Academic Journal of Science and Technology 
ISSN: 2771-3032 | Vol. 9, No. 2, 2024 

98 

Research	on	Misalignment	Detection	of	Paper	Machine	
Clothing	Based	on	Computer	Vision	
Lianhua Hu, Tian Wang*

College of Mechanical and Electrical Engineering, Shaanxi University of Science and Technology, Xi’an 710021, China 
* Corresponding author: Tian Wang (Email: 210511048@sust.edu.cn)

Abstract: Paper machine clothing (PMC) is a specialized material essential to the paper industry, forming an integral part of 
paper-making equipment. Deviations in the net carpet, in particular, can lead to pleats along the edge or in the middle of the net 
carpet, making it challenging to spread and often resulting in broken pleats. These pleats contribute to increased paper-making 
costs. To address this issue of paper machine clothing misalignment, this study proposes a monitoring approach based on 
computer vision. Initially, image filtering and edge enhancement are applied to the video images obtained from the complex 
manufacturing environment of a paper mill. The Canny operator is then utilized for edge detection to obtain a binary image of 
the edges. Subsequently, Hough transform is employed to detect the edges of the net blanket based on the straight-line 
characteristics of paper machine clothing. Additionally, regions of interest (ROI) are established to enhance processing speed 
and reliability. If broken straight line segments are observed at the edge of the mesh blanket, these are combined using least 
squares fitting to form a complete straight line. Finally, deviation in the paper machine clothing is monitored by analyzing the 
slope value of the fitted edge straight line and its distance from the x-axis of the outer roller edge. Test results demonstrate that 
using the method proposed in this study, it is possible to determine both the angle of net blanket deviation and its distance from 
the rollers' edges. Through validation using laboratory and mill data, the algorithm presented in this study effectively reflects 
these parameters. 

Keywords: Paper machine clothing, Computer vision, Misalignment detection, Hough linear detection. 

1. Introduction
The paper machine clothing serves a crucial function in 

supporting and transmitting pulp, paper blanks, and paper 
sheets, while also playing an important role in the forming, 
pressing, dewatering, and drying of paper sheets. With the 
acceleration of automation in the paper-making industry and 
the increase in the speed of paper machines, a series of issues 
affecting the normal operation of equipment have emerged, 
with misalignment of the paper machine clothing being a 
notable example[1, 2] .These cloths can be categorized into 
forming net, press felt, and drying net based on their location 
within the paper-making machine [3]. As the paper net 
blanket is in direct contact with the paper surface throughout 
the papermaking process, its condition has a direct impact on 
the quality of the paper products, as well as on the 
productivity and production costs of the papermaking 
machine [4, 5]. 

The causes of paper machine clothing misalignment are 
complex, the difference in tension on the transverse direction 
of the web, the rolls are not fully centered, the difference in 
the parallelism of each roll, the difference in roundness, all of 
which can cause the web to run out of shape. The 
misalignment of the net carpet will cause head breakage or 
paper diseases such as embossing and crimping, and when the 
misalignment is serious, it will tear the net carpet, affecting 
the normal operation of the paper machine, and causing great 
losses to the production of the production plant's output and 
quality of paper copying. 

Existing mesh blanket calibrators can be divided into 
constant contact, intermittent contact and non-contact 
according to the detection equipment, and non-contact 
calibrators can be divided into pneumatic and electric 
calibrators according to the actuators[6]. 

