the A / D conversion circuit 4. The digital signal output from the A / D conversion circuit 4 is input to the learning unit 5 described below in the learning stage, and is input to the diagnosis unit 6 in the operating state of the electrical equipment 2.

The learning unit 5 inputs detection signals of various feature amounts detected by the sensor 3 in a plurality of predetermined states (normal state, failure / abnormality, deterioration state) of the electrical equipment 2, and each feature The pattern corresponding to the quantity is parameterized based on the hidden Markov model, and each parameter is stored as diagnostic information.
In addition, since the parameter memorize | stored previously in the learning part 5 can be updated or exchanged, the versatility of the electrical equipment diagnostic system 1 can be improved.

  Further, the diagnosis unit 6 determines the probability (likelihood) of generating the feature amount pattern detected in the operating state (operating state) of the electrical equipment 2 based on each parameter stored in advance in the learning unit 5. And determining the parameter that maximizes the generation probability (likelihood), and diagnosing the electrical equipment 2 based on the diagnostic information corresponding to the parameter. Specify the degree.

  Note that the characteristic quantities used for diagnosis of the electrical equipment 2 include current, voltage, vibration, temperature, sound, and the like. The frequency spectrum of these feature quantities and the time series of the frequency spectrum are also included as feature quantities. In vector control widely used in electric motor systems, torque current and excitation current are created based on the current value and rotation angle by a drive control computer. Thus, the amount directly detected by the sensor 3 and the amount created by the computer based on the amount directly detected by the sensor 3 are also included as the feature amount. It is also possible to combine several feature amounts obtained from different sensors and use them as feature amounts. Further, it may be a multidimensional distribution created from these various feature quantities or a time series of multidimensional distributions. Further, a feature amount obtained by reducing a multi-feature amount (multi-dimension) to one dimension or the like by using a principal component analysis or a self-organizing map may be used.

  In the learning unit 5, when the feature quantity pattern is parameterized, calculation is performed by applying a calculation algorithm including a forward algorithm and a backward algorithm. In the diagnosis unit 6, when calculating the probability (likelihood) of generating the feature amount pattern using the parameters stored in the learning unit 5, a calculation algorithm including a forward algorithm is applied. To calculate.

  The display unit 7 displays a diagnosis result of the electrical equipment 2 diagnosed by the diagnosis unit 6 and is a display unit of a computer constituting the learning unit 5 and the diagnosis unit 6. The state of the electrical equipment 2 diagnosed by the diagnosis unit 6 is displayed on the computer display as described above, and can be transmitted to, for example, a central monitoring device of a factory or building by communication means.

  The A / D conversion circuit 4, the learning unit 5, and the diagnosis unit 6 described above are collectively used as a hidden Markov model unit (HMM unit) 10 in the following description.

  FIG. 3 shows an inverter 11 that supplies a drive current to an electric motor 2M that is an example of electrical equipment, a converter 12 that supplies a DC voltage to the inverter 11, and various sensors 3a and 3b for detecting various feature quantities. FIG.

  As the sensor 3a, a current sensor for detecting currents Iu and Iv flowing through the electric motor 2M, or a voltage sensor for detecting a voltage output from the inverter 11 to the electric motor 2M is used. Further, as the sensor 3b, there are a rotation sensor for detecting a rotation angle position and a rotation speed of the electric motor 2M, or a torque sensor, a temperature sensor, a vibration sensor, a microphone and the like (not shown), which are appropriately selected and used.

  The current sensor 3a shown in FIG. 3 detects the currents Iu and Iv flowing through the electric motor 2M, the vibration sensor (not shown) detects the vibration, and the microphone detects the noise as the feature quantity. By applying the hidden Markov model, the failure of the electric motor 2M, the inverter 11 and the converter 12 can be diagnosed. Further, by attaching a temperature sensor to the electric motor 2M, the inverter 11 and the converter 12, it is possible to diagnose each device abnormality from the temperature rise. In this case, a hidden Markov model is applied to the time-series pattern of temperature fluctuations, and abnormality of each device is diagnosed.

