KR20210082330A - Method and apparatus for estimating remaining useful life - Google Patents

Abstract

A method for predicting the remaining useful life of a test subject is provided. After acquiring time series data for a plurality of different characteristics from the test subject for a predetermined time, an integrity curve is obtained from the acquired time series data, and the remaining useful life of the test subject is evaluated by comparing the integrity curve stored in advance. The remaining useful life can be predicted without assumptions, and the present invention is robust against noise. Using this method, it is possible to evaluate the soundness of an epoxy injector for forming an epoxy seal of a display module of an electronic product.

Description

Method and apparatus for predicting remaining useful life {Method and apparatus for estimating remaining useful life}

The present invention relates to a method and apparatus for predicting the remaining useful life of an adhesive material injector, and more specifically, to an injector for extruding an adhesive material such as an epoxy resin to the periphery of a display panel in the process of manufacturing a display of an electronic product It relates to a method for predicting lifespan and an apparatus therefor.

Recently, a number of sensors have been installed to monitor the operation of devices at manufacturing sites. Accordingly, interest in maintenance is increasing, such as predicting device failure or predicting remaining useful life using data output from these sensors.

Various remaining useful life prediction methods have been proposed, but most are based on assumptions about how the device deteriorates. It is difficult to build a physical model for the degradation analysis of a complex device having a large number of components. Accordingly, many data-driven approaches assume that the deterioration trend is, for example, exponential deterioration. This method is particularly useful when there are no parameters to measure the state of the device. However, this assumption becomes invalid as, for example, when certain components of a device are nearing the end of their lifespan, the sensor data is initially symptomatic intermittently and then increases in a non-exponential trend over time.

In addition, the sensor data generated from the device has difficulties such as a lot of noise and often occurrence of measured values. Sensor measurements are affected by varying levels of environmental noise, and the magnitude of the noise may vary from sensor to sensor.

USUS10289509 B USUS20150227838 A

An object of the present invention is to solve the above problem, and an object of the present invention is to provide a method for predicting the remaining useful life without based on assumptions related to device deterioration.

Another object of the present invention is to provide a method that is robust against noise and can predict the remaining useful life even with measured values.

In order to solve the above problems, according to an aspect of the present invention, there is provided a method for predicting the remaining useful life of an inspection leader, (a) for a predetermined time, time-series data for a plurality of different characteristics from the inspection object obtaining; (b) obtaining a health curve for the predetermined time of the test subject from the time series data obtained in step (a); And, (c) comparing the health curve obtained in step (b) with one or more pre-stored health curves to evaluate the remaining useful life of the test subject is provided.

As another aspect of the present invention, there is provided a method for evaluating the soundness of an epoxy injector forming an epoxy sealing part of an electronic product display module using the above-described remaining life prediction method.

As another aspect of the present invention, there is provided an apparatus for predicting a remaining useful life, comprising: at least one processor; and at least one memory storing computer-executable instructions, wherein the computer-executable instructions stored in the at least one memory are, by the at least one processor, (a) for a predetermined time period, the test target obtaining time series data for a plurality of different characteristics from (b) obtaining a health curve for the predetermined time of the test subject from the time series data obtained in step (a); And, (c) comparing the soundness curve obtained in step (b) with one or more pre-stored HI curves to evaluate the remaining useful life of the test subject is provided.

As another aspect of the present invention, there is provided an apparatus for evaluating the soundness of an epoxy injector forming an epoxy sealing part of an electronic product display module using the above-described remaining life prediction apparatus.

As another aspect of the present invention, as a computer program for predicting the remaining useful life, stored in a non-transitory storage medium, by a processor, (a) for a predetermined time, time series for a plurality of different characteristics from the test subject acquiring data; (b) obtaining a health curve for the predetermined time of the test subject from the time series data obtained in step (a); And, (c) comparing the soundness curve obtained in step (b) with one or more pre-stored HI curves to evaluate the remaining useful life of the test subject A computer program for predicting the remaining useful life to be executed is provided

According to the present invention, a method for estimating the remaining useful life without based on assumptions related to deterioration of a device under inspection is provided.

