KR102210159B1 - Wire mismatch detection device and method using convolutional neural network and fault localization based on time-frequency domain reflectometry - Google Patents

Abstract

The present invention relates to a wire mismatch detection apparatus and method using a deep learning-based time-frequency domain reflected wave technique, and in more detail, a deep learning-based time-based time capable of detecting that a wire connected to a connector is incorrectly connected. It relates to a wire mismatch detection apparatus and method using a frequency domain reflected wave technique.

Description

Wire mismatch detection device and method using convolutional neural network and fault localization based on time-frequency domain reflectometry

The present invention relates to a wire mismatch detection apparatus and method using a deep learning-based time-frequency domain reflected wave technique, and more particularly, to a deep learning-based time that can detect that a wire connected to a connector is incorrectly connected. -It relates to a wire mismatch detection apparatus and method using a frequency domain reflected wave technique.

As various electronic parts are increasing in the automobile field, and in various fields such as the maintenance industry and the power industry, a very large number of connectors and matching wires are connected to be used for signal application and fault diagnosis.

Particularly, diagnosis is made periodically to determine whether the connection (cable) between the connector and the wire is properly connected, and a fault point (short circuit, disconnection, etc.) of the wire is usually detected to determine whether there is an abnormality.

Of course, detecting the fault point of the wire is also a very important detection, but more fundamentally, it must be determined whether the connector and the wire are properly connected (determining whether there is a mismatch).

Conventionally, since it was configured to include a small number of connectors and wires connected thereto, it was extremely rare that the connector and the wire were incorrectly connected.

However, as described above, as technology develops in various fields, a large number of connectors and wires connected thereto are configured, so that the necessity to detect whether the connection itself is properly made has emerged.

In detail, if the connector and the wire are incorrectly connected, it seems to operate normally for a short period of time, unless there is a fault point of the wire, but the allowable current for each wire type is different and the wire thickness is also different, so the wire may be damaged over a long period of time. It is natural that damage increases the risk of accidents.

In this regard, in Korean Patent Registration No. 10-1358047 ("Fault learning and discrimination device of electrical equipment using neural networks and ultrasonic signals"), ultrasonic band signals input from electrical equipment are converted into time-amplitude, frequency-amplitude signals. By converting and inputting it into a neural network, learning to determine the type of failure is performed, and a device for determining the presence or absence of a failure of an electrical facility and the type of failure is provided. It is not determined to be wrong.

Domestic Registration Patent No. 10-1358047 (Registration Date 2014.01.27)

The present invention was devised to solve the problems of the prior art as described above, and does not determine the fault point of the wire (breakdown, short circuit, etc.), but detects that the fault point of the wire does not exist but the connection itself is incorrect. By developing a signal design algorithm optimized for the frequency characteristics of each wire to prevent the risk of accidents, a wire mismatch detection device using a deep learning-based time-frequency domain reflected wave technique that can detect whether a wire is inconsistent, and its To provide a way.

A wire mismatch detection apparatus using a deep learning-based time-frequency domain reflected wave technique according to an embodiment of the present invention is an apparatus for detecting a mismatch between at least one interconnected connector and at least one wire, each It is configured to include an optimum signal setting unit 100 for setting a signal suitable for the characteristics of the wire, and an arbitrary waveform signal generator, and receiving signals set according to the characteristics of each wire from the optimum signal setting unit 100, A signal generator 200 that applies a signal set according to characteristics with a specific wire according to the control of, is located at the end point of the wire, and is composed of an oscilloscope, and is applied to a specific wire by the signal generator 200 A reflected signal acquisition unit 300 that acquires a signal reflected from an end point according to one signal, and a waveform conversion unit 400 that converts the reflected signal acquired from the reflected signal acquisition unit 300 into an energy waveform through a Wigner-Bill distribution. ), by applying a preset image processing technique, the image conversion unit 500 converting the energy waveform converted by the waveform conversion unit 400 into an image and the image converted by the image conversion unit 500 are received, It is preferable to include an abnormality determination unit 600 that detects whether a mismatch with a connector of a specific wire is applied to classify an image learned using a preset deep learning technique.