The contact calibrator installs a detecting device, the touch 
palm, on one side of the net blanket, and uses the direct 
contact between the touch palm and the edge of the net 
blanket to follow the movement of the net blanket to detect 
the change of air pressure, and changes the position of the 
guide rollers by adjusting the air bladder to keep the net 
blanket running in a certain position. The characteristics of 
the palm of the material requirements are relatively high, and 
wear and tear on the edge of the net carpet; intermittent 
contact calibrator on both sides of a palm to drive a 
mechanical-pneumatic reversing valve action, to achieve the 
air circuit switching, will be compressed air to the 
corresponding air spring, to achieve the automatic correction 
of the net carpet, the edge of the net carpet wear and tear is 
relatively small. In addition, the contact calibration system of 
the airbag, spring, air source failure caused by dislocation 
misalignment and other reasons is gradually eliminated, and 
non-contact calibration device can effectively overcome these 
problems gradually promote the application. Among them, 
the non-contact pneumatic calibrator consists of infrared 
detector, solenoid valve and so on [7]. Through the infrared 
detector will signal feedback to the solenoid valve, so as to 
control the actuator movement, but the infrared detector 
performance sometimes detects the signal is not stable. Non-
contact electric correction device, two sets of optical 
measuring heads is installed on one side of the net blanket to 
detect the position of the net blanket, and the controller 
controls the electric correction device through the frequency 
converter. No damage to the edge of the net blanket, high 
adjustment accuracy, but the investment is relatively large. 

Most of the above correction methods require the 
installation of various types of sensors on the paper machine, 
for the paper machine clothing in the copying process of the 
misalignment problem, the use of image processing methods 



 

99 

can be more intuitive to solve. At present, currently, closed-
circuit television monitoring systems (CCTV) act as an 
intuitive visual monitoring tool for industrial production, 
benefiting from a broad application scope, mature technology, 
and relatively low costs [8]. At the same time, the level of 
development of computer technology, high-speed digital 
image acquisition and processing technology have made it 
possible to monitor the misalignment of paper machine 
clothing through computer vision technology, but the 
misalignment recognition algorithm is the detection of the 
research difficulties. 

In recent years, a lot of research has been carried out on 
paper machine clothing, mainly focusing on the development 
of a new type of net carpet design, felt chemical cleaning and 
net carpet dewatering laboratory model, to prevent dehairing, 
and to improve the performance of the net carpet itself, etc. 
[9] [10] [11] [12]. The research on the monitoring of the 
overall operating condition of the felt in the actual working 
environment is very limited, which in turn greatly restricts the 
development of modern large-scale high-speed paper 
machines. 

This paper aims at study a high-efficiency method for 
detecting the misalignment of the paper machine clothing, 
and at the same time, It can detect abnormal situations with 
the net blanket in a timely and accurate manner and send out 
alarm signals accordingly., which can effectively prevent the 
occurrence of paper disease, and provide a theoretical basis 
for the design and development of the net blanket condition 
monitoring system with practical value [13]. 

2. Paper Machine Clothing 
Misalignment Detection Method 

The flow of the misalignment detection method for paper 
machine clothing proposed in this paper is shown in Fig. 1. 
Firstly, preprocess the collected images in the computer, 
including grayscale processing, noise removal, and enhancing 
image features. Then the canny operator is used to extract the 
edge features of the mesh carpet, filtering the useless 
information and retaining the important edge structure 
features. And then through the Region of Interest (ROI) 
selection the part of the image that needs to be detected in the 
mesh carpet edge line is extracted, while the other parts of the 
image are not further processed in order to reduce the 
processing of redundant image data. The Hough linear 
transformation algorithm is utilized, with an angular filter 
employed to eliminate interfering straight lines. Subsequently, 
the least squares method is utilized to fit the discontinuous 
mesh carpet edge straight lines. This approach facilitates the 
accurate extraction of the positional coordinates of the mesh 
carpet edge straight lines [14]. Finally, through the angle of 
inclination of the paper machine clothing and the distance 
between the two sides of the stick according to the net carpet, 
comprehensively determine whether the net carpet has 
misalignment faults and fault level, and carry out the 
corresponding fault alarm. 

2.1. Image acquisition 
The acquisition hardware of the computer vision-based 

paper machine clothing misalignment detection system 
includes, light source, industrial camera, lens, computer, data 
transmission equipment, etc. [15]. The structure of the image 
acquisition system is shown in Fig. 2, which adopts the LED 
ring lighting technology and image acquisition on the surface 

of the paper machine clothing through the CMOS surface 
array industrial camera, and then the image data is transmitted 
to the PC through the gigabit network cable, and the PC 
realizes the misalignment detection according to the image 
processing technology. 

