Construction site waste management method and system

Abstract

The present invention belongs to the field of intelligent construction, and provides a construction site waste management method and system, wherein various types of Internet of Things sensors are arranged on the construction site to collect real-time data of waste; the component ratio of the current construction site waste is determined based on the first model according to the real-time data; the parameters of the second model are adjusted based on the component ratio to improve the recognition weight and processing priority of the high-proportion components; and the construction site waste is sorted according to the instructions generated by the second model. The above scheme can improve the efficiency of sorting construction site waste.

Description

Technical Field

[0001] The present invention belongs to the field of intelligent construction, and in particular relates to a construction site waste management method and system. Background Art

[0002] With the acceleration of urbanization, the number and scale of construction projects are increasing, and the amount of construction site waste generated is also increasing. These wastes include concrete, bricks, steel bars, wood, glass, plastic and other components, which are complex, diverse and large in quantity. Traditional methods of handling construction waste mainly rely on manual sorting and simple mechanical separation, which is inefficient and prone to waste of resources. In addition, manual sorting is prone to misclassification or omissions, which affects the recycling and reuse of waste and may even have a negative impact on the environment.

[0003] The composition of waste generated at different construction stages of a construction site often varies significantly. For example, in the foundation construction stage, the waste mainly includes a large amount of concrete and steel bars; in the decoration and renovation stage, the waste is mainly bricks, tiles, gypsum boards, wood and plastic. Since the types and composition ratios of waste generated at different construction stages are different, it is often difficult to achieve ideal results using unified model parameters for sorting and identification. Specifically, when faced with waste with complex and diverse components, the model may experience a decrease in recognition accuracy and an increase in the misclassification rate, which in turn affects the overall sorting efficiency.

[0004] In existing construction waste management systems, model parameters are often fixed and cannot be adjusted dynamically according to changes in the construction phase. This type of fixed-parameter model performs poorly when dealing with changes in the proportion of waste components, especially when the proportion of components changes significantly, the model's recognition ability will drop significantly. For example, during the foundation construction phase, the model may have a high recognition accuracy for concrete and steel bars, but when faced with a large amount of wood and plastic generated during the decoration and renovation phase, the model may not be able to effectively recognize these materials because they account for a low proportion of the previous training data, resulting in low sorting efficiency. Summary of the invention

[0005] In order to solve the problems in the prior art, the present invention provides a construction site waste management method, which includes the following steps: arranging various types of Internet of Things sensors on the construction site to collect real-time data of waste; determining the component ratio of current construction site waste based on a first model according to the real-time data; adjusting the parameters of the second model based on the component ratio to improve the recognition weight and processing priority of high-proportion components; and sorting the construction site waste according to the instructions generated by the second model.

[0006] Another aspect of the present invention provides a construction site waste management system, which performs construction site waste management based on the above-mentioned method.

[0007] The present invention introduces a mechanism for dynamically adjusting model parameters, so that the system can adjust the weight of the recognition algorithm in real time according to the changes in the composition of waste generated at different construction stages. This adaptive adjustment mechanism significantly improves the accuracy of waste identification and sorting, especially in construction sites with complex and diverse components, and can effectively reduce misclassification and omission.

[0008] The present invention specifically considers the differences in waste composition generated at different construction stages, and dynamically adjusts model parameters to enable the system to adapt to the waste characteristics at different construction stages. This flexibility ensures that the system can maintain efficient sorting capabilities at all construction stages, thereby improving the overall efficiency of waste management.

[0009] The present invention uses a combination of convolutional neural networks (CNN) and reinforcement learning algorithms to automatically generate operating instructions for the manipulator, optimizing the grabbing and sorting paths of waste. The manipulator can perform sorting tasks accurately and efficiently based on the output of the model, reducing manual intervention and improving overall operating efficiency.