  Further, in a rotating machine such as an electric motor or a generator, the failure frequency of the bearing is high. In such a case, the failure of the bearing of the rotating machine can be diagnosed by using the vibration sensor and observing the vibration generated from the rotating machine. FIG. 4 shows an example of a vibration waveform when an abnormality occurs in the bearing of the rotating machine. FIG. 5 shows the relationship between the bearing abnormality and the vibration waveform. In addition, fo, fi, and fb in FIG. 5 are frequencies determined from structures and dimensions of inner rings, outer rings, and rolling elements constituting the bearing, a rotation frequency at that time, and the like. As shown in FIG. 5, when an abnormality is found in the bearing of the rotating machine, the vibration waveform becomes a characteristic pattern according to the abnormality form. Therefore, when the vibration is detected as a feature quantity and the feature quantity is input to the aforementioned hidden Markov model section (HMM section) 10, the hidden Markov model section (HMM section) 10 generates a feature quantity pattern. The probability (likelihood) to be calculated is calculated based on each parameter stored in the learning unit 5 in advance, and the parameter that maximizes the generation probability (likelihood) is determined. Based on the diagnosis information, the state of the bearing is diagnosed. If there is an abnormality, the abnormality is specified, and the diagnosis result is displayed on the display unit 7.

In addition, the frequency of the vibration obtained from the vibration sensor is analyzed once, and the bearing of the rotating machine can be diagnosed by applying a hidden Markov model to the frequency spectrum pattern and the frequency spectrum time series pattern. it can.
Furthermore, by attaching a microphone around the rotating machine, using the operating sound at that time as a feature value, and applying a hidden Markov model to the feature value pattern, the bearing of the rotating machine can be diagnosed. . Further, the bearing of the rotating machine can be diagnosed by using the frequency spectrum of the operating sound or the time change of the frequency spectrum as a feature amount and applying the hidden Markov model to the pattern.

FIG. 6 shows an example in which vibration sensors 3c and 3d and a rotational speed sensor 3e are attached to an electric motor 2M that is not driven by an inverter. If the object to be diagnosed is a bearing of the motor 2M, the vibration sensors 3c and 3d are preferably mounted at a plurality of locations such as on the bearing of the motor 2M and on the upper side of the motor 2M.
In addition, when an electric motor or a generator is a high voltage rotating machine, partial discharge may occur and vibration may occur. In order to detect vibration in this case, an ultrasonic sensor is attached to the main body of the rotating machine, the winding part, etc., and the vibration detected by the ultrasonic sensor is used as a feature value, and a hidden Markov model is applied to this feature value pattern. By applying, a high-voltage rotating machine can be diagnosed.

  In addition, in a rotating machine such as an electric motor or a generator, a pattern of feature values obtained at the time of abnormality is often correlated with a rotational position (or rotational speed). Therefore, it is preferable to attach a position (or speed) sensor for detecting the rotational position (or rotational speed) of the rotor of the rotating machine.

In addition to sound and vibration during operation of rotating machines such as electric motors and generators, the rotating machine is stopped and the part to be diagnosed is hit with a hammer etc., and the sound and vibration obtained at that time are featured Or a frequency spectrum of sound or vibration, a temporal change of the frequency spectrum, or the like can be used as a feature amount pattern for diagnosis.
Furthermore, since an abnormality in the bearing affects the current flowing through the rotating machine, this current is detected as a feature value, and the rotating machine is diagnosed by applying a hidden Markov model to this feature value pattern. Can do.
Also, the bearing of the rotating machine can be diagnosed by using the frequency spectrum with respect to the current or the time change of the frequency spectrum as a feature amount and applying a hidden Markov model to the pattern.
Furthermore, by attaching several temperature sensors to the rotating machine or the bearing part, the temperature can be used as a feature quantity, and the bearing of the rotating machine can be diagnosed by applying a hidden Markov model to the temperature fluctuation pattern. .

