KR20200043196A - Deep learning based analysis method and device for remaining useful lifetime of equipment or parts using vibration signals - Google Patents

Abstract

Disclosed is an apparatus of predicting a remaining useful lifetime (RUL), which has more efficient and precise performance than that of the conventional ones. According to the present invention, the apparatus of predicting an RUL comprises: a data conversion unit converting each of a plurality of one-dimensional vibration signals into a two-dimensional image; a prediction model generation unit using the plurality of two-dimensional images generated by the data conversion unit as input data to allow an artificial neural network (ANN) model, which outputs a health indicator (HI), to learn; and an RUL prediction unit using the ANN model, which has learned, to predict the RUL of a target facility for prediction or a target part for prediction. The data conversion unit performs continuous wavelet transform (CWT) on each of the one-dimensional vibration signals to generate the two-dimensional images.

Description

DEEP LEARNING BASED ANALYSIS METHOD AND DEVICE FOR REMAINING USEFUL LIFETIME OF EQUIPMENT OR PARTS USING VIBRATION SIGNALS}

The present invention relates to a technique for predicting a residual useful life (RUL) by analyzing vibration signals collected from a facility or a part, and in particular, it is possible to continuously collect signals by attaching a vibration sensor. The present invention can be applied to a facility or a part, and in particular, it is possible to predict and determine when to replace a part, such as a bearing, which measures a condition or life based on a vibration signal.

Residual life prediction technology is being actively researched in the field of health predictive management (PHM), and has recently emerged as a more important factor in the emergence of smart factories. Residual life prediction can be useful mainly in the aviation industry, nuclear power generation, automotive industry, high-tech industries such as semiconductor / display, etc. It is mainly applied to facilities or parts that are expected to be fatally damaged in the event of a sudden abnormality or failure, and continuously monitors to collect and analyze data to predict the failure and life of the machine to predict predictive maintenance (PdM). . Proper use of the residual life prediction technology can reduce unnecessary equipment maintenance, reduce maintenance costs, and predict failures to improve safety. In particular, vibration monitoring is widely used because it is simple and accurate to estimate the overall condition of the machine. Therefore, the present invention proposes a method and system for predicting residual life based on a vibration signal, and incorporates a deep learning technique that greatly contributes to improving the accuracy of recent data analysis to improve the shortcomings in existing studies. We want to improve the performance of the residual life prediction.

Republic of Korea Patent Publication No. 2018-0040452 (published April 20, 2018) United States Registered Patent No. 9,797,328 (October 24, 2017 registered) U.S. Registered Patent No. 8,903,750 (registered on March 4, 2014)

The technical problem to be achieved by the present invention is to provide a residual life prediction method using a vibration signal, which has more efficient and accurate performance than existing methods.

The residual life prediction apparatus according to an embodiment of the present invention uses a data conversion unit to convert each of a plurality of 1D vibration signals into a 2D image, and a plurality of 2D images generated by the data conversion unit as input data. Predictive model generator to train ANN (Artificial Neural Network) model that outputs a health indicator (HI), and residual life of the predicted equipment or predicted component using the trained ANN model (Remaining And a residual life prediction unit for predicting Useful Lifetime (RUL), wherein the data transformation unit generates the two-dimensional images by performing continuous wavelet transformation (CWT) on each of the one-dimensional vibration signals.

In addition, the residual life prediction method according to an embodiment of the present invention is performed by the residual life prediction apparatus, and the data conversion unit included in the residual life prediction apparatus converts each of a plurality of one-dimensional vibration signals into a two-dimensional image Step, learning a CNN model that outputs a health indicator (HI) by using a plurality of two-dimensional images generated by the data conversion unit as a prediction model generator included in the residual life prediction apparatus as input data, the residual And predicting a residual life (RUL) of a prediction target facility or a prediction target part using a CNN model in which the residual life prediction unit included in the life prediction apparatus is trained, and converting the 2D image into the 2D image comprises: Each of the vibration signals is characterized by generating a continuous wavelet transform (CWT) to generate the two-dimensional images.

According to the method for predicting residual life according to an embodiment of the present invention, by applying a deep learning technique, a feature required for prediction of residual life can be automatically searched and extracted from one signal.

In addition, in the present invention, a deep learning technique can be utilized more efficiently by converting and using a vibration signal into a two-dimensional image containing information in a time-frequency domain.