[0010] The present invention continuously optimizes the parameter settings of the model through real-time monitoring and feedback mechanisms, thereby improving the robustness and adaptability of the system in complex and ever-changing construction site environments. This self-learning and self-optimization function enables the system to maintain an efficient operating state as the type and quantity of waste change. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0012] Figure 1 is a flow chart of the method of the present invention. DETAILED DESCRIPTION

[0013] Below, the invention is preferably described in conjunction with the accompanying drawings and specific implementation methods.

[0014] This embodiment solves the above problem through the following steps:

[0015] In one embodiment, reference Figure 1The present invention provides a construction site waste management method, which refers to a technical solution for effectively managing various types of waste generated at a construction site (including but not limited to new construction, expansion, reconstruction, demolition and other engineering construction processes). The management method aims to classify, collect, transport, process and recycle the construction waste generated on the construction site through a series of technical means and operating procedures, so as to reduce the impact of waste on the environment, improve the recycling rate of waste, and reduce resource waste and management costs during the construction process.

[0016] Deploy various types of IoT sensors on construction sites to collect real-time data on waste. Specifically, it means deploying various types of IoT sensors at key locations on the construction site. These sensors are connected to the central data processing system through wireless or wired networks to monitor and collect relevant information on waste generation in real time.

[0017] IoT sensors refer to electronic devices used to sense and collect various information in the physical world (such as temperature, humidity, weight, location, composition, etc.). These sensors have networking capabilities and can transmit the collected data to the cloud or local server through the Internet of Things (IoT) technology for further analysis and processing. In construction site waste management, the types of IoT sensors may include but are not limited to visual sensors, weight sensors, spectral sensors, RFID sensors, position sensors, etc.

[0018] Various types of IoT sensors are deployed in different areas of the construction site (such as waste storage areas, construction operation areas, transportation channels, etc.). These sensors are distributed at various key points of the site according to different monitoring needs and functions. Visual sensors identify the type of waste by capturing images of the waste; weight sensors monitor the quality of the waste to count its quantity; spectral sensors determine its chemical composition by analyzing the spectral characteristics of the waste; RFID sensors are used to track and identify the source and flow of tagged waste; and position sensors record the spatial location and movement trajectory of the waste.

[0019] These sensors form an interconnected monitoring network that can collect and update relevant data on construction site waste in real time and continuously. Through this data, the system can instantly understand the generation of waste and provide data support for subsequent processing and management. This real-time data collection method based on IoT technology has significantly improved the scientificity and accuracy of construction site waste management, and helped to achieve efficient sorting and resource utilization of construction waste.

[0020] Various types of IoT sensors deployed on construction sites can collect specific data, including various attributes of waste and information related to site management:

[0021] 1. Waste type data

[0022] Visual data: Collect images and video data of waste through visual sensors (such as cameras, machine vision systems). This data can be used to identify the types of waste, such as concrete blocks, bricks, steel bars, wood, plastic, glass, etc.

[0023] Spectral data: Spectral characteristics of waste are collected using spectral sensors (such as near-infrared spectrometers and X-ray fluorescence spectrometers) to determine the chemical composition of the waste, such as different types of plastics, metals, and mineral components.

[0024] 2. Waste quantity and quality data

[0025] Weight data: The weight of each batch of waste is measured in real time through weight sensors installed at waste collection points, transport vehicles or sorting equipment. This data can be used to count the total amount of waste, the trend of the amount generated, and the distribution of different types of waste.

[0026] Volumetric data: Measure the volume of waste piles or individual waste items using volume sensors or laser rangefinders, especially for large or irregularly shaped waste items.

[0027] 3. Waste composition data

[0028] Chemical composition data: The chemical composition information of waste is obtained through spectral analysis instruments, such as metal content, mineral composition, organic matter content, etc. These data are crucial for identifying the reuse value of waste and the appropriate treatment methods.

[0029] Pollutant content data: Detects possible hazardous substances or pollutants in waste, such as heavy metals, volatile organic compounds (VOCs), asbestos, etc., so that appropriate protective measures can be taken during the treatment process.