Next, a case where a short-circuit failure of a stator winding in an electric motor or a generator occurs will be examined.
In a rotating machine such as an electric motor or a generator, a winding short circuit may occur due to a decrease in the dielectric strength of the stator winding (coil). When this winding short circuit occurs, the current waveform flowing through the rotating machine is deformed. FIG. 7 and FIG. 8 show a current waveform in a normal state and a current waveform when a winding short-circuit occurs in an electric motor not driven by an inverter. As can be seen from these figures, when the winding short-circuit occurs, not only the peak value of the current changes, but also the waveform itself is distorted. Also, the larger the gap between the shorted windings, the more the current waveform will be different from the normal waveform. By detecting this current as a feature amount and applying a hidden Markov model to the feature amount pattern, a winding short circuit can be diagnosed.
Furthermore, by using the frequency spectrum with respect to the current or the time change of the frequency spectrum as a feature amount and applying a hidden Markov model to the pattern, it is possible to diagnose a winding short circuit of the rotating machine.
In addition, by attaching a microphone around the rotating machine, using the operating sound at that time as a feature value, and applying a hidden Markov model to the feature value pattern, diagnose the winding short circuit of the rotating machine Can do. Furthermore, the short circuit of the rotating machine can be diagnosed by using the frequency spectrum of the operation sound or the time change of the frequency spectrum as a feature amount and applying a hidden Markov model to the pattern.
Moreover, by attaching a vibration sensor to the rotating machine, using the vibration at this time as a feature quantity, and applying a hidden Markov model to the pattern of the feature quantity, it is possible to diagnose a winding short circuit of the rotating machine. . Furthermore, the short circuit of the rotating machine can be diagnosed by using the frequency spectrum with respect to the vibration or the time change of the frequency spectrum as a feature amount and applying a hidden Markov model to the pattern.

  FIG. 9 shows how the sensors 3f, 3g, and 3h for detecting the current in the motor are attached. In the example of FIG. 9, the sensors 3f, 3g, and 3h are attached to each of the three phases. However, a current sensor may be attached to the two phases, and the current value of the remaining one phase may be obtained by calculation. Similarly, in the case of a generator, a winding short circuit can be diagnosed by using a current as a feature value and applying a hidden Markov model to the pattern of the feature value.

  In addition, by attaching several temperature sensors to the winding of the rotating machine or the main body of the rotating machine, the temperature is used as a feature quantity, and the rotating machine is diagnosed by applying a hidden Markov model to the temperature fluctuation pattern. be able to.

  Next, when the rotor bar or the end ring of the motor or the rotating machine of the generator is broken, the waveform pattern of the current flowing through the motor is changed from the normal current waveform shown in FIG. 10 to the abnormal current shown in FIG. Change to waveform. By detecting this current as a feature value and applying a hidden Markov model to this feature value pattern, it is possible to diagnose when the rotor bar or the end ring is broken. The characteristic frequency spectrum change at that time is shown in FIGS. Here, f represents a rotation frequency and s represents a slip. Diagnosing rotor bars and end rings can be performed by applying a hidden Markov model using the frequency spectrum and the time variation of the frequency spectrum as a feature amount pattern.

  In addition, by attaching several temperature sensors to the main body of the rotating machine and the windings of the stator, the detected temperature is used as a feature value, and by applying a hidden Markov model to the temperature fluctuation pattern, rotation Breakage of machine rotor bars and end rings can be diagnosed.

Furthermore, by attaching a vibration sensor to the main body of the rotating machine, the detected vibration is used as a feature value, and a hidden Markov model is applied to the pattern of the feature value. Can be diagnosed.
Moreover, the breakage of the rotor bar and end ring of the rotating machine can be diagnosed by using the frequency spectrum with respect to the detected vibration and the temporal change of the frequency spectrum as a feature amount and applying a hidden Markov model to the pattern. . Furthermore, by attaching a microphone around the rotating machine, using the operating sound at that time as a feature, and applying a hidden Markov model to the pattern of that feature, the rotor bar and end ring of the rotating machine Breakage can be diagnosed. In addition, the breakage of the rotor bar and end ring of the rotating machine is diagnosed by using the frequency spectrum of the operating sound at this time and the temporal change of the frequency spectrum as a feature amount and applying a hidden Markov model to the pattern. Can do.