In order to better understand the drawings cited in the detailed description of the present invention, detailed descriptions of each drawing are provided.
1 is a functional block diagram of an apparatus for predicting residual life according to an embodiment of the present invention.
2 shows an example of a vibration signal detected from a facility or component.
3 shows a contour plot of a wavelet power spectrum generated by a vibration signal and a continuous wavelet transform (CWT).
4 is a diagram for explaining a CNN model generated by the prediction model generator 300 shown in FIG. 1.
FIG. 5 is a diagram for explaining a residual life prediction operation of the predictive model generated by the predictive model generator shown in FIG. 1, which is an estimated health indicator (estimated HI) and a predicted health indicator (predicted HI) of a specific component. It shows.
FIG. 6 is a flowchart illustrating a residual life prediction method performed by the residual life prediction apparatus illustrated in FIG. 1.

Specific structural or functional descriptions of the embodiments according to the concept of the present invention disclosed in this specification are exemplified only for the purpose of explaining the embodiments according to the concept of the present invention, and the embodiments according to the concept of the present invention It can be implemented in various forms and is not limited to the embodiments described herein.

Embodiments according to the concept of the present invention can be applied to various changes and can have various forms, so the embodiments will be illustrated in the drawings and described in detail herein. However, this is not intended to limit the embodiments according to the concept of the present invention to specific disclosure forms, and includes all changes, equivalents, or substitutes included in the spirit and scope of the present invention.

Terms such as first or second may be used to describe various components, but the components should not be limited by the terms. The above terms are only for the purpose of distinguishing one component from other components, for example, without departing from the scope of rights according to the concept of the present invention, the first component can be referred to as the second component and similarly the second The component may also be referred to as the first component.

When a component is said to be "connected" or "connected" to another component, it is understood that other components may be directly connected to or connected to the other component, but other components may exist in the middle. It should be. On the other hand, when a component is said to be "directly connected" or "directly connected" to another component, it should be understood that no other component exists in the middle. Other expressions describing the relationship between the components, such as "between" and "immediately between" or "adjacent to" and "directly neighboring to," should be interpreted similarly.

The terms used in this specification are only used to describe specific embodiments, and are not intended to limit the present invention. Singular expressions include plural expressions unless the context clearly indicates otherwise. In this specification, terms such as “include” or “have” are intended to indicate that there are features, numbers, steps, operations, components, parts, or combinations thereof described herein, one or more other features. It should be understood that the presence or addition possibilities of fields or numbers, steps, actions, components, parts or combinations thereof are not excluded in advance.

Unless otherwise defined, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by a person skilled in the art to which the present invention pertains. Terms such as those defined in a commonly used dictionary should be interpreted as having meanings consistent with meanings in the context of related technologies, and should not be interpreted as ideal or excessively formal meanings unless explicitly defined herein. Does not.

Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings. However, the scope of the patent application is not limited or limited by these embodiments. The same reference numerals in each drawing denote the same members.

1 is a functional block diagram of an apparatus for predicting residual life according to an embodiment of the present invention.

Referring to FIG. 1, the residual life prediction apparatus 10 includes a data conversion unit 100, a prediction model generation unit 300, and a residual life prediction unit 500. According to an embodiment, the residual life prediction apparatus 10 may further include a storage unit 700. The residual life prediction apparatus 10 may generate a prediction model using a plurality of vibration signals, and predict the residual useful life (RUL) of a facility or component where a prediction target vibration signal is generated using the generated prediction model. have.

The data converter 100 may convert each of a plurality of vibration signals output from a vibration sensor installed in a specific facility or a specific component constituting the specific facility into a two-dimensional image (or two-dimensional image data). Specifically, the data converter 100 may extract a characteristic of the time-frequency domain region from a one-dimensional vibration signal, and convert the extracted characteristic into a two-dimensional image composed of a time axis and a frequency axis. Here, the characteristic may mean wavelet coefficients, and one vibration signal may be converted into one image.

Also, the data converter 100 may convert the predicted vibration signal into a two-dimensional image signal in order to predict the residual life (RUL).

The predictive model generator 300 may generate a predictive model capable of predicting a residual life (RUL) of equipment or parts. Specifically, the prediction model generator 300 may generate an ANN (Artificial Neural Network) model, and train the ANN model using the two-dimensional images generated by the data converter 100. The ANN may be a convolutional neural network (CNN), but the scope of the present invention is not necessarily limited to the type of artificial neural network. In addition, the prediction model generator 300 may predict a residual life of a facility or component using a predetermined prediction algorithm, for example, a GPR algorithm (Gaussian Process Regression algorithm). That is, the prediction model generated by the prediction model generator 300 may be understood as a concept including an ANN model (or CNN model) and a GPR algorithm.

