TWI426242B - Diagnosing device and an associated method for a motor device - Google Patents

Description

Power equipment abnormality detecting device and detecting method thereof

The invention relates to a monitoring and diagnosis for the operation of a power device, in particular to a power device abnormality detecting device and a detecting method thereof.

In general, power equipment often experiences performance degradation and energy consumption before failures occur, but these phenomena do not immediately affect the operation of the equipment, so they are often not discovered by users. In addition to increasing the probability of failure and shortening the life of equipment, the reduction of power equipment efficiency and energy consumption have a negative impact on industrial competitiveness and environmental protection.

Most of the diagnostic modes of power equipment (such as motors) use a sensor to sense the state of operation of the power equipment, and then transmit the sensed data to the back-end system for further analysis through wired or wireless transmission. However, this mode requires a large amount of data transmission bandwidth. The main reason is that in order to ensure the validity and stability of the analysis and diagnosis, the operational status data received by all the sensors is transmitted to the back-end system as much as possible.

Therefore, how can a method or means, in addition to maintaining the same accurate analytical diagnostic capabilities, first process the sensed operational status data to reduce the amount of data that needs to be transmitted to the back-end system, thereby effectively reducing the data transmission bandwidth. The effects of size, improved data transmission stability, shortened diagnostic update time, and reduced construction costs have long been the goal of related vendors.

In view of the above problems, the present invention provides a power plant abnormality detecting device and a detecting method thereof. By pre-processing the sensed operational status data, the data transmission bandwidth is effectively reduced, the data transmission stability is improved, the diagnostic update time is shortened, and the construction cost is reduced.

The power device abnormality detecting device disclosed in the present invention comprises a sensing module, a processing module, an optimization processing module and a classification diagnostic module. The sensing module is configured to sense a power device to obtain a plurality of operating signals. The processing module is connected to the sensing module to receive the operation signals, and sequentially obtain a plurality of feature values from the operation signals.

The optimization processing module is connected to the processing module to receive the feature values and classify the feature values to establish a plurality of factor groups. The optimization processing module can use the factor analysis method to classify the feature values into a plurality of factor groups according to the relevance. Each of the factor groups has a variation characteristic value representing the factor group, and the number of the variation feature values is less than the number of the feature values.

The classification diagnostic module is connected to the optimization processing module for receiving the group of factors, and sends a status signal to the group of factors according to a preset rule. The preset rule can be a category list, and the category list includes a normal item and an abnormal item. The normal item is the normal condition when the power equipment is running, and the abnormal item is the abnormal condition indicating the operation. For example, the abnormal item may include but is not limited to the imbalance condition, the misalignment condition, the lubrication condition, the resonance condition, the bearing damage condition, the shaft Bending, looseness, phase imbalance, potential imbalance, harmonic doubling, and short circuit conditions.

Therefore, the above-mentioned power device abnormality detecting device can be installed on a power device, and the operating signal sensed on the power device is simplified by the optimization processing module through the optimization processing module, and directly passed through the classification diagnostic module. To judge the operation of the power equipment, it is not necessary to send the sensed operation signal to the back-end system for direct and immediate processing, so as to shorten the diagnostic update time and reduce the construction cost. Furthermore, even if the back-end system is still needed for processing in the future (for example, the operation state of multiple power devices is integrated through a remote server), the number of eigenvalues summarized by the factor analysis method is lower than that of each. The number of characteristic values obtained by the operation signal can reduce the bandwidth of the data transmission and improve the stability of data transmission.

According to the power device abnormality detecting method disclosed in the present invention, the abnormal parameter detection and diagnosis are performed by detecting information on the operation of the power device. The power equipment abnormality detecting method firstly uses a signal processing method to obtain a plurality of running signals from the power device, and extracts a plurality of characteristic values from the running signal. Then, the feature values are further classified according to the relevance to establish a plurality of factor groups, and each of the factor groups has a variation feature value. Finally, the group of factors with the mutated feature value greater than 1 is obtained, and the operating state of the power device is obtained by using the neural network and the rule of thumb, and whether the operating state of the power device is abnormal according to a preset rule.

Wherein, when the operational state determined by the neural network and the rule of thumb are inconsistent, the model of the neural network is modified according to the group of factors until the operational state results judged by the two are consistent.

