KR20230083810A - Method for building predictive model of roll bearing life in the roll-to-roll process - Google Patents

Abstract

The present invention relates to a method for building a roll bearing life prediction model for a roll-to-roll process. The method comprises the steps of: measuring vibration data generated during the operation of the roll-to-roll process; building a feature dataset by extracting at least one piece of feature data from the vibration data; generating learning data by selecting feature data regarding monotonicity and correlation based on the feature dataset; and building a life prediction model to predict the life of a roll bearing in the roll-to-roll process by learning the learning data.

Description

Roll bearing life prediction model construction method of roll-to-roll process {METHOD FOR BUILDING PREDICTIVE MODEL OF ROLL BEARING LIFE IN THE ROLL-TO-ROLL PROCESS}

The present invention relates to bearing monitoring and diagnosis technology, and more particularly, to a method for building a roll bearing life prediction model in a roll-to-roll process for building a bearing life prediction model by collecting and statistically analyzing vibration data generated in bearings. it's about

In the roll-to-roll process, various rolls may operate, and roll bearings supporting each roll may be used. In this case, in the case of a roll bearing, the failure rate may increase as the operating time increases, and failure of the roll bearing may adversely affect the process.

At this time, the cause of failure of the roll bearing may include insufficient lubrication, use of an inappropriate lubricant, incorrect installation of the bearing, excessive deformation of the shaft system, etc. Recently, various techniques for diagnosing failure of the roll bearing have been developed.

However, although monitoring the operation of the roll bearing and predicting failure in advance are very important tasks in the process, technology for accurate diagnosis and prediction is still insufficient.

Korean Patent Registration No. 10-1823746 (2018.01.24)

One embodiment of the present invention is to provide a roll bearing life prediction model construction method of a roll-to-roll process for building a bearing life prediction model by statistically analyzing vibration data generated in bearings.

Among the embodiments, a method for building a roll bearing life prediction model of a roll-to-roll process includes measuring vibration data generated during an operation of a roll-to-roll process; constructing a feature dataset by extracting at least one feature data from the vibration data; generating learning data by selecting feature data related to monotonicity or correlation based on the feature dataset; and constructing a life prediction model for predicting the life of the roll bearing in the roll-to-roll process by learning the learning data.

Measuring the vibration data may include measuring vibration generated during an operation of the roll bearing through a 3-axis acceleration sensor attached to the roll bearing in the roll-to-roll process.

Measuring the vibration data may include measuring the vibration by attaching the three-axis acceleration sensor to the housing of the roll bearing in a vertical direction.

Building the feature dataset may include extracting a frequency component from the vibration data by applying an empirical mode decomposition method.

The constructing of the feature data set may include acquiring a natural frequency component of the defect of the roll bearing in a time domain among the frequency components.

The generating of the training data may include constructing the training data by extracting feature variables including kurtosis, skewness, standard deviation, and mean value from the feature data of the time domain.

The generating of the training data may include deriving an envelope of a specific signal based on the feature dataset; calculating slopes of a first inflection point and a second inflection point on the envelope; determining a directional deflection of the specific signal using the slopes; and determining an axis direction having the greatest directional bias and generating the learning data using feature data of the corresponding axis direction.

Generating the learning data may include calculating a directional deflection of the specific signal through the following equation.

[mathematical expression]

(Here, DRUL is a directional deflection, Yupper and Ylower are the first and second inflection points, respectively, and diff() is a differential function.)

Building the life prediction model may include building a bearing diagnosis model by applying a machine learning technique including a support vector machine to the learning data.

The method for constructing the roll bearing life prediction model includes the step of predicting the remaining life of the roll bearing based on a point in time when a predetermined threshold value is reached by inputting a health factor with respect to time to the life prediction model; can include more.

The disclosed technology may have the following effects. However, it does not mean that a specific embodiment must include all of the following effects or only the following effects, so it should not be understood that the scope of rights of the disclosed technology is limited thereby.

In the method for constructing a roll bearing life prediction model in a roll-to-roll process according to an embodiment of the present invention, a bearing life prediction model may be constructed by collecting vibration data generated from bearings and then statistically analyzing them.

