On-site intelligent sorting method for construction waste

Abstract

The present invention relates to the technical field of construction waste treatment. Provided is an on-site intelligent sorting method for construction waste. The method comprises: S10, identifying waste components, and quantifying the proportion of inert material in construction waste; S20, by means of a computer vision method, acquiring visual difference features between the inert material and non-inert material; S30, performing mechanical screening, wherein the construction waste is screened by means of a mechanical screening module; S40, on the basis of precise sorting by a robot, performing detection on waste on a conveyor belt by means of waste detection driven by machine vision, then performing waste positioning, and, according to the results of waste detection and positioning, performing accurate sorting by using multiple robotic arms. The advantages of the present invention comprises: the described method is suitable for compact construction site scenes, the required space for installation and deployment is small, and an intelligent solution is provided for construction waste classification at compact sites.

Description

On-site intelligent sorting method for construction waste Technical Field

[0001] The present invention relates to the technical field of construction waste management, and more particularly, to an on-site intelligent sorting method for construction waste. Background Art

[0002] While the construction industry plays a vital role in socioeconomic activities, it also generates large amounts of construction waste, causing serious environmental problems. Construction waste sorting involves separating inert and non-inert waste or specific materials for subsequent recycling or landfill disposal. Sorting on-site at construction sites allows waste to be sorted and processed at the source, reducing greenhouse gas emissions from subsequent waste transportation and reducing the burden on centralized sorting facilities. However, in many high-density cities (such as Hong Kong, Shanghai, and Shenzhen), on-site sorting is often difficult to achieve due to limited construction site space and high labor costs.

[0003] Domestic and international scholars and manufacturers have explored and developed mobile and vehicle-mounted construction waste sorting equipment for construction site applications. For example, Kleemann has developed a mobile jaw crusher for screening natural stone and aggregate for recycling and stockpiling; Steinert offers sorting instruments based on magnetism, conductivity, and near-infrared (NIR) spectroscopy for construction and demolition waste recycling. In recent years, Jianuo (Suzhou, China) has developed lightweight products for construction waste sorting.

[0004] However, these products are often large in size, making them unsuitable for sorting on compact construction sites. Furthermore, many existing technologies rely primarily on mechanical sorting, which only allows for coarse screening of materials. To ensure the quality and purity of recycled materials, this mechanical sorting requires manual selection of specific materials from the waste stream. This manual process is inefficient and prone to potential occupational health and safety hazards.

[0005] To address the limitations of manual sorting, robotics and artificial intelligence technologies are increasingly being deployed in waste recycling facilities. Examples include the Bulk Handling System (BHS)'s artificial intelligence robot "Max-AI" and ZenRobotics' robotic sorting system "Fast Picker." However, these intelligent sorting technologies are primarily used in centralized sorting facilities. Due to the more complex construction site environment (constrained spaces, vibrations, and outdoor conditions), existing intelligent sorting technologies are not directly applicable. Summary of the Invention

[0006] In order to overcome the shortcomings of the existing technology, the present invention provides an on-site intelligent sorting method for construction waste. The method is suitable for compact construction site scenarios, requires little space for installation and deployment, and provides an intelligent solution for the classification of construction waste in compact sites.

[0007] The technical solution adopted by the present invention to solve the technical problem is: a method for on-site intelligent sorting of construction waste, the improvement of which is that the method includes the following steps:

[0008] S10, identification of waste components and quantification of the proportion of inert materials in construction waste;

[0009] S20. Obtain visual difference features between the inert material and the non-inert material components using computer vision methods;

[0010] S30, mechanical screening, screening the construction waste through a mechanical screening module, the mechanical screening module including a pre-treatment module, a disc screening module, a magnetic drum sorting module and an air separation module vertically distributed from top to bottom;

[0011] S40, based on the precise sorting of robots, the waste on the conveyor belt is detected through waste detection driven by machine vision, and then the waste is located. Based on the results of the waste detection and positioning, multiple robot arms are used for precise sorting.

[0012] Furthermore, in step S10, the composition of construction waste, i.e., the quantitative ratio of inert components to non-inert components in the construction waste, is calculated by developing a composition estimation model, which includes the following steps:

[0013] S1011. Input the collected data to obtain the weight and calculation of the construction waste;

[0014] S1012, probability calculation, calculating the probability of each possible composition state;

[0015] S1013. Composition estimation to determine the composition of construction waste load.