The residual life prediction unit 500 may predict the residual life (RUL) of the predicted equipment or the predicted component based on the vibration signal measured by the vibration sensor installed in the predicted equipment or the predicted component. Specifically, the vibration signal of the prediction target facility or the prediction target component is converted into a two-dimensional image by the data conversion unit 100, and the residual life prediction unit 500 transmits the two-dimensional image to the prediction model generation unit 300. The residual life (RUL) of a predicted equipment or a predicted component can be predicted by using it as an input of the predictive model generated by the predicted model.

The storage unit 700 includes a plurality of vibration signals, a plurality of two-dimensional images generated by the data conversion unit 100, a prediction target vibration signal, a two-dimensional image generated from the prediction target vibration signal, and a prediction model generator The predictive model generated by the 300, the vibration signal generated from the evaluation target facility or the evaluation target component may be stored.

2 shows an example of a vibration signal detected from a facility or component.

The vibration signal shown in FIG. 2 is a vibration signal measured over the whole lifetime in a specific bearing. In the vibration signal of the time domain, the horizontal axis and the vertical axis represent time and vibration amplitude, respectively. The unit [g] representing the acceleration of gravity can be used to define the size. Where 1 g is equal to 9.81 m / s ² . The magnitude of the vibration signal tends to increase gradually with time, which means that the vibration signal contains a lot of information for diagnosis and prediction.

In the present invention, a continuous wavelet transform (CWT) is used to extract features (eg, image features) from a vibration signal. According to an embodiment, the continuous wavelet transform may be a Morlet-based CWT.

FIG. 3 shows a contour plot of the wavelet power spectrum generated by the vibration signal and continuous wavelet transform (CWT). The vibration signal sampled at 0.1 s intervals at a frequency of 25.6 kHz has 2560 data points.

FIG. 3A shows the vibration signal of a component in a normal condition, and FIG. 3C shows the vibration signal of a component in a fault condition. When the parts have reached the end of their service life, the magnitude of the vibration signal is distributed between -30 g and 30 g, which can be seen to be very large compared to the vibration signal in the steady state. In the wavelet power spectrum, the horizontal and vertical axes represent time and frequency, respectively. The color of each point represents the magnitude of wavelet coefficients in a time-frequency grid. For example, red color means that the energy level is high. Referring to FIG. 3 (b), it can be seen that most of the energy is concentrated around 4 kHz when the bearing is operating under normal conditions, and there is no clear periodicity in the energy distribution. On the other hand, referring to (d) of FIG. 3, energy bursts are shown in a range from about 0.25 kHz to a point exceeding 8 kHz. Shock occurs at regular intervals in the frequency band between 0.25 kHz and 1 kHz. When high energy is observed in the low frequency band, this may mean that the life of the bearing is close.

As described above, the data converter 100 may convert a one-dimensional vibration signal into a two-dimensional image. The two-dimensional image may be generated by displaying a wavelet coefficient in a color corresponding to its size in a rectangular coordinate system consisting of a time axis and a frequency axis. That is, the 2D image may mean a wavelet power spectrum.

4 is a diagram for explaining a CNN model generated by the prediction model generator 300 shown in FIG. 1.

Referring to FIG. 4, the prediction model generator 300 may generate a convolutional neural network (CNN) model. For feature extraction, the CNN model includes four alternately connected convolutional layers and four pooling layers. The filter size of the convolution layer is 3 × 3, and the filter size of the max pooling layer is 2 × 2. The channels of each of the four convolutional layers are 32, 64, 128, and 256. The final layer of the CNN model includes a flatten layer, two fully connected layers and an output layer. After going through the convolutional layers, the two-dimensional feature map is flattened and connected to the pulley connected layer. In the pulley connected layers, 2560 and 768 nodes are selected respectively. Finally, the last pulley connected layer and the output neuron are connected to obtain a health indicator (HI). The ReLU function is applied to the convolutional layer and the pulley connected layer, and the sigmoid function is used in the output layer to normalize the value of the health indicator (HI) to a value between 0 and 1 (may include 0 and 1). Can be. After the parameter setting is completed, the CNN model is trained to minimize the loss function represented by Equation (1). That is, the prediction model generator 300 may train the CNN model using the image data generated by the data converter 100.

는 시간

에서 CNN 모델의 예측값(predicted value)이고

는 실제값(actual value)이며,

는 고장 시간(즉, 설비 또는 부품의 수명)을 의미한다.In Equation 1,

The time

Is the predicted value of the CNN model

Is the actual value,

Means failure time (i.e., the life of the equipment or parts).

