Intelligent sorting and cyclic regeneration system for construction waste

Abstract

The invention relates to the technical field of fault prediction, and discloses a construction waste intelligent sorting and cyclic regeneration system which comprises an acquisition subsystem used for installing sensor arrays on the working face of a crusher and the edge of a vibrating screen of a construction waste sorting device and acquiring a feature data set; the simulation subsystem is used for establishing a simulation model of the construction waste sorting device and generating a simulation data set; the actual measurement subsystem is used for collecting an actual measurement abrasion data set with labels under the actual construction waste sorting working condition; the training subsystem is used for inputting the simulation data set and the actually measured wear data set with the label into an adaptive domain adversarial neural network for training to obtain an equipment wear degree prediction model; and the output subsystem is used for identifying wear fault characteristics through the equipment wear degree prediction model and outputting a fault closed-loop response scheme, so that the feeding rate, the crushing rotating speed, the screening frequency and the screening amplitude of the sorting device are dynamically adjusted.

Description

Technical Field

[0001] The present application relates to the technical field of fault prediction, and in particular to an intelligent sorting and recycling system for construction waste. Background Art

[0002] The effective sorting and recycling of construction waste can not only reduce environmental pollution, but also create significant economic value, and has become a hot area of ​​global concern. However, the existing construction waste treatment equipment faces severe challenges in the long-term operation process, especially the key parts such as the crusher working surface and the edge of the vibrating screen are prone to "excessive wear" during the long-term contact with irregular construction waste materials, resulting in reduced equipment stability, increased maintenance costs, and reduced processing efficiency, which seriously affects the sorting quality and the quality of recycled materials.

[0003] At present, there are obvious deficiencies in the fault prediction of construction waste sorting and recycling systems. Traditional mechanical equipment fault prediction mainly relies on empirical models or pure data-driven methods. The former oversimplifies the complexity of material-equipment interaction and cannot accurately reflect the irregular shape and diversity characteristics of construction waste materials; the latter requires a large amount of historical fault data as support, but it is difficult to obtain sufficient fault samples in actual engineering applications, resulting in poor generalization ability under small sample conditions. In addition, existing fault prediction methods often process data from different sources in isolation, lack an effective multi-source data fusion mechanism, and are difficult to fully capture the multi-dimensional characteristics of equipment wear, which in turn affects the accuracy of prediction. Summary of the invention

[0004] The present application provides an intelligent sorting and recycling system for construction waste, thereby realizing dynamic adjustment of the feed rate, crushing speed, screening frequency and screening amplitude of the sorting device, forming a complete fault closed-loop response solution, which effectively extends the service life of the equipment while ensuring the sorting efficiency.

[0005] The first aspect of the present application provides a construction waste intelligent sorting and recycling system, the construction waste intelligent sorting and recycling system comprising: A collection subsystem, which is used to install a sensor array on the crusher working surface and the edge of the vibrating screen of the construction waste sorting device, and collect a characteristic data set reflecting the wear status of the equipment; A simulation subsystem, used for establishing a simulation model of the construction waste sorting device according to the characteristic data set, and generating a simulation data set containing a combination of multiple working condition parameters based on the simulation model; The measurement subsystem is used to collect small sample measured data under actual construction waste sorting conditions and calibrate the wear level to obtain a labeled measured wear data set; A training subsystem, used for inputting the simulation data set and the measured wear data set with labels into an adaptive domain adversarial neural network for training, so as to obtain a prediction model for the degree of equipment wear; The output subsystem is used to identify wear fault characteristics through the equipment wear degree prediction model, and based on the wear fault characteristics, optimize the operating parameters of the construction waste sorting device and output a fault closed-loop response plan.

[0006] Compared with the prior art, the present application has the following beneficial effects: by configuring a multi-dimensional sensor array of pressure sensors, triaxial acceleration sensors, displacement sensors and deep vision camera modules on the working surface of the crusher and the edge of the vibrating screen, and combining wavelet transform noise reduction, time synchronization calibration and Kalman filter algorithm for multi-source data fusion, high-precision monitoring of the wear state of the equipment is achieved, and complex wear characteristics that cannot be recognized by traditional single sensors are captured. The simulation model constructed based on finite element analysis, Archard wear theory and discrete element method is combined with small sample measured data to form a mechanism-data hybrid driven fault prediction method, which effectively solves the problem of poor generalization ability of pure data driven methods for small sample data, and avoids the prediction bias caused by over-simplification of pure mechanism models. By designing an adaptive domain adversarial neural network including feature extractors, domain classifiers and wear predictors, effective knowledge transfer from the simulation data domain to the measured data domain is achieved, and a large amount of simulation data is fully utilized to improve the training effect of the model, while maintaining adaptability to actual working conditions, significantly improving the generalization ability of the fault prediction model. The deep vision camera module is used to collect point cloud data, and the three-dimensional convolutional neural network and variational autoencoder are combined for feature extraction and dimensionality reduction. The orthogonal constraint and KL divergence optimization are used to generate low-dimensional wear fault features with high discrimination. The Takagi-Sugeno fuzzy model is used to describe the nonlinear dynamic characteristics of the sorting system. The sensor acquisition frequency is grouped in the multi-rate sampling framework. The feedback control gain is determined by solving the linear matrix inequality optimization problem, which effectively copes with the uncertainty caused by sensor failure and time delay, and improves the stability and reliability of the system under complex working conditions. Based on the wear fault risk assessment results and the basic strategy of parameter optimization, the Bayesian optimization system is combined to solve the optimal operating parameter combination, and the dynamic adjustment of the feed rate, crushing speed, screening frequency and screening amplitude of the sorting device is realized, forming a complete fault closed-loop response plan, which effectively extends the service life of the equipment while ensuring the sorting efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0008] The structures, proportions, sizes, etc. illustrated in the drawings of this specification are only used to match the contents disclosed in the specification so as to facilitate understanding and reading by persons familiar with this technology. They are not used to limit the conditions under which this application can be implemented, and therefore have no substantive technical significance. Any structural modification, change in proportion or adjustment of size, without affecting the effects and purposes that can be achieved by this application, should still fall within the scope of the technical contents disclosed in this application.

