Intelligent Judgment of Wrong Wiring Method in Large-Scale

Training of Electrical Testing

JiaHeng Xu

, WeiWei Yang, Na Song, Wei Ren and Xiao Rong

Shandong Electric Power College, No. 500 South Second Ring Road, Jinan, Shandong Province, China

Keywords: Algorithm, Electrical Test, Wiring, Judgment.

Abstract: In the paper, in view of the complex wiring in the large-scale training process of electrical testing, the

trainer needs to frequently judge the wiring right and wrong status quo, and proposes an intelligent

judgment method based on the YOLO algorithm, through the logical binding of the digitized equipment

terminals and the connected wire relationship in the test, relying on the learning of standard wiring, the

automatic identification of the test connection line is realized, which greatly saves labor costs, reduces the

possibility of human error, and improves work efficiency. New ideas have been developed for the training

of new employees to conduct large-scale voltage transformer error tests.

1 INTRODUCTION

Electrical test is an important means for the power

grid company to judge the status of electrical

equipment, good or bad, but also a new employee

must master a class of important skills, in the

training process, the trainer needs to constantly

check whether their wiring is correct or not, due to

the large amount of training, the trainer in the

inspection process will inevitably appear paralyzed,

visual fatigue, resulting in low efficiency, there is

such or that error, which brings greater safety risks

to the training process, for this reason, The author

proposes an automatic determination method for

electrical test wiring based on YOLO algorithm,

which greatly improves the labor efficiency of

wiring inspection, reduces the possibility of false

positive, and thus ensures the safety of the training

process. Below we take the field test of current

transformer as an example to illustrate the

application of this method.

2 TEST METHOD FOR ERROR

OF CURRENT TRANSFORMER

The current transformer error test is using the

comparative method, in order to measure the error of

the current transformer under test, it is necessary to

use a standard current transformer with a high

accuracy level, compared with the test product, in

the rated load and the lower limit load of the two

cases, respectively, the error of the current

transformer under test is measured, and the specific

wiring is shown in Figure 1.

3 DESIGN IDEAS

Based on YOLO V4 target detection, intelligent

barcode recognition, and data intelligent verification

technology, an intelligent judgment system for

misconnection of current transformer field error test

is developed.

3.1 Digital Preprocessing

3.1.1 Digitization of Test Equipment

All the test equipment participating in the test, paste

the equipment asset code, and enter the equipment

information and barcode into the system through the

data acquisition terminal that supports the camera

function (hereinafter referred to as the acquisition

terminal).

Xu, J., Yang, W., Song, N., Ren, W. and Rong, X.

Intelligent Judgment of Wrong Wiring Method in Large-Scale Training of Electrical Testing.

DOI: 10.5220/0011909500003613

In Proceedings of the 2nd International Conference on New Media Development and Modernized Education (NMDME 2022), pages 229-234

ISBN: 978-989-758-630-9

 2023 by SCITEPRESS – Science and Technology Publications, Lda. Under CC license (CC BY-NC-ND 4.0)

229

Figure 1: Wiring diagram of current transformer error test.

Seven devices are used in Figure 1, namely

power supply box, voltage regulator, lifting current

instrument, standard current transformer, current

transformer tested, load cases and transformer

calibrators, using a total of 16 wires, including 12

secondary wires, 3 grounding wires, and 1

high-current wire. A total of 16 pairs of terminals are

involved, of which wire 1K1-1S1 and wire 1K1-K

share a terminal 1K1.

3.1.2 Digitization of Test Equipment

Terminals

Digitize the terminals of the test equipment, adopt

the recording method of pasting a two-dimensional

code, and enter the system

through the acquisition

terminal, hereinafter referred to as the terminal

code, and the terminal code is bound to the test

equipment code, as shown in Figure 2 and Figure

Caliator terminal T

Caliator terminal

Caliator terminal T

earth terminal

Figure 2: Terminal code.

Figure 3: Device terminal code digitization.

3.1.3 Digitization of the Terminals of the

Device Tested

The terminals of the tested transformers are

digitized, and the recording method of sticking

codes is adopted. The terminal codes of the

transformers are bound to the equipment codes of

the transformers and are entered into the system

through the acquisition terminal, as shown in Figure

4 and Figure 5.

Figure 4: Device primary terminal code digitization.

NMDME 2022 - The International Conference on New Media Development and Modernized Education

230

Figure 5: Device secondary terminal code digitization.

3.1.4 Digitization of Test Lines

The two terminals of the test line are coded in rings,

used in pairs, and entered into the system through

the acquisition terminal, as shown in Figure 6,

Figure 7 and Figure 8.

Current line lo

o 1 Current line lo

Current line lo

o 3 Groundin

wire lo

o 1

Figure 6: Test line labeling digitization.

Figure 7: Digitization of the primary line.

Figure 8: Secondary line digitization.