Next, when, for example, the U-phase positive current energizing power transistor attached to the inside of the inverter 11 as shown in FIG. 3 does not turn on, the current waveform which is a sine wave in the normal state is shown in FIG. As shown in FIG. 4, the U-phase current waveform is deformed so that the positive current portion (upper half) disappears. That is, when this failure occurs, it can be seen that the waveform pattern of the current flowing through the motor 2M changes. Therefore, by using this current as a feature amount and applying a hidden Markov model to the feature amount pattern, it is possible to diagnose whether the transistor is normal or abnormal and to determine where it is abnormal if it is abnormal. it can.
Further, it is possible to diagnose whether the transistor is normal or abnormal by using a frequency spectrum with respect to the current or a time change of the frequency spectrum as a feature amount and applying a hidden Markov model to the pattern of the feature amount. Furthermore, it is possible to diagnose whether the transistor is normal or abnormal by attaching a vibration sensor to the motor, using the vibration as a feature value, and applying a hidden Markov model to the pattern of the feature value. Further, by using a frequency spectrum with respect to vibration or a time change of the frequency spectrum as a feature amount and applying a hidden Markov model to the pattern, it is possible to diagnose whether the transistor is normal or abnormal.
Further, a microphone is attached around the motor 2M, the operation sound at that time is used as a feature value, and a hidden Markov model is applied to the feature value pattern to diagnose whether the transistor is normal or abnormal. be able to. Further, by using the frequency spectrum of the operation sound or the time change of the frequency spectrum as a feature amount and applying a hidden Markov model to the pattern, it is possible to diagnose whether the transistor is normal or abnormal.
In addition, when an abnormality occurs in the transistor, pulsation is observed in the torque current and excitation current during vector control. Therefore, it is possible to diagnose whether the transistor is normal or abnormal by using the torque current or the excitation current as a feature amount and applying a hidden Markov model to the feature amount pattern.

Next, the deterioration diagnosis of the smoothing capacitor in the converter 12 as shown in FIG. 3 will be described.
Inside the converter 12, a rectifying diode for converting a three-phase AC voltage into a DC voltage and a smoothing capacitor for smoothing the DC voltage are installed, and when the smoothing capacitor is deteriorated. Increases the pulsation with a smoothed DC voltage. Therefore, by using the smoothed DC voltage as a feature amount and applying a hidden Markov model to the feature amount pattern, it is possible to diagnose deterioration or breakage of the smoothing capacitor in the converter 12. Further, the current flowing in the electric motor 2M at that time is used as a feature amount, and the hidden Markov model is applied to the feature amount pattern, so that the smoothing capacitor in the converter 12 can be diagnosed for deterioration or breakage. . Further, by using the frequency spectrum with respect to the current and voltage, the time change of the frequency spectrum, and the like as the feature amount and applying the hidden Markov model to the pattern, it is possible to diagnose deterioration or breakage of the smoothing capacitor.
When the smoothing capacitor is deteriorated, pulsation is observed in the torque current and excitation current during vector control. Therefore, deterioration and breakage of the smoothing capacitor in the converter 12 can be diagnosed by using the torque current or excitation current as a feature value and applying a hidden Markov model to the feature value pattern.
Further, by using a frequency spectrum with respect to the current or a time change of the frequency spectrum as a feature amount and applying a hidden Markov model to the pattern, deterioration or breakage of the smoothing capacitor can be diagnosed.
Furthermore, by attaching a temperature sensor to the smoothing capacitor, the detected temperature is used as a feature value, and a hidden Markov model is applied to the pattern of the feature value, so that the smoothing capacitor is not deteriorated or damaged. Can be diagnosed.
Further, by attaching a vibration sensor to the electric motor, using the vibration as a feature value, and applying a hidden Markov model to the feature value pattern, it is possible to diagnose deterioration or breakage of the smoothing capacitor. Furthermore, by using a frequency spectrum with respect to vibration or a time change of the frequency spectrum as a feature amount and applying a hidden Markov model to the pattern, deterioration or breakage of the smoothing capacitor can be diagnosed.
In addition, a microphone is attached around the motor 2M, the operating sound at that time is used as a feature value, and a hidden Markov model is applied to the pattern of the feature value, thereby diagnosing deterioration and breakage of the smoothing capacitor. can do. Furthermore, by using the frequency spectrum of the operating sound or the time change of the frequency spectrum as a feature amount and applying the hidden Markov model to the pattern, it is possible to diagnose the deterioration or breakage of the smoothing capacitor.