 

 

Figure 1. Flowchart of paper machine clothing deviation 
detection 

 

 
Figure 2. Schematic structure of image acquisition system 

 



 

100 

  
(a) Original image (b) Pre-processing(c) Edge feature extraction (d) 
ROI mask area (inside the green quadrilateral)(e) Region of interest 
(ROI) (f) Hough linear transformation (g) Edge detection of mesh 
blanket (h) Graph of deviation detection results 

Figure 3. Net blanket image processing results 

2.2. Image Preprocessing  

(Figure 2a) is an original image of the paper machine clothing 
collected. Through its analysis, it is known that the original 
image contains more irrelevant information from other 
equipment, there are a lot of dust and water vapor in the 
working environment of the field, and there is uneven 
brightness. In order to eliminate irrelevant information and 
suppress noise in the image, it is necessary to preprocess the 
original image. This paper proposes a simple and efficient 
image preprocessing algorithm, including grayscale 
processing, image enhancement, Gaussian blur, and erosion 
and expansion operations [16]. The preprocessing results are 
shown in (Fig. 2b). 

2.3. Edge Feature Extraction 
In digital image processing, the operators that can be used 

to perform edge detection are usually Robert, Log, Canny, 
Sobel etc[17]. Following a comparison of various operators, 
the Canny operator is selected for binarizing the image and 
detecting edges in this study. The thresholding process used 
in Canny operator edge detection helps to generate a complete 
and continuous edge curve, which can eliminate the effect of 
noise as much as possible and has good detection effect on 
edge points [18]. The implementation steps of the edge 
feature extraction algorithm are shown in Fig. 4. The 
minimum and maximum thresholds set in this paper are 180 
and 250, respectively, and the detection detection effect is 
shown in (Fig. 2c). 

 

 
Figure 4. Canny edge detection implementation flow 

2.4. Extract Region of Interest (ROI) 
Although the image has been preprocessed and Canny edge 

detection, there are still some other environmental 
information in the image that affects the accurate recognition 
of the edge of the mesh carpet, and it is necessary to make the 
picture contain only the information about the edge of the 
mesh carpet as far as possible, and also in order to improve 
the accuracy of the algorithm's detection and reduce the 
complexity of the algorithm's computation. So a region of 
interest is designed to include all the information related to 
the edge of the mesh carpet in the picture. From the image 
information after Canny edge detection, a quadrilateral region 
can be used to include all the information about the edge of 
the mesh carpet. Firstly, a quadrilateral mask is set manually 
according to the position information of the mesh carpet 
edges after Canny edge detection, and the mask area set in this 
paper is shown in the green quadrilateral in Fig. 2(d), and then 
the mask is used to cover the mesh carpet edges information 
in the image, and the extraction of the region of interest is 
finally realized, and the result is shown in Fig. 2(e). 

2.5. Mesh Blanket Edge Detection 
2.5.1. Hough Linear Transform 

The mesh carpet edge detection is a straight line detection 
of the edge features extracted from the region of interest 
(ROI). The principle of the Hough transform is to use the one-
to-one correspondence between the straight line in the 
Cartesian coordinate system and the point in the parameter 
space to convert a straight line or a curve from the image 
space to the parameter space, and the transformed straight line 
maps into a peak point in the parameter space. The problem 
of determining the presence or absence of a straight line in the 
image space of a detected Cartesian coordinate system is 
converted into the problem of counting the number of times a 
peak point appears in the parameter space [19]. On a planar 
image, the distance from any point on a straight line in the 
coordinate system to the origin: 

 
 𝑟 𝑥 cos 𝜃 𝑦 sin 𝜃            (1) 

 
𝜃 - the azimuth of the line; 
x, y - coordinates of any point on the line. 
The image space and parameter space are shown in Fig. 5, 

for 1 straight line in the image coordinate space, after this 
transformation, any point on the line can be mapped to 1 
sinusoidal curve in the parameter space, and for 2 points on 1 
straight line in the image space, their corresponding curves in 
the parameter space intersect at the same point, so that 
detecting a straight line in the image space can be changed 
into finding the maximum number of sinusoids in the polar 
coordinate parameter space that passes through the point ( r,θ) 
with the maximum number of sinusoidal curves. 

