Method for detecting tear and foreign object of conveyor belt

Abstract

A kind of tear and foreign matter identification detection method of conveying belt, the method includes: the image collected by image acquisition equipment is preprocessed;The belt image after pre-processing is identified by the belt tear and foreign matter identification detection model established in advance, and belt tear detection and foreign matter identification are completed;The belt tear and foreign matter identification detection model uses the image enhancement method based on Retinex, and realizes multi-scale target detection according to the detection idea of image feature pyramid.The method of the present application can reduce the belt tear length to the greatest extent, has great promoting effect on the optimization and promotion of converter process, and has great economic benefits and production significance.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of conveying belt detection, and in particular to a tearing and foreign matter identification detection method for a conveying belt. BACKGROUND

[0002] Most steel plants still rely on old detection devices or simply rely on manual regular inspection as a belt tearing detection means, which is difficult to meet the production needs, has limited emergency handling capacity for sudden accidents, and seriously threatens the safe production operation of enterprises.

[0003] The lime kiln feeding system belt conveying device of the steelmaking operation has a belt length of 660 meters and a conveying distance of 330 meters, supplies multiple converters, and the normal transportation of lime materials, the working state and equipment safety are controllable. However, when there are metal or other sharp objects and other inclusions in the materials, the belt will be damaged, and then the belt tearing will be caused.

[0004] The belt is expensive, and the cost can account for more than 50% of the entire belt conveyor. At the present stage, the transverse tensile strength and carrying capacity of the belt are significantly improved, but the longitudinal tear resistance is not significantly improved. Investigation shows that 90% of the belt tearing is longitudinal tearing. If the entire belt tears, it will cause material spilling, damage to the reducer and motor and other equipment, and in serious cases, it will even damage the rack structure, threatening the personal safety of the on-site personnel.

[0005] The improvement of the belt conveying monitoring and identification technology is the inherent demand of production process and safety. At present, the inspection of the belt conveying system mainly relies on old detection devices and manual regular inspection as a belt tearing detection means, which has been difficult to meet the production needs, has limited emergency handling capacity for sudden accidents, and seriously threatens the safe production operation of enterprises. The improvement of the belt conveying monitoring and identification technology is the inherent demand of production process and safety.

[0006] Since the 1970s, research and testing in related fields have been carried out at home and abroad. The detection idea has gradually developed from single to intelligent and from contact to non-contact. In recent years, machine vision detection has become a new direction of industrial detection. Industrial cameras are used to replace human eyes to collect data, and the detection results are fed back through processors. It has been researched and utilized in many fields. However, there is no related identification detection technology on the lime kiln feeding system belt conveying device of the steelmaking operation. SUMMARY

[0007] The purpose of the present application is to overcome the defects of the prior art and provide a tearing and foreign matter identification detection method for a conveying belt.

[0008] In order to achieve the above purpose, the present application provides a tearing and foreign matter identification detection method for a conveying belt, which comprises:

[0009] preprocessing the image collected by the image collection device;

[0010] identifying the preprocessed belt image through a pre-established belt tear and foreign matter identification detection model to complete belt tear detection and foreign matter identification;

[0011] The belt tear and foreign matter identification detection model adopts a Retinex-based image enhancement method, and realizes multi-scale target detection according to the detection idea of the image feature pyramid.

[0012] As an improvement of the above method, the image collection device includes a first line array camera and a second line array camera, wherein the first line array camera is arranged above the belt to photograph the material transported on the belt, and the second line array camera is arranged below the belt to photograph the belt.

[0013] As an improvement of the above method, the preprocessing includes filtering and denoising and open-close operation processing of the images collected by the first line array camera and the second line array camera.