The method for constructing a roll bearing life prediction model in a roll-to-roll process according to an embodiment of the present invention predicts the life state of the bearings of the rolls used in the roll-to-roll process to build a smart production facility system, and monitors the life of the bearings to detect bearing defects. Productivity can be improved by preventing product quality deterioration due to

1 is a diagram illustrating a life prediction model building system according to the present invention.
FIG. 2 is a diagram explaining the system configuration of the life prediction model building device of FIG. 1 .
FIG. 3 is a diagram for explaining the logical configuration of the device for building a life prediction model of FIG. 1 .
4 is a diagram illustrating a method of building a roll bearing life prediction model in a roll-to-roll process according to the present invention.
5 is a diagram illustrating vibration data of a bearing according to the present invention.
6A and 6B are diagrams for explaining axial vibration data of bearings.
7A and 7B are diagrams illustrating the life prediction results of bearings according to the present invention.
8 is a flowchart illustrating a method of constructing a life prediction model according to the present invention.

Since the description of the present invention is only an embodiment for structural or functional description, the scope of the present invention should not be construed as being limited by the embodiments described in the text. That is, since the embodiment can be changed in various ways and can have various forms, it should be understood that the scope of the present invention includes equivalents capable of realizing the technical idea. In addition, since the object or effect presented in the present invention does not mean that a specific embodiment should include all of them or only such effects, the scope of the present invention should not be construed as being limited thereto.

Meanwhile, the meaning of terms described in this application should be understood as follows.

Terms such as "first" and "second" are used to distinguish one component from another, and the scope of rights should not be limited by these terms. For example, a first element may be termed a second element, and similarly, a second element may be termed a first element.

It should be understood that when an element is referred to as being “connected” to another element, it may be directly connected to the other element, but other elements may exist in the middle. On the other hand, when an element is referred to as being "directly connected" to another element, it should be understood that no intervening elements exist. Meanwhile, other expressions describing the relationship between components, such as “between” and “immediately between” or “adjacent to” and “directly adjacent to” should be interpreted similarly.

Expressions in the singular number should be understood to include plural expressions unless the context clearly dictates otherwise, and terms such as “comprise” or “having” refer to an embodied feature, number, step, operation, component, part, or these. It should be understood that it is intended to indicate that a combination exists, and does not preclude the possibility of the presence or addition of one or more other features, numbers, steps, operations, components, parts, or combinations thereof.

In each step, the identification code (eg, a, b, c, etc.) is used for convenience of explanation, and the identification code does not describe the order of each step, and each step clearly follows a specific order in context. Unless otherwise specified, it may occur in a different order than specified. That is, each step may occur in the same order as specified, may be performed substantially simultaneously, or may be performed in the reverse order.

The present invention can be implemented as computer readable code on a computer readable recording medium, and the computer readable recording medium includes all types of recording devices storing data that can be read by a computer system. . Examples of computer-readable recording media include ROM, RAM, CD-ROM, magnetic tape, floppy disk, and optical data storage devices. In addition, the computer-readable recording medium may be distributed to computer systems connected through a network, so that computer-readable codes may be stored and executed in a distributed manner.

All terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present invention belongs, unless defined otherwise. Terms defined in commonly used dictionaries should be interpreted as consistent with meanings in the context of the related art, and cannot be interpreted as having ideal or excessively formal meanings unless explicitly defined in the present application.

1 is a diagram illustrating a life prediction model building system according to the present invention.

Referring to FIG. 1 , the life prediction model building system 100 may include a process device 110 , a life prediction model building device 130 and a database 150 .

The process device 110 may correspond to one component of a roll-to-roll system that performs a roll-to-roll process, and may be implemented by including at least one roll including a roll bearing. That is, the processing device 110 may correspond to a device for operating rolls organically coupled with a roll bearing in a roll-to-roll process, and may include a movable unit performing a roll operation and an arithmetic unit controlling the operation of the movable unit. can The process device 110 may operate in connection with the life prediction model building device 130, and may implement the roll bearing life prediction model building method of the roll-to-roll process according to the present invention.