[0009] Figure 1 It is a flow chart of the intelligent sorting and recycling system for construction waste provided in the embodiment of the present application. DETAILED DESCRIPTION

[0010] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0011] The flowcharts shown in the accompanying drawings are only examples and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may also be decomposed, combined or partially merged, so the actual execution order may change according to actual conditions.

[0012] It should also be understood that the terms used in this application specification are only for the purpose of describing specific embodiments and are not intended to limit the application. As used in this application specification and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0013] It should be further understood that the term "and / or" used in the specification and appended claims of this application refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations. Figure 1 In the embodiment of the present application, an embodiment of the intelligent sorting and recycling system 10 for construction waste includes: The acquisition subsystem 11 is used to install a sensor array on the crusher working surface and the edge of the vibrating screen of the construction waste sorting device, and to collect a characteristic data set reflecting the wear status of the equipment; In this embodiment, a variety of sensing devices are reasonably arranged on the inner wall of the crusher working surface and the edge of the vibrating screen of the construction waste sorting device, including pressure sensors, three-axis acceleration sensors, displacement sensors and deep vision camera modules, to obtain a multi-dimensional sensor array to obtain the changes of different physical quantities during the operation of the equipment. The pressure sensor is used to monitor the force changes on the inner wall of the crusher, so as to evaluate the working state and wear degree of the crusher; the three-axis acceleration sensor captures the vibration information of the equipment in all directions, which is helpful for analyzing the stability of the equipment operation and abnormal vibration caused by wear; the displacement sensor is used to measure the relative displacement of key components, which is helpful for detecting problems such as loosening and offset of mechanical components caused by wear; the deep vision camera module monitors the surface morphology of the equipment by acquiring three-dimensional point cloud data, especially at the edge of the vibrating screen, and the mesh changes and structural damage caused by wear can be more intuitively observed through visual information. Differentiated sampling frequencies are set for the multi-dimensional sensor array to ensure the collection efficiency and accuracy of various sensor data. The acquisition subsystem sets the sampling frequency according to the characteristics of each type of signal, so that the sampling frequency of the pressure sensor is moderate and the trend of the force change of the equipment is captured. The three-axis acceleration sensor uses a higher sampling frequency to accurately record the high-frequency vibration characteristics of the equipment. The sampling frequency of the displacement sensor is set according to the speed and change amplitude of the equipment movement. The deep vision camera module adjusts the frame rate according to the movement state of the equipment to obtain the original data set of equipment wear, including pressure data, vibration data, displacement data and three-dimensional point cloud data. The original data of equipment wear is subjected to wavelet transform denoising. Wavelet transform is an effective signal processing method. By multi-scale decomposition of the signal, the signal is decomposed into components of different frequencies, and the noise component is filtered out by setting an appropriate threshold to retain the effective signal characteristics. When performing wavelet transform denoising, the appropriate wavelet basis function is selected according to the characteristics of different sensor data. For example, for vibration data, a wavelet basis that can finely distinguish high-frequency signals is selected, while for pressure and displacement data, a wavelet basis with better smoothness is selected to obtain a denoised data segment set. Time synchronization calibration is performed on the denoised data segment set. The acquisition subsystem uses time synchronization calibration technology to align data from different sources to a unified time axis, achieve accurate correspondence of each sensor data in the time dimension, and generate a standardized data set. Features that effectively reflect the wear status of the equipment are extracted from the standardized data set, including time domain features, frequency domain features, and time-frequency domain features. Time domain features include the mean, variance, peak value, peak-to-peak value, and signal energy of the signal, reflecting the changes in physical quantities of the equipment during operation. For example, the peak-to-peak value change of the vibration signal indicates looseness or damage of mechanical parts. Frequency domain features obtain the energy distribution of the signal on different frequency components by performing Fourier transform on the signal. For example, by analyzing the spectral characteristics of the vibration signal, the resonance phenomenon of a specific frequency caused by wear can be identified.The time-frequency domain features use short-time Fourier transform or wavelet transform to achieve joint analysis of signals in both time and frequency dimensions. They are suitable for processing non-stationary signals, such as vibration signals of equipment when the load changes. Time-frequency analysis can be used to capture the changes in frequency characteristics over time during equipment wear. The Kalman filter algorithm is used to fuse multi-source data for multi-dimensional feature vectors. Kalman filtering is a recursive estimation method that effectively eliminates noise during multi-dimensional data fusion and predicts the wear status of the current equipment through a dynamic model. Through Kalman filtering, data from different sensors are effectively fused in the spatial and temporal dimensions to generate a more stable and accurate feature data set.

