KR101818394B1 - Method and apparatus for diagnosing facility condition - Google Patents

Abstract

The present invention relates to a method and an apparatus for diagnosing a facility condition. More specifically, the method comprises the steps of: receiving data obtained from a facility as a signal; generating a feature map image by extracting a feature from the received signal; and classifying a facility condition by using a pre-learned feature map image based on the generated feature map image and a two-dimensional convolution neutral network (CNN).

Description

TECHNICAL FIELD The present invention relates to a method and apparatus for diagnosing facility conditions,

The present invention relates to a method and apparatus for diagnosing a facility condition, and more particularly, to a method and apparatus for diagnosing a facility condition based on a feature map of a signal.

As industrial equipment and machining systems are speeded up and automated, production speed of products increases and productivity is improved, and thus stability is required. As equipment facilities become more dependent on the equipment, damage to equipment, such as large-scale accidents caused by breakdowns, and production losses resulting from shutdowns, increase. The failure modes of industrial equipment and mechanical systems vary, but they are particularly common in bearings, which are essential for many vibrating rotors.

Causes of bearing failure include insufficient lubrication, improper use of lubricant, misalignment of bearings, and excessive deformation of shaft. In the past, experienced technicians diagnosed these problems and judged whether they were faulty. However, most of them have a long diagnosis time, subjective, and sometimes have to stop the operation of the equipment system. In recent years, a system capable of diagnosing a failure of a bearing has been required while maintaining the operation of a device system, and thus the bearing operation state is continuously monitored and developed as a form of technology capable of detecting an abnormality before a failure.

Generally, in order to monitor the operating condition of a bearing, a method of analyzing a signal acquired from a facility, and detecting a mismatch phenomenon based on the signal, is used.

In particular, when the original signal is converted into a two-dimensional image to diagnose the state of the equipment, the vibration signal has a low sampling rate and thus shows a constant texture pattern, so that it is easy to classify the state. However, There is a problem that a situation of misdiagnosing a facility state may occur.

US Patent 6,922,482

SUMMARY OF THE INVENTION Accordingly, the present invention has been made in view of the above problems, and it is an object of the present invention to provide a facility state diagnosis method and apparatus capable of more accurately diagnosing a facility state based on a feature map of a signal and a two-dimensional CNN (Convolution Neural Network) .

It is another object of the present invention to provide a facility state diagnosis method and apparatus capable of correctly diagnosing a facility state even when a signal having a high sampling rate is used by detecting a pattern according to facility state using a discrete wavelet-based feature map.

It is another object of the present invention to provide a facility state diagnosis method and apparatus capable of correctly diagnosing a facility state by using signals acquired from an equipment in operation and thereby detecting an abnormality before a failure.

According to an aspect of the present invention, there is provided a method of diagnosing a condition of a facility, comprising: inputting data acquired from a facility as a signal; extracting features from the input signal to generate a feature map image And classifying the facility state using the generated feature map image and the feature map image that has been learned based on the two-dimensional CNN (Convolution Neural Network).

The step of generating the feature map image may include extracting a subband signal from the input signal, calculating a feature value from the extracted subband signal, converting the matrix of the calculated feature value into a feature map image And transforming the transformed feature map image into an input size of the CNN and transforming the transformed vector into a vector, and then normalizing the transformed vector by assigning an intermediate value of the transformed vector to the feature map image.

The subband signal can be extracted based on Discrete Wavelet Packet Transform (DWPT).

The feature value may be any one of a Root Mean Square (RMS), a Shape Factor, a kurtosis value, and a Square Mean Root (SMR).

The signal may be at least one of sound, vibration, current, magnetic flux, thermal image, and ultrasonic waves.

As a means for solving the above problems, a facility state apparatus according to another embodiment of the present invention includes an input unit for inputting data acquired from a facility as a signal, a generation unit for generating a feature map image by extracting a feature from the input signal, And a classifying unit classifying the facility state using the generated feature map image and the feature map image that has been learned based on the two-dimensional CNN (Convolution Neural Network).