[0016] Furthermore, in step S20, the DenseNet169 model is used to extract key high-level visual features for inert content measurement.

[0017] Furthermore, after step S20, the two single recognition results output by the component estimation model and the DenseNet169 model are sorted and weighed using an ensemble learning method, including the following steps:

[0018] S201, finding base classifiers with independent errors through experimental comparison;

[0019] S202, training the selected base classifier;

[0020] S203: Merge the results of the base classifiers. The merging method includes voting or stacking.

[0021] Furthermore, in step S30, the pre-processing module includes a vibrating conveyor belt, a spreading plate, a crusher, and a shredder;

[0022] The vibrating conveyor belt is inclined, and the spreading plate, crusher and shredder are arranged above the vibrating conveyor belt along the inclined direction. The spreading plate is used to evenly distribute the construction waste on the vibrating conveyor belt. After passing through the crusher and shredder, the size of the construction waste is controlled to be reduced to less than 150 mm.

[0023] Furthermore, in step S30, the disc screening module is disposed below the vibrating conveyor belt, and a first conveyor belt is disposed on one side of the disc screening module, the first conveyor belt being used to receive the construction waste conveyed by the vibrating conveyor belt and convey it to the disc screening module;

[0024] The disc screening module includes a plurality of rotating discs with gaps reserved between the discs to remove fine heavy materials with a size less than 50 mm, including sand, soil and dirt.

[0025] Furthermore, in step S30, an inclined second conveyor belt is provided below the disc screening module, and the magnetic drum sorting module is located above the bottom end of the second conveyor belt to separate metals from the construction waste;

[0026] The air separation module includes an air separator, and the air separator is located below the bottom end of the second conveyor belt to achieve the separation of light objects;

[0027] A feeding conveyor belt is provided below the air separator for receiving the construction waste dropped from the second conveyor belt.

[0028] Furthermore, in step S40, the object detection model YOLO is used to perform waste detection to determine the type of waste.

[0029] Furthermore, in step S40, after the type of the waste is detected, the three-dimensional position of the waste on the conveyor belt needs to be calculated; a binocular vision system is used to locate the waste and calculate the three-dimensional spatial coordinates of the waste.

[0030] Furthermore, in step S40 , a group of robotic arms are used to perform collaborative classification based on the results of waste detection and positioning, allowing multiple robotic arms to work collaboratively to effectively classify waste.

[0031] The present invention provides a method for intelligent on-site construction waste sorting suitable for compact construction sites. Its vertical design, with various physical screening devices arranged vertically, effectively reduces device volume and makes it suitable for small, confined construction sites typical of high-density cities. This method improves the efficiency of on-site construction waste sorting, promotes its widespread application, and positively impacts construction waste management systems, thereby contributing to environmental protection. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] FIG1 is a schematic flow chart of an on-site intelligent sorting method for construction waste according to the present invention.

[0033] FIG2 is a schematic structural diagram of an on-site intelligent sorting device for construction waste according to the present invention.

[0034] FIG3 is a front view of the mechanical screening module of the present invention.

[0035] FIG4 is a side view of the mechanical screening module of the present invention.

[0036] FIG5 is a top view of the mechanical screening module of the present invention. Implementation Method

[0037] The present invention will be further described below with reference to the accompanying drawings and examples.

[0038] The following will clearly and completely describe the concept, specific structure and technical effects of the present invention in combination with the embodiments and drawings, so as to fully understand the purpose, characteristics and effects of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention. In addition, all the connection / connection relationships involved in the patent do not refer to the direct connection of components, but refer to the fact that a better connection structure can be formed by adding or reducing connection accessories according to the specific implementation situation. The various technical features in the invention can be combined interactively without conflicting with each other.