[0014] The simulation subsystem 12 is used to establish a simulation model of the construction waste sorting device according to the characteristic data set, and generate a simulation data set containing a combination of multiple working condition parameters based on the simulation model; In this embodiment, based on the characteristic data set, the simulation subsystem uses the finite element analysis method to establish a digital model of the equipment structure of the crusher and the vibrating screen. In the process of model building, the key components of the modeled equipment include the rotor, hammer head, liner of the crusher, and the screen and support structure of the vibrating screen. In order to improve the accuracy of the model, the equipment structure model takes into account the material properties, geometry, connection method and boundary conditions of the actual equipment, and discretizes the complex structure into finite units through meshing. These units reflect the stress and strain distribution inside the equipment during the force and deformation process. After the digital model of the equipment structure is established, the simulation subsystem constructs an equipment wear progression model based on the Archard wear theory and cumulative damage theory. Archard wear theory points out that the amount of wear is proportional to the contact pressure, relative movement distance and the inverse of the material hardness. When establishing the wear model, parameters such as wear coefficient (K), contact pressure (P), sliding distance (L) and material hardness (H) are introduced. The wear amount (W) is expressed as . On this basis, combined with the cumulative damage theory, the cumulative effect of material fatigue and micro cracks in the long-term operation of the equipment is considered, and a prediction model for the progress of equipment wear is established, so that the model is not only suitable for short-term wear prediction, but also can evaluate the life attenuation trend of the equipment under long-term use. At the same time, in order to accurately simulate the interaction between construction waste and equipment in the simulation, discrete element modeling of construction waste materials is carried out. In the simulation subsystem, the physical properties of construction waste such as density distribution, particle size distribution, shape factor, elastic modulus and friction coefficient are parameterized to form a discrete element model of construction waste materials. The discrete element method regards construction waste as a large number of interacting particle systems. By defining the contact mechanics model and collision recovery coefficient between particles, the movement, collision, crushing and screening process of materials in the crusher and vibrating screen are simulated. And by introducing particle shape factors (such as spherical, lamellar, prismatic, etc.), the force and motion behavior of different types of construction waste inside the equipment are more realistically reproduced to ensure the credibility of the simulation results. The discrete element method is used to simulate the dynamic contact process of the discrete element model of construction waste materials and the digital model of equipment structure. During the simulation process, the local stress distribution and contact time data generated by the contact between particles and the equipment surface are used to analyze the degree of wear on the equipment surface material caused by the impact and friction of construction waste. The simulation subsystem generates stress distribution cloud maps, contact wear heat maps, and material damage accumulation curves of the equipment under different working conditions through multi-step iterative calculations. In order to achieve full process automation and multi-condition evaluation of simulation calculations, the simulation subsystem constructs a complex interactive process simulation system, which includes functional modules such as material input module, crushing dynamics module, screening dynamics module, wear calculation module, equipment response module, parameter mapping module, and data output module. The material input module imports the material parameters of construction waste with different characteristics into the simulation environment; the crushing dynamics module simulates the collision and crushing behavior of materials in the crusher; the screening dynamics module is used to calculate the screening efficiency and particle distribution of materials in the vibrating screen; the wear calculation module dynamically evaluates the wear of key components of the equipment based on the Archard wear model and the cumulative damage model; the equipment response module uses finite element analysis to provide real-time feedback on the deformation and stress response of the equipment during the force process; the parameter mapping module matches the equipment operating parameters with the wear characteristics to achieve quantitative analysis of the equipment wear state; the data output module outputs the equipment wear data, material distribution data, and equipment stress response data obtained by simulation calculation as a standardized simulation data set. The simulation subsystem designs multiple sets of different working condition parameter combinations through the control variable method. In different simulation experiments, the equipment operating parameters (such as crusher speed, vibrating screen frequency, feed speed, etc.) and material characteristic parameters (such as density, particle size, shape factor, etc. of construction waste) are adjusted respectively. By comparing and analyzing the wear degree of the equipment under different working conditions, a simulation data set containing equipment operating parameters, material characteristic parameters and corresponding equipment wear degree is generated.This dataset contains the wear data of the equipment under normal operating conditions, as well as the wear prediction results under extreme conditions.