 

 
(a) Image space (b) Parameter space 

Figure 5. Image space and parameter space 



 

101 

The flow of Hough straight line detection is shown in Fig. 
6, set the accumulator A′ to map every point in the image 
space to the parameter space, if there is a point that satisfies 
Eq. (1), add 1 to the value of the accumulator A′, if there is 1 
straight line in a region of the image, the accumulator A′ will 
be extremely large in this region, and by x detecting the 
maximum value of the A′ in each region, the presence of the 
straight line segments. 

 

 
Figure 6. Hough linear detection process 

2.5.2. Improved Hough linear transform 
Figure 7 shows the straight lines detected by the Hough 

transform for the edges of the enlarged mesh carpet, and it can 
be seen that all edges with straight line features will be 
detected as straight lines, including not only the edges of the 
mesh carpet, but also the edges of other objects with straight 
line geometrical features except the edges of the mesh carpet 
in the captured image, and these straight lines become the 
noise points that affect the detection results of the mesh carpet. 

 

 
Figure 7. Hough transform detection 

 

In order to quickly find out the edge position of the mesh 
carpet, reduce the interference of noise, and improve the real-
time detection, the Hough transform is improved as follows, 
taking into account the characteristics of the mesh carpet in 
the pixel space. 

1) Limit the range of the slope variation of the left and right 
edge lines. This paper takes the image captured by the camera 
at the location as the processing sample, and through a large 

number of image analysis, it is known that the clip angle of 
the edge of the mesh carpet is not random, and these clip 
angles are generally distributed in a range. Generally 
speaking, the angle between the left and right edge lines of 
the mesh carpet and the horizontal direction in the image 
ranges from 60°<θ<90°. Based on this characteristic of the 
mesh carpet edges, a linear screening method based on linear 
angle constraints is used. The edge lines that are not within 
the angle constraints are eliminated, and the mesh carpet 
edges that are within the angle constraints are retained. As a 
result, the slopes of all discrete small straight line segments 
of the mesh blanket edges also exist in a certain range.  

 
𝜃 arctan  

60° 𝜃 90°    Recognized as paper machine clothing edge line
𝑂𝑡ℎ𝑒𝑟                                                        𝐸𝑥𝑐𝑙𝑢𝑠𝑖𝑜𝑛

 

(2) 
 

Where𝑥 ，𝑥 ，𝑦 ，𝑦  are the endpoint coordinates of the 
line segments detected by Hough. By comparing the angles of 
the line segments, those line segments whose angles are not 
within the set range are eliminated. 

2) Remove straight line segments that deviate significantly 
from the average slope value. Considering the slope 
constraint, all discrete line segments may not be the edge lines 
of the mesh blanket. Calculate the average slope of the 
remaining set of line segments. Based on a large number of 
experiments, setting the threshold for slope deviation to 0.3, 
the final set of left and right edge lines of the mesh blanket is: 

 
𝐿 𝐿 /𝐿 ∈ 𝑘 0.3, 𝑘 0.3 , 𝑖 0,1,2,3 ⋯ (3) 

 
𝐿 𝐿 /𝐿 ∈ 𝑘 0.3, 𝑘 0.3 , 𝑗 0,1,2,3 ⋯  (4) 

 
Where: 𝑘  represents the average slope of the straight 

line segments in the collection 𝐿   of the left lift paper 
machine clothing. 𝑘  represents the average slope of the 
straight line segments in the collection 𝐿  of the right lift 
paper machine clothing. 