[0039] 1 , the present invention discloses a method for intelligent on-site sorting of construction waste. In this embodiment, the method is implemented based on an intelligent on-site sorting device for construction waste as shown in FIG2 , which includes a truck 0-1, a detachable platform 0-2, a hydraulic support 0-3, an integrated control engine 0-4, an electronic scale 1-1, a camera 1-2, a laser sensor 1-3, a detachable load bucket 1-4, a bucket outlet 1-5, a conveyor belt 1-6, a grille 1-7 (250 mm), a vibrating conveyor belt 2-1, a spreading plate 2-2, a crusher 2-3, a shredder 2-4, a disc screening module 2-5, a magnetic drum sorting module 2-6, an air separation module 2-7, a feed conveyor belt 3-1, a robot arm 3-2, and a re-sorting conveyor belt 3-3.

[0040] Specifically, the method includes the following steps:

[0041] S10, identification of waste components and quantification of the proportion of inert materials in construction waste;

[0042] In this embodiment, construction waste enters the sorting process from a removable loading bucket at the rear of the device. This bucket is equipped with sensors such as a camera, scale, and laser sensor (or rangefinder). These sensors collect information such as the surface image characteristics, weight, and depth of the waste mixture. Based on this information, the composition of the landfill can be inferred using big data analysis and computer vision.

[0043] The camera used here is a Hikvision MV-CA050-11UC CCD industrial camera, the scale is a Zhengfeng Electronics 5t industrial floor scale, and the laser sensor is a machine element infrared ranging probe. The specific product can be flexibly selected according to the actual application situation and is not limited by this patent.

[0044] Construction waste composition identification involves quantifying the proportion of inert materials in construction waste, known as the inter-ratio (IR). If a construction waste pile consists entirely of inert materials, it should be sent directly to a public fill facility. Waste with an IR below 50% can be sent to a landfill. Therefore, if the IR at a waste dump is too high (e.g., 100%) or too low (e.g., <5%), on-site sorting is economically unwise, as waste can be sent directly to an appropriate landfill disposal facility. Therefore, the purpose of the Waste Composition Identification module is to determine whether large volumes of waste warrant on-site sorting by efficiently and reliably measuring the IR.

[0045] Accurate assessment of IR is achieved by integrating big data analysis and computer vision. In terms of big data analysis, bulk density is used as a characteristic indicator to characterize the volume of construction waste in each disposal record. In materials science, bulk density is defined as the mass of many particles of a material divided by the total volume they occupy. The bulk density of construction waste volume is equal to the value of the waste weight divided by the waste volume. Although the bulk density of construction waste is a highly variable indicator, it should follow a certain probability distribution pattern in a specific area. With a sufficiently large data set, the distribution pattern can be recovered. With a detailed probability distribution of the bulk density of inert and non-inert construction waste, the composition of construction waste can be estimated according to the joint probability law.

[0046] In this embodiment, in step S10, the composition of construction waste, i.e., the quantitative ratio of inert components to non-inert components in the construction waste, is calculated by developing a composition estimation model, which includes the following steps:

[0047] S1011. Input the collected data to obtain the weight and calculation of the construction waste;

[0048] S1012, probability calculation, calculating the probability of each possible composition state;

[0049] S1013. Composition estimation to determine the composition of construction waste load.

[0050] The main steps for using this model to estimate composition are as follows: First, measure the weight and volume of construction waste. Using an electronic scale and laser sensor mounted on the rear of a vehicle, combined with the pre-recorded bucket bottom area, the volume and weight of the construction waste can be easily determined. Based on field observations in industry practice, using currently available technology and equipment, collecting weight and volume typically takes about one minute. The obtained weight and volume are then input into the developed model to calculate the composition of the construction waste, namely, the quantitative ratio of inert to non-inert components in the waste.

[0051] S20. Obtain visual difference features between the inert material and the non-inert material components using computer vision methods;

[0052] In this example, computer vision methods are used to identify inert materials (such as concrete and bricks) and non-inert materials (such as wood, plastic, and bamboo) in construction waste. Therefore, a photograph of a pile of construction waste can provide important visual cues for estimating its composition. A convolutional neural network (CNN) is used to extract the visual feature differences between inert and non-inert components.

[0053] This paper uses DenseNet169 to extract key high-level visual features for inert content measurement. Transfer learning techniques are applied to train a classification model whose input is a top-down photo of a truck captured by a surveillance camera. The model outputs a label indicating whether the truck transaction is completed. The model consists of a DenseNet169 structure pre-trained on ImageNet at the bottom, and three custom-designed layers at the top: a global average pooling layer and two fully connected layers (with 4,096 and 1 neurons, respectively). The neurons in the last fully connected layer use a sigmoid activation function and output a scaled value between 0 and 1. Based on this scale, the corresponding input image is assigned a class label: if the scale is at least 0.5, the input image belongs to the "Completed" class (indicating a composition with more than 50% inert content); otherwise, the image belongs to the "Rejected" class (indicating less than 50% inert content).