[0015] The measurement subsystem 13 is used to collect small sample measured data under actual construction waste sorting conditions and calibrate the wear level to obtain a labeled measured wear data set; In this embodiment, multiple target devices are selected as monitoring objects on the actual construction waste sorting line, and these target devices include key equipment such as crushers, vibrating screens, and conveying devices. The actual measurement subsystem configures sensors according to the installation scheme of the sensor array, including installing pressure sensors, triaxial acceleration sensors, displacement sensors, and depth vision camera modules at key parts of the equipment such as the inner wall of the crusher working surface and the edge of the vibrating screen, to obtain an actual measurement monitoring system. After completing the sensor configuration, the actual measurement subsystem sets multiple groups of different typical working condition parameters for the actual measurement monitoring system, such as the running speed of the equipment, feed amount, material type, equipment load, etc., to cover various working conditions encountered by the construction waste sorting equipment in actual production. The actual measurement subsystem collects the surface morphology data of the equipment and various sensor data in real time, including pressure data, vibration data, displacement data, and three-dimensional point cloud data, by continuously monitoring the operation of the equipment under different working conditions, to ensure the comprehensiveness and multi-dimensional characteristics of data collection. In order to ensure the continuity and accuracy of the data, the system sets a regular recording mechanism during the operation of the equipment, such as automatically recording data through periodic triggering or event-driven methods, to obtain the measured original data under multiple working conditions. Time domain features are extracted from the measured raw data to obtain the time series characteristics of signal changes during equipment operation. Time domain feature extraction includes the mean, variance, peak, root mean square value, skewness, kurtosis and envelope of the signal. These time domain features reflect the dynamic changes in the equipment wear process. At the same time, in order to capture the frequency characteristics and multi-scale change information of the equipment wear state, the measured subsystem performs fast Fourier transform and wavelet packet decomposition on the raw data to obtain frequency domain features and time-frequency domain feature sets. Fast Fourier transform converts time domain signals into frequency domain signals and analyzes the energy distribution of signals at different frequencies. Wavelet packet decomposition provides characteristic information of signals in two dimensions, time and frequency, through multi-scale decomposition, which helps to identify the changing characteristics of non-stationary signals during equipment wear. The three-dimensional point cloud data obtained by the deep vision camera module is used to perform a refined analysis of the surface morphology of the equipment. By calculating the surface roughness index in the point cloud data, the roughness change of the equipment surface caused by wear is quantified; by analyzing the main curvature distribution of the point cloud model, the local details of the surface morphology change of the equipment, such as pits and scratches caused by wear, are detected; by calculating the area and depth distribution of the wear area, the material loss of the equipment is quantified. For example, for a crusher hammer or a vibrating screen, the distribution of wear depth is used to evaluate the service life of the equipment and determine when the equipment needs to be replaced or repaired. These features extracted from the three-dimensional point cloud data are classified into a depth feature set. The most representative features are selected from the time domain feature set, the frequency domain and time-frequency domain feature set, and the depth feature set. In the feature selection process, the measurement subsystem uses statistical analysis, principal component analysis, or information gain-based methods to ensure that the final selected features provide the most effective input for the equipment wear prediction model.In order to convert these characteristic data into the actual degree of equipment wear, the measurement subsystem combines visual inspection and laser scanning data to calibrate the degree of equipment wear. During the visual inspection process, the wear level of the equipment is preliminarily evaluated based on the wear of the equipment surface (such as color change, cracks, deformation, etc.). At the same time, the laser scanning equipment obtains the changes in the geometric morphology of the equipment surface through high-precision scanning data, and compares it with the three-dimensional point cloud data to provide a quantitative basis for the evaluation of the wear level. For example, by analyzing the surface profile of the equipment obtained by scanning, measuring the depth change of the wear area, and comparing these data with the standard wear level, the degree of equipment wear is finally divided into different levels (such as light, moderate, severe wear, etc.).