In addition, the process apparatus 110 may include at least one roll, and the corresponding roll may operate in combination with a roll bearing. The process device 110 may be implemented by including various sensors for measuring information about the operation of the roll bearing. For example, the process device 110 may include an acceleration sensor, a pressure sensor, a temperature sensor, a current sensor, a sound sensor, a frequency sensor, and the like. In one embodiment, the process device 110 may be implemented by including at least one vibration sensor capable of measuring the vibration of the roll bearing, in which case the vibration sensor may be directly connected to the life prediction model building device 130. there is.

The life prediction model building device 130 is a server corresponding to a computer or program that collects vibration data about a roll bearing and selects and learns feature data about monotonicity and correlation to build a life prediction model. can be implemented as The life prediction model building device 130 may be connected to the process device 110 through a wired or wireless network, and may transmit/receive data with the process device 110 through the network. In addition, the life prediction model building device 130 may be implemented to operate in conjunction with a separate external system (not shown in FIG. 1) to collect data or provide additional functions.

The database 150 may correspond to a storage device for storing various information necessary for the operation of the life prediction model building device 130 . The database 150 may store vibration data and characteristic data related to a roll bearing, and may store learning data and a learning algorithm related to a life prediction model, but are not necessarily limited thereto, and build a roll bearing life prediction model in a roll-to-roll process. In the process, information collected or processed in various forms can be stored.

FIG. 2 is a diagram explaining the system configuration of the life prediction model building device of FIG. 1 .

Referring to FIG. 2 , the device for building a life prediction model 130 may be implemented by including a processor 210, a memory 230, a user input/output unit 250, and a network input/output unit 270.

The processor 210 may execute a procedure for processing each step in the operation of the life prediction model building device 130, manage the memory 230 read or written throughout the process, and Synchronization time between volatile memory and non-volatile memory in 230 can be scheduled. The processor 210 may control the overall operation of the life prediction model building device 130, and is electrically connected to the memory 230, the user input/output unit 250, and the network input/output unit 270 to prevent data flow between them. You can control it. The processor 210 may be implemented as a central processing unit (CPU) of the life prediction model building device 130 .

The memory 230 is implemented as a non-volatile memory such as a solid state drive (SSD) or a hard disk drive (HDD) and may include a secondary storage device used to store all data required for the life prediction model building device 130. and may include a main memory implemented as a volatile memory such as RAM (Random Access Memory).

The user input/output unit 250 may include an environment for receiving user input and an environment for outputting specific information to the user. For example, the user input/output unit 250 may include an input device including an adapter such as a touch pad, a touch screen, an on-screen keyboard, or a pointing device, and an output device including an adapter such as a monitor or touch screen. In one embodiment, the user input/output unit 250 may correspond to a computing device connected through a remote connection, and in such a case, the life prediction model building device 130 may be implemented as an independent server.

The network input/output unit 270 includes an environment for connecting to an external device or system through a network, and includes, for example, a local area network (LAN), a metropolitan area network (MAN), a wide area network (WAN), and a VAN ( An adapter for communication such as Value Added Network) may be included.

FIG. 3 is a diagram for explaining the logical configuration of the device for building a life prediction model of FIG. 1 .

Referring to FIG. 3, the life prediction model building device 130 includes a vibration data measurement unit 310, a feature dataset building unit 330, a learning data generating unit 350, a life prediction model building unit 370, a residual It may include a life predicting unit 390 and a control unit (not shown in FIG. 3).

The vibration data measurement unit 310 may measure vibration data generated during the operation of the roll-to-roll process. Although it has been described here that vibration data is specified and collected, it is not necessarily limited thereto, and the vibration data measuring unit 310 may collect various data during the operation of the roll-to-roll process and store them in the database 150. In particular, the vibration data measurement unit 310 may operate in conjunction with a vibration sensor that measures vibration data, and may operate in connection with various sensors depending on the type and number of data to be collected.

In one embodiment, the vibration data measuring unit 310 may perform a predetermined preprocessing operation on the collected vibration data. For example, the vibration data measurer 310 may perform a sampling operation on the vibration data, and may collect only vibration data within a predetermined effective range by applying a range filter. Also, the vibration data measuring unit 310 may normalize the measured vibration data to data within a predetermined range.