[0054] After step S20, the two single recognition results output by the component estimation model and the DenseNet169 model are sorted and weighed using an ensemble learning method, including the following steps:

[0055] S201, finding base classifiers with independent errors through experimental comparison;

[0056] S202, training the selected base classifier;

[0057] S203: Merge the results of the base classifiers. The merging method includes voting or stacking.

[0058] More specifically, in this embodiment, a voting method is used to merge the results of the big data model and the machine vision model.

[0059] Through integrated learning using big data and machine vision, the IR of construction waste can be accurately estimated. If the IR is approximately 100% or below 5%, the removable bucket can be removed from the device, and the waste inside does not need to undergo further screening. Otherwise, the outlet at the bottom of the bucket opens, allowing the waste to fall onto the conveyor below, which then guides the waste to the mechanical screening module. A 250mm mesh screen is installed at the bottom of the loading hopper to prevent large pieces of waste from entering the bucket and being processed by the mechanical screening module.

[0060] S30, mechanical screening, screening the construction waste through a mechanical screening module, the mechanical screening module including a pre-treatment module, a disc screening module, a magnetic drum sorting module and an air separation module vertically distributed from top to bottom;

[0061] Mechanical screening plays a vital role in centralized waste treatment facilities, where the large off-site spaces allow for extensive deployment of heavy machinery. However, available space on construction sites is often limited, making it difficult to accommodate the screening equipment used in traditional off-site environments. To address this issue, a novel vertical layout solution was proposed.

[0062] In this embodiment, a vertical layout is adopted, and four waste screening modules can be accommodated in a confined space. The four modules are a pre-treatment module, a disc screening module, a magnetic drum sorting module, and an air separation module.

[0063] As shown in Figures 1, 3, and 5, in this embodiment, the pre-processing module includes a vibrating conveyor belt 2-1, a spreading plate 2-2, a crusher 2-3, and a shredder 2-4. The vibrating conveyor belt 2-1 is inclined, and the spreading plate 2-2, crusher 2-3, and shredder 2-4 are arranged above the vibrating conveyor belt 2-1 along the inclination. The spreading plate 2-2 evenly distributes the construction waste onto the vibrating conveyor belt 2-1. After passing through the crusher 2-3 and shredder 2-4, the size of the construction waste is reduced to less than 150 mm. Therefore, the pre-processing module is designed to pre-process the incoming waste stream by applying mechanical force so that it can be more easily processed through the following procedures. Because the grid screen in the component identification module has already sorted out the bucky waste, the waste entering the pre-processing module is within 250 mm in size. The module first uses a scraper to flatten the incoming waste pile, evenly distributing it on the belt. After passing through the crusher and shredder, the size of the garbage is greatly reduced to less than 150 mm, which can prevent large pieces of garbage from getting stuck in the machine.

[0064] Furthermore, the disc screening module 2-5 is arranged below the vibrating conveyor belt 2-1, and a first conveyor belt is arranged on one side of the disc screening module 2-5. The first conveyor belt is used to receive the construction waste conveyed by the vibrating conveyor belt 2-1 and convey it to the disc screening module 2-5; the disc screening module 2-5 includes a plurality of rotating discs, and gaps are reserved between the discs to remove fine heavy materials with a size less than 50 mm, including sand, soil and dirt.

[0065] In addition, an inclined second conveyor belt is provided below the disc screening module 2-5, and the magnetic drum sorting module 2-6 is located above the bottom end of the second conveyor belt to realize the separation of metals in construction waste; the air separation module 2-7 includes an air separator, and the air separator is located below the bottom end of the second conveyor belt to realize the sorting of lightweight items; a feed conveyor belt 3-1 is provided below the air separator to receive construction waste dropped on the second conveyor belt.

[0066] S40, based on the precise sorting of robots, the waste on the conveyor belt is detected through waste detection driven by machine vision, and then the waste is located. Based on the results of the waste detection and positioning, multiple robot arms are used for precise sorting.