3.2 Terminal Wiring Detection

The target detection method of YOLO v4 is adopted,

the terminal and wiring are set as the target

recognition area, multiple barcodes in each

recognition frame are collected, and the barcode

image is processed by the image processing function

of the collection terminal, and the computer reads

the image in the image file format, and then identify

them after image preprocessing (Redmon 2016,

Bochkovskiy 2020), bind the interconnected

terminal codes and line labels, identify the terminal

codes and single or multiple line labels in the area,

and record them in the temporary database.

In this way, all terminals are photographed and

identified in turn, and all terminal codes and the

corresponding wire labels of the terminals are

identified in turn, and recorded in the comparison

library. Here, YOLOv4 is required to intelligently

distinguish the mutual binding relationship between

different terminals and line labels (Yu 2019).

3.3 Architecture of the Algorithm

For a target detection algorithm, it can usually be

divided into 4 general modules. The YOLOv4

model includes the input end. CSPDarknet53 is used

as the benchmark network BackBone, SPP is used as

the additional module of Neck, PANet is used as the

feature fusion module of Neck, and YOLOv3 is used

as the Head. The network structure analysis diagram

is shown in Figure 9.

Intelligent Judgment of Wrong Wiring Method in Large-Scale Training of Electrical Testing

231

Figure 9: Analysis of YOLOv4 network structure.

The input side refers to the picture representing

the input. This stage usually includes an image

preprocessing stage, that is, scaling the input image

to the input size of the network, and performing

operations such as normalization. In the network

training stage, YOLOv4 uses Mosaic data

enhancement operations to improve the training

speed of the model and the accuracy of the network;

it uses cmBN and SAT self-adversarial training to

improve the generalization performance of the

network;

The benchmark network is usually a network of

classifiers with excellent performance, which is used

to extract some general feature representations.

YOLOv4 uses CSPDarknet53 as the benchmark

network. On the one hand, Concat is used instead of

Add to extract richer features. After the Concat

operation, the size of the feature map will remain

unchanged, and the depth will increase, while the

size and depth will not change after the Add

operation. In this sense, using Concat instead of Add

can extract richer features. On the other hand, the

transition layer (1 * 1conv + 2 * 2pooling) is

introduced to extract features, reduce the amount of

calculation, and improve the speed. Another point is

to fuse the Base layer into two parts to extract richer

features. Divide the Base layer into two parts, and

perform the Concat operation on one part of the

output obtained through a similar residual network

and the other part, and pass the result of the

operation through the Transition Layer.

In fact, CSPNet is based on the idea of Densnet,

copying the feature map of the base layer, and

sending a copy to the next stage through the dense

block, thereby separating the feature map of the

base layer. This can effectively alleviate the

gradient vanishing problem (it is difficult to reverse

the lost signal through a very deep network),

support feature propagation, and encourage the

network to reuse features, thereby reducing the

number of network parameters.

The Mish activation function is used to replace

the original RELU activation function; a Dropblock

block is added to this module to further improve the

generalization ability of the model.

The Neck network is usually located in the

middle of the benchmark network and the head

network, and it can be used to further improve the

diversity and robustness of features. YOLOv4 uses

the SPP module to fuse feature maps of different

scales. The full name of SPP is Spatial Pyramid

Pooling, that is, spatial pyramid pooling. The

purpose of using in YOLOv4 is to increase the

receptive field of the network; using the top-down

FPN feature pyramid and self- The bottom-up PAN

feature pyramid is used to improve the feature

extraction capability of the network. PANet (Path

Aggregation Network) is used instead of FPN for

parameter aggregation to be suitable for target

NMDME 2022 - The International Conference on New Media Development and Modernized Education

232

detection at different levels. The method used in the

PANet paper is Addition, and the YOLOv4

algorithm will The fusion method was changed from

addition to Concatenation.

Head output - Head is used to complete the

output of target detection results. For the detection

head part, YOLOv4 continues to use the detection

head of the YOLOv3 algorithm (Cao 2021). For

different detection algorithms, the number of

branches at the output end varies, usually including a

classification branch and a regression branch.

YOLOv4 uses CIOU_Loss to replace the Smooth L1

Loss function, and uses DIOU_nms to replace the

traditional NMS operation, thereby further

improving the detection accuracy of the algorithm.

All equipment, terminals, and connecting wires

are made with QR codes or barcode digital labels.

OPENCV is combined with cameras to collect the

target area, interpret the QR code information on the

collected photos, and bind the information

accordingly. Log into a temporary database.

The data detection function interprets the photos

detected by YOLOv4 that need to be judged by the

QR code interpretation algorithm, and compares the

interpreted information with the binding relationship

in the previous database to determine whether the

wiring is wrong.

4 DESIGN OF INTELLIGENT

JUDGMENT SYSTEM FOR

TRANSFORMER WIRING

The core of this paper is to realize the intelligent

judgment system of transformer wiring. First,

through the image (video) acquisition equipment,

combined with the YOLOv4 target detection

algorithm, set the target recognition area, and collect

the barcodes of all equipment, terminals and wiring

(Gao 2021); then use the computer to check the

barcodes. Perform identification and analysis to

obtain relevant information and record it in a

temporary database for relational binding to

determine wiring connection rules. After the picture

to be detected is sent to the system for a series of

analysis operations, the actual wiring relationship is

compared with the information in the database to

judge whether the wiring is correct and complete the

intelligent judgment of wiring.

