KR102601072B1 - Method and apparatus for generating an image for motor fault diagnosis, and method and apparatus for motor fault diagnosis using said image - Google Patents

Abstract

The motor failure diagnosis method according to the present invention includes the step of learning a neural network model using learning data obtained from the phase current signal of the motor - where the neural network model is used to identify the failure mode of the motor and to each failure mode. a severity estimation module that estimates the severity of a failure, and feature data extracted from the failure diagnosis module is input to the severity estimation module; and determining the failure mode and failure severity of the motor using the learned neural network model.

Description

Method and apparatus for generating an image for motor fault diagnosis, method and apparatus for motor fault diagnosis using the image {Method and apparatus for generating an image for motor fault diagnosis, and method and apparatus for motor fault diagnosis using said image}

The present invention relates to a method and device for generating an image for diagnosing a motor failure, and a method and device for diagnosing a motor failure using the image.

Motors are widely used in a variety of applications requiring rotational power due to their low cost and high reliability. Motors are subject to mechanical and electrical failure as they are exposed to unexpected stresses such as damage and environmental conditions during use. Since a decrease in motor performance leads to a decrease in product quality, it is very important to diagnose the condition of the motor.

Existing motor failure diagnosis technologies using phase current have been mainly developed in a constant-velocity, constant-load environment. However, motors are mostly operated in environments where speed and load change in the field. Therefore, technologies for diagnosing motor failures in variable environments are being developed. Typically, time-frequency representation techniques can represent time-series signals as two-dimensional images using time-frequency representation. Since each element of the image represents the size of the frequency component over time, the signal of the shifting condition can also be expressed as a two-dimensional image. It is possible to express. However, time-frequency representation images have a limitation in that phase information is missing. In the case of phase current signals, failure characteristics are expressed in amplitude and phase, so missing phase information can be a critical limitation.

In addition, technology for performing fault diagnosis using deep learning through various signal processing technologies such as time-frequency analysis of phase current signals is being researched. However, in order to apply signal processing technology, motor-related domain knowledge (e.g., number of poles, bearing information, etc.) and selection of various parameters are required, requiring considerable expert knowledge, and existing deep learning-based methods lack physical meaning. There are limitations, such as missing phase information that causes fault components to occur.

The technical problem to be achieved by the present invention is to provide an image generation method for diagnosing motor failures that reduces the influence of variable operating environments, considers both amplitude and phase information of signals, and does not require motor-related domain knowledge and complex parameter selection, and The aim is to provide a device and a motor failure diagnosis method and device using the image.

The image generation method for diagnosing motor failure according to the present invention to solve the above technical problem includes extracting instantaneous phase and instantaneous amplitude from the phase current signal of the motor through Hilbert transform; calculating a phase residual by removing an ideal phase component from the instantaneous phase; calculating an amplitude residual by removing driving-related components from the instantaneous amplitude; and generating an image for motor failure diagnosis using the phase residual and amplitude residual.

In the step of generating an image for diagnosing a motor failure, the phase residual and amplitude residual may be mapped onto a two-dimensional plane.

The method of generating an image for diagnosing a motor failure may further include scaling the phase residual and the amplitude residual, respectively.

In calculating the phase residual, the ideal phase component may be a linear component from -π to π.

In calculating the amplitude residual, the driving-related component may be extracted from the instantaneous amplitude using linear regression, moving average, or low-pass filter.

The motor failure diagnosis method according to the present invention for solving the above technical problem may include diagnosing a motor failure using an image generated through the image generation method for motor failure diagnosis.

The step of diagnosing a failure of the motor may use a convolutional neural network model that inputs an image for diagnosing the motor failure.

An image generating device for diagnosing a motor failure according to the present invention to solve the above technical problem includes an instantaneous phase and instantaneous amplitude extractor for extracting the instantaneous phase and instantaneous amplitude from the phase current signal of the motor through Hilbert transformation; a phase residual calculation unit that calculates phase residual by removing ideal phase components from the instantaneous phase; an amplitude residual calculation unit that calculates an amplitude residual by removing drive-related components from the instantaneous amplitude; and an image generator that generates an image for motor failure diagnosis using the phase residual and amplitude residual.

The image generator may map the phase residual and amplitude residual to a two-dimensional plane.