[0067] In this embodiment, as shown in FIG. 2 , a robot arm 3 - 2 is used to sort the waste, and the unsorted waste is returned via a re-sorting conveyor belt 3 - 3 for mechanical screening again.

[0068] The robotic sorting module uses computer vision to detect and locate individual garbage items on the conveyor belt, and uses a group of robotic arms to accurately sort the detected garbage items. Although pre-processed by the mechanical screening module, the size, shape and appearance of the waste still have large variability, which poses a challenge to the robotic arms to perform sorting operations. In order to meet this challenge, a flat push plate gripper is used to push the detected garbage into the corresponding garbage bin, thereby improving the accuracy and efficiency of sorting waste materials of different shapes and sizes. This embodiment uses the Pbox-8208G on-board dual-GPU workstation I9-9900K CAN-FD 2080S embedded industrial computer to realize the collaborative computing of multi-robotic arm recognition and manipulation. In actual applications, specific products can be flexibly selected according to the situation, and the patent of this invention is not limited to this.

[0069] The operation of the robotic sorting module is achieved by three interrelated tasks, namely machine vision-driven waste detection, waste positioning and multi-robot collaborative control.

[0070] (1) Machine vision driven scrap detection.

[0071] Before a robotic arm can accurately sort waste materials, it is necessary to first detect the waste on the conveyor belt. Compared to other sensing technologies such as hyperspectral imaging (HSI) and X-rays, machine vision offers the advantages of cost-effectiveness, ease of maintenance, and applicability to a wide range of waste materials. This patent utilizes the object detection model YOLO (You Only Look Once) for waste detection. To train the model, transfer learning techniques are used to reduce the number of required training samples, requiring only a limited number of new images of waste materials on the conveyor belt for model fine-tuning.

[0072] (2) Waste positioning.

[0073] After detecting the type of construction waste, the 3D position of the waste on the conveyor belt needs to be calculated so that the robot arm can subsequently grab it. A binocular vision solution is used to achieve waste positioning. A classic binocular vision system consists of two cameras that take pictures of the same scene (for example, the same garbage on a conveyor belt) from different positions and / or different perspectives. Based on the principle of triangulation positioning, the 3D spatial coordinates of the waste material can be calculated.

[0074] (3) Multi-robot collaborative sorting.

[0075] Based on the results of waste detection and localization, a group of robotic arms performs collaborative sorting. A robust collaborative control system was developed, allowing multiple robotic arms to work together to efficiently sort waste. The conveyor belt is controlled by an industrial computer, and its speed can be adaptively adjusted to meet different requirements, enhancing the robustness of sorting. The fundamental idea of ​​this solution is to avoid collisions when calculating the motion trajectories of the robotic arms. In our approach, the system is designed as a multi-agent reinforcement learning process, in which a decentralized policy is trained to control each robotic arm. This design is computationally more efficient than typical centralized methods. Furthermore, the reinforcement learning-based approach exhibits better scalability. The training process remains the same regardless of the number of robotic arms (less than 10).

[0076] Based on this, the present invention provides an on-site intelligent sorting method for construction waste, which is suitable for the classification of construction waste on compact construction sites. The device is driven by cutting-edge technologies such as artificial intelligence, computer vision, and big data analysis, and includes three interconnected modules: garbage component identification, mechanical screening, and robot-based precision sorting. The entire device is integrated on a detachable platform and can be easily installed on a common tonnage trailer, making it easy to move to various construction sites for screening. It adopts a vertical design and arranges various physical screening equipment vertically, which effectively saves the volume of the device and can be applied to typical small construction sites in high-density cities. The integration of artificial intelligence and robotic arm technology replaces manual quality control to ensure the accuracy and robustness of sorting.

[0077] This invention can improve the efficiency of on-site construction waste sorting, promote its widespread application, and positively impact construction waste management systems, thereby contributing to environmental protection. This invention surpasses existing technologies / products in two key respects. First, its compact layout and detachable design make it particularly suitable for work sites with limited space. Second, it integrates traditional mechanical physical screening technology with artificial intelligence-driven robotic sorting technology into a unified system; the efficiency of mechanical screening complements the precision of robotic sorting, ensuring the quality of recycled construction materials in a fully automated manner.