Wherein the generation unit includes an extraction unit that extracts a subband signal from the input signal, a calculation unit that calculates a characteristic value from the extracted subband signal, a conversion unit that converts the matrix of the calculated characteristic value into a characteristic map image, And a normalizing unit that divides and arranges the size of the converted feature map image into the input size of the CNN and converts the converted vector into a vector and then substitutes the intermediate value of the transformed vector into the feature map image and normalizes the transformed vector.

The subband signal can be extracted based on Discrete Wavelet Packet Transform (DWPT).

The feature value may be any one of a Root Mean Square (RMS), a Shape Factor, a kurtosis value, and a Square Mean Root (SMR).

The signal may be for at least one of acoustic, vibration, current, magnetic flux, thermal imaging, and ultrasound.

According to the present invention, it is possible to more accurately diagnose a facility state based on a characteristic map for signals and a two-dimensional CNN (Convolution Neural Network).

Further, according to the present invention, a facility state can be accurately diagnosed even when a signal having a high sampling rate is used by detecting a pattern according to a facility state using a discrete wavelet-based feature map.

Further, according to the present invention, an abnormality can be detected in advance before a failure by correctly diagnosing the equipment state using a signal acquired from an equipment in operation.

1 is a block diagram showing a facility condition diagnosis apparatus according to an embodiment of the present invention;
2 is a flowchart showing a method of diagnosing a facility condition according to an embodiment of the present invention,
3 is a block diagram showing a generator for generating a feature map image according to an embodiment of the present invention;
4 is a flow chart illustrating a method for generating a feature map image according to an embodiment of the present invention;
5A is a diagram illustrating an example of a subband signal extracted from a signal according to an embodiment of the present invention,
FIG. 5B is a view showing an example of a feature map image generated according to an embodiment of the present invention;
FIG. 5C illustrates an example of a normalized feature map image according to an embodiment of the present invention; FIG.
6A is a diagram illustrating an example of a feature map image for a steady state generated according to an embodiment of the present invention;
FIG. 6B is a view showing an example of a feature map image for an outer ring defect state generated according to an embodiment of the present invention;
FIG. 6C illustrates an example of a feature map image for a roller defect state generated according to an embodiment of the present invention; FIG.
6D is a diagram showing an example of a feature map image for the inner ring defect state generated according to the embodiment of the present invention,
FIG. 6E is a view showing an example of a feature map image for the outer ring and roller defect states generated according to the embodiment of the present invention; FIG.
FIG. 6F is a view showing an example of a feature map image for the outer ring and inner ring defect states generated according to the embodiment of the present invention;
FIG. 6G is a view showing an example of a feature map image for the inner ring and roller defect states generated according to the embodiment of the present invention;
6H is a diagram showing an example of a feature map image for an outer ring, a roller, and an inner ring defect state generated according to an embodiment of the present invention.

Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings. Will be described in detail with reference to the portions necessary for understanding the operation and operation according to the present specification. In describing the embodiments of the present invention, description of technical contents which are well known in the technical field to which the present invention belongs and which are not directly related to the present specification will be omitted. This is for the sake of clarity without omitting the unnecessary explanation and without giving the gist of the present invention.

In describing the components of the present specification, the same reference numerals may be given to components having the same name, and the same reference numerals may be given to different drawings. However, even in such a case, it does not mean that the corresponding component has different functions according to the embodiment, or does not mean that it has the same function in different embodiments, and the function of each component is different from that of the corresponding embodiment Based on the description of each component in FIG.

The present invention generates a feature map image of a signal acquired from a facility and diagnoses the facility condition through the generated feature map image and the feature map image learned based on the two-dimensional CNN (Convolution Neural Network) Even when a high signal is used, the equipment state is accurately diagnosed. Hereinafter, the present invention will be described in detail with reference to examples and drawings.

1 is a block diagram showing a facility condition diagnosis apparatus according to an embodiment of the present invention.

1, the equipment condition diagnosis apparatus 100 according to an embodiment of the present invention includes an input unit 110, a generating unit 130, and a classifying unit 150. As shown in FIG.

The input unit 110 receives the data acquired from the facility as a signal. Here, the signal may be acoustic, vibration, current, magnetic flux, thermal image, and ultrasonic wave, and a plurality of signals may be input.

The generation unit 130 extracts a feature from a signal input by the input unit 110 and generates a feature map image.