Furthermore, the optimal signal setting unit 100 extracts the frequency attenuation characteristics of each wire using a preset parameter selection algorithm, compares and analyzes the frequency attenuation characteristics of each wire, and selects the frequency section with the greatest difference for each wire. Thus, it is preferable to select a parameter for the selected frequency section and set a signal suitable for the characteristics of each wire.

Further, the abnormality determination unit 600 is a learning unit 610 and the learning unit 610 for learning the converted images for the reflected signal when the connector and the wire are normally connected using a preset deep learning technique. And an image comparison unit 620 configured to receive the image converted by the image conversion unit 500 using the learning result of, classify by group, and detect whether the wire is normally connected or defective according to the classification result. It is desirable.

A wire mismatch detection method using a deep learning-based time-frequency domain reflected wave technique according to an embodiment of the present invention is a method of detecting a mismatch between one or more interconnected connectors and one or more wires, the optimal signal In the setting unit, the optimum signal setting step (S100) of setting a signal suitable for the characteristics of each wire, and in the signal generator, the signal set according to the characteristics of each wire received from the optimum signal setting step (S100) is used. The signal generated by the signal generation step (S200) of controlling the arbitrary waveform signal generator to apply a signal set according to the characteristic with a specific wire, and the signal applied by the signal generation step (S200) using an oscilloscope in the reflected signal acquisition unit According to the reflected signal acquisition step (S300) of acquiring a signal reflected from the end point of a specific wire, the waveform conversion unit converts the reflected signal acquired in the reflected signal acquisition step (S300) into an energy waveform through the Wigner-Bill distribution In the waveform conversion step (S400), in the image conversion unit, an image conversion step (S500) of converting the energy waveform converted in the waveform conversion step (S400) into an image by applying a preset image processing technique and an abnormality determination unit. , An abnormality determination step (S600) of detecting whether a mismatch with a connector of a specific wire is detected according to the learned image classification by applying a preset deep learning technique using the image converted in the image conversion step (S500). It is desirable.

Further, in the optimal signal setting step (S100), the frequency attenuation characteristic of each wire is extracted using a preset parameter selection algorithm, and the frequency section with the greatest difference for each wire is selected by comparing and analyzing it. Thus, it is preferable to select a parameter for the selected frequency section and set a signal suitable for the characteristics of each wire.

Further, the abnormality determination step (S600) is a learning step (S610) and the learning step (S610) of learning the converted images for the reflected signal when the connector and the wire are normally connected using a preset deep learning technique. Using the learning result of, the image converted in the image conversion step (S500) is received, classified by groups, and a comparison step (S620) of detecting whether a wire is normally connected or defective according to the classification result is preferably performed. .

A wire mismatch detection apparatus and method using the deep learning-based time-frequency domain reflected wave technique of the present invention having the above configuration, by determining the mismatch between the connector and the wire itself, can occur over a long period of time There is an advantage that can solve normal defects.

In particular, in the automotive field in which many electronic parts are recently added, the possibility of causing a problem of misconnection is increasing due to the many wirings of the electronic parts that have become complicated. Therefore, there is an advantage in that integrated management and diagnosis are possible.

1 is a block diagram illustrating an apparatus for detecting a wire mismatch using a deep learning-based time-frequency domain reflected wave technique according to an embodiment of the present invention.
2 is a diagram illustrating a configuration of an apparatus for detecting a wire mismatch using a deep learning-based time-frequency domain reflected wave technique according to an embodiment of the present invention.
6 is an exemplary diagram of an image learning process in a wire mismatch detection apparatus using a deep learning-based time-frequency domain reflected wave technique according to an embodiment of the present invention.
7 is a flowchart illustrating a method of detecting a wire mismatch using a deep learning-based time-frequency domain reflected wave technique according to an embodiment of the present invention.

Hereinafter, a wire mismatch detection apparatus and method using a deep learning-based time-frequency domain reflected wave technique of the present invention will be described in detail with reference to the accompanying drawings. The drawings introduced below are provided as examples in order to sufficiently convey the spirit of the present invention to those skilled in the art. Accordingly, the present invention is not limited to the drawings presented below and may be embodied in other forms. In addition, the same reference numbers throughout the specification indicate the same elements.