[0014] As an improvement of the above method, the input of the belt tear and foreign matter identification detection model is the preprocessed image, and the output is the belt tear identification result and / or the foreign matter identification result; the belt tear and foreign matter identification detection model includes a backbone network, a feature extraction module, an RPN network, a RoI pooling layer and a classifier connected in sequence; wherein,

[0015] The backbone network is a restnet18 network;

[0016] The feature extraction module is used to complete feature extraction, and includes 7*7 convolution kernel, 5*5 convolution kernel, two 3*3 convolution kernels and four inverse convolution operations connected in sequence;

[0017] The RPN network is used to generate a plurality of proposal windows;

[0018] The RoI pooling layer is used to generate a fixed-size feature map;

[0019] The classifier is used to obtain the identification result of whether the belt is torn and / or whether there is foreign matter.

[0020] As an improvement of the above method, the method further includes a training step of the belt tear and foreign matter identification detection model, and the detection classification probability and the detection bounding box regression in the classifier are jointly trained until the training requirement is met, so as to obtain the trained belt tear and foreign matter identification detection model.

[0021] A kind of tear and foreign matter identification detection system of conveying belt, the system includes: belt tear and foreign matter identification detection model, pre-processing module and detection identification module;Wherein,

[0022] The pre-processing module is used to pre-process the image collected by the image acquisition device;

[0023] The detection identification module is used to identify the belt image after pre-processing by the pre-established belt tear and foreign matter identification detection model, complete belt tear detection and foreign matter identification;

[0024] The belt tear and foreign matter identification detection model adopts the image enhancement method based on Retinex, and realizes multi-scale target detection according to the detection idea of image feature pyramid.

[0025] Compared with the prior art, the advantages of the present application are that:

[0026] 1、The method of the present application can quickly detect the inclusions, tears and other defects in the belt material, accurately and timely issue an alarm and stop, and greatly reduce the belt tear length, greatly promote the optimization and improvement of the converter process, and have great economic benefits and production significance;

[0027] 2、The inclusion identification of the present application can identify non-raw material foreign matter on the belt in real time and efficiently by studying the belt foreign matter detection method, prevent belt tearing, maintain the safe operation of the belt machine and ensure the stable production of steel enterprises, which has important social significance and economic value;

[0028] 3、The belt longitudinal tear and defect detection of the present application can early warn the production line personnel of fault occurrence and minimize the on-site loss. BRIEF DESCRIPTION OF DRAWINGS

[0029] Figure 1 It is a conveying belt and image acquisition device installation schematic diagram;

[0030] Figure 2 It is the structure schematic diagram of belt tear and foreign matter identification detection model of the present application. DETAILED DESCRIPTION

[0031] Based on the above problems, it is urgent to develop a system that can quickly detect the inclusions, tears and other defects in the belt material, accurately and timely issue an alarm and stop, and greatly reduce the belt tear length, greatly promote the optimization and improvement of the converter process, and have great economic benefits and production significance.

[0032] The application is a belt crack and foreign matter detection method based on a deep convolutional network. Raw data is first obtained, and preprocessing including extraction of a region of interest, image filtering and noise reduction is performed. Then, a belt tear and inclusion identification algorithm based on semantic segmentation is designed. Finally, according to the inclusion identification and belt tear detection results, combined with the belt tear analysis and prediction algorithm, timely warning is performed.

[0033] Related abbreviations and key terms:

[0034] Image feature pyramid: The image pyramid is a kind of multi-scale expression of an image, which is an effective but simple concept structure to explain the image in multiple resolutions. The image pyramid of an image is a set of image resolutions derived from the same original image, which is gradually reduced in a pyramid shape (from bottom to top). It is obtained by hierarchical downsampling until a certain termination condition is reached. We compare the image layer by layer to a pyramid, and the higher the level, the smaller the image and the lower the resolution.