Further, according to the present invention, there is provided a method that is robust against noise and can predict the remaining useful life even with measured values.

1 is a view schematically showing a display module of an electronic product to be inspected;
Figure 2 (a) is a view showing a case in which the state of the epoxy sealing portion of the display module shown in Figure 1 is not good.
Figure 2 (b) is a view showing another case in which the state of the epoxy sealing portion of the display module shown in Figure 1 is not good.
3 is a view showing an example of an inspection log for the display module shown in FIG.
4 is a diagram schematically illustrating a structure for performing a remaining useful life prediction method according to the present invention.
Fig. 5 shows a typical RNN encoder-decoder.
6 is a diagram illustrating an input of an RNN encoder-decoder and a signal restored thereby.

Hereinafter, an embodiment according to the present invention will be described in detail with reference to the accompanying drawings. The same or similar reference numerals are assigned to the same or similar components, and overlapping descriptions thereof are omitted. In describing the embodiments disclosed in the present specification, if it is determined that detailed descriptions of related known technologies may obscure the gist of the embodiments disclosed in this specification, the detailed description thereof will be omitted. The accompanying drawings are only for easy understanding of the embodiments disclosed in the present specification, and the technical spirit disclosed in the present specification is not limited by the accompanying drawings, and all changes and equivalents included in the spirit and scope of the present invention It should be understood to include what is or replaces.

Terms including an ordinal number, such as first, second, etc., may be used to describe various elements, but these terms are used only for the purpose of distinguishing one element from other elements, and the corresponding elements are defined by these terms. not limited by The singular expression includes the plural expression unless the context clearly dictates otherwise.

As used herein, terms such as “comprises”, “comprises” or “have” are to be understood as limiting the existence of a feature, step, element, or combination thereof described in the specification, and one or more other It is not intended to exclude the possibility that features, steps, components, or combinations thereof may be present or added.

When a component is described as being “connected” or “connected” to another component, it may be directly connected or connected to the other component, but it is understood that another component may exist in between. should be On the other hand, when it is mentioned that a certain element is "directly connected" or "directly connected" to another element, it should be understood that there is no other element in the middle.

1 shows a display module 10 of an electronic product as an inspection target for applying the remaining useful life prediction method according to the present invention, and FIGS. 2 ( a ) and 2 ( b ) show the display module of the inspection target A case of failure is shown with respect to the epoxy seal 12 applied along the perimeter.

The illustrated display module 10 is formed by bonding members forming a plurality of layers. In the illustrated embodiment for joining the members, an epoxy resin was used. The epoxy resin joins the members and at the same time seals the space between the members to prevent foreign substances from entering. This epoxy resin is injected along the periphery of the display module by an injector having a syringe shape. The epoxy seal 12 should maintain a substantially uniform width without breaking, and there are minimum and maximum values to be satisfied. Fig. 2 (a) shows a state in which there is a break in the epoxy sealing part, and Fig. 2 (b) shows a state in which the width of the epoxy sealing part becomes less than the minimum value and is determined to be defective.

Whether the injection process is properly performed may be checked by, for example, a camera. In the illustrated embodiment, such an inspection operation is performed by a single camera disposed on the display module. 3 shows a part of a log according to the inspection operation of the camera. As shown, the log contains timestamps including date and time, messages related to the task, and metrics. The measurements represent the width (w) of the epoxy seal, the height of the epoxy seal, and the inter-bubble distance within the epoxy seal, respectively. In general, if the width (w) of the epoxy sealing part is 0.5 or less of the normal value, it is a sign that there is an error in the remaining amount of epoxy resin remaining in the injector. It is a sign that there is On the other hand, it is known that the measurement of the width (w) of the epoxy sealing part is accurate, but the height of the epoxy sealing part and the distance between cells are not accurate.

In the present invention, these measurements are used to predict the remaining useful life of the epoxy injector. 3 schematically illustrates an RNN-based model 100 used for prediction of the remaining useful life.