Next, damage diagnosis of the rectifying diode in the converter 12 will be described.
When the rectifying diode is damaged and short-circuited or opened, the smoothed DC voltage pulsation increases. Therefore, by using this DC voltage as a feature amount and applying a hidden Markov model to the feature amount pattern, it is possible to diagnose the breakage of the rectifying diode in the converter 12.
Further, the current flowing in the rectifying diode at that time is used as a feature amount, and the hidden Markov model is applied to the feature amount pattern, so that the breakage of the rectifying diode in the converter 12 can be diagnosed. Further, the frequency spectrum with respect to these currents and voltages, the time change of the frequency spectrum, and the like are used as feature amounts, and the hidden Markov model is applied to the pattern, whereby the breakage of the rectifying diode can be diagnosed.
Further, the current flowing in the motor 2M at that time is used as a feature value, and the hidden Markov model is applied to the pattern of the feature value, whereby the breakage of the rectifying diode in the converter 12 can be diagnosed. Further, when the rectifying diode is broken and short-circuited or opened, pulsation is observed in the torque current and excitation current during vector control. Therefore, by using the torque current or the excitation current as a feature amount and applying the hidden Markov model to the feature amount pattern, the breakage of the rectifying diode can be diagnosed.
Further, by using the frequency spectrum with respect to these currents and the time change of the frequency spectrum as a feature amount and applying the hidden Markov model to the pattern, the breakage of the rectifying diode can be diagnosed.
Further, the vibration sensor is attached to the electric motor, the vibration is used as a feature amount, and the hidden Markov model is applied to the feature amount pattern, so that the breakage of the rectifying diode can be diagnosed. Furthermore, by using a frequency spectrum with respect to vibration or a time change of the frequency spectrum as a feature amount and applying a hidden Markov model to the pattern, damage of the rectifying diode can be diagnosed.
In addition, a microphone is attached to the periphery of the motor 2M, the operating sound at that time is used as a feature amount, and a hidden Markov model is applied to the feature amount pattern, thereby diagnosing the breakage of the rectifying diode. Can do. Furthermore, by using the frequency spectrum of the operating sound or the time change of the frequency spectrum as a feature amount and applying the hidden Markov model to the pattern, the breakage of the rectifying diode can be diagnosed.
In some cases, either or both of the rectifying diode and the smoothing capacitor are incorporated in the inverter 12 described above. In this case as well, deterioration and breakage can be diagnosed in the same manner as described above.

  In addition, by providing a frequency analysis unit such as fast Fourier transform, window Fourier transform, and wavelet in parallel with the hidden Markov model 10 shown in FIG. 2, the waveform of a specific frequency component of the frequency spectrum or time-frequency spectrum can be obtained. If the failure state and deterioration state of the electrical equipment are determined in combination, the state of each electrical equipment can be determined more accurately in conjunction with the diagnosis of the hidden Markov model 10.

As described above, in the failure of the motor system shown in FIG. 3 described above, a characteristic pattern of each failure mode appears in the current flowing through the motor 2M. Therefore, by using the current sensor 3a, the current flowing through the motor 2M is detected as a feature amount, and a fault in the motor system can be diagnosed by applying a hidden Markov model to the feature amount pattern. Further, the current value obtained from the current sensor may be subjected to frequency analysis once, and the hidden Markov model may be applied to the frequency spectrum or the time series pattern of the frequency spectrum. Although there are a large number of failures in the motor system that can be diagnosed by detecting the current, examples other than the above include the following.
・ Crack of rotor end ring ・ High resistance connection of winding rotor (contact failure)
・ Air gap difference ・ Magnetic imbalance (magnetic center error)
・ Rotor bending ・ Rotor eccentricity

  Next, with respect to semiconductor elements including FETs (field effect transistors) including power transistors and rectifier diodes used in the inverter 11 and converter 12 of the electric motor system shown in FIG. The semiconductor element can be diagnosed by attaching a sensor, using the temperature detected by the temperature sensor as a feature amount, and applying a hidden Markov model to the temperature change pattern. The temperature sensor may be attached to the surface of the semiconductor element. In many cases, the power transistor and the rectifying diode are attached so as to be in close contact with the heat sink. In this case, a temperature sensor may be attached between the power transistor or the rectifying diode and the heat sink, or directly on the heat sink, and the temperature variation detected by the temperature sensor may be used as the feature amount.