[0030] 4. Waste location and source data

[0031] Location information: GPS devices or real-time location system (RTLS) tags are used to collect location data of waste within the construction site and track its entire life cycle from generation to final disposal. This is particularly suitable for large-scale construction sites or multi-regional waste management.

[0032] Source tracking data: RFID tags can be used to record and track the source of waste, including specific construction areas, work teams or equipment. This helps to understand the causes and responsibilities of waste generation and optimize construction processes and material usage.

[0033] 5. Time and frequency data

[0034] Generation time data: record the time when each batch of waste is generated, analyze the peak hours and periodic characteristics of waste generation, and optimize the treatment and transportation arrangements of waste.

[0035] Processing cycle data: Collect the time interval from waste generation to sorting and processing, evaluate the response efficiency of the waste management system, and identify potential delays or bottlenecks.

[0036] 6. Environmental monitoring data

[0037] Temperature and humidity data: Temperature and humidity sensors are used to monitor the climate conditions of the construction site environment. This data can be used to adjust treatment strategies, especially when handling waste materials that may be affected by environmental factors (such as wood and gypsum board).

[0038] Air quality data: Monitor the pollutant content (such as dust, PM2.5, volatile organic compounds) in the air on the construction site, especially in the waste treatment area. This data can be used to assess the impact of waste treatment on the environment and take necessary protective measures.

[0039] 7. Operation status and equipment data

[0040] Equipment operation status data: Real-time monitoring of the operation status of waste treatment equipment, including operating time, load conditions, energy consumption data, etc. This data is used for equipment maintenance, operation optimization and efficiency improvement.

[0041] Sensor self-test and calibration data: Collect sensor self-test and calibration data to ensure the accuracy and reliability of the collected data and warn of possible sensor failures.

[0042] The main components of the current construction site waste are determined based on the first model according to the real-time data.

[0043] The core of this step is to use the pre-trained and optimized first model to analyze and identify the components of waste generated at the construction site through multi-dimensional data collected in real time.

[0044] Optionally, the first model can be a machine learning model, which processes various types of real-time data collected by sensors, such as images, spectra, weight, location information, etc., to accurately determine the main components of the waste and their distribution ratios, thereby providing a scientific basis for subsequent waste classification, sorting and treatment.

[0045] The first model can be implemented using a multimodal deep learning model (such as a combination of a convolutional neural network (CNN) and a multilayer perceptron (MLP). This model can simultaneously process and fuse data from different types of sensors, automatically learn the characteristics of waste, and identify the components of waste based on these characteristics.

[0046] Model input:

[0047] Image data (visual sensor): The image data of waste is processed through a convolutional neural network (CNN) to extract its visual features such as shape, color, and texture.

[0048] Spectral data (spectral sensor): Spectral analysis technology is used to process the near-infrared spectrum or X-ray fluorescence spectrum data of waste to identify the chemical composition of the material.

[0049] Weight data (weight sensor): This helps the model determine the type and density of waste by processing the weight of the waste.

[0050] Location information (location sensor / GPS / RTLS): Use location information, combined with the geographical distribution characteristics of the construction site, to infer the source of waste and assist in component identification.

[0051] Source tracking data (RFID sensor): Through RFID tag information, historical data of waste is associated with its source, thereby improving identification accuracy.

[0052] Model output:

[0053] Main component categories of waste: The output of the model is the main component categories of waste, such as concrete, bricks and tiles, steel bars, wood, plastics, glass, etc.

[0054] Component ratio: The model can also output the proportion of each type of component to help further sorting and processing.

[0055] At a construction site, the system collected data on waste in real time through a variety of sensors and input the data into the first model for analysis. The site was demolishing concrete structures, which generated a large amount of mixed waste.

[0056] The visual sensor captured an image of a pile of waste, and the system used the CNN model to analyze the image and extract the edge, shape and color features of the object. Based on these features, the model initially determined that the waste may contain concrete blocks, bricks and tiles, and a small amount of metal materials (such as steel bars).

