1 Introduction

Chemical fiber products have been widely applied in traditional textile processing, aerospace, national defense and military industry, transportation, energy and environmental protection, safety protection, medical and health [1]. It is impossible for the development of chemical fiber industry to continue the growth mode of relying on quantity in the past, which shall be transformed and upgraded, and extended to such steps as R&D, design, brand, marketing and service; and the mode of production shall be transformed to be flexible, intelligent, digital, refined and green production. With the progress of computer technology, network and communication as well as related hardware and software technology, equipment manufacturing and other basic industrial technology, especially in programmatic documents like Made in China 2025, Guiding Opinions on Actively Promoting the “Internet+” Action, and Development Plan for the New Generation of Artificial Intelligence, the manufacturing technology of chemical fiber industry in China has developed from automation, digitalization and informatization to intelligence. It is feasible for chemical fiber equipment to move forward from automation equipment to intelligence through digitization and informatization [2].

The polyester filament spinning cake is formed in the way that the PET melt is extruded from the spinneret to form melt streams; after cooling and blowing, the filament is formed by cluster oiling, and then the spinning cake is formed by the winding head of the winding section through the spinning shaft [3]. The filament is composed of dozens or even hundreds of fine hairlike filaments, and the filament is wound into the spinning cake at the speed of several thousand meters per minute. This process requires strict process control. Slight fluctuation of process parameters may lead to abnormal spinning, which will cause end breakage if not found and handled in time, thus reducing the product high-grade rate and then the enterprise efficiency. In the past, this problem can only be handled by manual inspection, which is not timely and even causes false inspection and missing inspection. It is considered to develop automatic inspection robots to periodically monitor the abnormal wire floating and wrong wire in the spinning process, identify the abnormal data in real time, and guide relevant staff to deal with the problem in time.

Intelligent inspection robots act as an execution platform with certain mobile sensing ability. With the help of a variety of sensors, they can detect specific states, carry out customized services, and achieve more programming operations [4]. The intelligent inspection robots based on machine vision enjoy widespread application, such as substation inspection robots [5], meter inspection robots [6] and so on. For exception identified manually by naked eyes in spinning inspection, machine vision sensors and computers can be used to simulate the visual function of human eyes, that is, to obtain and judge the environmental information through vision sensors, so as to realize the perception, recognition and understanding of the three-dimensional scene of the objective world [7]. The information display detection technology based on machine vision is significantly different from the traditional detection technology. The technology is sensitive to light conditions, and usually has to be equipped with auxiliary light source for lighting; according to the characteristics of specific detection target, the appropriate detection algorithm and targeted training are selected, and the algorithm is specific; with wide application field, it enjoys huge application market potential [8]. Characterized by high detection speed, high intelligence, and high detection accuracy, this technology can complete the detection task better than human power.

Against industry and technical background above, this paper introduces the machine vision, mobile bearer and data visualization technology, focuses on the inspection technology for spinning production, which is an important part of the chemical fiber industry.

2 Technical Proposal of Spinning Inspection Robot

During the spinning workshop production, the spinning equipment will encounter some abnormal phenomena, such as broken wire, string wire, wire hook deviation, etc. Therefore, the on-site production must be inspected manually every two hours on average. First, the harsh environment (high temperature, high humidity and high noise) on site will cause damage to the inspectors. Second, manual inspection only depends on the naked eye recognition, which may lead to problems like subjective misjudgment and fatigue identification. Third, manual inspection is time consuming, and short of record tracing.

In order to solve the exception during the production and problems with manual inspection, the method of “mobile robot + intelligent recognition visual algorithm system + visualized data platform” can be adopted for intelligent inspection robots to improve the process of on-site inspection (see Fig. 1) and the inspection effect.

Fig. 1.
figure 1

(a) Schematic and (b) image of the spinning inspection robot

Full size image

2.1 Site Environment and Test Items

This spinning workshop has 2 production lines, as shown in Fig. 2(a), with the passage length of about 66 m, and the length of about 1.6 m. The production line passage has two production lines, each of which has 36 stations, and 72 stations need to be inspected. Each station is about 1.5 m long, and the station is shown in Fig. 2(b). The detection elements in the station are wire hook, oil nozzle and wire. The wire hooks and oil nozzles are arranged up and down and in a front-and-back staggered way, and they are installed one by one and relatively perpendicular to the ground, each 20.

Fig. 2.
figure 2

Spinning inspection site

Full size image

The main content of manual detection of abnormal spinning is broken wire, string wire, wire hook (nozzle) deflection, floating impurity, etc., as shown in Fig. 3. The broken wire refers to spinning breakage, and the spinning position instantly forms floccules. The string wire means the wire on one side is dislocated to the other side, which will be in the shape of△. The wire hook (nozzle) deflection is the deflection of the angle of the hook (nozzle), and the floating impurity means there is waste wire hanging on the hook (nozzle).