[0016] A training subsystem 14 is used to input the simulation data set and the measured wear data set with labels into the adaptive domain adversarial neural network for training to obtain a prediction model for the degree of equipment wear; In this embodiment, in the design subsystem, an overall neural network architecture including a feature extractor, a domain classifier and a wear predictor is constructed. As the core part of the network, the feature extractor maps the input multi-dimensional feature data to a high-dimensional feature space. Its internal structure consists of multiple fully connected layers, batch normalization layers and nonlinear activation functions (such as ReLU or Leaky ReLU). The fully connected layer extracts the deep features of the data through linear transformation, the batch normalization layer is used to stabilize the training process and avoid the problem of gradient disappearance or explosion, and the activation function introduces nonlinearity so that the model learns complex feature representations. The multi-level feature extractor can effectively extract common features from simulation data and measured data, and retain unique information about the wear state of the equipment. After completing the network structure design, the adversarial unit combines the wear prediction task with the domain adversarial task by defining a multi-task loss function to achieve multi-objective optimization of the model. The wear prediction loss calculates the difference between the predicted value of the equipment wear degree and the actual labeled value through the mean square error or cross entropy loss function. The goal of this loss term is to improve the prediction accuracy of the model for labeled measured data. The domain adversarial loss predicts the source of the input features (simulated data or measured data) through the domain classifier, and uses the gradient reversal layer to achieve adversarial training. The gradient reversal layer does not change the data flow during forward propagation, but reverses the gradient sign during back propagation, so that the feature extractor tends to produce feature representations that "confuse" the domain classifier during training, thereby reducing the difference between the feature distribution of simulated data and measured data, and achieving the effect of domain adaptation. In order to ensure the stability and generalization ability of model training, the partitioning unit divides the data set, divides the simulated data set into the first training set and the validation set to ensure that the network can fully learn features on a larger data set, and evaluates the generalization ability of the model on the validation set. At the same time, the labeled measured wear data set is divided into the second training set and the test set. The second training set is used for joint training of the model, and the test set is used to evaluate the wear prediction ability of the model under real working conditions. In the partitioning process, the random distribution of data and the consistency of data distribution are guaranteed to avoid the risk of data leakage and overfitting. After completing the data set partitioning, the joint unit uses the first training set to pre-train the neural network. The pre-training stage focuses on the basic feature learning ability of the feature extractor. Through a large amount of simulation data training, the feature extractor learns to extract common features related to the wear status of the equipment. When the pre-training is completed, the joint unit extends the model training to the labeled measured wear data and starts the joint training process by inputting the second training set into the network. In the joint training, the model continues to optimize the wear prediction task and gradually introduces domain adversarial tasks. Through the dynamic weight adjustment strategy, the domain adversarial loss weight is gradually increased, so that the model can maintain adaptability to the characteristics of simulation data while learning the characteristics of measured data.The dynamic weight adjustment strategy adopts a linear increase or exponential increase method, which makes the domain adversarial task gradually occupy a larger proportion in the later stage of training, ensuring that the model achieves the domain adversarial effect on the basis of maintaining the prediction accuracy. During the entire training process, the optimization goal of the model is to minimize the wear prediction loss and maximize the error rate of the domain classifier at the same time. This adversarial training mechanism enables the feature extractor to continuously adjust the output features to achieve the purpose of "confusing" the domain classifier, and finally achieve a seamless fusion of the feature distribution of simulation data and measured data. When the training process is completed, the model verifies its prediction accuracy and generalization ability through the test set, and obtains an equipment wear degree prediction model with good domain adaptability and high-precision wear prediction ability.

[0017] The joint unit initializes the parameters of the adaptive domain adversarial neural network, which involves initializing all weights and bias parameters of the feature extractor, domain classifier, and wear predictor. Xavier initialization or He initialization method is used to make the initial parameters of the model within a reasonable distribution range, so as to avoid the initial state of the network deviating too much from the optimal solution, affecting the convergence speed of training and model performance. During the initialization process, the initial reversal coefficient is set for the gradient reversal layer to minimize the impact of the domain adversarial loss in the early stage of training, so that the model can preferentially learn the effective features in the wear prediction task. After completing the model parameter initialization, the joint unit inputs the first training set into the adaptive domain adversarial neural network in batches, and only updates the parameters of the feature extractor and wear predictor. The parameters of the domain classifier remain frozen at this stage. The core goal of this pre-training stage is to enable the feature extractor to learn effective features related to the wear state of the device by minimizing the wear prediction loss without domain adversarial interference. By adopting mini-batch stochastic gradient descent or Adam optimizer, the model updates the parameters of the feature extractor and wear predictor in each batch training, gradually reducing the verification loss. During the training process, when the validation loss of the model shows stability on the validation set, it means that the feature extractor has the preliminary wear feature extraction capability, and the pre-training model is considered to be completed. After completing the pre-training model, the joint unit constructs a data loader, marks the first training set as source domain data, marks the second training set as target domain data, and mixes the source domain and target domain data to generate batch data streams by setting the same proportion of sampling strategies. The mixed batches in this process ensure that each batch contains samples from simulation data (source domain data) and measured data (target domain data), so that the model can learn the characteristics of both data distributions in subsequent adversarial training. The data loader uses random shuffling and batch balancing techniques to make the source domain and target domain data in each batch as evenly distributed as possible, thereby avoiding the model from being biased towards a certain data domain during the training process. In the domain adversarial training stage, the joint unit inputs each batch of data in the domain adversarial training data stream into the adaptive domain adversarial neural network, and converts the input data into high-dimensional feature vectors through the feature extractor. These feature vectors are then simultaneously input into the domain classifier and wear predictor to obtain domain prediction results and wear prediction results, respectively. The goal of the domain classifier is to determine whether the features of the input data are from the source domain or the target domain, while the wear predictor predicts the degree of wear of the equipment. In this process, the feature extractor plays an "adversarial role". Under the action of the gradient reversal layer, the feature extractor must ensure the accuracy of the wear prediction and "confuse" the domain classifier so that the domain classifier cannot accurately determine the source domain of the data, thereby achieving domain adaptation. When calculating the loss function, the joint unit combines the wear prediction loss and the domain adversarial loss to calculate the total loss.The wear prediction loss uses the mean square error or cross entropy loss function to measure the deviation between the wear degree predicted by the model and the actual label, while the domain adversarial loss is calculated based on the cross entropy loss between the output of the domain classifier and the true domain label. In the loss function, the weight coefficient (λ) of the domain adversarial loss is not fixed, but dynamically adjusted according to the current training progress. In the initial stage of model training, the value of λ is small, so that the model prioritizes the wear prediction task. As the training progresses, the value of λ gradually increases, and the model gradually increases its attention to the domain adversarial task through a linear or exponential growth strategy. During the training process, after every 50 rounds of training, the joint unit decays the learning rate to the original 0.1. The learning rate decay strategy is adopted to effectively prevent the model from oscillating when it is close to convergence, which helps the model find a better local minimum. At the same time, after each learning rate decay, the model calculates the wear prediction error and the domain classification confusion matrix on the test set, and judges the training status of the model through evaluation indicators. When the wear prediction error is less than the set threshold and the domain classification accuracy is close to the random guessing level (i.e., close to 50%), it means that the feature extractor has successfully "confused" the domain classifier and achieved the ideal domain adaptation effect. When the model shows a stable low wear prediction error on the test set and the domain classifier cannot effectively distinguish between the source domain and the target domain data, the joint unit obtains a dynamically adjusted cross-domain migration model. This model can accurately predict the degree of wear of the equipment under actual working conditions and has strong generalization ability. It can transfer the features learned from the simulation data to the measured data, thereby realizing the intelligent, automated and efficient monitoring and prediction of equipment wear in the intelligent sorting and recycling system of construction waste.