[0035] Convolutional neural network: Its English is Convolutional Neural Networks, abbreviated as CNN. It is a kind of feedforward neural network containing convolution calculation and having a deep structure, and is one of the representative algorithms of deep learning. Convolutional neural network has representation learning ability, and can perform translation invariant classification on input information according to its hierarchical structure,

[0036] Deep learning: It is a kind of machine learning, and machine learning. Deep learning is derived from the research of artificial neural network, which is a kind of multilayer perceptron. It learns the internal rules and representation levels of sample data, and obtains information in the learning process. It is effective in recognizing text, images and sound. Deep learning can use combined low-level features to form more abstract representations of high-level attributes to discover distributed features of data. The concept of deep learning is similar to the neural network that simulates the brain to analyze and learn. It can simulate the mechanism of the human brain to explain data.

[0037] Attention mechanism: English abbreviation Attention, the application of this attention mechanism is called self-focus or internal focus mechanism. It can be proposed in the process of using encoder-decoder structure for neural machine translation, and can be applied to similar tasks such as generating a descriptive sentence according to a picture and summarizing the content of a text. From a high level, it allows the decoder to select the required part from multiple context vectors, so that the encoder is freed from the constraint of only being able to compress context information into a fixed-length vector, and thus can represent more information. Attention mechanism is very common in deep learning models, not limited to encoder-decoder hierarchy, and can solve tasks such as text classification or representation learning

[0038] The technical solutions of the present application will be described in detail below in combination with the drawings and examples.

[0039] Example 1

[0040] Example 1 of the present application proposes a method for identifying and detecting tearing and foreign matter of a conveying belt.

[0041] As Figure 1 shown, the image acquisition device acquires a belt image containing a laser line in real time, and after noise reduction and filtering by the internal circuit, the collected data is transmitted to the signal processing transmission device through the gigabit Ethernet network. The number of linear array cameras is determined according to the width of the belt conveyor belt, and two are taken as an example, but not limited to two. One is placed above the belt to identify the material, and the other is placed below the belt to identify the belt tearing, ensuring that the entire belt data is completely collected.

[0042] The transmission processing device is a signal receiving and forwarding device in the belt tearing detection device based on visual technology, and its function is similar to that of a relay station. Multiple image acquisition devices transmit real-time collected data to the signal processing transmission device. Each device captures 110Mbyte of data per second, and the data generated by multiple devices working simultaneously is very large. It is difficult for a general Ethernet network to realize real-time data transmission, and the maximum transmission distance of the Ethernet network is 100 meters, so the signal processing transmission device needs to convert the Ethernet data into optical fiber transmission to the total control console in the central control room.

[0043] In the design of protective equipment and tooling, effective dustproof measures need to be taken to ensure that the data acquisition device is not affected by the dust on site and can effectively prevent the accumulation of flying dust.

[0044] 1) The belt conveyor speed is typically between 1.6m / s and 2.5m / s. When selecting the core vision camera and laser, performance factors such as exposure time and frame rate must be fully considered. The selection of accessories such as lenses, filters, and light sources needs to be customized based on the operating conditions. Additionally, protective fixtures or automated equipment for the core acquisition device need to be designed. High temperatures and dust conditions can be addressed through both hardware modifications and software optimizations to minimize adverse effects.

[0045] 2) Material and inclusion identification

[0046] This paper adopts a Retinex-based image enhancement method. Considering the characteristics of belt conveyor monitoring images, an improved method is designed based on Retinex theory. Based on the detection approach of Feature Pyramid Network (FPN), a novel feature fusion method is designed and used to achieve multi-scale target detection through the superposition of multi-scale feature maps, supporting the detection and identification of impurities of different sizes. It is suitable for the operating environment of conveyors with high dust and moisture content and uneven lighting.

[0047] 3) Detection and identification of belt tears and foreign objects

[0048] The machine vision-based belt tear detection system mainly consists of several parts: data acquisition, data transmission, data processing, and auxiliary devices. The data acquisition part primarily uses image acquisition equipment, installed between the carrying and return belts of the belt conveyor and fixed to the conveyor's idler frame. After acquiring image data of the belt and materials, the image acquisition equipment transmits the data to the central control console via transmission and processing equipment. The central control console uses a uniquely designed belt tear detection algorithm to process and analyze the acquired data. Through a designed and implemented belt tear and foreign object identification detection model, it determines the current belt condition and identifies and detects foreign objects, feeding the results back to the belt conveyor control system to complete the belt tear detection and foreign object identification detection.