In one embodiment, the vibration data measuring unit 310 may measure vibration generated during the operation of the roll bearing through a 3-axis accelerometer attached to the roll bearing in the roll-to-roll process. Vibration data measurement unit 310 can measure the sensing values in the x-axis, y-axis and z-axis through the 3-axis acceleration sensor, and the velocity and displacement (i.e., the amount of change in position) in the time domain, and in each axis of vibration data can be collected. The vibration data measurement unit 310 may collect vibration data from each of the roll bearings in the roll-to-roll process, and in this case, it may be stored after connecting identification information about the corresponding roll bearing with vibration data.

In one embodiment, the vibration data measuring unit 310 may measure vibration by attaching a 3-axis acceleration sensor to the housing of the roll bearing in a vertical direction. In this case, the vertical direction may correspond to the z-axis direction of the 3-axis acceleration sensor, and accordingly, the x-axis and the y-axis may correspond to vertical and horizontal movements of the roll bearing, respectively. On the other hand, the vibration of the roll bearing can be divided into the main axis direction and the vertical/horizontal direction, and can be expressed as the x-axis, y-axis, and z-axis according to the direction of the sensor attached to the housing.

The feature dataset builder 330 may construct a feature dataset by extracting at least one feature data from vibration data. For example, the feature dataset builder 330 determines time domain features, frequency domain spectral, and time-frequency domain features based on vibration data collected from a 3-axis acceleration sensor. (time-frequency domain features) may be collected, classified for each data type, and stored in the database 150. That is, the feature dataset may correspond to a feature set for building a learning model, and may be composed of a set of partial feature datasets classified according to a predetermined classification criterion. Also, the feature dataset may be used to generate training data for building a learning model.

In an embodiment, the feature dataset builder 330 may extract a frequency component from vibration data by applying an empirical mode decomposition method. Here, the empirical mode decomposition method is one of the decomposition analysis techniques and may correspond to a method of decomposing a signal into a function called an intrinsic mode function according to the degree of local frequency. The empirical mode decomposition method is a calculation method based on an empirical algorithm and can be applied to signals having non-stationarity in that wave data inherent in a signal can be easily and data adaptively extracted. The feature dataset builder 330 may extract a specific frequency component from the time domain of the vibration data to build a separate independent dataset.

In an embodiment, the feature dataset builder 330 may acquire a natural frequency component of a defect in a roll bearing in the time domain among frequency components. That is, the feature dataset builder 330 can classify the feature data into various frequency domains in the time domain, and extract natural frequency components related to defects in the roll bearing to construct a separate and independent dataset. Meanwhile, defects of roll bearings may include flaking, peeling, scoring, smearing, fracture, cracks, cage, denting, pitting, wear, creep, and the like. The feature dataset builder 330 may extract a frequency component unique to each defect, and build an independent dataset by classifying each defect. In particular, the feature dataset builder 330 may determine a combination of frequency components as needed and then build a dataset according to the combination.

The learning data generation unit 350 may generate learning data by selecting feature data related to monotonicity or correlation based on the feature dataset. The learning data generation unit 350 may selectively extract feature data about monotonicity or correlation from the feature dataset and use it as learning data for constructing a learning model. In particular, considering the fact that the vibration data of the roll bearing is utilized, the characteristic data on forgeability can be utilized with the highest priority, and when other characteristic data of the roll bearing are utilized, the characteristic associated with the characteristic data can be applied with the highest priority. Of course you can.

In one embodiment, the learning data generator 350 configures the learning data by extracting feature variables including kurtosis, skewness, standard deviation, average value, etc. from the feature data in the time domain. can do. That is, the learning data generator 350 may generate learning data by combining some of the statistical data of the feature data in consideration of the learning purpose and the characteristics of the learning algorithm. For example, in the case of learning data composed of kurtosis, skewness, standard deviation and average values for feature data, it can be used for a diagnostic model for diagnosing defects (or abnormalities) of roll bearings, In the case of learning data consisting of the median value, RMS, and Peak to RMS of the feature data having correlation, it can be used for a prediction model for predicting the remaining life of the roll bearing.

In one embodiment, the learning data generation unit 350 derives an envelope for a specific signal based on the feature dataset, calculates slopes of a first inflection point and a second inflection point on the envelope, and uses the slopes to determine the value of the specific signal. A directional bias may be determined, an axis direction having the largest directional bias may be determined, and learning data may be generated using feature data of the corresponding axis direction.