[0018] The output subsystem 15 is used to identify the wear fault characteristics through the equipment wear degree prediction model, and based on the wear fault characteristics, optimize the operating parameters of the construction waste sorting device and output the fault closed-loop response plan.

[0019] In this embodiment, the filtering unit collects point cloud data of the working surface of the construction waste sorting device through a deep vision camera module to reflect the three-dimensional morphology of the equipment surface, including the working surface of the crusher, the vibrating screen mesh and the geometric characteristics of other key components. The filtering unit performs adaptive voxel filtering on the original three-dimensional point cloud data, and simplifies multiple points in each voxel into a representative point by dividing the space into a cube (voxel) grid of the same size, effectively reducing the amount of point cloud data, and can eliminate noise and redundant points, thereby improving the efficiency and stability of data processing. After the voxel filtering is completed, the filtering unit estimates the normal vector of the point cloud data, and obtains local geometric features (such as curvature, flatness, edge characteristics, etc.) by calculating the normal vector direction of each point on the point cloud surface, and obtains a regularized voxel grid and local geometric features. The feature extraction unit uses a feature extraction algorithm to convert the spatial geometric information of the point cloud into a high-dimensional feature vector, which contains the geometric shape information of the point cloud and integrates the spatial relationship and surface characteristics between the point clouds. The dimensionality reduction processing unit uses a variational autoencoder to reduce the dimensionality of the high-dimensional feature vector of the point cloud. The variational autoencoder compresses high-dimensional input features into a low-dimensional latent space through an encoder-decoder structure. In this process, by introducing the KL divergence regularization term, the model retains the main distribution information of the feature data during dimensionality reduction, while avoiding the distribution of the latent space being too concentrated or sparse. After the variational autoencoder dimensionality reduction process, the low-dimensional feature vector obtained retains the main information of the wear fault characteristics. The clustering unit combines the prediction results of the equipment wear degree prediction model to perform wear pattern clustering analysis on the wear fault characteristics. Clustering analysis uses unsupervised learning algorithms such as K-means, DBSCAN or Gaussian mixture model to divide the wear characteristics of the equipment surface into different wear patterns, such as mild wear, uniform wear, local severe wear, abnormal wear and other types. In the clustering process, the model assigns data points to the closest cluster center according to the similarity of the wear fault characteristics, and generates a wear fault risk assessment result. By analyzing the distribution of various wear patterns, the current wear status and potential failure risks of the equipment are evaluated, and the stability and safety of the equipment in future operation are predicted. The construction unit further constructs the Takagi-Sugeno (TS) fuzzy model to describe the nonlinear dynamic characteristics of the construction waste sorting device. The TS fuzzy model is a hybrid model that can decompose a complex nonlinear system into a set of linear sub-models. Through the fuzzy logic reasoning mechanism, it can achieve accurate modeling of the equipment operation status. When constructing the model, a multi-rate sampling framework is designed to reasonably configure the grouped sensor acquisition frequency so that different types of sensors can dynamically adjust the sampling rate in different working modes, thereby ensuring the timeliness and effectiveness of the sampling data. The construction unit determines the feedback control gain by solving the linear matrix inequality optimization problem. This process can minimize the control error and ensure the stability of the equipment control system.After the feedback control gain is determined, the construction unit obtains the basic strategy for parameter optimization. Based on the wear failure risk assessment results and the basic strategy for parameter optimization, the solution unit combines the Bayesian optimization system to solve the optimal operating parameter combination of the construction waste sorting device. In the Bayesian optimization process, the model builds an agent model (such as a Gaussian process model) based on the current equipment status and historical data to predict the operating effect of the equipment under different parameter combinations and calculate the corresponding objective function values ​​(such as equipment wear, sorting efficiency, energy consumption, etc.). Through iterative updates, Bayesian optimization continuously selects parameter combinations that can maximize the acquisition function, thereby finding the optimal operating parameters with the least number of experiments. After obtaining the optimal parameter combination, the solution unit applies these parameters to the construction waste sorting device, dynamically adjusts the feed rate, crushing speed, screening frequency, and screening amplitude, and realizes adaptive operation optimization of the equipment. The solving unit feeds back the optimal parameter settings to the equipment control system through the controller, forming a closed-loop control path, so that the equipment can automatically adjust the operating parameters according to the real-time monitored wear status and failure risk, ensuring that the equipment operates in the best state, thereby extending the service life of the equipment, improving the efficiency of construction waste sorting, reducing maintenance costs, and realizing efficient, safe and stable operation of the intelligent sorting and recycling system.