In this case, unless there are other definitions in the technical terms and scientific terms used, they have the meanings commonly understood by those of ordinary skill in the art to which this invention belongs, and the gist of the present invention in the following description and accompanying drawings A description of known functions and configurations that may unnecessarily obscure will be omitted.

An apparatus for detecting a wire mismatch using a deep learning-based time-frequency domain reflection wave technique according to an embodiment of the present invention, and the method thereof, is an apparatus for detecting whether a wire connected to a connector is mismatched, that is, that a wire is incorrectly connected to the connector And a method.

In detail, since the allowable current for each wire type is different and the wire thickness is different, if it is incorrectly connected to the connector, it will operate normally in a short period, but the wire is damaged over a long period of time, further increasing the risk of an accident.

Therefore, in the wire mismatch detection apparatus and method using a deep learning-based time-frequency domain reflected wave technique according to an embodiment of the present invention, it is possible to determine whether a connector and a wire are normally connected.

Through this, there is an advantage that it can be applied as a solution for integrated management and diagnosis of problems that may occur due to a large number of wirings of a complex electric system in the automotive industry that becomes autonomous driving and electric vehicles in the future.

1 is a configuration diagram showing a wire mismatch detection apparatus using a deep learning-based time-frequency domain reflected wave technique according to an embodiment of the present invention, and FIG. 2 is a deep learning-based time-frequency according to an embodiment of the present invention. It is an exemplary configuration of a wire mismatch detection apparatus using a region reflection wave technique. An apparatus for detecting a wire mismatch using a deep learning-based time-frequency domain reflected wave technique according to an embodiment of the present invention will be described in detail with reference to FIGS. 1 and 2.

The wire mismatch detection apparatus using a deep learning-based time-frequency domain reflected wave technique according to an embodiment of the present invention includes an optimal signal setting unit 100, a signal generation unit 200, and a reflected signal as shown in FIG. It is preferable to include an acquisition unit 300, a waveform conversion unit 400, an image conversion unit 500, and an abnormality determination unit 600.

A wire mismatch detection apparatus using a deep learning-based time-frequency domain reflected wave technique according to an embodiment of the present invention is an apparatus for detecting a mismatch between at least one interconnected connector and at least one wire, wherein the signal generator It is preferable that 200 is connected to one end of the wire, and the reflected signal acquisition unit 300 is configured to be located at the end point of the wire.

To learn more about each configuration,

It is preferable that the optimum signal setting unit 100 sets a signal suitable for the characteristics of each wire.

In detail, the optimal signal setting unit 100 may extract a frequency attenuation characteristic of each wire using a preset parameter selection algorithm.

By comparing and analyzing the extracted frequency attenuation characteristics for a plurality of wires, a frequency section with the greatest difference is selected for each wire.

By selecting a parameter for a frequency section selected for each wire, a signal suitable for the characteristics of each wire can be set through this.

In this case, the frequency attenuation characteristic of each wire is preferably extracted through a network analyzer provided in advance.

For the operation of the optimal signal setting unit 100, for example,

Assuming that there are four types of wires, the'magnitude' and'phase' parameters as shown in FIG. 3 can be extracted through a network analyzer equipped with each wire in advance. When a signal is applied to one side of the wire, the parameters are indices indicating whether the channel characteristics of the wire are different, that is, the magnitude and phase of the signal through the wire are different through the signal from the other side.

Through this, by analyzing the extracted parameters of each wire, the optimal signal setting unit 100 may select (set) a section in which characteristics are most different, that is, a frequency section with the greatest difference.

Through this, a signal suitable for the characteristics of each wire can be set, and the parameter of the signal suitable for the characteristic is most preferably set through control of the center frequency, frequency band, and time width.