More specifically, the learning data generation unit 350 may derive an envelope for the characteristic data of the x-axis among the characteristic data collected through the 3-axis acceleration sensor. At this time, the characteristic data of the x-axis may be divided into upper data (or positive data) and lower data (or negative data) according to the vibration of the roll bearing, and accordingly, the learning data generator 350 may use the characteristic data of the x-axis Two different envelopes can be derived for

Then, the learning data generation unit 350 may calculate slopes of the first inflection point and the second inflection point on the envelope. In this case, the first inflection point may correspond to a positive vertex present on the first envelope of upper data, and the second inflection point may correspond to a negative vertex present on the second envelope of lower data.

Then, the learning data generator 350 may determine a directional bias of a specific signal using the slopes. Here, the directional deflection may correspond to a characteristic value indicating a degree of change in characteristic data (eg, vibration data). That is, the greater the directional deflection, the greater the change in vibration in the corresponding direction, and may correspond to greater monotonicity. The learning data generation unit 350 calculates the directional deflection instead of utilizing both the characteristic data in the main axis direction and the radial direction of the vibration data, and then selectively extracts only the characteristic data in the direction in which the directional bias is large, thereby learning for a life prediction model. data can be determined.

In one embodiment, the learning data generator 350 may calculate the directional deflection of a specific signal through the following equation.

[mathematical expression]

Here, D _RUL is a directional bias, Y _upper and Y _lower are the first and second inflection points, respectively, and diff() is a differential function. That is, the learning data generator 350 derives the envelope of the characteristic data based on the specific direction of the vibration data, calculates the slopes at the first and second inflection points for each envelope, and calculates the average value of the difference between the slopes. It can be determined as a directional bias for the corresponding direction. For example, in the case of vibration data, directional deflections may be calculated in the main axis direction and the radial direction, respectively, and various numbers of directional deflections may be derived according to characteristic data.

The life prediction model building unit 370 may build a life prediction model that predicts the life of the roll bearing in the roll-to-roll process by learning the learning data. The life prediction model builder 370 may learn the learning data generated by the learning data generator 350 to build a life prediction model for predicting the life of the roll bearing. At this time, the lifespan prediction model building unit 370 may selectively apply a learning algorithm according to characteristic data, and may build a plurality of lifespan prediction models by independently applying a plurality of learning algorithms as needed.

In one embodiment, the life prediction model building unit 370 may build a bearing diagnosis model by applying a machine learning technique including a support vector machine to learning data. A support vector machine is one of the machine learning algorithms and is a supervised learning model for pattern recognition and data analysis, and is mainly used for classification and regression analysis. The life prediction model building unit 370 may build a bearing diagnosis model as a model independent of the life prediction model, and may perform an operation of diagnosing an abnormality of the roll bearing in the case of the bearing diagnosis model. That is, the bearing diagnosis model can perform an operation of detecting an abnormality of the bearing or predicting the type of abnormality (ie, defect) from measured vibration data by learning characteristics of frequency components related to defects of the roll bearing.

The remaining life predicting unit 390 may predict the remaining life of the roll bearing based on a time point when a predetermined threshold value is reached by inputting a health indicator for time to the life prediction model. That is, the point at which the threshold value is reached may correspond to the point at which the remaining life of the roll bearing ends, and the remaining life predicting unit 390 predicts the point at which the roll bearing reaches the critical value based on the current time point. The remaining life of the roll bearing can be determined by calculating the time from the present point to the point of arrival.

In one embodiment, the remaining life predicting unit 390 may predict the remaining life of the roll bearing by applying the bearing diagnosis model and the life prediction model step by step. More specifically, the remaining life predicting unit 390 may firstly detect a defect in the roll bearing through a bearing diagnosis model, and secondarily predict the remaining life of the roll bearing through a life prediction model when a defect is detected. can

In one embodiment, the remaining life predicting unit 390 may generate and provide a notification when an abnormality related to the roll bearing is detected or the remaining life is predicted to be less than a predetermined critical life. In this case, the corresponding notification may be delivered to the process device 110, and may be output through an output unit of the process device 110 and delivered to the user.