[0057] The spectral sensor scanned the waste, and the model used the spectral data to further confirm the chemical composition of the waste. The model detected specific spectral peaks, indicating that it did contain concrete (by detecting its silicate content), bricks and tiles (by detecting their aluminum oxide content), and steel bars (by detecting iron).

[0058] Weight sensors measured the total weight of the waste at 500 kilograms. Combining the visual and spectral data, the model inferred that most of this weight probably came from concrete blocks, while bricks and steel bars accounted for relatively small amounts.

[0059] Location information showed that the waste came from the southwest corner of the construction site, where the concrete wall was being demolished. RFID tag records showed that the waste had been transported through a specific route, which further verified the composition of the waste.

[0060] The model ultimately identified that the main component of the waste pile was concrete (about 70%), followed by bricks and tiles (about 20%) and steel bars (about 10%).

[0061] The output data shows that concrete accounts for the largest proportion, so in subsequent sorting, concrete should be separated for recycling.

[0062] Based on the component identification results output by the model, the system can automatically adjust the parameters of the sorting equipment, such as increasing the priority of concrete sorting and adding a magnetic separation process to steel bar sorting to improve sorting efficiency.

[0063] After identifying the main components, the waste can be sent to different recycling processes separately, such as crushing concrete blocks into aggregates, returning steel bars to the furnace for smelting, and using brick and tile fragments for paving base materials.

[0064] Through the above examples, the application of the first model in construction site waste management can effectively improve the accuracy of waste component identification and provide strong support for waste classification, treatment and resource utilization.

[0065] Adjusting the parameters of the second model based on the main component to increase the recognition weight and processing priority of the main component;

[0066] The core of this step is to dynamically adjust the parameter settings of the second model according to the main components of the waste identified by the first model, so as to optimize the strategy of waste sorting and treatment. The second model is mainly used to control and adjust the operating parameters of the sorting equipment, so as to improve the identification weight and treatment priority of these components in the sorting process according to the main components of the waste, so as to achieve efficient classification and resource processing of waste.

[0067] The second model uses an adaptive control model, combined with rule algorithms and reinforcement learning technology. The main function of this model is to adjust the operating parameters of the sorting equipment in real time based on the component identification results output by the first model to optimize the sorting and processing efficiency.

[0068] The overall architecture of the second model includes the following main parts:

[0069] Input layer: Receives real-time images of waste on the conveyor belt.

[0070] Recognition weight adjustment module: adjusts the weights in CNN based on the recognition results of the first model to improve the recognition accuracy of the main components

[0071] Image processing layer: Use convolutional neural network (CNN) to process and identify images and extract the characteristics of waste.

[0072] Decision-making layer: Based on the recognition results and combined with reinforcement learning, it generates operating instructions for controlling the robot.

[0073] Output layer: controls the motion parameters of the robot, including grasping position, force, and sorting path.

[0074] Specifically:

[0075] The input layer is responsible for receiving real-time image data of waste on the conveyor belt and passing this data to the subsequent image processing module.

[0076] A camera mounted above the conveyor belt captures high-resolution images at regular intervals (e.g., multiple frames per second). These images contain a real-time view of the waste on the conveyor belt.

[0077] In the input layer, the collected images are first subjected to basic preprocessing operations, including:

[0078] Normalize pixel values ​​to the range [0,1] to reduce data fluctuations.

[0079] If the images captured by different cameras have different sizes, they are uniformly adjusted to a predetermined input size (for example, 224x224 pixels).

[0080] Perform operations such as image rotation, flipping, and color adjustment to enhance the robustness of the model.

[0081] Represent the preprocessed image data as a tensor Where H is the image height, W is the image width, and C is the number of channels (for example, C=3 in RGB images).

[0082] The recognition weight adjustment module dynamically adjusts the weights in the convolutional neural network (CNN) based on the main component recognition results provided by the first model to improve the recognition accuracy of the main components.