In detail, the'magnitude' parameter extracted through the network analyzer is a variable that indicates the degree of attenuation of the signal, and since the characteristics of each type of wire are remarkably different as the frequency increases, a good frequency band to distinguish the type of wire However, on the contrary, since attenuation of the signal becomes severe, there is a problem that it is difficult to detect when a reflected signal is acquired. Accordingly, in the optimal signal setting unit 100, by fixing the frequency band and the time width to arbitrary values and then increasing the center frequency, as shown in FIG. 4A, the applied signal and the reflected signal are It is most desirable to increase the center frequency by analyzing the similarity (TFCC value, Time-Frequency Cross Correlation value), and to select the largest center frequency among values with a TFCC value of 0.3 or more.

In this case, the TFCC value is an index indicating similarity, and it is generally determined that the similarity of the signal decreases when the value goes down to 0.3 or less, but this is also only an example.

In this way, after selecting the center frequency, it is preferable to change the frequency band after fixing the time width to an arbitrary value with the center frequency set to the selected value.

At this time, the maximum value of the frequency band is twice the center frequency, and the width of the TFCC curve is proportional to the v.o.p (Velocity Of Propagation, meaning the propagation speed of the signal)/frequency band. That is, as the frequency band increases, the width of the TFCC curve decreases, so the resolution that can be distinguished is improved. The peak value of the TFCC indicates the location of the signal similar to the applied signal. In order to select the maximum frequency band, as shown in b) of FIG. 4, it is most preferable to select the frequency band with the maximum value excluding the dc component that does not exceed twice the center frequency and is 0 Hz.

Finally, after selecting the center frequency and frequency band, it is desirable to select the time width. According to the uncertainty principle theorem, the time width should be designed so that the product of the time width and the frequency band exceeds 1/2. . Based on this, the time width can be designed through the selected frequency band. However, when the applied signal and the reflected signal overlap more than half, it is difficult to distinguish between the applied signal and the reflected signal, so that a blind spot problem occurs. Accordingly, as shown in c) of FIG. 4, in order to design a signal capable of distinguishing the starting point and the end point of each wire, it is most preferable to design the wire to satisfy the length> v.o.p * time width/2. That is, it is most preferable to calculate the maximum and minimum values of the time width and select the time width.

The signal generator 200 is preferably configured to include an arbitrary waveform generator, and a signal suitable for the characteristics (frequency characteristics, etc.) of each wire received from the optimal signal setting unit 100 By using, a signal is generated according to the characteristics of the wire and applied to the wire. That is, the target wire generates and applies a signal set according to the characteristics of the wire.

In other words, the signal generation unit 200 receives a signal optimized for the frequency characteristics of each wire from the optimum signal setting unit 100, and is defined in the time-frequency domain according to the corresponding signal. A Gaussian-enveloped linear chirp signal can be generated and applied with a wire.

At this time, it is desirable to operate the application to the wire according to the control of the outside (administrator, etc.), and a signal set according to the characteristics of the corresponding wire by selecting a desired wire (specific wire) among a number of wires according to external control. Is applied.

As described above, the reflected signal acquisition unit 300 includes an oscilloscope (digital phosphor oscilloscope) located at the end point of the wire, through which, as shown in FIG. 5, the signal generation unit 200 It is preferable to acquire (collect) the reflected signal generated at the impedance discontinuity point according to the applied signal, and the waveform conversion unit 400 is based on the length of the incident signal of the reflected signal acquired by the reflected signal acquisition unit 300. It is preferable to extract a section of the obtained signal and convert it into a Wigner-Ville distribution to convert it into an energy waveform, and the image conversion unit 500 includes a preset image processing technique (commercial image processing technique). By applying, it is preferable to convert the energy waveform converted by the waveform conversion unit 400 into an image.

The abnormality determination unit 600 receives the image converted by the image conversion unit 500 and applies it to the learned image classification using a preset deep learning technique to detect whether there is a mismatch with a connector of a specific wire. I can.

In detail, it is preferable that the abnormality determination unit 600 includes a learning unit 610 and an image comparison unit 620 as shown in FIG. 1.

It is preferable that the learning unit 610 converts the reflected signal when normally connected between the connector and the wire into an energy waveform using a preset deep learning technique, and then learns the converted images.