The classifying unit 150 classifies the facility state using the feature map image generated by the generating unit 130 and the feature map image based on the two-dimensional CNN (Convolution Neural Network).

2 is a flowchart showing a method of diagnosing a facility condition according to an embodiment of the present invention.

The input unit 110 receives the data acquired from the facility as a signal (S201).

The generation unit 130 extracts a feature from the signal input by the input unit 110 to generate a feature map image (S203).

The classifying unit 150 classifies the facility state using the feature map image generated by the generating unit 130 and the feature map image based on the two-dimensional CNN (Convolution Neural Network) (S205).

3 is a block diagram showing a generator for generating a feature map image according to an embodiment of the present invention, and more specifically, a generator 130 of the equipment condition diagnosis apparatus 100. As shown in FIG.

3, the generation unit 130 for generating a feature map image according to an embodiment of the present invention includes an extraction unit 131, a calculation unit 133, a conversion unit 135, and a normalization unit 137 ).

The extracting unit 131 extracts a subband signal from the input signal. At this time, the extracting unit 131 extracts a subband signal of each level by applying Discrete Wavelet Packet Transform (DWPT) to the input signal.

The calculation unit 133 calculates a feature value from the extracted subband signal. At this time, the feature value can be any one of 1) RMS (Root Mean Square), 2) Shape factor, 3) kurtosis value, and 4) SMR (Square Mean Root).

1) When RMS is calculated as a feature value, it is based on the following Equation (1). RMS is the magnitude of a change in the signal, which means the magnitude and / or magnitude of a sinusoidal waveform, such as a sinusoidal waveform.

&Quot; (1) "

2) When the shape coefficient is calculated as the feature value, it is based on the following formula (2). Indicates the ratio of the DC component to the AC component, and refers to the ratio of the magnitude of the continuous waveform that alternates between the sound and the amount of the signal.

&Quot; (2) "

3) When the kurtosis value is calculated as the feature value, it is based on Equation (3) below. The distribution of the signal values increases as the distribution of the signal values converges near the specific value and becomes sharp.

&Quot; (3) "

4) When SMR is calculated as a feature value, it is based on Equation (4) below. It is the same as RMS, which means the amplitude or the magnitude of the difference between the tone and the amount of the continuous waveform. It is more sensitive to the signal size than RMS.

&Quot; (4) "

는 신호의 i번째 시간 인덱스에 해당하는 진폭 값을 나타내고,

는 신호의 평균 값을 정의된다.On the other hand, the conversion unit 135 converts the matrix of the calculated characteristic values into a characteristic map image. The variables of Equations (1) to (4)

Represents the amplitude value corresponding to the i < th > time index of the signal,

Is defined as the average value of the signal.

The normalization unit 137 normalizes the size of the converted feature map image to the input size of the CNN.

For example, the normalization unit 137 divides the feature map image of the N x M size into CNN input sizes (N'x M '), and aligns the respective divided parts to convert them into vectors. Here, each of the divided portions can be calculated based on Equation (5) below.

Equation (5)

Then, the intermediate value of the transformed vector is extracted, and the intermediate value of each extracted part is substituted into the feature map image to normalize it.

4 is a flowchart illustrating a method of generating a feature map image according to an embodiment of the present invention.

The extraction unit 131 extracts sub-band signals at each level by applying a Discrete Wavelet Packet Transform (DWPT) to the input signal (S401).

The calculating unit 133 calculates a feature value from the extracted subband signal (S403).

The conversion unit 135 converts the matrix of the calculated feature values into a feature map image (S405).

The normalization unit 137 normalizes the size of the feature map image to the input size of the CNN (S407). Specifically, the size of the transformed feature map image is divided into CNN input size, each divided part is sorted and converted into a vector, and the median value of the transformed vector is substituted into the feature map image and normalized.

FIG. 5A is a diagram illustrating an example of a subband signal extracted from a signal according to an embodiment of the present invention, and is an example of a subband signal extracted by the extraction unit 131 shown in FIG. 3 described above.

FIG. 5B is a view showing an example of a feature map image generated according to an embodiment of the present invention, and is an example of a feature map image converted by the conversion unit 135 shown in FIG. 3 described above.