[0049] Belt tear and foreign object detection model

[0050] The FPN-ResNet algorithm is used to identify belt tears and foreign objects. The specific structure of the belt tear and foreign object identification and detection model is as follows: Figure 2 As shown, the model incorporates an attention mechanism, which improves recognition efficiency.

[0051] The belt tear and foreign object detection model takes a preprocessed image as input and outputs the belt tear detection result and / or foreign object detection result. The model comprises a backbone network, a feature extraction module, an RPN network, a RoI pooling layer, and a classifier connected sequentially.

[0052] The backbone network is a restnet18 network;

[0053] The feature extraction module is configured to complete feature extraction, including sequentially connected 7*7 convolution kernels, 5*5 convolution kernels, two 3*3 convolution kernels, and four reverse convolution operations.

[0054] The RPN network is configured to generate a plurality of proposal windows.

[0055] The RoI pooling layer is configured to generate a fixed-size feature map.

[0056] The classifier is configured to obtain an identification result of whether the belt is torn and / or whether there is a foreign object.

[0057] Specific steps: input the detection image; input the entire image into the model to obtain feature0 through the backbone network (restnet18), then obtain feature1 through 7*7 convolution kernels, then obtain feature2 through 5*5 convolution kernels, then obtain feature3 through 3*3 convolution kernels, then obtain feature4 through 3*3 convolution kernels (the stride changes from 1 to 2), then perform reverse convolution operation on feature4 to obtain feature5, perform reverse convolution operation on feature5 to obtain feature6 (realize new feature fusion), perform reverse convolution operation on feature6 to obtain feature7, perform reverse convolution operation on feature7 to obtain feature8, then stack feature0 and feature8 to obtain feature9, and complete feature extraction; generate proposal windows (proposals) through the RPN, and generate 500 proposal windows for each image; map the proposal windows to the last layer of the CNN feature map; make each RoI generate a fixed-size feature map through the RoI pooling layer; and train the classification probability and the bounding box regression (Bounding box regression) jointly through SoftmaxLoss (detection classification probability) and Smooth L1 Loss (detection bounding box regression). When the training requirements are met, a trained belt tearing and foreign object identification detection model is obtained.

[0058] 4) Automatic triggering

[0059] When the material enters the detection area, the posture and trajectory are automatically adjusted to facilitate data acquisition.

[0060] 5) Abnormal situation early warning

[0061] According to the lime kiln production management system, when obvious abnormal conditions such as material mixing and belt damage occur, active early warning is carried out, the warning light flashes, and the whistle sound reminds the on-site workers to further confirm.

[0062] Embodiment 2

[0063] Embodiment 2 of the present application proposes a tearing and foreign matter identification detection system of a conveying belt, which is realized based on the method of embodiment 1, and the system comprises a belt tearing and foreign matter identification detection model, a preprocessing module and a detection and identification module; wherein,

[0064] The preprocessing module is used for preprocessing the image collected by the image acquisition device;

[0065] The detection and identification module is used for identifying the preprocessed belt image through the pre-established belt tearing and foreign matter identification detection model, and completing belt tearing detection and foreign matter identification;

[0066] The belt tearing and foreign matter identification detection model adopts an image enhancement method based on Retinex, and realizes multi-scale target detection according to the detection idea of the image feature pyramid.

[0067] The present application can replace artificial regular patrol inspection as a belt tearing detection means, reduce sudden safety accidents, and improve the enterprise safety production operation capability. It can quickly detect the defects such as inclusions and tearing in the belt material, accurately and timely issue an alarm and stop, and greatly reduce the belt tearing length, which has a great promoting effect on the optimization and improvement of the converter process, has great economic benefits and production significance. At the same time, it helps enterprises to complete the systematic and overall safety management improvement from the aspects of physics, network, system, information and management, and meets the internal needs of production process and safety.