2.5.3. Edge line fitting 
Since the edge of the mesh carpet detected by the Hough 

straight line may be a continuous line segment, it can be seen 
from Fig. 7 that the detected straight lines of the edge of the 
mesh carpet are 2, that is, the whole edge is not detected as 1 
complete straight line, but it can be seen from Fig. 7 that the 
2 straight lines are trending on a straight line. In order to 
merge the 2 straight lines into 1 straight line, the 2 straight 
lines are fitted using least squares, and the principle of least 
squares is shown in Fig. 8. According to a given set of data 
points P ( xi, yi ) consisting of n points, i = 0, 1, 2, ⋯, n, seeks 
a straight line or curve that minimizes the sum of the squares 
of the distances of the given n points, so as to minimize the 
sum of the squares of the distances of these n points to the 
straight line or curve. Least square are a mathematical 
approximation and optimization method that does not require 
that the curve in question passes through all n points. 



 

102 

 
Figure 8. Principle of least squares fitting 

 

Because the algorithm program gets 2 endpoints of each 
straight line, 2 straight lines that is, 4 endpoints, using least 
squares fitting for these 4 points, the slope of the straight line 
and the intercept of the straight line on the y-axis can be 
obtained, so that the parameters of the straight line are 
obtained, and the corresponding equation of the straight line 
can be obtained. The final edge-fitted straight line is obtained 
as in (Fig. 3f). It can be seen that the fitted straight line is 
completely close to the edge of the paper machine clothing, 
the effect is more satisfactory. 

2.6. Determination of Misalignment  
Paper machine clothing misalignment failure is usually 

manifested in two forms: ① edge of the net carpet deflection, 
as shown in (Fig 9a), ② net carpet overall misalignment, as 
shown in (Fig 9b). 
 

 
a) Edge deflection b) Overall deflection 
Figure 9. Two types of misalignment 

 
(1) Paper machine clothing edge deflection 
Edge straight line features extracted using the improved 

Hough algorithm, the results will retain the starting 
coordinates of the detected edge straight line, assuming that 
the starting coordinates of the straight line are (𝑥  ,𝑦  ), the 
ending coordinates are (𝑥 ,𝑦 ) and the angle of the net carpet 
running direction clip angle is 𝜃 , then the edge straight line 
corresponding to the angle of the misalignment of the running 
angle 𝜃  is: 

 

𝜃 arctan 𝜃            (5) 

 
When the paper machine clothing misalignment occurs, the 

angle between the detected straight line and the original web 
running direction is the misalignment angle 𝜃 , and two angle 
thresholds 𝜃  and 𝜃  are set simultaneously and 𝜃 ＜𝜃 . 

A net blanket is considered not to have undergone 
misalignment when there is a range of slight controllable 
misalignments that are allowed to occur during normal 
operation, i.e., 𝜃 ＜𝜃 ; 

When 𝜃 ≤𝜃 ≤𝜃 , it is considered that the mesh carpet has 
a slight misalignment fault; when 𝜃 ＞𝜃  , it is considered 
that the mesh carpet has a serious misalignment fault.At this 
time, the system automatically in the host computer detection 
interface pop-up grading alarm prompts, and on-site paper 
machine equipment issued sound and light alarms, to remind 
the relevant personnel to the scene in a timely manner for 

maintenance and repair. 
(2) Overall misalignment of paper machine clothing 
When the entire mesh carpet experiences deviation, 

assuming that the average value of the transverse coordinates 
of the right edge line of the mesh carpet is 𝑥  and the average 
value of the transverse coordinates of the left edge line is 𝑥 , 
then the transverse coordinate of the center line of the mesh 
carpet, d = |𝑥 𝑥 | /2. Typically, when the net blanket as 
a whole undergoes misalignment, the difference in the 
transverse coordinates of the centerline relative to the 
transverse coordinates of the misalignment exceeds 5% of the 
bandwidth. 

 
|𝑑 𝑑 | 0.05𝑤             (6) 

 
(where d  represents the coordinates of the center position 

of the two edges in the image of the paper machine clothing, 
and 𝑤 represents the width of the paper machine clothing). 