The running signal can be the vibration signal, temperature signal and magnetic communication of the power equipment. Number, current signal or voltage signal. The processing module transmits the sensed operation signal through a time domain conversion process or a multiscale entropy (MSE) operation to obtain a feature value, and the feature value may be a frequency doubling peak or a characteristic frequency value of the vibration signal. The time domain conversion process may employ a Discrete Fourier Transform (DFT), a Fast Fourier Transform (FFT), a Discrete Cosine Transformation (DCT), and a Discrete Cotley Transformation (DCT). Discrete Hartley Transform (DHT), Wavelet Transform (WT) or Power Spectrum.

The preset rule for determining whether the operating state of the power device is abnormal may be a category list, and the category list includes a normal item and an abnormal item. The normal item is the normal condition when the power equipment is running, and the abnormal item is the abnormal condition indicating the operation. For example, the abnormal item may include but is not limited to the imbalance condition, the misalignment condition, the lubrication condition, the resonance condition, the bearing damage condition, the shaft Bending, looseness, phase imbalance, potential imbalance, harmonic doubling, and short circuit conditions.

In this case, after obtaining the operating state of the power device through the neural network, it may be determined according to the preset rule that the power device may be abnormal. The neural network can use a Back Propagation Network (BPN), a Hopfield Neural Network (HNN), and a Radial Basis Function Network. , RBFN), a fuzzy neural network (Fuzzy Neural Network, FNN) or a functional Link Neural Network (FLNN). The rule of thumb is a characteristic spectrum, a critical threshold, a trajectory map, an envelope, a harmonic analysis, or a combination thereof.

Therefore, the power device abnormality detecting method of the present invention can improve the operation signal of a power device and simplify the number and size of the feature values obtained by the self-running signal through the factor analysis method, and can be applied to the resource consumption. In the microprocessor, the operation process is directly performed to determine the operating state of the power device, and the sensed operation signal is not transmitted to the back-end system, and the back-end system only needs to receive the judgment result, thereby reducing the data transmission bandwidth and the lifting. Data transmission stability, shortened diagnostic update time and reduced implementation costs.

The features, implementations, and effects of the present invention are described in detail below with reference to the drawings.

Please refer to FIG. 1 and FIG. 1 is a schematic diagram of the power plant abnormality detecting device of the present invention. The power device abnormality detecting device 100 can be an embedded system chip or a data processing device such as a personal digital assistant (PDA). The power device abnormality detecting device 100 is disposed on a power device 200 and includes a sensing module. 110. A processing module 120, an optimization processing module 130, and a classification diagnostic module 140.

The sensing module 110 is configured to sense the power device 100 to obtain a plurality of operating signals. Taking the motor as an example, the sensing module 110 senses the vibration signal data of the power device 100 through the degree of vibration, and also obtains temperature and magnetic. Pass, current, speed, electricity The operation signal when the motor is operating.

The processing module 120 uses the signal processing method to extract a plurality of feature values from the operation signal according to the operation signal sensed by the sensing module 110. In the case of the vibration signal, the processing module 120 converts the operation signal from the time domain to the frequency domain by using the fast Fourier transform processing method, and converts the fundamental frequency of the vibration signal by the rotation speed, and then sequentially extracts the vibration signal from the frequency domain. Multiplier, 1x to 12x, the multiplied signal is the characteristic value of the corresponding vibration signal.

The optimization processing module 130 is connected to the processing module 120 to receive the feature values and classify the feature values to establish a plurality of factor groups. The optimization processing module 130 may use a factor analysis method to classify feature values into a plurality of factor groups according to relevance. Each of the factor groups has a variation characteristic value representing the factor group, and the number of the variation feature values is less than the number of the feature values.

The classification diagnostic module 140 is connected to the optimization processing module 130 for receiving the group of factors, and sends a status signal to the group of factors according to a preset rule to determine the operation of the power device. The preset rule can be a category list, and the category list includes a normal item and an abnormal item. The normal item is the normal condition when the power equipment is running, and the abnormal item is the abnormal condition indicating the operation. For example, the abnormal item may include but is not limited to the imbalance condition, the misalignment condition, the lubrication condition, the resonance condition, the bearing damage condition, the shaft Bending, looseness, phase imbalance, potential imbalance, harmonic doubling, and short circuit conditions.

Please refer to FIG. 2A, and FIG. 2A is a schematic view showing an embodiment of the power plant abnormality detecting device of the present invention. Power equipment anomaly detection device The device includes a warning device 150 for receiving the status signal, and is used to notify the user when the status signal is the abnormality. The warning device 150 can be, but is not limited to, a vibration module, a light module, a display module, an audio module, or a combination thereof, for notifying the use of a vibration warning, a light warning, a message warning, or an audible alarm. The operation of the power device 200 is abnormal.