The image generating device for diagnosing motor failure may further include a scaling unit that scales the phase residual and the amplitude residual, respectively.

The ideal phase component may be a linear component from -π to π.

The driving-related component may be extracted from the instantaneous amplitude using linear regression, moving average, or low-pass filter.

The motor failure diagnosis device according to the present invention for solving the above technical problem may include a motor failure diagnosis unit that diagnoses a motor failure using an image generated by the image generating device for motor failure diagnosis.

The motor failure diagnosis unit may use a convolutional neural network model that inputs an image for diagnosing the motor failure.

According to the present invention described above, an image generation method and device for diagnosing motor failure, which reduces the influence of variable driving environments, considers both amplitude and phase information of signals, and does not require motor-related domain knowledge and complex parameter selection; Additionally, a motor failure diagnosis method and device using the image can be provided.

In addition, since the image for motor failure diagnosis reflects the failure physics of phase current, qualitative failure diagnosis is possible with the image itself, and can be used as input data for deep learning to provide high failure diagnosis accuracy.

Figure 1 shows a block diagram of an image generating device for diagnosing a motor failure and a motor failure diagnosing device including the same according to an embodiment of the present invention.
Figure 2 shows a flowchart of an image generation method for motor failure diagnosis and a motor failure diagnosis method including the same according to an embodiment of the present invention.
Figure 3 is a diagram schematically showing an image generation method for diagnosing motor failure according to this embodiment.
Figure 4 shows an example of an image for motor failure diagnosis generated according to an embodiment of the present invention.
Figure 5 shows an example of a variable driving speed profile used in a fault diagnosis experiment.
Figure 6 shows an example of an image for diagnosing a motor failure generated according to an embodiment of the present invention under a variable driving environment.
Figure 7 shows an example of a convolutional neural network model used to diagnose motor failure in an embodiment of the present invention.
Figure 8 shows the results of an experiment comparing the failure diagnosis accuracy of the motor failure diagnosis method according to an embodiment of the present invention and the existing technology.

Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the drawings. In the following description and the accompanying drawings, substantially the same components are denoted by the same symbols to avoid redundant description. Additionally, when describing the present invention, if it is determined that a detailed description of a related known function or configuration may unnecessarily obscure the gist of the present invention, the detailed description thereof will be omitted.

Figure 1 shows a block diagram of an image generating device for diagnosing a motor failure and a motor failure diagnosing device including the same according to an embodiment of the present invention. The device according to this embodiment includes an instantaneous phase and instantaneous amplitude extraction unit 110, a phase residual calculation unit 120, an amplitude residual calculation unit 130, a scaling unit 140, and a motor failure diagnosis unit 150. do.

The instantaneous phase and instantaneous amplitude extraction unit 110 extracts the instantaneous phase and instantaneous amplitude through Hilbert transformation from the phase current signal of the motor acquired in a variable operating environment.

The phase residual calculation unit 120 calculates the phase residual by removing the ideal phase component from the instantaneous phase. The amplitude residual calculation unit 130 calculates the amplitude residual by removing driving-related components from the instantaneous amplitude. Since the fault-related components in the phase current signal are expressed in the form of modulation in each of the amplitude and phase, the phase residual and amplitude residual are obtained by removing the influence of the variable operating environment and extracting the fault-related components from the instantaneous amplitude and phase, respectively. Since the phase current signal is basically dominated by driving-related components, it is important to highlight the fault-related components by removing the influence of the variable driving environment.

And in order to remove the influence of the load condition, the scaling unit 140 scales the phase residual and the amplitude residual respectively to a certain range.

The image generator 150 maps the scaled phase residual and amplitude residual onto a two-dimensional plane to generate an image for motor failure diagnosis.

The motor failure diagnosis unit 160 diagnoses a motor failure using the image for motor failure diagnosis generated by the image generation unit 150.

Figure 2 shows a flowchart of an image generation method for motor failure diagnosis and a motor failure diagnosis method including the same according to an embodiment of the present invention. Figure 3 is a diagram schematically showing an image generation method for diagnosing motor failure according to this embodiment.

In step 210, the instantaneous phase and instantaneous amplitude extractor 110 receives a phase current signal. The motor's phase current signal can be measured by installing an alternating current sensor in the motor power supply or at the output terminal of the current controller inside the control device that drives the motor.