That is, as shown in FIG. 6, an image that is easily classified through an image pre-processing process is input to a convolutional neural network (CNN) and trained, so that the corresponding image (input image) is A learning model (a learning model group) can be created as a group (a wire group classified in advance according to frequency characteristics).

In other words, the reflected signal acquired by the reflected signal acquisition unit 300 through the waveform conversion unit 400 and the image conversion unit 500 is subjected to an image pre-processing process, and the learning unit 610 performs convolution. It is preferable to generate a learning model by inputting into the tional neural network and performing training in advance.

The image comparison unit 620 selects a newly entered input image, that is, a desired wire (a specific wire) and applies a signal set according to the characteristics of the corresponding wire, thereby performing an image pre-processing process on the acquired reflected signal. It is preferable to apply is applied to the image group (learning model) previously learned by the learning unit 610, and determine that the wire is incorrectly connected if it is not the previously learned image group.

Specifically, the wire mismatch detection apparatus using the deep learning-based time-frequency domain reflected wave technique according to an embodiment of the present invention, through the optimal signal setting unit 100, the frequency of each wire with a network analyzer. By measuring the attenuation characteristics, it is possible to design a correction filter for each wire according to the measured frequency attenuation characteristics.

At this time, the correction filter is to set the optimum signal by setting the center frequency, the frequency band, and the time width as described above.

Thereafter, through the signal generation unit 200, a signal optimized for the frequency characteristics of each wire set by the optimum signal setting unit 100 is applied to the target wire (target wire).

After that, it is preferable to collect the signal reflected from the end point through the reflected signal acquisition unit 300 and analyze the similarity between the applied signal and the reflected signal (TFCC value) to calculate the impedance discontinuities distance. .

At this time, it is preferable to compare the difference between the impedance discontinuity distance and the length of the target wire by an absolute value, limiting it to a case that is smaller than a preset threshold, and to perform an operation to be described later, and if it is greater than a preset threshold, It is desirable to judge that the connection itself is not faulty, but that a fault exists.

In this case, the threshold value η1 is usually set to 0.3 and used, but this is determined through a number of experiments and is not limited to 0.3.

Through this, it is determined that less than 0.3 is noise, and if there is a value greater than 0.3, it is determined that there is a signal similar to the applied signal.

Thereafter, the reference signal collected at the end point of the target wire through the waveform conversion unit 400 is converted into an energy waveform through the Wigner-Bill distribution, and then converted into an image through the image conversion unit 500. .

Thereafter, the images converted by the image conversion unit 500 are learned using a convolutional neural network through the learning unit 610, and an input image and a learning group newly entered through the image comparison unit 620 By comparison, when there is a matching learning group, it is determined that the connector and the wire are normally connected, and when there is no matching learning group, it can be determined that the connector and the wire are abnormally connected.

7 is a flowchart illustrating a method of detecting a wire mismatch using a deep learning-based time-frequency domain reflected wave technique according to an embodiment of the present invention. A method of detecting a wire miss match using a deep learning-based time-frequency domain reflected wave technique according to an embodiment of the present invention will be described in detail with reference to FIG. 7.

A method for detecting a wire mismatch using a deep learning-based time-frequency domain reflected wave technique according to an embodiment of the present invention is as shown in FIG. 7, an optimal signal setting step (S100), a signal generating step (S200), and a reflected signal. It is preferable that it consists of an acquisition step (S300), a waveform conversion step (S400), an image conversion step (S500), and an abnormality determination step (S600).

To learn more about each step,

In the optimum signal setting step (S100), it is preferable that the optimum signal setting unit 100 sets a signal suitable for the characteristics of each wire.

In detail, a frequency attenuation characteristic of each wire may be extracted using a preset parameter selection algorithm.

By comparing and analyzing the extracted frequency attenuation characteristics for a plurality of wires, a frequency section with the greatest difference is selected for each wire.

By selecting a parameter for a frequency section selected for each wire, a signal suitable for the characteristics of each wire can be set through this.

In this case, the frequency attenuation characteristic of each wire is preferably extracted through a network analyzer provided in advance.