Fig. 3.
figure 3

Example of spinning faults

Full size image

2.2 Visual Inspection System Scheme

The visual inspection system mainly includes the image acquisition system and the image recognition and analysis system. The image acquisition system takes the pictures of the wire hook, the oil nozzle and the high wire as the inspection points. While ensuring the normal operation of the frame, we make corresponding transformation on the shooting background of the inspection station to achieve the inspection conditions, and configure the corresponding light source for optical compensation, as shown in Fig. 4. The parameters in the figure have to be adjusted according to the field situation. The image recognition and analysis system visually identify the judgment items, and the results are transmitted to the visualized data platform (see Fig. 4).

Fig. 4.
figure 4

Schematic diagram of camera and light source scheme

Full size image

The image recognition detects the exception with different geometric measurement algorithms [9,10,11] according to the ROI regions of the collected image. Firstly, the gray value of the image is converted to generate the corresponding gray matrix. Secondly, the threshold value of the corresponding algorithm is set. Then, the eigenvalues of the matrix are calculated and compared with the threshold value, and the normal and abnormal detection results are provided.

The broken wire is judged by the proportion parameter of black and white pixels in the collected picture (≥50%), as shown in Fig. 3(a). As for the string wire, the width of a section of wire is calculated to judge the consistency of the wire width. The wire width greater than 1 is regarded as abnormal string wire, as shown in Fig. 5(a). The wire hook (nozzle) deviation is judged by calculating the angle parameter (≥3°) of the wire hook (nozzle), as shown in Fig. 5(b). The floating impurity is judged by calculating the proportion parameter ([36%,50%)) of the black-and-white pixel block on the position of the wire hook (nozzle), as shown in Fig. 5(c).

Fig. 5.
figure 5

Schematic diagram of abnormal spinning in visual inspection

Full size image

2.3 Mobile Robot Scheme

AGV with the upper system is used for the mobile bearing machine to carry out automatic inspection. The AGV parameters are shown in Table 1. The AGV is provided with laser navigation and multiple protection such as sound and light alarm [12, 13]. Through the two-dimensional code precise positioning, the RCS system schedules the AGV mobile task, supports the specified position movement, stops and waits for 60 s when encountering obstacles to perform the avoidance function. The shovel truck has priority to pass and perform the task.

The hardware of AGV is protected by:

  1. a.

    The AGV forward direction is equipped with safety laser to detect the obstacles in 180° range in the front;

  2. b.

    The AGV is equipped with contact type of anti-collision touchdowns around it, which can trigger and stop moving instantaneously;

  3. c.

    AGV is provided with the emergency stop button, and the emergency stop button can be pressed manually at any time to stop the car.

AGV is equipped with laser detection to avoid obstacles. The detection mechanism and safety control are as follows (the following distance can be adjusted):

  1. a.

    The first gear is 1910—830 mm, the maximum speed is 600 mm/s, the deceleration distance is actual distance - 1080 mm;

  2. b.

    The second gear is 830—460 mm, the maximum speed is 100 mm/s, and the deceleration distance is: actual distance - 370 mm;

  3. c.

    The third gear is 460—250 mm, the maximum speed is 0 mm/s, the deceleration distance is: actual distance - 210 mm.

AGV divides the three laser detection areas into the far area, the middle area and the stop area respectively, with the former including the latter in turn. When the obstacle enters the far area and the middle area, it will slow down and but not stop. When it enters the stop area, it will slow down and stop.

Table 1. AGV parameters
Full size table

2.4 Scheme of Visualized Data Platform

The visualized data platform (see Fig. 6) connects the intelligent recognition visual algorithm system and the mobile bearing robot. In case of any abnormality or fault found in the inspection, the alarm information will be sent to the workers and management personnel on site in real time, so that the on-site personnel can locate the abnormal or fault information. At the same time, the manager can check the on-site production situation remotely. The client-side display of the visualized data platform mainly includes the field operator terminal, the dedicated handheld mobile device terminal and the remote view PC terminal. According to the authority classification, the field situation can be viewed and exception recovery operation can be carried out.

Fig. 6.
figure 6

Scheme of visualized data platform

Full size image

2.5 Network Architecture of Robot

The spinning inspection robot conducts the communication and interaction of various information through the network. The information includes the distribution of inspection tasks, the feedback of task execution results, the display of statistical data, etc. The network connection objects involve servers, robots, switches, wireless AP, clients, monitors, etc. See Fig. 7 for the schematic diagram of network connection. The database service and RCS server are deployed in the central computer room (in this project, the server is placed in the field cabinet). The active-standby mode can be adopted for the database server and the RCS server according to the needs to improve the disaster recovery ability. AGV, the robot upper computer and other intelligent devices are connected to the network through the wireless AP to communicate with RCS. The web client and monitoring client can be of ordinary PCs, connected to the server through the switch.