Please refer to FIG. 2B, and FIG. 2B is a schematic view showing another embodiment of the power plant abnormality detecting device of the present invention. The power device abnormality detecting device further includes a transmission module 160. The transmission module 160 is connected to the classification diagnostic module 140 for receiving the status signal, and transmitting the status signal to the warning device 150 through a wired or wireless transmission manner. .

Referring to FIG. 2C, FIG. 2C is a schematic view showing still another embodiment of the power plant abnormality detecting device of the present invention. The power device abnormality detecting device further includes a memory module 170 for storing the operation signals of the power device 200. When the user needs to read the operation signal of the power device for further analysis, The memory module 170 is accessed to obtain the desired operation signal. The memory module 170 can be used to set a memory card to store the sensing operation signal. The memory card can be a compact flash (CF) memory card, a micro hard disk (MD) memory card, and a secure digital position. (Secure Digital, SD) memory card, a micro SD digital memory card, a multimedia (Multi Media Card, MMC) memory card, a long (Memory Stick, MS) memory card or a micro strip (Micro MS) Memory card.

Therefore, with the power device abnormality detecting device described above, the detecting device can be set It is placed on a power equipment, and the operation signal sensed by the automatic force equipment is simplified by the factorization analysis method through the optimization processing module, and the operation condition of the power equipment is directly judged through the classification diagnosis module, without sensing the sensed The operation signal is sent to the back-end system for direct and immediate processing to reduce diagnostic update time and reduce construction costs. Furthermore, even if the back-end system is still needed for processing in the future (for example, the operation state of multiple power devices is integrated through a remote server), the number of eigenvalues summarized by the factor analysis method is lower than that of each. The number of characteristic values obtained by the operation signal can reduce the bandwidth of the data transmission and improve the stability of data transmission.

Please refer to FIG. 3, and FIG. 3 is a flow chart showing the steps of the power plant abnormality detecting method of the present invention. The power device abnormality detecting method is applied to a power device, comprising: step S300: obtaining a plurality of operation signals from the power device by using a signal processing method; and step S310: obtaining a plurality of operation signals corresponding to the operation signals from the operation signals Feature value; Step S320: grouping the feature values to establish a plurality of factor groups, each of the factor groups having a variation feature value; Step S330: determining, according to the factor groups, using a neural network a first operational state of the power device operation; step S340: determining, according to the characteristic values, a second operational state of the power device operation by using a rule of thumb; Step S350: comparing whether the first operating state and the second operating state are the same; step S360: when the first operating state is different from the second operating state, correcting the neural network according to the factor group, Until the first operational state is the same as the second operational state; step S370: when the first operational state is the same as the second operational state, determining whether the first operational state is abnormal according to a preset rule; step S380: If it is determined that the first operating state is abnormal, an abnormal signal is sent; and step S390: if it is determined that the first operating state is normal, a normal signal is sent.

Please refer to FIG. 4A, and FIG. 4A is a flowchart of an embodiment of step S310 in FIG. The plurality of characteristic values corresponding to the operation signals are obtained from the operation signals, and the operation signals include vibration signals, temperature signals, magnetic communication numbers, current signals or voltage signals. Step S310 includes: step S311: sensing the power device to obtain the operation signals; and step S312: converting a time domain data of the operation signal into a frequency domain data by using a time domain conversion process; and step S313: The frequency domain data captures multiple feature values.

The time domain conversion process may be a discrete Fourier transform process, a fast Fourier transform process, a discrete cosine transform process, a discrete Hartley conversion process, a wavelet transform process, or a power frequency process.

In the vibration signal of the motor, when the vibration signal is subjected to fast Fourier transform processing, the fundamental frequency (harmonic) can be calculated by the following formula: the first fundamental frequency position = ((1 * speed * data length of the operation signal) / (60) * Spectrum acquisition frequency)); Second fundamental frequency position = ((2 * speed * data length of the operation signal) / (60 * spectrum acquisition frequency)); and so on.

For example, suppose that a motor with a frequency of 1800 rpm is sensed to obtain a 16-kilobit operation signal, and the frequency of the frequency is 12 kHz, and the first fundamental frequency can be 40.