[0083] Assume that the first model identifies the main components of the waste and their proportions, for example, concrete accounts for 70%, steel bars account for 20%, and bricks and tiles account for 10%.

[0084] According to these proportions, the convolution kernel weights corresponding to these components in CNN are adjusted. For example, the adjusted weights It can be calculated by the following formula:

[0085] ,

[0086] in, is the adjusted weight, is the basic weight, is the adjustment factor, is the proportion of the i-th component.

[0087] These adjusted weights are directly applied to the convolutional layers of the CNN, allowing the model to more accurately identify and classify the main components when processing images.

[0088] The image processing layer uses a convolutional neural network (CNN) to process and identify images on the conveyor belt and extract the features of the waste.

[0089] The image features are extracted through multiple convolutional layers, and each convolutional layer uses multiple convolution kernels 𝐾 to perform convolution operations on the input image.

[0090] Each convolutional layer is usually followed by a nonlinear activation function (such as ReLU) and a pooling layer (such as max pooling). The pooling layer is used to reduce the dimension of the feature map and retain the main features.

[0091] The pooled feature map is flattened and input into the fully connected layer, and finally the probability distribution of each type of waste is output through the Softmax layer.

[0092] The decision layer generates operating instructions to control the robot arm based on the recognition results of the image processing layer and combined with reinforcement learning.

[0093] Current Status By identification results (i.e. probability distribution of waste categories ) and the current operating parameters of the robot composition:

[0094] ,

[0095] Define the action space of the manipulator , each action contains the grasping position , Grasping strength , Sorting Road :

[0096] ,

[0097] in Indicates the current operating parameters. Indicates the current crawl position. Indicates the current grab strength. Indicates the current sorting route.

[0098] Setting the reward function , used to evaluate the effect of each operation. The reward function includes positive rewards for successful grasping, penalties for operation time, and penalties for wrong operations:

[0099] ,

[0100] in is the positive reward for a successful capture, It is a penalty for operation time. It is a punishment for misoperation

[0101] Use the deep Q network to update the strategy and select the best action for the current state:

[0102] ,

[0103] in, is the current Q value, is the historical maximum Q value, and the current Q value determines the current optimal action , is the weight.

[0104] The output layer converts the instructions generated by the decision layer into specific operations of the robot, controlling the robot to perform the sorting task. According to the optimal action, specific instructions for controlling the robot are generated, including moving to a specified position, adjusting the gripping force, and selecting a sorting path.

[0105] For example, on a construction site waste disposal line, mixed construction waste is transported on a conveyor belt. The main components include concrete, steel bars, and bricks and tiles. The first model identifies that concrete accounts for 70%, steel bars account for 20%, and bricks and tiles account for 10%. The following is the specific implementation process of the second model:

[0106] The image data received by the input layer is sent to CNN for processing after preprocessing. At the same time, the input layer receives the main component proportion data P={0.7,0.2,0.1} identified by the first model as a reference.

[0107] The recognition weight adjustment module adjusts the recognition weight of concrete from the basic weight of 0.5 to 0.85 based on the high proportion of concrete components (70%). This adjustment ensures that CNN has a higher recognition accuracy for concrete when processing images.

[0108] The image processing layer uses the adjusted CNN weights to process the image

[0109] Convolution is performed to extract the features of the waste. After multiple layers of convolution, pooling, and full connection, the model outputs the classification probability of the waste {0.75, 0.15, 0.1}, which means that the model recognizes that there is a 75% probability that it is concrete.

[0110] The decision layer calculates the Q value of each possible action based on the recognition results and the current state of the robot through a reinforcement learning algorithm and selects the optimal grasping action.

[0111] The output layer will select the optimal action

[0112] It is converted into specific control instructions, directing the robot to move to the grasping position, apply appropriate grasping force, and sort the concrete to the designated area along the predetermined path.

[0113] Through this series of operations, the robot is able to prioritize and sort the main component (concrete), improving the efficiency and accuracy of waste treatment.