As an industrial production network, AGV needs to use the frequency band exclusively, 2.4G or 5.8G band independently. At the same time, in order to ensure the uniqueness of IP address management, the MAC address binding method is adopted to manage AGV. Through AP installation, the network signal strength is greater than −55 dB, and the Ping 1500 byte packet delay is less than 100 ms; For the edge of warehouse, non-trunk channel and other non-core business areas and AGV non-cluster areas, the signal strength is recommended to be greater than −65 dB and not less than −68 dB, and the delay of the Ping 1500 byte packet is less than 200 ms.

Fig. 7.
figure 7

Network topology deployment

Full size image

3 Scheme Test

3.1 Test Index

This paper tests the effect and efficiency of inspection. The inspection efficiency refers to the time for the robot to inspect a production line except for avoiding obstacles. The inspection effect refers to the statistics of detection rate and false detection rate of spinning exception recognition. The detection rate refers to the percentage of correct detection quantity of each exception by the robot and the manual detection quantity, As shown in Eq. (1). The false detection rate refers to the percentage of the number of errors detected by the robot for each exception to the total number detected by the robot, as shown in Eqs. (2) and (3).

$$ DR = {{RCDN} \mathord{\left/ {\vphantom {{RCDN} {MDN}}} \right. \kern-\nulldelimiterspace} {MDN}} \times 100\% $$
(1)

Here DR is Detection rate,RCDN is robot correct detection number and MDN is manual detection number

$$ FDR = \, {{(TRDN - RCDN)} \mathord{\left/ {\vphantom {{(TRDN - RCDN)} {TRDN}}} \right. \kern-\nulldelimiterspace} {TRDN}} $$
(2)
$$ TRDN = RCDN + RFDN $$
(3)

Here FDR is False detection rate, TRDN is total robot detection number and RCDN is robot false detection number.

3.2 Test Plan

A program is written in the robot, which records the start time and end time (including moving time) for the robot to reach the spinning position for inspection. By detecting 500 spinning positions, the average detection time of each spinning position is calculated, and the inspection effect is obtained, as shown in Table 2. According to Table 2, the slowest time 40 s, the fastest time 83 s, and the average time 63 s of the inspection efficiency are calculated.

Table 2. Statistical table of spinning position detection time (program output)
Full size table

In this paper, the corresponding test plan is made for the artificial exception test and the actual inspection task, and the exception detection rate and false detection rate are recorded and calculated. The artificial exception test is first to randomly select 30 spinning positions, then artificially make each exception 5 times, and then inspect the robot designated spinning position, and the exception detection situation is recorded, as shown in Table 3. According to Eqs. (2) and (3), the detection rate is about 99%, and the false detection rate is about 3%. The actual inspection task test is to manually follow and detect three periodic tasks randomly selected from the 24-h continuous inspection tasks for 7 consecutive days. The test statistics are shown in Table 4, where the detection rate is about 98%, and the false detection rate is about 5%

Table 3. Statistical table of artificial exception test
Full size table
Table 4. Actual inspection task test statistics
Full size table

3.3 Test Results

According to the test index given in Sect. 3.1, we conduct the inspection according to the test plan in Sect. 3.2. Actual good inspection effect is achieved, the detection rate is x, and the false detection rate is x. As shown in Fig. 8, the main reasons for false detection in the test are: (1) complicated shooting background; (2) Interference of ambient light; (3) Mechanical vibration; (4) Waste wire suspending. As shown in Fig. 9, the main reasons for exception detection failures are: (1) the spinning process is a dynamic and fast winding state, which leads to the “unreal transformation” of the detection target partially; (2) Metal reflection causes excessive exposure of the image taken, which cannot be detected; (3) Because the wire is too thin, the camera cannot capture the wire and it cannot be detected.

Fig. 8.
figure 8

Examples of false inspection pictures

Full size image
Fig. 9.
figure 9

Examples of undetected pictures

Full size image

4 Conclusions

Based on the AGV chassis equipped with high-definition visual recognition system, the spinning inspection robot is designed to realize the periodic inspection task of the planning path; the exception detection with the help of the visual inspection system based on geometric calculation possesses the advantages of high accuracy, simple maintenance and low cost; and the network transmission mode is adopted to realize real-time data collection, statistics and feedback. Through the visual platform, the on-site operator can understand the exception situation in real time and remove the abnormal situation.