Further, in the motor system under inverter control as shown in FIG. 3, when a failure occurs in the motor 2M, the inverter 11, the converter 12, etc., the PWM command value (pulse wide modulation command value) changes. The motor system can be diagnosed by applying the hidden Markov model to the change in the PWM command value and the feature value pattern obtained based on the command value.
In addition, when a failure occurs in the motor 2M, the inverter 11, the converter 12, etc., the torque current and the excitation current also change. Therefore, these current values are used as feature amounts, and they are hidden from the feature amount pattern. The motor system can be diagnosed by applying the Markov model.

Next, the diagnosis of the earth leakage breaker will be described with reference to FIG.
FIG. 15 shows the earth leakage breaker 20 connected to the single-phase 100V electric wire as the electrical equipment to be diagnosed, diagnoses the contact connection state of the contact portions 20a and 20b of the earth leakage breaker 20, and sets the state. The electrical equipment diagnostic system to judge is shown. When a contact failure occurs in the contact portions 20a and 20b of the earth leakage breaker 20, the voltage between the power supply side and both ends of the load side of each of the contact portions 20a and 20b of the earth leakage breaker 20 is changed to a voltage waveform detected by the voltage sensor 21. Spike-like components come to be seen, and when this defect progresses, a cuprous oxide proliferation phenomenon occurs. When this cuprous oxide growth phenomenon occurs, the contact portions 20a and 20b may generate high heat and may cause a fire.

  FIG. 16 is a waveform diagram of the detection voltage when the cuprous oxide proliferation phenomenon occurs. FIG. 17 is a waveform of a current detected by the current sensor 22 in a state where the cuprous oxide proliferation phenomenon has not occurred, and FIG. 18 shows a state in which the cuprous oxide proliferation phenomenon has occurred. It is a current waveform. As is apparent from these drawings, when the cuprous oxide growth phenomenon occurs, the voltage and current become characteristic waveforms.

  From the above, using the voltage and current as feature amounts, the feature amount patterns are parameterized based on the hidden Markov model, and the respective parameters are stored in the learning unit 5 described above. At the time of diagnosis, since the voltage and current are input to the diagnosis unit 6 described above, the diagnosis unit 6 stores the probability (likelihood) of generating the feature amount pattern in the learning unit 5 in advance. It is calculated based on each parameter, and the parameter that maximizes the generation probability (likelihood) is determined, and whether or not a cuprous oxide growth phenomenon has occurred is diagnosed based on diagnostic information corresponding to the parameter. If it has occurred, the diagnosis result is displayed on the display unit 7 described above. In this way, it is possible to determine the occurrence of the cuprous oxide proliferation phenomenon, and it is possible to prevent fires caused by it. Or it is good also considering the line voltage detected by the secondary side of the earth leakage circuit breaker 20 as a feature-value. In addition, it is possible to make a diagnosis by using a frequency spectrum of voltage or current or a time series pattern of the frequency spectrum as a feature amount and applying a hidden Markov model to the pattern of the feature amount. Furthermore, by attaching a temperature sensor to the earth leakage breaker 20, the detected temperature can be used as a feature quantity, and the earth leakage breaker 20 can be diagnosed by applying a hidden Markov model to the feature quantity pattern. it can.

Next, diagnosis for a lightning arrester used in the power system will be described.
The lightning arresters used in the power system include a lightning arrester with a series gap and a gapless zinc oxide lightning arrester. The leakage current of these lightning arresters can be measured, and the insulation state of the lightning arresters can be determined from the change in the leakage current in the normal operation state. In the case of a lightning arrester with a series gap, it is possible to check the insulation characteristics and shunt resistance of the series gap part, and in the case of a gapless zinc oxide lightning arrester, the leakage current of the ZnO element can be checked at any time without stopping the equipment. It has become.

  As shown in FIG. 19, the leakage current measurement of the lightning arrester 30 to be diagnosed includes a method of inserting an ammeter 31 into the lightning arrester grounding circuit 32 and a leakage current detection CT 34 on the lightning arrester grounding wire 33 as shown in FIG. Is inserted at the time of inspection. However, FIG. 19 shows an example in which an ammeter disconnector 35 is connected. The above two methods can be applied to both a series gap arrester and a gapless zinc oxide type arrester.