FIG. 5 is a diagram for describing a residual life prediction operation of the prediction model generated by the prediction model generation unit illustrated in FIG. 1. That is, Figure 5 shows the estimated health index (estimated HI) and the predicted health index (predicted HI) of a particular component.

When the health index (HI) by the CNN model is obtained (calculated), a GPR algorithm (Gaussian Process Regression algorithm) is used to predict the residual life (RUL). The GPR algorithm (or GPR model) is a kind of kernel-based machine learning technique. In the present invention, the average and covariance functions can be assumed to be 0 and a linear function, respectively. The hyper-parameters for the covariance function can be optimized by maximizing the log-likelihood function. The GPR model can predict the future health indicator (HI) from the estimated health indicator (HI) up to the current time.

In FIG. 5, the health indicator HI up to the current time obtained by the CNN model is indicated by a dot. In FIG. 5, solid lines indicate estimated health indicators (HI) until the current time point and predicted health indicators (HI) after the current time point by the GPR model. The dotted lines indicate a confidence interval of 95%. Residual life (RUL) can be calculated as the difference between the time when the health indicator (HI) reaches a threshold (threshold) and the current time. Here, the threshold may be determined by the engineer's domain knowledge or may be determined through experimentation. In addition, it is obvious that the threshold value is not a fixed value, but can be changed depending on the situation or depending on the parts or equipment applied.

FIG. 6 is a flowchart illustrating a residual life prediction method performed by the residual life prediction apparatus illustrated in FIG. 1. In describing the remaining life prediction method, the description will be omitted with respect to the contents overlapping with those described above.

Referring to FIG. 6, the data conversion unit 100 of the residual life prediction apparatus 10 converts each of the one-dimensional vibration signals into a two-dimensional image (S100). That is, the data converter 100 may generate a two-dimensional image by displaying (or expressing) the characteristics (eg, wavelet coefficients) obtained from the one-dimensional vibration signal in a rectangular coordinate system including a time axis and a frequency axis. .

The prediction model generator 300 of the residual life prediction apparatus 10 may generate a prediction model using the two-dimensional images generated by the data converter 100 (S300). Specifically, the predictive model generator 300 generates a trained CNN model by using the two-dimensional images generated by the data converter 100 as training data, and the future from the output of the trained CNN model. A predictive model can be generated by combining a GPR model (or GPR algorithm) that predicts health indicators (HI) and predicts residual life (RUL). That is, the prediction model is understood as a concept including a CNN model and a GPR model, but according to an embodiment, the prediction model may mean only a CNN model. In this case, the prediction model generator 300 generates only the trained CNN model, and the residual life prediction through the GPR model may be performed by the residual life predictor 500.

In step S500, the residual life (RUL) of the predicted equipment or the predicted component is predicted. Specifically, the one-dimensional vibration signal from the prediction target facility or the prediction target component to the current viewpoint is converted into a two-dimensional image by the data converter 100. In addition, the residual life prediction unit 500 may predict the residual life (RUL) of the prediction target facility or the prediction target component using the prediction model generated by the prediction model generation unit 300. According to an embodiment, the prediction model may include only a CNN model. In this case, the residual life prediction unit 500 calculates the health index (HI) of the prediction target facility or the prediction target part using a prediction model, that is, a CNN model, and uses the GPR model to calculate the residual life (RUL). Predictable.

The device described above may be implemented as a hardware component, a software component, and / or a set of hardware components and software components. For example, the devices and components described in the embodiments include, for example, a processor, a controller, an Arithmetic Logic Unit (ALU), a digital signal processor (MIC), a microcomputer, a field programmable array (FPA), and a PLU (Programmable Logic Unit), microprocessor, or any other device capable of executing and responding to an instruction, may be implemented using one or more general purpose computers or special purpose computers. The processing device may perform an operating system (OS) and one or more software applications running on the operating system. In addition, the processing device may access, store, manipulate, process, and generate data in response to the execution of the software. For convenience of understanding, a processing device may be described as one being used, but a person having ordinary skill in the art, the processing device may include a plurality of processing elements and / or multiple types of processing elements. It can be seen that may include. For example, the processing device may include a plurality of processors or a processor and a controller. In addition, other processing configurations, such as parallel processors, are possible.

The software may include a computer program, code, instruction, or a combination of one or more of them, and the processing device may be configured to operate as desired or processed independently or collectively. You can command the device. Software and / or data may be interpreted by a processing device, or to provide instructions or data to a processing device, of any type of machine, component, physical device, virtual equipment, computer storage medium or device. , Or may be permanently or temporarily embodied in the transmitted signal wave. The software may be distributed on networked computer systems, and stored or executed in a distributed manner. Software and data may be stored in one or more computer-readable recording media.