[0078] The above is a specific description of the preferred implementation of the present invention, but the invention is not limited to the embodiments. Those skilled in the art can make various equivalent modifications or substitutions without violating the spirit of the present invention. These equivalent modifications or substitutions are all included in the scope defined by the claims of this application.

Claims

1. A method for intelligent on-site sorting of construction waste, characterized in that: The method comprises the following steps: S10, identification of waste components, quantification of the proportion of inert materials in construction waste; S20. using computer vision methods to obtain visual difference features between inert material and non-inert material components; S30, mechanical screening, screening the construction waste through a mechanical screening module; S40, based on the precise sorting of robots, the waste on the conveyor belt is detected through waste detection driven by machine vision, and then the waste is located, and according to the results of the waste detection and positioning, multiple robot arms are used for precise sorting.

2. The method for on-site intelligent sorting of construction waste according to claim 1, characterized in that: In the step S10, the composition of the construction waste, i.e., the quantitative ratio of the inert components to the non-inert components in the construction waste, is calculated by developing a composition estimation model, which includes the following steps: S1011, input the collected data to obtain the weight and calculation of the construction waste; S1012, probability calculation, calculating the probability of each possible composition state; S1013. Composition estimation to determine the composition of construction waste load.

3. The method for on-site intelligent sorting of construction waste according to claim 2, characterized in that: In step S20, the DenseNet169 model is used to extract key high-level visual features for inert content measurement.

4. The method for on-site intelligent sorting of construction waste according to claim 3, characterized in that: After step S20, the two single recognition results output by the component estimation model and the DenseNet169 model are sorted and weighed by using an integrated learning method, including the following steps: S201, finding base classifiers with independent errors through experimental comparison; S202, training the selected base classifier; S203, merging the results of the base classifiers, wherein the merging method includes voting or stacking.

5. The method for on-site intelligent sorting of construction waste according to claim 1, characterized in that: In step S30, the mechanical screening module includes a pre-treatment module, a disc screening module, a magnetic drum sorting module and an air separation module which are vertically distributed from top to bottom; The pre-processing module includes a vibrating conveyor belt, a spreading plate, a crusher and a shredder; the vibrating conveyor belt is inclined, and the spreading plate, the crusher and the shredder are arranged above the vibrating conveyor belt along the inclined direction. The construction waste is evenly distributed on the vibrating conveyor belt through the spreading plate, and after passing through the crusher and the shredder, the size of the construction waste is controlled to be reduced to less than 150 mm.

6. The method for on-site intelligent sorting of construction waste according to claim 5, characterized in that: In the step S30, the disc screening module is arranged below the vibrating conveyor belt, and a first conveyor belt is arranged on one side of the disc screening module, and the first conveyor belt is used to receive the construction waste conveyed by the vibrating conveyor belt and convey it to the disc screening module; The disc screening module includes a plurality of rotating discs with gaps reserved between the discs to remove fine heavy materials with a size less than 50 mm, including sand, soil and dirt.

7. The method for on-site intelligent sorting of construction waste according to claim 6, characterized in that: In the step S30, an inclined second conveyor belt is arranged below the disc screening module, and the magnetic drum sorting module is located above the bottom end of the second conveyor belt to separate the metals in the construction waste; The air separation module includes an air separator, and the air separator is located below the bottom end of the second conveyor belt to achieve the separation of light objects; A feeding conveyor belt is arranged below the air separator to receive the construction waste dropped from the second conveyor belt.

8. The method for on-site intelligent sorting of construction waste according to claim 1, characterized in that: In step S40, the object detection model YOLO is used to perform waste detection to determine the type of waste.

9. The method for on-site intelligent sorting of construction waste according to claim 8, characterized in that: In the step S40, after the type of waste is detected, the three-dimensional position of the waste on the conveyor belt needs to be calculated; a binocular vision system is used to locate the waste and calculate the three-dimensional spatial coordinates of the waste.

10. The method for on-site intelligent sorting of construction waste according to claim 9, characterized in that: In step S40, a group of robotic arms are used to perform collaborative classification based on the results of waste detection and positioning, allowing multiple robotic arms to work collaboratively to effectively classify the waste.

Images