The controller (not shown in FIG. 3) controls the overall operation of the life prediction model building device 130, and includes a vibration data measuring unit 310, a feature data set building unit 330, a learning data generating unit 350, A control flow or data flow between the life prediction model building unit 370 and the remaining life prediction unit 390 may be managed.

4 is a diagram illustrating a method of building a roll bearing life prediction model in a roll-to-roll process according to the present invention.

Referring to FIG. 4 , the life prediction model building device 130 may measure vibration data generated during the operation of the roll-to-roll process through the vibration data measuring unit 310 (step S410). The life prediction model building apparatus 130 may construct a feature dataset by extracting at least one feature data from the vibration data through the feature dataset builder 330 (step S430).

In addition, the life prediction model building device 130 may generate learning data by selecting feature data related to monotonicity and correlation based on the feature dataset through the learning data generator 350. (Step S450). The life prediction model building device 130 can predict the life of the roll bearing in the roll-to-roll process by learning learning data through the life prediction model building unit 370 (step S470).

5 is a diagram illustrating vibration data of a bearing according to the present invention.

Referring to FIG. 5 , the life prediction model building device 130 may collect vibration data as characteristic data through a 3-axis acceleration sensor attached to a vertical direction of a roll bearing. The lifespan prediction model building device 130 may extract and selectively utilize a component having monotonicity from feature data of the signal in order to improve the accuracy of the lifespan prediction model.

In FIG. 5 , the life prediction model building device 130 may derive an envelope of a signal related to vibration data, and calculate slopes of Y _upper and Y _lower corresponding to inflection points on each envelope. The life expectancy model building apparatus 130 may calculate the characteristic value D _RUL of the signal by calculating an average value of the difference between the slopes. In this case, the characteristic value of the signal may correspond to a directional deflection. The life prediction model building device 130 may extract feature data such as kurtosis, skewness, standard deviation, average value, median value, RMS, Peak to RMS, etc. using axial data having large feature values, and based on the feature data A regression model of can be used to build a life prediction model.

6A and 6B are diagrams for explaining axial vibration data of bearings.

Referring to FIGS. 6A and 6B , the life prediction model building device 130 may collect vibration data of a roll bearing and classify the vibration data according to each axial direction. 6A may correspond to vibration data in the x-axis direction, and FIG. 6B may correspond to vibration data in the y-axis direction. In particular, the life prediction model building device 130 may calculate the directional deflection for each axis, and the test results shown in Table 1 below may be derived through the vibration test on the roll bearing.

Axis D _RUL X axis 0.417 Y-axis 0.443

Referring to Table 1, the directional deflection in the y-axis may be greater than the directional deflection in the x-axis of the roll bearing. Accordingly, the lifespan prediction model building apparatus 130 may generate learning data for a lifespan prediction model by selectively using only feature data in the Y-axis direction.

7A and 7B are diagrams illustrating the life prediction results of bearings according to the present invention.

Referring to FIGS. 7A and 7 , the life prediction model building apparatus 130 may predict the remaining life of the roll bearing using a previously built life prediction model. Specifically, the lifespan prediction model construction apparatus 130 may predict the remaining lifespan reaching a predetermined threshold value by inputting a health indicator with respect to time to the lifespan prediction model.

As shown in FIG. 7A, when data of 138 days is input, the predicted remaining lifespan may be calculated as about 310 days, and the confidence interval may correspond to 250 to 340 days. In addition, as shown in FIG. 7B, when data of 736 days is input, the predicted remaining lifespan may be calculated as about 52 days, and the confidence interval may correspond to 5 to 110 days. Meanwhile, in the experiment, the actual remaining lifespan may appear as 32 days, and since the actual remaining lifespan and the predicted remaining lifespan exist in the confidence interval, it can be derived that the remaining lifespan can be predicted.

Actual RUL(R _A ) Predict RUL(R _P ) Computation time Previous method 5,730s 4,723 s (83.57%) 128min Proposed method 4,480s 4,148 s (92.59%) 68min

On the other hand, as shown in Table 2 above, as a result of constructing a life prediction model by obtaining a directional bias of data, it is possible to predict that life changes according to size change during principal component analysis. In the point where the actual life expectancy reliability curve and the predicted life reliability curve overlap on the final predicted life span α-λ graph, the use of a soundness factor using a directional bias rather than raw data may be more appropriate. The accuracy used in the experiment can be expressed as in Equation 1 below.