[0020] The dimensionality reduction processing unit constructs a variational autoencoder model consisting of an encoder and a decoder, in which the encoder consists of three layers of fully connected layers with 256, 128 and 64 nodes respectively. This decreasing node design can gradually compress the dimension of the input data, so that the feature vector can extract more compact and critical feature information in each layer. Each fully connected layer is connected to a batch normalization layer and a ReLU activation function. The batch normalization layer can stabilize the model training process and prevent the gradient from disappearing or exploding, while the ReLU activation function introduces nonlinear transformations to enable the model to learn complex features. After completing the encoder structure design, the high-dimensional feature vector of the point cloud is input into the variational autoencoder for feature processing. The encoder maps the input high-dimensional feature data layer by layer to a lower-dimensional space, and the feature vector output in the last fully connected layer is called the encoded output feature. The key difference between the variational autoencoder and the traditional autoencoder is that the variational autoencoder does not directly output a deterministic potential representation, but instead maps the encoded output features to a mean vector (μ) and a variance vector (σ²), which define a multidimensional normal distribution in the latent space. In this process, the variational autoencoder calculates the mean and variance of the features through the fully connected layer, making the potential representation not only compact but also able to capture the uncertainty of the data distribution. In order to effectively sample in the latent space, the dimensionality reduction processing unit introduces reparameterization technology. , where ε is random noise sampled from a standard normal distribution. This reparameterization technique enables the back propagation of gradients and keeps the potential representation continuous and smooth during model training, thereby obtaining a reduced-dimensional potential representation. The dimensionality reduction processing unit imposes orthogonal constraints on the reduced-dimensional potential representation to make the correlation between potential features as close to zero as possible, which helps to eliminate feature redundancy and improve the generalization ability of the model. The specific implementation method is to calculate the difference between the covariance matrix Σ of the latent variable and the identity matrix I, and minimize its norm (such as the Frobenius norm). By adding the orthogonal constraint loss term to the total loss function, the model will automatically adjust the parameters during training so that the potential features meet the orthogonality requirements as much as possible, thereby obtaining orthogonal potential features. The dimensionality reduction processing unit maps the orthogonal potential features back to the original feature space through the decoder to obtain a reconstructed feature vector. The decoder structure is similar to the encoder, consisting of multiple layers of fully connected layers and activation functions. By gradually enlarging the dimensions of the potential features, the low-dimensional potential representation is restored to a high-dimensional feature vector with the same dimension as the original input feature. In order to evaluate the reconstruction effect, the dimensionality reduction processing unit calculates the mean square error between the reconstructed feature vector output by the decoder network and the original input feature, and uses it as the reconstruction loss. The goal of the reconstruction loss is to make the reconstructed features as close to the original features as possible, so as to ensure that the original information is retained as much as possible during the dimensionality reduction process. At the same time, in order to maintain the distribution characteristics of the potential representation, the variational autoencoder calculates the KL divergence between the distribution of the latent variable and the standard normal distribution, and uses it as the regularization loss. KL divergence can quantify the difference between two probability distributions. The variational autoencoder minimizes the KL divergence to make the distribution of the latent variable close to the standard normal distribution. This regularization process avoids the potential representation from being too concentrated or too sparse, and enhances the generalization ability of the model on new data. The reconstruction loss and the regularization loss are combined to form a total loss function. During the model training process, the dimensionality reduction processing unit performs iterative optimization based on the total loss function, and continuously updates the parameters of the model through back propagation and gradient descent methods to minimize the total loss function. After multiple rounds of iterative optimization, the model gradually converges and outputs the wear fault characteristics.