In summary, the judgment formula of whether the conveyor 
belt misalignment: 

𝜃 arctan
𝑦 𝑦
𝑥 𝑥

𝜃  

|𝑑 𝑑 | 0.05𝑤 
When any one of the above 2 situations is met, it can be 

judged that the net carpet misalignment, and the 
corresponding warning or alarm processing can be carried out. 

3. Experimental Results 

In order to verify the accuracy and robustness of the 
implementation of the algorithm in this paper, we evaluate the 
misalignment detection method in the paper through two parts 
of experiments, static acquisition and dynamic acquisition, 
the static acquisition is the image of the web carpet in non-
operational state in the laboratory, and the dynamic 
acquisition is the image of the web carpet monitoring video 
frames in operation at a real industrial site. Since the press 
section of a paper machine is subjected to continuous 
mechanical stresses and will undergo roll pressing, wear, 
compaction and contamination with the highest risk of 
misalignment [20], we used the Press felt in the press section 
of a paper machine as an example for experimental validation. 

Build an experimental platform according to the structure 
shown in Figure 1, and the algorithms were compiled in the 
PyCharm environment, written in Python 3.8 and 
implemented using OpenCV 4.5.1, using a computer with the 
following hard environments: an Intel Core i5 11400H 
processor, 16 GB of RAM, an NVIDIA GeForce RTX 3050, 
and a 4 GB RAM graphics card. 

3.1.  Laboratory experiments 
Press felt data for the laboratory experiments were obtained 

from a medium-sized inclined long web test paper machine at 
the College of Light Industry Science and Engineering, 
Shaanxi University of Science and Technology, with a design 
speed of 35 m/min. 5-megapixel 2/3" CMOS Gigabit Ethernet 
industrial face-mounted camera from Hikvision, model 
number MV-CA050-10GC, was used for the laboratory 
experiments. The resolution of the paper machine Press felt is 
2445*2048, in order to meet the experimental needs, the 
collected images are cropped by AOI area [21], and the final 
resolution of the processed image is 1296*864, the 
experimental scene is shot with a variety of interference 
situations, such as the interference of other equipment in the 



 

103 

surrounding and the linear features in the experimental 
environment, and so on. 

After the algorithm proposed in this paper, the tilt angle of 
the net blanket, the distance between the left and right sides 
of the net blanket near the stick, and the running time of the 
algorithm can be obtained, as shown in Table 1. By artificially 
creating misalignment, the images of four groups of paper 
machines with different positions were detected and analyzed. 
Firstly, based on the position and tilt angle of the normal 
operation of the paper machine net blanket, the analysis and 
calculation of the unmisaligned pressing net images was 
performed. At the pixel coordinate system, the distance 

between the left and right sides of the net blanket near the 
stick was found to be 35 pixels, and the standard pressing net 
operation angle was 85.5 degrees. Then, the maximum 
misalignment angle and maximum offset were set (in this 
experiment, the maximum offset is 2.5% of the bandwidth, 
the maximum offset angle is 2.5 degrees, i.e., the threshold 
for the lateral offset of the net blanket is 5 pixels, and if the 
tilt angle exceeds 2.5 degrees, it is considered to be 
misaligned). This paper analyzes the angle at the pixel level, 
and through the correspondence relationship between pixels 
and actual distance [22], can effectively detect the real size. 