According to this, when step S311 obtains an operation signal representing the motor vibration signal, the operation signal is converted from the time domain data to the frequency domain data through the step S312, and then the frequency signal is extracted from the frequency domain, and the frequency is extracted by 0.5 times and 1 time. The frequency multiplied signal is the characteristic value of the corresponding vibration signal. The number of these characteristic values is 24, and is defined as 0.5x, 1x, 1.5x according to the size of the multiplication frequency. , 2x, 2.5x, 3x, 3.5x, 4x, 4.5x, 5x, 5.5x, 6x, 6.5x, 7x, 7.5x, 8x, 8.5x, 9x, 9.5x, 10x, 10.5x, 11x, 11.5x And 12x.

Please refer to FIG. 4B, and FIG. 4B is a flow chart of another embodiment of step S310 in FIG. Compared with the embodiment of step S310 shown in FIG. 4A, the embodiment of step S310 shown in FIG. 4B obtains the feature value of the operation signal through multi-scale entropy (MSE) operation. This embodiment includes: S314: sensing the power device to obtain the operation signals; and step S315: transmitting the operation signals after removing the noise through a multi-scale entropy (Multiscale Entropy, MSE) operation to obtain the characteristic values corresponding to the operation signal.

Step S316 and S315 may further include a step S316, and step S316: performing noise processing on the operation signals by using wavelet transform. The main reason is that the operation signal obtained through sensing may have noise, and the effect of suppressing noise can be achieved by wavelet conversion processing.

Please refer to the "Fig. 3" and "4C" diagrams, and the "4C diagram" is a flow chart of the steps of step S320 in Fig. 3. To obtain the plurality of feature values corresponding to the operation signal (step S310), the power device abnormality detecting method simplifies the feature values by the factor analysis method, and group the feature values according to step S320 to establish multiple Each of the factor groups has a variability feature value, and step S320 includes: step S321: grouping the feature values by a factor analysis method to establish the factor groups; and step S322: calculating each of the factors The feature values in the factor group to obtain a variation feature value; and step S323: retaining the variation feature values having the variation feature value greater than 1.

By the above steps S321 to S323, and matching the characteristic value of the corresponding vibration signal obtained in the above step S310, the result of the following list 1 can be obtained.

Table I

Table 1 is taken as an example to illustrate how step S322 obtains the variation characteristic values of each factor group. First, 24 octave eigenvalues (0.5x, 1x, 1.5x, 2x, 2.5x, 3x, 3.5x, 4x, 4.5x, 5x, 5.5x, 6x, 6.5x, 7x, 7.5x, 8x) , 8.5x, 9x, 9.5x, 10x, 10.5x, 11x, 11.5x, and 12x) calculate the sample covariate matrix S, and the covariate matrix is expressed as follows:

Where var represents the number of variances and cov represents the number of variances.

Then, 24 eigenvalues λ ₁ ,..., λ ₂₄ are calculated from the common variable matrix, respectively.

Solution. Therefore, λ ₁ = 7.431, λ ₂ = 3.257, λ ₃ = 1.258, λ ₄ = 1.206, λ ₅ = 1.124, ..., λ ₂₄ = 0.029 can be solved, and the variation eigenvalue is greater than 1 according to the calculation result. The principle of the number of factor groups, taking Table 1 as an example, a total of five factor groups are selected as input variables of the neural network.

And eigenvalue grouping is performed by factor analysis (step S321), since each octave eigenvalue has its load value under five factor groups, by selecting the maximum load value under a certain factor group, To indicate the factor group to which the feature value belongs. Taking the 1X octave characteristic value as an example, the load of factor 1 is -0.260, the load of factor 2 is 0.899, the load of factor 3 is -0.038, the load of factor 4 is -0.092, and the load of factor 5 is The value of -1015, in which the load value of 1X under factor 2 is the largest among the five factor groups, that is, 1X belongs to the factor group of factor two. Therefore, in the same way, the remaining 23 feature values can be classified into five factor groups according to the maximum load value, as listed in Table 1, so the factors include 1.5X, 4X, 4.5X, 5X. , 5.5X, 6X, 6.5X, 7X, 8.5X, 12X; factor two includes 1X, 2X, 2.5X, 3X, 3.5X, 7.5X, 8X, 9X; factor three includes 10X, 10.5X; factor four includes 11X , 11.5X; factor five contains 0.5X, 9.5X.

As described in step S330, a combination of five factors is used to bring into the neural network to obtain the first operational state. The establishment of a neural network is known to those skilled in the art and will not be described herein. The neural network can use a Back Propagation Network (BPN), a Hopfield Neural Network (HNN), and a Radial Basis Function Network. , RBFN), a fuzzy neural network (FNN) or a functional Link Neural Network (FLNN). The rule of thumb is a characteristic spectrum, a critical threshold, a trajectory map, an envelope, a harmonic analysis, or a combination thereof.