In step 220, the instantaneous phase and instantaneous amplitude extractor 110 performs Hilbert Transform on the phase current signal, and extracts the instantaneous phase and instantaneous amplitude from the Hilbert transform in steps 230 and 240, respectively. Phase current Hilbert transform of signal is expressed as the following equation, so that the instantaneous phase and instantaneous amplitude can be extracted.

where k is discrete time, k=1, 2, 3, ..., n, is the instantaneous phase, is the instantaneous amplitude.

In step 233, the phase residual calculation unit 120 calculates an ideal phase component from the instantaneous phase. The ideal phase component can be calculated as a linear component from -π to π, independent of speed or load.

In step 235, the phase residual calculation unit 120 calculates the phase residual by removing the ideal phase component from the instantaneous phase. Phase Residual can be calculated by the following equation.

here is the ideal phase component, and M is the data length, a value determined according to the speed of the unit instantaneous phase from -π to π. The physical meaning of phase residual is the degree of phase irregularity independent of speed and load.

In step 243, the amplitude residual calculation unit 130 extracts driving-related components from the instantaneous amplitude. Drive-related components can be defined by key trends based on shift profile and load torque level. Size of phase current instantaneous amplitude ( ) is the output torque of the motor ( ), and the output of the motor governs the driving of the rotary system as shown in the following equation.

here is the inertia of the rotating system, is the rotation speed, is the coefficient of friction, is the load torque, represents the change in load torque induced by the fault.

Therefore, the driving-related components can be defined as follows:

In an embodiment of the present invention, driving-related components may be extracted from the instantaneous amplitude using linear regression, moving average, or low-pass filter.

In step 245, the amplitude residual calculation unit 130 calculates the amplitude residual by removing the driving-related component from the instantaneous amplitude. Envelope Residual can be calculated by the following equation.

The physical meaning of amplitude residual is the degree of amplitude irregularity, which is less dependent on variable speed and load torque conditions.

In steps 237 and 247, the scaling unit 140 scales the phase residual and the amplitude residual to be within a certain range (for example, from ??1 to 1). Various methods such as tanh and sigmoid functions can be used for scaling, and in the case of the tanh function, the following equation can be used.

In step 250, the image generator 150 maps the scaled phase residual and amplitude residual onto a two-dimensional plane to generate an image for motor failure diagnosis. The scaled phase residual and amplitude residual form a one-to-one corresponding residual conjugate in the time domain. In order to map these residual conjugates to a two-dimensional plane, the size of the two-dimensional matrix N (for example, 50) can be determined, the rows and columns are divided into N equal parts from ??1 to 1, and the range can be assigned to N2 columns. Each of the N2 cells corresponds to a residual conjugate included in the corresponding range, and the more residual conjugates included in the range, the higher the density of the specific cell. By accumulating the residual conjugates and distributing them on a two-dimensional plane, an image for motor failure diagnosis is created. Hereinafter, for convenience, this will be referred to as the instantaneous residual current image.

Figure 4 is an example of an instantaneous residual current image generated according to an embodiment of the present invention, showing images corresponding to normal (a), winding deterioration (b), and unbalanced (c) states. The image of winding deterioration (b) shows the results of amplitude residuals, and the image of imbalance (c) shows the results of amplitude residuals and phase residuals.

Referring to FIG. 4, it can be seen that in the normal state, the residual signal is small and distributed near the origin (center), while in the fault state, the residual signal spreads along each axis. At this time, the shape of the spread on the horizontal and vertical axes varies depending on the size of the amplitude residual and the phase residual, and this distribution can be used as a fault diagnosis characteristic factor. Using a convolutional neural network, a more precise fault diagnosis algorithm can be created by using the instantaneous residual current image as input data. In addition, it is possible to analyze convolutional neural network models learned by applying technologies for analyzing convolutional neural networks in the field of computer vision (e.g. GradCAM, Filter visualization, etc.).

In step 260, the motor failure diagnosis unit 160 diagnoses a motor failure using the instantaneous residual current image generated in step 250. In the instantaneous residual current image, the higher the density of the origin (center), the more likely it is to be judged as a failure if it is normal or spreads widely along the horizontal or vertical axis. This is because the size of the residual becomes larger when there is a failure. For more precise failure diagnosis, the motor failure diagnosis unit 160 can diagnose motor failure using a convolutional neural network model that uses the instantaneous residual current image as an input. By learning the characteristics of the instantaneous residual current image for each fault type through a convolutional neural network that uses the instantaneous residual current image as input data, a highly accurate fault diagnosis convolutional neural network model can be built.