Specifically, the optimal signal setting step (S100), assuming that there are four types of wires, through a network analyzer equipped with each wire in advance, the'magnitude' and'phase' parameters as shown in FIG. Can be extracted. When a signal is applied to one side of the wire, the parameters are indices indicating whether the channel characteristics of the wire are different, that is, the magnitude and phase of the signal through the wire are different through the signal from the other side.

Through this, the optimal signal setting step (S100) may select (set) a section in which the characteristics are most different, that is, a frequency section with the greatest difference, by analyzing the extracted parameters of each wire.

Through this, a signal suitable for the characteristics of each wire can be set, and the parameter of the signal suitable for the characteristic is most preferably set through control of the center frequency, frequency band, and time width.

In detail, the'magnitude' parameter extracted through the network analyzer is a variable that indicates the degree of attenuation of the signal, and since the characteristics of each type of wire are remarkably different as the frequency increases, a good frequency band to distinguish the type of wire. However, on the contrary, since the attenuation of the signal becomes severe, there is a problem that it is difficult to detect when a reflected signal is acquired. Accordingly, in the optimal signal setting unit 100, by fixing the frequency band and the time width to arbitrary values and then increasing the center frequency, as shown in FIG. 4A, the applied signal and the reflected signal are It is most desirable to increase the center frequency by analyzing the similarity (TFCC value, Time-Frequency Cross Correlation value), and to select the largest center frequency among values with a TFCC value of 0.3 or more.

In this case, the TFCC value is an index indicating similarity, and it is generally determined that the similarity of the signal decreases when the value goes down to 0.3 or less, but this is also only an example.

In this way, after selecting the center frequency, it is preferable to change the frequency band after fixing the time width to an arbitrary value with the center frequency set to the selected value.

At this time, the maximum value of the frequency band is twice the center frequency, and the width of the TFCC curve is proportional to the v.o.p (Velocity Of Propagation, meaning the propagation speed of the signal)/frequency band. That is, as the frequency band increases, the width of the TFCC curve decreases, so the resolution that can be distinguished is improved. The peak value of the TFCC indicates the location of the signal similar to the applied signal. In order to select the maximum frequency band, as shown in b) of FIG. 4, it is most preferable to select the frequency band with the maximum value excluding the dc component that does not exceed twice the center frequency and is 0 Hz.

Finally, after selecting the center frequency and frequency band, it is desirable to select the time width. According to the uncertainty principle theorem, the time width should be designed so that the product of the time width and the frequency band exceeds 1/2. . Based on this, the time width can be designed through the selected frequency band. However, when the applied signal and the reflected signal overlap more than half, it is difficult to distinguish between the applied signal and the reflected signal, and thus a blind spot problem occurs. Accordingly, as shown in c) of FIG. 4, in order to design a signal capable of distinguishing the starting point and the end point of each wire, it is most preferable to design the wire to satisfy the length> v.o.p * time width/2. That is, it is most preferable to calculate the maximum and minimum values of the time width and select the time width.

In the signal generation step (S200), by using a signal suitable for the characteristics of each wire received from the optimal signal setting step (S100) using an arbitrary waveform signal generator, the characteristics of the wire It generates a signal according to and applies it with a wire.

In other words, it is possible to receive a signal suitable for the characteristics of each wire, generate a Gau time envelope linear chirp signal defined in the time-frequency domain according to the corresponding signal, and apply it to the wire.

At this time, it is desirable to operate the application to the wire according to the control of the outside (administrator, etc.), and a signal set according to the characteristics of the corresponding wire by selecting a desired wire (specific wire) among a number of wires according to external control. Is applied.

In the reflected signal acquisition step (S300), it is preferable that the reflected signal acquisition unit 300 acquires a reflected signal according to the signal applied by the signal generation step (S200) using an oscilloscope.

In the waveform conversion step (S400), it is preferable that the waveform conversion unit 400 converts the reflected signal obtained in the reflected signal acquisition step (S300) into an energy waveform through a Wigner-Bill distribution.

In the image conversion step (S500), it is preferable that the image conversion unit 500 converts the energy waveform converted in the waveform conversion step (S400) into an image by applying a preset image processing technique.