Figure 10: Flow chart of design of intelligent judgment system for transformer wiring.

Specific function realization:(1) Data acquisition

function: use microcomputer Raspberry Pi with

camera as video picture acquisition terminal, use

LINUX system, install OPENCV environment,

implant YOLOv4, barcode recognition, information

comparison and other algorithms. (2) barcode

Information binding: The barcode recognition

algorithm is implanted in the computer, the data is

locally analyzed and processed, the barcode is

parsed, and the corresponding relationship is bound

and recorded in the temporary database. (3) Data

detection function: use YOLOv4 to detect the

pictures that need to be judged, interpret the

barcodes in them, compare the interpreted

information with the database, and get the results

(Wang 2021). The process is shown in Figure 10.

The system uses a PC host as the management

platform host, which is used for information

comparison and equipment management of multiple

acquisition terminals. The test platform software

includes management platform software and

acquisition terminal. The function of the

management platform is to set up and manage

multiple collection terminals, and manage the

comparison data in a unified manner. The

acquisition terminal is embedded with a variety of

artificial intelligence target detection algorithms,

barcode recognition algorithms, and data intelligent

verification algorithms. It has the function of

automatically outputting assessment results, error

prompting, built-in camera and display screen,

which is convenient for handheld detection and

bracket fixed detection. The secondary development

interface is convenient for users to expand

functions.

Intelligent Judgment of Wrong Wiring Method in Large-Scale Training of Electrical Testing

233

5 WIRING INTELLIGENT

JUDGMENT APPLICATION

5.1 Learning About Standard Wiring

Standard wiring: The teacher holds the collection

terminal, identifies and scans the standard wiring

terminal code and line label code, obtains the

corresponding relationship between each wiring port,

and saves the name as the standard wiring method

(Liu 2022).

5.2 Wiring Judgment Detection

Wiring detection: Before the students connect, they

should watch the operation training video. After the

students have completed the wiring, they will hold

the acquisition terminal to perform target detection

and identification on all the test instruments and

transformers in turn, automatically capture the

equipment terminal code and line code, and detect

the unscanned data. The missing terminals will be

prompted until all scanning and verification are

completed, and the corresponding relationship

between each port will be obtained, and it will be

automatically compared with the standard wiring

method of the test item. If it is correct, it will display

"wrong wiring", and display the name of the

wrongly connected terminal, and the system will not

be powered on. This item is changed from manual

inspection by on-site teachers to intelligent

inspection by software system, the specific test

circuit diagram is shown in Figure 11, through the

experiment, the efficiency of the trainer to check the

wiring is improved, and the time is shortened.

Figure 11: Digital verification of the error test terminal of

the current transformer.

6 CONCLUSION

This project is mainly divided into two stages. The

first stage is to collect data through the acquisition

terminal and bind the corresponding relationship;

Wiring is checked. The barcode size, model

parameters, database size and other settings are

combined with the actual scene, adjusted to

appropriate values, accurately detect and identify

the target, record equipment information in time,

and conduct wiring inspection through an intelligent

detection system to improve detection efficiency

and reduce errors. The method proposed in this

paper is suitable for the automatic determination of

wiring of various electrical tests. especially when it

comes to training projects, which can greatly reduce

labor intensity and misjudgment rate, and improve

the labor efficiency of users.

REFERENCES

Bochkovskiy A, Wang C Y, Liao H Y M. YOLOv4:

Optimal Speed and Accuracy of Object Detection[J].

Computer Vision and Pattern Recognition,

2020,17(9): 198-215.

Cao Chunjian, Zang Qiang, Wang Zejia, et al. Improved

YOLOv3 target detection algorithm [J]. Chinese

Science and Technology Papers,

2021,16(11):1195-1201.

Gao Xinbo, Mo Mengjingcheng, Wang Haitao, etc.

Research progress of small target detection [J]. Data

Collection and Processing, 2021, 36(3): 391-417.

Liu Xinyu. Research on image target detection algorithm

based on YOLO deep learning model [J]. Computer

Programming Skills and Maintenance,

2022(7):131-134.

Redmon J , Divvala S , Girshick R , et al. You Only Look

Once: Unified, Real-Time Object Detection[J]. IEEE,

2016:779-788.

Wang Miao, Zhang Longxin. Image fault detection of

train locking plate based on YOLOv4 [J]. Modern

Information Technology, 2021, 5(20): 42-43, 49. DOI:

10.19850/j.cnki.2096-4706.2021 .20.012.

Yu Xiuping, Lv Shuping, Chen Zhitao. Multi-class target

recognition based on YOLO algorithm [J]. Laboratory

Research and Exploration, 2019,38(3):34-36,76.

NMDME 2022 - The International Conference on New Media Development and Modernized Education

234