[0068] Finally, it should be explained that the above embodiments are only used to illustrate the technical solutions of the present application and are not limited. Although the present application has been described in detail with reference to the embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced by equivalents without departing from the spirit and scope of the present application, and they should be covered in the scope of the claims of the present application.

Claims

1. A method for detecting belt tear and foreign matter identification of a conveying belt, which is used in an environment with high temperature dust, large water vapor and uneven light intensity, the method comprising: preprocessing images collected by an image collection device; identifying the preprocessed belt images through a pre-established belt tear and foreign matter identification detection model to complete belt tear detection and foreign matter identification; the belt tear and foreign matter identification detection model adopts a Retinex-based image enhancement method and realizes multi-scale target detection according to the detection idea of an image feature pyramid; the input of the belt tear and foreign matter identification detection model is the preprocessed images, and the output is a belt tear identification result and / or a foreign matter identification result; the belt tear and foreign matter identification detection model comprises a backbone network, a feature extraction module, an RPN network, an RoI pooling layer and a classifier connected in sequence; wherein, the backbone network is a ResNet18 network; The feature extraction module is used for completing feature extraction, including 7 7 convolution kernels, 5 5 convolution kernels, two 3 3 convolution kernels and four reverse convolution operations; the RPN network is configured to generate a plurality of proposal windows; the RoI pooling layer is configured to generate a fixed-size feature map; the classifier is configured to obtain the identification result of whether the belt is torn and / or whether there is foreign matter.

2. The method of claim 1, wherein the step of detecting the tear and foreign object of the conveyor belt is characterized by, The image collection device comprises a first linear array camera and a second linear array camera, wherein the first linear array camera is arranged above the belt to capture the materials conveyed on the belt, and the second linear array camera is arranged below the belt to capture the belt.

3. The method of claim 2, wherein the step of detecting the tear and foreign object is performed by a sensor. The preprocessing comprises filtering and denoising and open-close operation processing of the images collected by the first linear array camera and the second linear array camera.

4. The method of claim 1, wherein the step of detecting the tear and foreign object in the conveyor belt is characterized by, The method further comprises a training step of the belt tear and foreign matter identification detection model, which is jointly trained according to the detection classification probability and detection bounding box regression in the classifier until the training requirements are met, thereby obtaining the trained belt tear and foreign matter identification detection model.

5. A kind of tear and foreign matter identification detection system of conveying belt, for high temperature dust, water vapor is larger, light illumination is uneven, it is characterized in that, The system comprises a belt tear and foreign matter identification detection model, a preprocessing module and a detection and identification module; wherein, the preprocessing module is configured to preprocess the images collected by the image collection device; the detection and identification module is configured to identify the preprocessed belt images through the pre-established belt tear and foreign matter identification detection model to complete belt tear detection and foreign matter identification; the belt tear and foreign matter identification detection model adopts a Retinex-based image enhancement method and realizes multi-scale target detection according to the detection idea of an image feature pyramid; the input of the belt tear and foreign matter identification detection model is the preprocessed images, and the output is a belt tear identification result and / or a foreign matter identification result; the belt tear and foreign matter identification detection model comprises a backbone network, a feature extraction module, an RPN network, an RoI pooling layer and a classifier connected in sequence; wherein, the backbone network is a ResNet18 network; The feature extraction module is used for completing feature extraction, including 7 7 convolution kernels, 5 5 convolution kernels, two 3 3 convolution kernels and four reverse convolution operations; the RPN network is configured to generate a plurality of proposal windows; the RoI pooling layer is configured to generate a fixed-size feature map; the classifier is configured to obtain the identification result of whether the belt is torn and / or whether there is foreign matter.

Images