In order to confirm the effectiveness of the present invention, a fault diagnosis experiment based on a convolutional neural network according to an embodiment of the present invention was performed under various driving conditions.

Figure 5 is an example of a variable driving speed profile used in a fault diagnosis experiment, showing a trapezoidal profile and a triangular profile. For each speed profile, the load torque was driven at 0, 30, 50, 70, and 100% of the motor's rated load.

Figure 6 is an example of an instantaneous residual current image generated according to an embodiment of the present invention under a variable operation environment, showing failure type (normal, winding deterioration, imbalance), load torque (30%, 70%), and variable operation speed profile ( Trapezoid, triangle) represents a star image. Referring to FIG. 6, it can be seen that while the difference depending on operating conditions is small regardless of normal or malfunction, it widens along the horizontal axis in case of winding deterioration failure compared to normal, and in the case of unbalanced failure, it widens not only on the horizontal axis but also on the vertical axis. Therefore, it can be seen that the instantaneous residual current image according to the present invention shows failure characteristics robustly to operating conditions.

Figure 7 shows an example of a convolutional neural network model used to diagnose motor failure in an embodiment of the present invention. The convolutional neural network model consists of three residual blocks and GAP ((Global Average Pooling), and each residual block consists of batch normalization, ReLU activation, and convolution. In Figure 7, tsne ( t-distributed stochastic neighbor embedding) results are also shown. Referring to the tsne results in Figure 7, it can be seen that the fault type is classified to some extent from the input data to the instantaneous residual current image itself, making deep learning model learning easier. , it can be seen that the failure type is better clustered as it passes through the residual block.

Figure 8 shows the results of an experiment comparing the failure diagnosis accuracy of the motor failure diagnosis method according to an embodiment of the present invention and the existing technology. Among the existing technologies, the time-frequency expression typically uses a spectrogram using a short-time Fourier transform, and the other three technologies (Markov transition field, Gramian angular field, and recurrence plot) are imaging technologies for time series signals, and the input data is in the form of 50x50. unified with the same precision. As for the evaluation data, the first uses 20% of the total data at random, the second and third use data with different speed profiles, and the fourth to sixth use data with different load torques. The same convolutional neural network model was applied to all, and the average and standard deviation of fault diagnosis accuracy were calculated through 10 repeated experiments. Referring to FIG. 8, compared to the existing technology, the present invention showed high failure diagnosis accuracy even when data from different operating conditions were used as learning data and evaluation data, and the present invention showed that data from different operating conditions that were not learned were used for evaluation. It can be seen that in some cases, the accuracy is overwhelmingly high.

According to the present invention, expert knowledge such as motor-related domain knowledge or complex parameter selection is not required, and since it reflects the fault physics of phase current, qualitative analysis of the fault is possible compared to the existing time series signal imaging technique, and time-frequency expression technique is possible. Fault characteristic components, including contrast phase information, can be highlighted more effectively and high fault diagnosis accuracy can be provided under various operating conditions.

Devices according to embodiments of the present invention include a processor, memory for storing and executing program data, permanent storage such as a disk drive, a communication port for communicating with an external device, a touch panel, keys, and buttons. It may include user interface devices such as the like. Methods implemented as software modules or algorithms may be stored on a computer-readable recording medium as computer-readable codes or program instructions executable on the processor. Here, computer-readable recording media include magnetic storage media (e.g., ROM (read-only memory), RAM (random-access memory), floppy disk, hard disk, etc.) and optical read media (e.g., CD-ROM). ), DVD (Digital Versatile Disc), etc. The computer-readable recording medium is distributed among computer systems connected to a network, so that computer-readable code can be stored and executed in a distributed manner. The media may be readable by a computer, stored in memory, and executed by a processor.