 
Table 1. Results of laboratory experiments 

Group Left tilt angle/° Right tilt 
angle/° 

Left 
distance/pixel 

value 
Right distance/pixel 

value 
Detection 

time/s result 

1 -85.00 83.50 18 52 0.715 Misalignment 
2 -85.07 83.65 14 56 0.814 Misalignment 
3 -85.11 85.99 31 38 0.697 normal 
4 -85.08 83.17 26 44 0.688 Misalignment 

 
According to the experimental results, it can be seen that 

for a single static paper machine clothing image, the 
algorithm can accurately measure the angle of the two edge 
straight lines and the pixel distance from the edge of the roll, 
and can quantitatively and truly reflect the misalignment of 
the value netting at the pixel level through the set threshold. 
In addition, the average running time of the algorithm 
described in this paper is 0.728s in four groups of experiments, 
which has good real-time processing efficiency, and if it is 
transplanted to a digital signal processor (DSP, Digital signal 
processing) [22], its running time will be further shortened. 

3.2. Industrial field experiment 
In order to verify the effectiveness of this paper's algorithm 

in real industrial scenarios, we further use the data of paper 
machine Press felt from industrial site for experimental 
analysis. And to determine whether the paper machine felt at 
this moment has a misalignment fault. The experimental 
samples for the industrial field experiments are from the 
crescent-type tissue machine designed by Dingchang Paper 
Making Machinery of the Famensi Paper Industry in Shaanxi 
Province, which reaches a speed of 520 m/min and a width of 
2600 mm. 

The final field test results of net run deviation detection are 
shown in Figure 10. (Fig 10a) are the original image obtained 
from pressing net, which undergoes preprocessing to obtain 
Figure 10(b). Figure 10(c) shows the result of edge feature 
extraction using the Canny operator. Figure 10(d) shows the 
image containing the net edge after setting the ROI. Figure 
10(e) is the result of Hough line transformation. The final 
display of edge pixel values and edge position is shown in 
Figure 10(f), where the green lines on both sides are the 
detected net edges, and the red numbers are the calculated 
pixel offsets. It can be seen that the detected net edges 
coincide with the actual edge positions. 

The images collected during high-speed operation in the 
factory may be blurry, and this algorithm can also recognize 
them well. As shown in Figure 11, the image with added 
filtering blur can still accurately extract the edge of the mesh 
blanket, which has good robustness. 

 

 
a) Original image b) Preprocessing c) Edge feature extraction d) 

Region of interest (ROI) e) Mesh blanket edge detection f) 
misalignment diagram 

Figure 10. Field test of net blanket misalignment detection 
 

Due to the development of large-scale and wide format 
paper machines, the shooting field of view is large, and 
multiple cameras need to be installed to see the entire paper 
machine mesh blanket image. According to the results in 
Tables 1 and 2, it can be seen that when the mesh blanket 
deviates, both the left and right edges of the deviated mesh 
will undergo angle and position changes. Therefore, it is 
feasible to only capture images of one side of the mesh 
blanket to determine the deviation. 
 

 
a) Blurred image b) Experimental results of blurred images 

Figure 11. Experimental results of blurred images 



 

104 

3 Conclusion 

The proposed computer vision-based improved Hough 
straight line detection method for paper machine clothing 
misalignment detection is able to accurately identify the mesh 
area and realize the extraction of straight line features of the 
mesh edge;A method for determining the net carpet 
misalignment of paper machine clothing is given, and the 
position coordinates of the extracted straight lines at the edge 
of the net carpet are used to comprehensively determine 
whether the net carpet misalignment occurs through the tilting 
angle of the net carpet of the paper machine clothing and the 
distance of the net carpet according to the two sides of the 
stick;Through the laboratory and factory experiments, the 
algorithm in this paper can effectively identify the 
misalignment, and provides an effective method for the 
correction of paper machine clothing. This technology opens 
up a new way for continuous online inspection in paper-
making industry. 

Acknowledgment 
This work is supported by grants from the Key R&D Plan 

of Shaanxi Province (2023-YBGY-331) and Scientific 
Research Fund of Shaanxi University of Science and 
Technology (Doctoral Fund, Research on Data Fusion 
Technology in Paper Industry, BJ15-19). 

References 
[1] Wang Guodong, Z.Y., Forming Fabrics for Modern Paper 

Machines. China Pulp & Paper Industry, 2022. 43(22): p. 12-
28. 