The method according to the embodiment may be implemented in the form of program instructions that can be executed through various computer means and may be recorded on a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, or the like alone or in combination. The program instructions recorded on the medium may be specially designed and configured for an embodiment or may be known and available to those skilled in computer software. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tapes, magnetic media such as optical media such as CD-ROMs and DVDs, and floppy disks. Includes hardware devices specifically configured to store and execute program instructions such as magneto-optical media, ROM, RAM, flash memory, and the like. Examples of program instructions include high-level language code that can be executed by a computer using an interpreter, etc., as well as machine language codes produced by a compiler. The hardware device described above may be configured to operate as one or more software modules to perform the operations of the embodiments, and vice versa.

The present invention has been described with reference to the embodiments shown in the drawings, but these are merely exemplary, and those skilled in the art will understand that various modifications and other equivalent embodiments are possible therefrom. For example, the described techniques are performed in a different order than the described method, and / or the components of the described system, structure, device, circuit, etc. are combined or combined in a different form from the described method, or other components Alternatively, even if replaced or substituted by equivalents, appropriate results can be achieved. Therefore, the true technical protection scope of the present invention should be determined by the technical spirit of the appended claims.

10: residual life prediction device
100: data conversion unit
300: prediction model generator
500: residual life prediction unit
700: storage unit

Claims (10)

A data conversion unit that converts each of the plurality of 1D vibration signals into a 2D image;
A predictive model generator that trains an artificial neural network (ANN) model that outputs a health indicator (HI) using a plurality of two-dimensional images generated by the data converter as input data; And
A residual life prediction unit for predicting a residual life (RUL) of a predicted facility or a predicted part using a trained ANN model is included,
The data conversion unit generates the two-dimensional images by performing continuous wavelet transformation (CWT) on each of the one-dimensional vibration signals,
Residual life prediction device.
According to claim 2,
The ANN model is a CNN (Convolutional Neural Network),
Residual life prediction device.
According to claim 2,
In the two-dimensional image, each of wavelet coefficients obtained from a one-dimensional vibration signal in a rectangular coordinate system consisting of a time axis and a frequency axis is an image represented by a color corresponding to a size,
Residual life prediction device.
According to claim 3,
The data converter generates a predicted target image by continuously performing wavelet transform (CWT) on a predicted target vibration signal generated from the predicted target facility or the predicted target component,
The residual life prediction unit calculates a health index (HI) up to the current time point by inputting the predicted target image into the learned CNN model, and estimates a health index up to the current time point using a GPR (Gaussian Process Regression) algorithm. To predict future health indicators,
Residual life prediction device.
According to claim 4,
The residual life prediction unit determines a difference between a time point at which the estimated future health indicator (HI) reaches a threshold and a current time point as the residual life (RUL) of the predicted facility or the predicted part.
Residual life prediction device.
The method of claim 5,
The health indicator (HI) has a value between 0 and 1,
Residual life prediction device.
In the residual life prediction method performed by the residual life prediction device,
Converting each of a plurality of one-dimensional vibration signals into a two-dimensional image by a data conversion unit included in the residual life prediction apparatus;
Learning a CNN model outputting a health indicator (HI) by using a plurality of 2D images generated by the data conversion unit as a prediction model generator included in the residual life prediction apparatus;
And predicting a residual life (RUL) of the predicted equipment or the predicted part using the CNN model in which the residual life predicting unit included in the residual life predicting apparatus is trained,
The converting into the 2D image may include generating the 2D images by continuously performing wavelet transform (CWT) on each of the 1D vibration signals,
How to predict residual life.
The method of claim 7,
In the two-dimensional image, each of the wavelet coefficients obtained from the one-dimensional vibration signal in a rectangular coordinate system consisting of a time axis and a frequency axis is an image represented by a color corresponding to the size,
How to predict residual life.
The method of claim 8,
The step of predicting the residual life (RUL),
Generating, by the data conversion unit, a predictive vibration signal generated from the predicted target facility or the predicted target component by performing wavelet transform;
Calculating, by the residual life prediction unit, a health index (HI) up to a current time point by inputting the prediction target image into the learned CNN model;
Estimating the health index up to the present time and predicting the future health index using the GPR algorithm; And
And determining a difference between a time point at which the residual health prediction unit estimates the estimated future health index (HI) reaches a threshold and a current time point as the residual life (RUL) of the prediction target facility or the prediction target component,
How to predict residual life.
The method of claim 9,
The health indicator (HI) has a value between 0 and 1,
How to predict residual life.

Images