In the abnormality determination step (S600), a connector of a specific wire is applied to the learned image classification by using the image converted in the image conversion step (S500) by the abnormality determination unit (600) using a preset deep learning technique. It is possible to detect whether there is a mismatch with and.

In detail, it is preferable that the abnormality determination step (S600) further includes a learning step (S610) and a comparison step (S620).

In the learning step (S610), it is preferable to use a preset deep learning technique to convert the reflected signal when the connector and the wire are normally connected to an energy waveform, and then learn the converted images.

That is, as shown in FIG. 6, an image that is easily classified through an image pre-processing process is input into a convolutional neural network to be trained.

In other words, through the waveform conversion step (S400) and the image conversion step (S500), the reflected signal acquired in the reflected signal acquisition step (S300) is subjected to an image preprocessing process, and through the learning step (S610) It is preferable to generate a learning model by inputting into the tional neural network and performing training in advance.

In the comparison step (S620), by selecting a new input image, that is, a desired wire (a specific wire) and applying a signal set according to the characteristics of the corresponding wire, the acquired reflected signal is subjected to an image pre-processing process. It is preferable to apply to an image group (learning model) that has been learned in advance in the learning step (S610), and determine that it is a wire connected incorrectly if it is not a pre-trained image group.

That is, in other words, the wire mismatch detection method using a deep learning-based time-frequency domain reflected wave technique according to an embodiment of the present invention measures the frequency attenuation characteristics of each wire with a network analyzer, and the measured frequency attenuation It is possible to design a correction filter for each wire according to the characteristics.

At this time, the correction filter is to set the optimum signal by setting the center frequency, the frequency band, and the time width as described above.

After that, a signal optimized for the frequency characteristics of each set wire is applied to the target wire (target wire).

After that, it is preferable to collect the reflected signal at the end point, analyze the similarity between the applied signal and the reflected signal (TFCC value), and calculate the impedance discontinuities distance.

At this time, it is preferable to compare the difference between the impedance discontinuity distance and the length of the target wire by an absolute value, limiting it to a case that is smaller than a preset threshold, and to perform an operation to be described later. It is desirable to judge that the connection itself is not faulty, but that a fault exists.

Thereafter, the reference signal collected at the end point of the target wire is converted into an energy waveform through the Wigner-Ville distribution, and then converted into an image.

After that, the transformed images are learned using a convolutional neural network, the newly entered input image and the learning group are compared, and if there is a matching learning group, it is determined that the connector and the wire are normally connected, and the matching If there is no learning group to do, it can be determined that the connector and wire are connected abnormally.

As described above, in the present invention, specific matters such as specific components, etc. and limited embodiments have been described, but this is provided only to aid in a more general understanding of the present invention, and the present invention is limited to the above-described embodiment. It is not, and those of ordinary skill in the field to which the present invention belongs can make various modifications and variations from this description.

Therefore, the spirit of the present invention is limited to the described embodiments and should not be determined, and all things equivalent or equivalent to the claims as well as the claims to be described later belong to the scope of the spirit of the present invention. .

100: optimal signal setting unit
200: signal generator
300: reflected signal acquisition unit
400: waveform conversion unit
500: image conversion unit
600: abnormality determination unit

Claims (6)