At this time, the transformed feature map image is transformed based on a matrix of feature values calculated from each subband signal extracted by the extracting unit 131.

5C is an example of a normalized feature map image according to an embodiment of the present invention, and is an example of a feature map image normalized by the normalizing unit 137 shown in FIG. 3 described above.

At this time, the normalized feature map image is obtained by dividing the transformed feature map image by the transform unit 135 into the input size of CNN, aligning the divided parts, converting the transformed vector into a vector, Map &lt; / RTI &gt;

6A to 6H illustrate an example of a feature map image for each facility state generated according to an embodiment of the present invention. Referring to FIG. 6A to FIG. 6H, an acoustic signal acquired from a bearing is input, It is.

6C is a roller defect state, FIG. 6D is an inner ring defect state, FIG. 6E is an outer ring and roller defect state, FIG. 6F is an outer ring and inner ring defect state, FIG. Inner ring and roller defect states, and Figure 6h is an image of each feature map generated for the case of outer ring, roller and inner ring fault conditions.

6A to 6D, it is possible to diagnose a single defect as well as to diagnose a composite defect as shown in FIGS. 6E to 6H.

These defect states are classified based on the feature map image generated as described above and the feature map image learned through the artificial neural network algorithm such as 2-dimensional CNN.

As shown in FIGS. 6A to 6H, each feature map image has a distinct pattern for each state, so that the state of the bearing can be more accurately classified.

In the above description, the case where the acoustic signal of the bearing is input is described as an example, but the present invention is not limited thereto.

It will be apparent to those skilled in the art that various modifications and variations can be made in the present invention without departing from the spirit or essential characteristics thereof. Therefore, the embodiments disclosed in the present invention are intended to illustrate rather than limit the scope of the present invention, and the scope of the technical idea of the present invention is not limited by these embodiments. The scope of protection of the present invention should be construed according to the following claims, and all technical ideas within the scope of equivalents should be construed as falling within the scope of the present invention.

100: Equipment condition diagnosis device
110: input unit
130:
150:
131:
133:
135:
137: normalization unit

Claims (10)

Inputting data acquired from a facility as a signal;
Extracting a feature from the input signal to generate a feature map image; And
Classifying the facility state using the generated feature map image and a feature map image that has been learned based on a two-dimensional CNN (Convolution Neural Network)
Wherein the step of generating the feature map image comprises:
Extracting a subband signal from the input signal;
Calculating a feature value from the extracted subband signal;
Converting the matrix of the calculated feature values into a feature map image; And
Dividing and rearranging the converted feature map image into an input size of the CNN and transforming the vector into a vector and then normalizing the transformed vector by substituting an intermediate value of the transformed vector into the feature map image. Diagnostic method.
delete The method according to claim 1,
The sub-
And extracting based on DWPT (Discrete Wavelet Packet Transform).
The method according to claim 1,
The feature value may be,
A root mean square (RMS), a shape factor, a kurtosis value, and a square mean root (SMR).
The method according to claim 1,
The signal,
Wherein the method is for at least one of sound, vibration, current, magnetic flux, thermal image, and ultrasonic wave.
An input unit for receiving data acquired from the facility as a signal;
A generator for extracting a feature from the input signal to generate a feature map image; And
And a classifying unit classifying the facility state using the generated feature map image and the feature map image that has been learned based on a two-dimensional CNN (Convolution Neural Network)
Wherein the generation unit comprises:
An extracting unit for extracting a subband signal from the input signal;
A calculating unit for calculating a feature value from the extracted subband signal;
A transform unit for transforming the matrix of the calculated feature values into a feature map image; And
And a normalization unit for dividing and arranging the transformed feature map image into the input size of the CNN and transforming the vector into a vector and then substituting the intermediate value of the transformed vector into a feature map image and normalizing the transformed feature map image. Device.
delete The method according to claim 6,
The sub-
And extracting based on DWPT (Discrete Wavelet Packet Transform).
The method according to claim 6,
The feature value may be,
Wherein the apparatus is one of a root mean square (RMS), a shape factor, a kurtosis value, and a square mean root (SMR).
The method according to claim 6,
The signal,
Wherein the apparatus is for at least one of sound, vibration, current, magnetic flux, thermal image, and ultrasonic wave.

Images