[2] Xiao, S., W. Lei, and Z. Weimin, Empirical Discussion on 
Abnormal Application Analysis of Paper Machine Felt. China 
Pulp & Paper 2022. 41(002): p. 041. 

[3] Adanur, S., Paper Machine Clothing. 1997, Technomic 
Publishing Company, Inc.: USA. 

[4] Paulapuro, H., Papermaking Part 1, Stock Preparation and 
Wet End. 2000, Finland: Finnish Paper Engineers’ Association. 

[5] Holmberg, H., M. Tuomaala, and P. Ahtila, Papermaking Part 
2, Drying Book 9. 2010, Finland: Finnish Paper Engineers’ 
Association. 

[6] LIU, Q. Principle application and maintenance of high speed 
paper machine clothing corrector. in Proceedings of the 2019 
China Pulp and Paper Automation Technology and Intelligent 
Manufacturing Symposium. 2019. 

[7] WEI Guo-hua, et al., Application of contactless correction 
device for dry fabric in high-speed paper machines. China Pulp 
& Paper Industry, 2019. 40(8): p. 42-45. 

[8] Cheng, J.C.P. and M.Z. Wang, Automated detection of sewer 
pipe defects in closed-circuit television images using deep 
learning techniques. Automation in Construction, 2018. 95: p. 
155-171. 

[9] Hakala, T., P.J.F. Heikkilä, and T.i.E. Europe, Press felts 
coated with electrospun nanofibres. 2011. 19(1): p. 84. 

[10] Reczulski, M., Intensification of Press Felt Dewatering for 
Tissue Machines. Fibres and Textiles in Eastern Europe, 2021. 
29(1(145)): p. 92-97. 

[11] Frischmon, C., J. Xu, and S. Ramaswamy. Three dimensional 
visualization and characterization of paper machine fabrics 
and their relationship to their properties and dewatering 
performance. in TAPPICon Virtual 2021. 2021. TAPPI Press. 

[12] Reczulski, M., Intensification of Press Felt Dewatering for 
Tissue Machines. Fibres & Textiles in Eastern Europe, 2021. 
29(1): p. 92-97. 

[13] Kumar, R., S. Rani, and M.A. Awadh, Exploring the 
Application Sphere of the Internet of Things in Industry 4.0: A 
Review, Bibliometric and Content Analysis. Sensors, 2022. 
22(11). 

[14] Wu, X., et al., Research on Belt Deviation Fault Detection 
Technology of Belt Conveyors Based on Machine Vision. 
Machines, 2023. 11(12): p. 5. 

[15] Yang, Y., et al., On-line conveyor belts inspection based on 
machine vision. Optik, 2014. 125(19): p. 5803-5807. 

[16] Yu, B.C., et al., Dual band infrared detection method based on 
mid-infrared and long infrared vision for conveyor belts 
longitudinal tear. Measurement, 2018. 120: p. 140-149. 

[17] Jing, J., et al., Recent advances on image edge detection: A 
comprehensive review. Neurocomputing, 2022. 503(2): p. 259-
271. 

[18] Marzougui, M., et al., A Lane Tracking Method Based on 
Progressive Probabilistic Hough Transform. IEEE Access, 
2020. 8(7): p. 84893-84905. 

[19] Zheng, F., et al., Improved Lane Line Detection Algorithm 
Based on Hough Transform. Pattern Recognition and Image 
Analysis, 2018. 28(2): p. 254-260. 

[20] Choi, K.-H., et al., Methods to Detect Road Features for Video-
Based In-Vehicle Navigation Systems. Journal of Intelligent 
Transportation Systems, 2010. 14(1): p. 13-26. 

[21] Poczekajło, P. and R. Suszyński, Modern computing methods 
for digital signal processing engineering systems. Procedia 
Computer Science, 2021. 192: p. 3534-3541. 

[22] Karanam, M., et al. Object and it’s dimension detection in real 
time. in E3S Web of Conferences. 2023. EDP Sciences.