[Equation 1]

Here, R _A is Actual Remaining Useful Life (RUL), and R _p is Predict RUL.

8 is a flowchart illustrating a method of constructing a life prediction model according to the present invention.

Referring to FIG. 8 , the life prediction model building device 130 can build a roll bearing life prediction model through three major steps. That is, the life prediction model construction step may proceed to a feature extraction step (S810), a feature selection step (S830), and a model construction step (S850).

First, in the feature extraction step (S810), vibration data is collected through a sensor attached to a roll bearing, and features including a directional bias are extracted from the vibration data to generate a feature set. may correspond to In this case, the feature dataset may be composed of various feature data, and includes time domain features, frequency domain spectral, time-frequency domain features, and the like. can include

The feature selection step (S830) may correspond to a step of selecting features based on monotonicity and correlation based on the feature dataset and generating training data about the selected features. can In this case, the learning data may be expressed as a combination of selected features.

In addition, the model building step (S850) may correspond to a step of building a life prediction model by learning learning data. In one embodiment, the life expectancy model may be defined based on a Weibull distribution.

Although the above has been described with reference to preferred embodiments of the present invention, those skilled in the art will variously modify and change the present invention within the scope not departing from the spirit and scope of the present invention described in the claims below. You will understand that it can be done.

100: life prediction model building system
110: process device 130: life prediction model building device
150: database
210: processor 230: memory
250: user input/output unit 270: network input/output unit
310: vibration data measurement unit 330: feature data set construction unit
350: learning data generation unit 370: life prediction model building unit
390: remaining life prediction unit

Claims (10)

Measuring vibration data generated during an operation of a roll-to-roll process;
constructing a feature dataset by extracting at least one feature data from the vibration data;
generating learning data by selecting feature data related to monotonicity or correlation based on the feature dataset; and
Building a life prediction model for predicting the life of the roll bearing in the roll-to-roll process by learning the learning data;
The method of claim 1, wherein measuring the vibration data comprises
Measuring vibration generated during the operation of the roll bearing through a three-axis acceleration sensor attached to the roll bearing of the roll-to-roll process.
The method of claim 2, wherein measuring the vibration data comprises
A method for building a life prediction model of a roll bearing in a roll-to-roll process, comprising the step of measuring the vibration by attaching the three-axis acceleration sensor in the vertical direction to the housing of the roll bearing.
The method of claim 1, wherein the step of building the feature dataset
A method for building a roll bearing life prediction model in a roll-to-roll process, comprising the step of extracting a frequency component by applying an empirical mode decomposition method from the vibration data.
The method of claim 4, wherein the step of building the feature dataset
A method for building a life prediction model of a roll bearing in a roll-to-roll process, comprising the step of acquiring a natural frequency component related to the defect of the roll bearing in the time domain among the frequency components.
The method of claim 5, wherein generating the learning data
Constructing the learning data by extracting feature variables including kurtosis, skewness, standard deviation, and average value from the feature data of the time domain, respectively. Roll-to-roll process rolling bearing life prediction model How to build.
The method of claim 1, wherein generating the learning data
deriving an envelope for a specific signal based on the feature dataset;
calculating slopes of a first inflection point and a second inflection point on the envelope;
determining a directional bias of the specific signal using the slopes; and
Determining an axial direction having the largest directional deflection and generating the learning data using feature data of the corresponding axial direction.
The method of claim 1, wherein generating the learning data
A method for building a roll bearing life prediction model in a roll-to-roll process, comprising calculating a directional bias of the specific signal through the following equation.
[mathematical expression]

(Here, DRUL is a directional deflection, Yupper and Ylower are the first and second inflection points, respectively, and diff() is a differential function.)
The method of claim 6, wherein the step of building the life prediction model
Building a bearing diagnosis model by applying a machine learning technique including a support vector machine to the learning data.
According to claim 1,
and predicting the remaining life of the roll bearing based on a point in time when a predetermined threshold value is reached by inputting a health indicator for time into the life prediction model. A method for constructing a model for predicting the life of roll bearings in a process.

Images