[0114] The construction site waste is sorted according to the second model.

[0115] In the above steps, the operating instructions for controlling the manipulator are generated. These instructions include key parameters such as the manipulator's grasping position, grasping force, and transportation path. After receiving these instructions, the manipulator accurately performs grasping and sorting operations, separates different types of waste and transports them to the designated processing area. The entire process is continuously optimized through real-time monitoring and feedback mechanisms to ensure that the waste sorting process is efficient and accurate, and can adapt to the changes in different batches of waste, thereby realizing the automated and efficient management of construction site waste.

[0116] On the other hand, the present invention also provides a construction site waste management system, which uses the above method to perform waste sorting management on the construction site.

[0117] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

[0118] The present invention does not particularly specify the structure of some modules, which shall be subject to the contents recorded in the prior art. The prior art mentioned in the above background technology section and the specific embodiment section of the present invention can be used as part of the present invention to understand the meaning of some technical features or parameters. The scope of protection of the present invention shall be subject to the contents actually recorded in the claims.

Claims

1. A method for managing construction site waste, characterized in that The method comprises the following steps: arranging various types of IoT sensors on a construction site to collect real-time data on waste; Determine the composition ratio of current construction site waste based on the first model according to the real-time data; Adjusting the parameters of the second model based on the component ratio to increase the recognition weight and processing priority of the high-ratio components; sorting the construction site waste according to the instructions generated by the second model; The real-time data includes: Waste type data, waste quantity and quality data, waste composition data, waste location and source data, timing and frequency data, environmental monitoring data, and operating status and equipment data; The first model is a multimodal deep learning model; The first model input is the real-time data; The output of the first model is the component categories and component ratios of the waste; The architecture of the second model includes the following parts: Input layer: receiving real-time images of waste on the conveyor belt; Recognition weight adjustment module: adjusts the weights in CNN based on the recognition results of the first model to improve the recognition accuracy of high-proportion components; Image processing layer: Use convolutional neural network (CNN) to process and identify images and extract features of waste; Decision layer: Based on the recognition results, combined with reinforcement learning, the operation instructions for controlling the robot are generated; Output layer: Controls the motion parameters of the robot, including the grasping position, force and sorting path; The real-time image is a high-resolution image collected at fixed time intervals by a camera installed above the conveyor belt; The recognition weight adjustment module performs weight adjustment using the following formula: In new,i =in base,i +α×p i Among them, w new,i is the adjusted weight, w base,i is the basic weight, α is the adjustment coefficient, p i is the proportion of the i-th category component; The decision layer generates operation instructions for controlling the robot based on the recognition results of the image processing layer and combined with reinforcement learning; Current Status t By identification results and the current operating parameters of the robot a t composition: Define the action space of the manipulator A = {a1, a2, ..., a k }, each action includes the grasping position x, grasping force f, and sorting path π: a t =(x t ,f t ,p t ) Set the reward function R(s t ,a t ) is used to evaluate the effect of each operation. The reward function includes positive rewards for successful grasping, penalties for operation time, and penalties for misoperation: R(s t ,a t )=r success -c time -c error where r success is the positive reward for successful capture, c time is the penalty for operation time, c error It is the punishment for wrong operation; Use the deep Q network to update the strategy and select the best action for the current state: Among them, Q(s t ,a t ) is the current Q value, is the historical maximum Q value, and the current Q value determines the current optimal action a t ; The output layer converts the instructions generated by the decision layer into specific operations of the robot, controls the robot to perform sorting tasks, and generates specific instructions to control the robot based on the optimal actions, including moving to the specified position, adjusting the grasping force, and selecting the sorting path.

2. The construction site waste management method according to claim 1, characterized in that: After receiving specific instructions, the robot performs grabbing and sorting operations to separate different types of waste and transport them to designated processing areas.

3. A construction site waste management system, characterized in that The system performs waste sorting according to the method according to any one of claims 1-2.

Images