Embodiments of the invention may be represented by functional block configurations and various processing steps. These functional blocks may be implemented in various numbers of hardware or/and software configurations that execute specific functions. For example, embodiments may include integrated circuit configurations such as memory, processing, logic, look-up tables, etc., capable of executing various functions under the control of one or more microprocessors or other control devices. can be hired. Similar to the fact that the components of the invention can be implemented as software programming or software elements, embodiments may include various algorithms implemented as combinations of data structures, processes, routines or other programming constructs, such as C, C++, , may be implemented in a programming or scripting language such as Java, assembler, etc. Functional aspects may be implemented as algorithms running on one or more processors. Additionally, embodiments may employ conventional technologies for electronic environment settings, signal processing, and/or data processing. Terms such as “mechanism,” “element,” “means,” and “configuration” may be used broadly and are not limited to mechanical and physical configurations. The term may include the meaning of a series of software routines in connection with a processor, etc.

Specific implementations described in the embodiments are examples and do not limit the scope of the embodiments in any way. For the sake of brevity of the specification, descriptions of conventional electronic components, control systems, software, and other functional aspects of the systems may be omitted. In addition, the connections or connection members of lines between components shown in the drawings exemplify functional connections and/or physical or circuit connections, and in actual devices, various functional connections or physical connections may be replaced or added. Can be represented as connections, or circuit connections. Additionally, if there is no specific mention such as “essential,” “important,” etc., it may not be a necessary component for the application of the present invention.

So far, the present invention has been examined focusing on its preferred embodiments. A person skilled in the art to which the present invention pertains will understand that the present invention may be implemented in a modified form without departing from the essential characteristics of the present invention. Therefore, the disclosed embodiments should be considered from an illustrative rather than a restrictive perspective. The scope of the present invention is indicated in the claims rather than the foregoing description, and all differences within the equivalent scope should be construed as being included in the present invention.

Claims (14)

Extracting the instantaneous phase and instantaneous amplitude from the phase current signal of the motor through Hilbert transformation;
calculating a phase residual by removing ideal phase components from the instantaneous phase;
calculating an amplitude residual by removing driving-related components from the instantaneous amplitude; and
An image generating method for motor failure diagnosis, comprising the step of generating an image for motor failure diagnosis using the phase residual and amplitude residual. According to paragraph 1,
The step of generating an image for motor failure diagnosis is characterized in that the phase residual and the amplitude residual are mapped to a two-dimensional plane. According to paragraph 1,
An image generation method for diagnosing a motor failure, further comprising scaling the phase residual and the amplitude residual, respectively. According to paragraph 1,
In the step of calculating the phase residual, the ideal phase component is a linear component from -π to π. An image generation method for diagnosing a motor failure. According to paragraph 1,
In the step of calculating the amplitude residual, the drive-related component is extracted from the instantaneous amplitude using linear regression, moving average, or low-pass filter. A motor failure diagnosis method comprising the step of diagnosing a motor failure using an image generated through the image generation method for motor failure diagnosis according to any one of claims 1 to 5. According to clause 6,
The step of diagnosing a failure of the motor is a motor failure diagnosis method, characterized in that using a convolutional neural network model that inputs an image for diagnosing the motor failure. An instantaneous phase and instantaneous amplitude extraction unit that extracts the instantaneous phase and instantaneous amplitude from the phase current signal of the motor through Hilbert transformation;
a phase residual calculation unit that calculates phase residual by removing ideal phase components from the instantaneous phase;
an amplitude residual calculation unit that calculates an amplitude residual by removing drive-related components from the instantaneous amplitude; and
An image generating device for diagnosing a motor malfunction, comprising an image generator that generates an image for diagnosing a motor malfunction using the phase residual and the amplitude residual. According to clause 8,
The image generator is characterized in that the phase residual and the amplitude residual are mapped to a two-dimensional plane. According to clause 8,
An image generating device for diagnosing a motor failure, further comprising a scaling unit that scales the phase residual and the amplitude residual, respectively. According to clause 8,
An image generating device for diagnosing motor failure, characterized in that the ideal phase component is a linear component from -π to π. According to clause 8,
The drive-related component is extracted from the instantaneous amplitude using linear regression, moving average, or low-pass filter. A motor failure diagnosis device comprising a motor failure diagnosis unit that diagnoses a motor failure using an image generated by the image generating device for motor failure diagnosis according to any one of claims 8 to 12. . According to clause 13,
The motor failure diagnosis unit is a motor failure diagnosis device, characterized in that it uses a convolutional neural network model that inputs an image for the motor failure diagnosis.

Images