As described above, the inspection of the epoxy seal is made by three measurements: the width, height, and inter-bubble distance of the epoxy seal. Although these three measurements are obtained by one sensor, that is, a camera, they can be viewed as separate instances because they are values for different physical quantities. Accordingly, there are three instances in this embodiment. These measurements are parsed from the log data shown in FIG. 3 and given as time series data (x). In general, since the time until the end of the remaining useful life of the epoxy injector is long, the size of the time series data (x) is very large. Accordingly, the time series data is divided into a plurality of windows w having a predetermined constant length (dimension).

An embedding having a fixed dimension for the time series data of each window is obtained by the RNN encoder 110 . The RNN encoder 110 is learned in an unsupervised manner using an RNN encoder-decoder (hereinafter referred to as 'RNN-ED'). RNN-ED is trained using time-series data acquired over the entire operating life of the epoxy injector, including both normal and abnormal operations. When a specific window is closed, the embedding of that window is used to determine whether the epoxy injector is good or not. The RNN encoder incorporates important features of the epoxy injector behavior into its embeddings. Thus, it is possible to distinguish the normal operation of the epoxy injector from the operation in the case of deterioration.

Hereinafter, each step of a method for predicting the remaining useful life will be described separately.

Acquisition of embeddings using RNN encoder-decoder

The RNN encoder-decoder is based on sequence-to-sequence (seq2seq) learning. In general, a seq2seq model consists of a pair of multilayer RNNs that are trained together, that is, an encoder RNN and a decoder RNN. Fig. 5 shows the operation of the encoder-decoder pair for time series data {x1, x2, x3}. Given the input time series data, the encoder RNN receives the time series data and calculates the final hidden state z. The final hidden state z is given by concatenating the hidden state vectors of all layers of the encoder. The decoder RNN has the same structure as the encoder. The decoder uses the last hidden state of the encoder as the initial hidden state, and restores the input time series data. The process of the RNN encoder can be thought of as a non-linear mapping that transcribes the input multivariate time series data into a fixed-dimensional vector representation (embedding), while the process of the decoder is another non-linear mapping that transcribes the fixed-dimensional vector representation into multivariate time series data. can be thought of as

In general, in addition to the final hidden state, an additional input is given to the decoder RNN. This additional input is the output from the previous time step of the decoder RNN. However, this additional input is not provided to the decoder RNN in the present invention. By doing this, the final hidden state of the encoder ensures that the decoder holds all the information needed to recover the time series data. This approach for learning robust embeddings is known to be useful for classifying time series data.

6 shows an example of a representative input and output of the RNN-ED. The smoothed reconstructed signal shows that the required pattern has been obtained from the input and the noise has been removed.

Acquisition of HI curves using embeddings

Many of the approaches to determine whether a device is in good health based on data acquired by a sensor use a health index (HI). The trend of HI over time is called the HI curve, and the remaining useful life (RUL) is evaluated by comparing this curve with the trend until the end of the device's lifespan. In general, the remaining useful life is finally determined by comparing a plurality of HI curves until the end of the lifespan with the HI curve to be inspected. The HI indicator generally has a value between 0 and 1, with 0 indicating very poor health and 1 indicating perfectly normal. This section describes a method for evaluating device health from embedding time series data.

Since the RNN encoder extracts important patterns from the input time-series data, the embeddings thus obtained can be used to distinguish between normal and degraded regions in the data. In FIG. 4 , reference numeral 210 denotes a set of embeddings corresponding to time-series data obtained when the device behaves normally (hereinafter, referred to as a 'normal embedding storage unit'). As the device operates, the health deteriorates over time, and embeddings corresponding to the measured time-series data move away from the normal embeddings stored in the normal embedding storage 210 . A health index can be calculated from the difference between the normal embedding and the embedding of the measured time series data. The HI evaluation unit 120 performs this task. An HI curve can be obtained from the health index calculated at each measurement time.

Remaining useful life prediction from HI curve

Like the normal embedding storage 210 , an HI curve storage 220 is provided. In order to predict the remaining useful life, the remaining useful life evaluation unit 130 compares the HI curve calculated as in the previous item with the HI curves stored in the HI curve storage unit 220 . Time delays are taken into account so that comparisons can be made if necessary. The similarity between HI curves may be evaluated by various metrics. Finally, the remaining useful life may be predicted from the HI curve determined to be closest to the measured HI curve among the HI curves stored in the HI curve storage unit 220 to be compared.