Similarly, in step S340, the feature value is used to determine the second operational state of the power device using the rule of thumb. The rule of thumb is based on the rules derived from the basic theory. The vibration signal is taken as an example. The rule of thumb is the vibration characteristic rule derived from the basis of mechanical vibration. The most common is the matching of the eigenvalues (ie, the characteristic spectrum) of the spectrum. The combination of thresholds can be calculated, and the eigenvalues can be further decoupled and analyzed using trajectory maps and envelopes. Common harmonic analysis can also assist in obtaining sideband data.

Here, the rule of thumb adopted in step S340 is a method of setting the feature value and the threshold, and when the feature value exceeds the threshold setting value, it can be judged that the operation of the power device is abnormal.

For example, suppose the peak value of the 1X multiplication characteristic value and the 2X multiplication characteristic value are set at a threshold of 5 millimeters per millimeter (millimeter/second, mm/s), and the peak value of the 3X characteristic value is 2 per second. The millimeter is the threshold setting value, as shown in Fig. 5, and Fig. 5 is a logic flow diagram of an embodiment in which the second operational state of the power unit is judged by the rule of thumb in step S340 in Fig. 3. If the 1X multiplier characteristic value is less than 5, the power device is judged to be normal. Otherwise, if the 1X multiplier characteristic value is greater than 5, and the 2X and 3X multiplier characteristic values are less than 5 and 2 at the same time, the operating state of the power device is judged. An imbalance has occurred.

Assuming that the first operational state obtained by the neural network is the same as the second operational state obtained by the rule of thumb (step S340), the preset rule may be a classification list including a normal item and an abnormal item. The normal item is the normal condition when the power equipment is running, and the abnormal item is the abnormal condition indicating the operation. For example, the abnormal item may include but is not limited to the imbalance condition, the misalignment condition, the lubrication condition, the resonance condition, the bearing damage condition, the shaft Bending, looseness, phase imbalance, potential imbalance, harmonic doubling, and short circuit conditions.

Step S380 determines whether the first operating state is an operation recorded in the abnormal item according to the preset rule, so as to determine whether the power device has an abnormal state. On the other hand, if the first operational status is the operation recorded in the normal project, it is judged that the operation state of the power device is normal.

Therefore, the power device abnormality detecting method of the present invention can improve the operation signal of a power device and simplify the number and size of the feature values obtained by the self-running signal through the factor analysis method, and can be applied to the resource consumption. In the microprocessor, the operation process is directly performed to determine the operating state of the power device, and the sensed operation signal is not transmitted to the back-end system, and the back-end system only needs to receive the judgment result, thereby reducing the data transmission bandwidth and the lifting. Data transmission stability, shortened diagnostic update time and reduced implementation costs.

Although the embodiments of the present invention are disclosed above, it is not intended to limit the present invention, and those skilled in the art, regardless of the spirit and scope of the present invention, the shapes, structures, features, and spirits as described in the application scope. The scope of the invention is to be determined by the scope of the appended claims.

100‧‧‧Power equipment anomaly detection device

110‧‧‧Sensing module

120‧‧‧Processing module

130‧‧‧Optimized processing module

140‧‧‧Classification Diagnostic Module

150‧‧‧Warning device

160‧‧‧Transmission module

170‧‧‧Memory Module

200‧‧‧Power equipment

Fig. 1 is a schematic view showing a power plant abnormality detecting device of the present invention.

Fig. 2A is a schematic view showing an embodiment of the power plant abnormality detecting device of the present invention.

Fig. 2B is a schematic view showing another embodiment of the power plant abnormality detecting device of the present invention.

Fig. 2C is a schematic view showing still another embodiment of the power plant abnormality detecting device of the present invention.

Fig. 3 is a flow chart showing the steps of the power plant abnormality detecting method of the present invention.

Figure 4A is a flow chart of an embodiment of step S310 in Figure 3.

Figure 4B is a flow chart of another embodiment of step S310 in Figure 3.

Fig. 4C is a flow chart showing the steps of step S320 in Fig. 3.

Figure 5 is a flow chart of an embodiment of step S340 in Figure 3.