An apparatus for detecting a mismatch between one or more interconnected connectors and one or more wires,
An optimum signal setting unit 100 for setting a frequency signal section suitable for a frequency attenuation characteristic of each wire;
It is configured to include an arbitrary waveform signal generator, receives a signal set according to the frequency attenuation characteristics of each wire from the optimal signal setting unit 100, and receives the signal set according to the characteristics corresponding to the specific wire according to external control- A signal generator 200 for generating and applying a Gaussian envelope linear chirp signal defined in a frequency domain;
A reflected signal acquiring unit 300 located at an end point of the wire, configured to include an oscilloscope, and acquiring a signal reflected from the end point according to a signal applied from the signal generator 200 to a specific wire;
A waveform conversion unit 400 for converting the reflected signal acquired by the reflected signal acquisition unit 300 into an energy waveform through a Wigner-Bill distribution;
An image conversion unit 500 for converting an energy waveform of the reflected signal converted by the waveform conversion unit 400 into an image by applying a preset image processing technique; And
An abnormality determination unit 600 that receives the image converted by the image conversion unit 500 and applies it to the learned image classification using a preset deep learning technique to detect whether there is a mismatch with a connector of a specific wire;
Consists of including,
The abnormality determination unit 600
Using a preset deep learning technique, a signal set according to the frequency attenuation characteristic of the wire is transmitted from the normally connected connector and wire, and then a Gaussian envelope linear chirp signal defined in the time-frequency domain is generated and acquired according to the applied signal. A learning unit 610 that learns transformed images for the energy waveform generated through the reflected signal and generates a learning model into image groups classified according to a set frequency attenuation characteristic; And,
The image conversion unit 500 receives the converted image for the energy waveform of the reflected signal, applies it to the learning model generated by the learning unit 610, and classifies the group according to the set frequency attenuation characteristics, An image comparison unit 620 for detecting a mismatch between the connector and the wire;
A wire mismatch detection apparatus using a deep learning-based time-frequency domain reflected wave technique, further comprising a.
The method of claim 1,
The optimal signal setting unit 100 is
Using a preset parameter selection algorithm,
By extracting the frequency attenuation characteristics of each wire, comparing and analyzing them, selecting a frequency section with the greatest difference for each wire, selecting a parameter for the selected frequency section, and setting a signal suitable for the characteristics of each wire Wire mismatch detection apparatus using a deep learning-based time-frequency domain reflected wave technique, characterized in that.
delete In a method for detecting a mismatch between one or more interconnected connectors and one or more wires,
In the optimum signal setting unit, the optimum signal setting step (S100) of setting a frequency signal section suitable for the frequency attenuation characteristic of each wire;
In the signal generator, the arbitrary waveform signal generator is controlled by using the signal set according to the frequency attenuation characteristics of each wire received from the optimal signal setting step (S100), and the time-frequency domain set according to the characteristics corresponding to the specific wire A signal generation step of generating and applying a Gaussian envelope linear chirp signal defined in S200;
In the reflected signal acquisition unit, a reflected signal acquisition step (S300) of acquiring a signal reflected from an end point of a specific wire according to the signal applied by the signal generation step (S200) using an oscilloscope;
In the waveform conversion unit, a waveform conversion step (S400) of converting the reflected signal obtained in the reflected signal acquisition step (S300) into an energy waveform through a Wigner-Bill distribution;
An image conversion step (S500) of converting an energy waveform of the reflected signal converted in the waveform conversion step (S400) into an image by applying a preset image processing technique; And
In the abnormality determination unit, an abnormality determination step of detecting whether a mismatch with a connector of a specific wire is detected according to the learned image classification by applying a preset deep learning technique using the image converted in the image conversion step (S500) ( S600);
Consists of
The abnormality determination step (S600)
Using a preset deep learning technique, a signal set according to the frequency attenuation characteristic of the wire is transmitted from the normally connected connector and wire, and then a Gaussian envelope linear chirp signal defined in the time-frequency domain is generated and acquired according to the applied signal. A learning step (S610) of learning images of the energy waveform generated through the reflected signal and generating a learning model into an image group classified according to a set frequency attenuation characteristic; And,
In the image conversion step (S500), the converted image for the energy waveform of the reflected signal is received, applied to the learning model generated in the learning step (S610), and according to the result of classification into groups suitable for a set frequency attenuation characteristic, A comparison step (S620) of detecting whether there is a mismatch between the connector and the wire;
A wire miss match detection method using a deep learning-based time-frequency domain reflected wave technique further comprising.

The method of claim 4,
The optimal signal setting step (S100)
Using a preset parameter selection algorithm,
By extracting the frequency attenuation characteristics of each wire, comparing and analyzing them, selecting a frequency section with the greatest difference for each wire, selecting a parameter for the selected frequency section, and setting a signal suitable for the characteristics of each wire A wire mismatch detection method using a deep learning-based time-frequency domain reflected wave technique, characterized in that. delete

Images