The remaining useful life of the device made according to the above-described steps is robust to noise and the accuracy is improved compared to the conventional method. By applying the method according to the present invention to time series data on the width, height, and distance between bubbles of the epoxy sealing part of the display module of the electronic product to be inspected, the remaining useful life of the epoxy injector can be predicted.

10: display module of electronic products
12: epoxy seal
100: Remaining useful life prediction structure
110: RNN encoder
120: HI evaluation unit
130: remaining useful life evaluation unit
210: normal embedding storage
220: HI curve storage unit

Claims (11)

As a method for predicting the remaining useful life of the inspector,
(a) acquiring time series data for a plurality of different characteristics from the inspection target for a predetermined time;
(b) obtaining a health curve for the predetermined time of the test subject from the time series data obtained in step (a); and;
(c) evaluating the remaining useful life of the test subject by comparing the health curve obtained in step (b) with one or more pre-stored health curves
Remaining useful life prediction method comprising a.
The method according to claim 1, wherein the step (b),
(b1) calculating the RNN embedding from the time series data obtained in step (a) by the RNN encoder; and;
(b2) comparing the RNN embedding calculated in step (b1) with one or more previously stored normal embeddings to obtain a health curve for the predetermined time of the test subject
Remaining useful life prediction method comprising a.
The method according to claim 1, wherein the step (c),
Evaluating the remaining useful life of the test subject based on the closest one to the health curve obtained in step (b) among the health curves stored in advance
Remaining useful life prediction method, characterized in that.
A method for evaluating the soundness of an epoxy injector for forming an epoxy sealing part of an electronic product display module using the remaining life prediction method according to claim 1 .
5. The method according to claim 4,
As time series data for evaluating the soundness of the epoxy injector, data on the width of the epoxy seal, the height of the epoxy seal, and the distance between the bubbles are used.
A method for evaluating the soundness of an epoxy injector, characterized in that.
A device for predicting the remaining useful life, comprising:
at least one processor; and
at least one memory storing computer-executable instructions
including,
The computer-executable instructions stored in the at least one memory are executed by the at least one processor,
(a) acquiring time series data for a plurality of different characteristics from the inspection target for a predetermined time;
(b) obtaining a health curve for the predetermined time of the test subject from the time series data obtained in step (a); and;
(c) evaluating the remaining useful life of the test subject by comparing the soundness curve obtained in step (b) with one or more pre-stored HI curves
Remaining useful life prediction device that allows to run.
The method according to claim 6, wherein the step (b),
(b1) calculating the RNN embedding from the time series data obtained in step (a) by the RNN encoder; and;
(b2) comparing the RNN embedding calculated in step (b1) with one or more previously stored normal embeddings to obtain a health curve for the predetermined time of the test subject
Remaining useful life prediction device comprising a.
The method according to claim 1, wherein the step (c),
Evaluating the remaining useful life of the test subject based on the closest one to the health curve obtained in step (b) among the health curves stored in advance
Remaining useful life prediction device, characterized in that.
An apparatus for evaluating the soundness of an epoxy injector for forming an epoxy sealing part of an electronic product display module using the remaining life prediction device according to claim 6 .
10. The method of claim 9,
As time series data for evaluating the soundness of the epoxy injector, data on the width of the epoxy seal, the height of the epoxy seal, and the distance between the bubbles are used.
Device for evaluating the soundness of the epoxy injector, characterized in that.
A computer program for predicting remaining useful life, comprising:
It is stored in a non-transitory storage medium, and by the processor,
(a) acquiring time series data for a plurality of different characteristics from the inspection target for a predetermined time;
(b) obtaining a health curve for the predetermined time of the test subject from the time series data obtained in step (a); and;
(c) evaluating the remaining useful life of the test subject by comparing the soundness curve obtained in step (b) with one or more pre-stored HI curves
A computer program for estimating the remaining useful life that will cause it to run.

Images