Claims (19)

A power device abnormality detecting device includes: a sensing module that senses a power device to obtain a plurality of operating signals; and a processing module that is connected to the sensing module to receive the operating signals And sequentially extracting a plurality of feature values from each of the operation signals; an optimization processing module is connected to the processing module to receive the feature values, and classify the feature values to establish a plurality of factor groups, Each of the factor groups has a variation characteristic value representing the factor group, the number of the variation feature values is less than the number of the feature values; and a classification diagnosis module is connected to the optimization process The module is configured to receive the group of factors under the mutated feature values, and send a status signal to the group of factors according to a preset rule. The power device abnormality detecting device according to claim 1 further includes a warning device for receiving the status signal, and when the status signal is abnormal, notifying the power device that an abnormality occurs. The power device abnormality detecting device according to claim 2, further comprising a transmission module connected to the classification diagnostic module for receiving the status signal and transmitting the status signal by wire or wireless transmission A status signal is sent to the alert device. The power device abnormality detecting device according to claim 1, further comprising a memory module, wherein the memory module stores the operation signals of the power device. A power device abnormality detecting method includes the following steps: Acquiring a plurality of operation signals from the power device by using a signal processing method; obtaining, by the operation signals, a plurality of feature values corresponding to the operation signals; grouping the feature values to establish a plurality of factor groups, Each of the factor groups has a variability eigenvalue, and the number of the eigenvalues is less than the number of the eigenvalues; according to the group of factors, one of the neural networks is used to judge that the power device operates one of the first An operating state; determining, according to the characteristic values, a second operating state of the power device operation by using a rule of thumb; comparing whether the first operating state and the second operating state are the same; when the first operating state and the second operating state When the operating states are different, the neural network is modified according to the group of factors until the first operating state is the same as the second operating state; when the first operating state is the same as the second operating state, according to one The preset rule determines whether the first operating state is abnormal; if it is determined that the first operating state is abnormal, sending an abnormal signal; and if determining the first operational state As normal, a normal signal is sent. The power device abnormality detecting method according to claim 5, wherein the operation signal is a vibration signal. The power device abnormality detecting method according to claim 5, wherein the operation signal is a temperature signal. The power device abnormality detecting method according to claim 5, wherein the operation signal is a magnetic communication number. The power device abnormality detecting method according to claim 5, wherein the operation signal is a current signal. The power device abnormality detecting method according to claim 5, wherein the operation signal is a voltage signal. The power device abnormality detecting method according to claim 5, wherein the operation signal is a rotation speed signal. The method for detecting an abnormality of a power device according to the fifth aspect of the invention, wherein the step of obtaining the operation signal from the power device by using the signal processing method further comprises: sensing the power device to obtain the a running signal; converting a time domain data of the operational signal into a frequency domain data by using a time domain conversion process; and extracting the characteristic values from the frequency domain data. The power device abnormality detecting method according to claim 12, wherein the time domain conversion processing is a discrete Fourier transform processing, a fast Fourier transform processing, a discrete cosine transform processing, a discrete Hartley conversion processing, A wavelet conversion process or a power frequency process. The power device abnormality detecting method according to claim 5, wherein the step of obtaining the operation signals from the power device by using the signal processing method further includes: The power device is sensed to obtain the operation signals; and the operation signals after the noise removal are transmitted through a multiscale entropy (MSE) operation to obtain the characteristic values corresponding to the operation signals. The power device abnormality detecting method according to claim 14, wherein the step of sensing the power device to obtain the operation signals and the operation signal after removing the noise are transmitted through the multi-scale The entropy operation is performed between the steps of obtaining the characteristic values corresponding to the operation signals, and further includes: performing noise processing on the operation signals by using wavelet transform. The power device abnormality detecting method according to claim 5, wherein the step of grouping the feature values to establish the factor groups comprises: grouping the feature values by using a factor analysis method, To establish the group of factors; sequentially calculating the feature values in each of the factor groups to obtain the variation feature value; and retaining the variation feature values having the variation feature value greater than 1. The power device abnormality detecting method described in claim 5, wherein the neural network is a Back Propagation Network (BPN) and a Hopfield Neural Network (Hopfield Neural Network). , HNN), a Radial Basis Function Network (RBFN), a Fuzzy Neural Network (FNN) or a functional Link Neural Network (FLNN) ). The power device abnormality detecting method according to claim 5, wherein the preset rule is a category list, the category list includes a normal item and an abnormal item, and the abnormal item includes an imbalance condition and an unbalanced condition. Lubrication, resonance, bearing damage, shaft bending, looseness, phase imbalance, potential imbalance, harmonic doubling and short circuit. The power device abnormality detecting method according to claim 5, wherein the rule of thumb is a characteristic spectrum, a threshold setting value, a trajectory map, an envelope or a harmonic analysis.