In the case of the zinc oxide lightning arrester, as is clear from the equivalent circuit shown in FIG. 21, the total leakage current Io of the ZnO element is a combined current of the capacitance current Ic and the resistance current Ir. When the ZnO element deteriorates, both the capacitance current Ic and the resistance current Ir tend to increase. Therefore, the insulation state of the lightning arrester 30 can be diagnosed by using the measured total leakage current as a feature amount and applying a hidden Markov model to the feature amount pattern. In particular, since the increase of the resistance current Ir becomes significant due to the deterioration of the ZnO element, the measurement of the resistance current Ir is effective for diagnosis. Therefore, the resistance current Ir is used as a feature value, and the pattern of the feature value is displayed. On the other hand, the insulation state of the lightning arrester 30 can be diagnosed by applying the hidden Markov model. Further, the capacitance current Ic may be used as the feature amount.
FIG. 22 is a characteristic diagram illustrating the characteristics of the capacitance current Ic, the resistance current Ir, and the total leakage current.

  In the embodiment described above, as an example of the electrical equipment, a power semiconductor element of a rotating machine drive system or a control device that drives and controls the rotating machine, components related thereto, an earth leakage breaker, or a lightning arrester are shown. Other than these, for example, pump facilities, electric furnaces, air-conditioning facilities, etc., general equipment driven by electric energy, generally used semiconductor parts, and the like are subject to diagnosis.

It is a conceptual explanatory view of a hidden Markov model. It is a systematic diagram of an electrical equipment diagnostic system. It is a systematic diagram of an electric motor system. It is a vibration waveform figure of the bearing of an electric motor. It is explanatory drawing of the relationship between the abnormality of the bearing of an electric motor, and a vibration waveform. It is various sensor arrangement | positioning figures of an electric motor. FIG. 6 is a normal current waveform diagram of a stator winding of an electric motor. It is an electric current waveform figure at the time of the short circuit of the stator winding | coil of an electric motor. It is an arrangement view of a current sensor of an electric motor that is not driven by an inverter. It is an electric current waveform figure at the time of normal of the rotor bar etc. of an electric motor. It is a current waveform figure at the time of abnormality, such as a rotor bar of an electric motor. It is a frequency spectrum figure in the normal time of the rotor bar etc. of an electric motor. It is a frequency spectrum figure at the time of abnormality, such as a rotor bar | burr of an electric motor. It is a current waveform figure at the time of abnormality of the power transistor of an inverter. It is a systematic diagram of the electrical equipment diagnostic system in the case of diagnosing an earth-leakage circuit breaker. It is a voltage waveform diagram when the cuprous oxide multiplication phenomenon occurs in the contact part of the earth leakage breaker. It is a current waveform figure in the state where the cuprous oxide multiplication phenomenon has not occurred in the contact part of the earth leakage breaker. It is a current waveform figure in the state where the cuprous oxide multiplication phenomenon has occurred in the contact part of the earth leakage breaker. It is the circuit diagram which showed the leakage current measuring means of the lightning arrester. It is the circuit diagram which showed the leakage current measuring means of the lightning arrester. It is an equivalent circuit for measuring the leakage current of a lightning arrester. It is a leakage current characteristic figure of a lightning arrester.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Electrical equipment diagnostic system 2 Electrical equipment 3 Sensor 4 A / D conversion circuit 5 Learning part 6 Diagnostic part 7 Display part 10 Hidden Markov model part

Claims (3)

  Learning means for detecting the feature quantity in each predetermined state of the electrical equipment to be diagnosed and storing each parameter obtained by parameterizing each feature quantity pattern based on a hidden Markov model; Diagnosis of determining and determining the state of the electrical equipment based on the parameter that maximizes the generation probability while calculating the generation probability of the feature amount pattern detected in the operating state of the equipment And an electrical equipment diagnostic system.   2. The electrical equipment diagnosis system according to claim 1, wherein the electrical frequency is based on a waveform of a specific frequency component in a frequency spectrum of the feature quantity or a waveform of a specific frequency component in a time series of the frequency spectrum of the feature quantity. An electrical equipment diagnosis system comprising frequency analysis means for determining the state of equipment.   The electrical equipment diagnosis system according to claim 1 or 2, wherein the parameters stored in the learning means can be updated or exchanged.

Images