[0021] Optionally, the acquisition subsystem 11 is specifically used for: A pressure sensor, a three-axis acceleration sensor, a displacement sensor and a depth vision camera module are installed on the inner wall of the crusher working surface and the edge of the vibrating screen of the construction waste sorting device to obtain a multi-dimensional sensor array; Differentiated sampling frequencies are set for the multi-dimensional sensor array to collect pressure data, vibration data, displacement data and three-dimensional point cloud data respectively to obtain the original data of equipment wear; Perform wavelet transform noise reduction on the original data of equipment wear to obtain a noise-reduced data segment set, and implement time synchronization calibration on the noise-reduced data segment set to obtain a standardized data set; Time domain features, frequency domain features and time-frequency domain features are extracted from the standardized data set to obtain a multidimensional feature vector. The Kalman filter algorithm is used to perform multi-source data fusion on the multidimensional feature vector to obtain a feature data set that reflects the current wear status of the equipment.

[0022] In this embodiment, various types of sensors are installed on the inner wall of the crusher working surface and the edge of the vibrating screen, including pressure sensors, three-axis acceleration sensors, displacement sensors and depth vision camera modules, to construct a multi-dimensional sensor array to achieve comprehensive collection of different physical quantities during the operation of the equipment. The pressure sensor is used to measure the change in the force on the inner wall of the equipment when crushing materials, and the obtained pressure data is recorded as ,in Indicates pressure, is time; the three-axis acceleration sensor obtains the device at , , The vibration data in three directions are recorded as ,in represents acceleration; displacement sensors are used to monitor the displacement changes of key components of equipment, and the displacement data is recorded as ,in Indicates displacement; the depth vision camera module collects three-dimensional point cloud data on the surface of the device, and the data is expressed as ,in In the data collection process, differentiated sampling frequencies are set for the multi-dimensional sensor array to ensure that various signals are accurately collected within their specific frequency ranges, and the original data set of equipment wear is obtained, including , , and The original data of equipment wear is subjected to wavelet transform noise reduction. Wavelet transform effectively separates the noise component in the signal by decomposing the signal into sub-bands of different frequencies. For example, for the vibration signal Perform wavelet decomposition to represent the signal as coefficients of different scales and frequencies:

[0023] in, It is wavelet coefficients, is the wavelet basis function, is the number of decomposition scales. In the noise reduction process, the low-amplitude noise coefficient is reduced by setting a threshold. Set to zero, then perform wavelet reconstruction to obtain the vibration data after noise reduction ,in Represents the acceleration data after noise reduction. Other data such as , and Using the same method for noise reduction, we get , and Represent the pressure, displacement and 3D visual data after noise reduction, respectively. Perform time synchronization calibration on the noise-reduced data segment set, and use interpolation methods (such as linear interpolation or spline interpolation) to align the sampling data of different frequencies to a unified time axis , and obtain a standardized data set. For example, the pressure data after interpolation is , the vibration data is , the displacement data is , the three-dimensional point cloud data is ,in is a unified time point after standardization. Time domain, frequency domain and time-frequency domain features are extracted from the standardized data set to form a multidimensional feature vector. The time domain features include the mean of the signal , Standard Deviation , Peak and RMS value , these features are calculated by the following formula:

[0024]

[0025] in, represents the number of sampling points, It's in time The frequency domain features are obtained by fast Fourier transform, mainly including the center frequency of the spectrum and bandwidth :

[0026] in is the spectrum of the signal, is the frequency. The time-frequency domain features are obtained through short-time Fourier transform or continuous wavelet transform, which can describe the dynamic change characteristics of the signal in time and frequency. The Kalman filter algorithm is used to fuse multi-source data for multi-dimensional feature vectors. Assume that the multi-dimensional feature vector is , predict the current characteristic state of the device through the state equation:

[0027] in, is the state transition matrix, is the control input matrix, is the control input. In the observation equation, the actual measured eigenvector Combined with the predicted feature vector, through the Kalman gain Update the fused features:

[0028] in, is the observation matrix, It is the Kalman gain, which can dynamically adjust the weight of feature fusion and finally obtain a feature data set that reflects the current wear status of the equipment.

[0029] Optionally, the simulation subsystem 12 is specifically used for: Based on the characteristic data set, the finite element analysis system is used to establish the digital model of the crusher and vibrating screen equipment structure; According to Archard wear theory and cumulative damage theory, the equipment wear progression model of wear coefficient and equipment material hardness is established; The density distribution, particle size distribution, shape factor, elastic modulus and friction coefficient of different types of construction waste are parameterized to obtain a discrete element model of construction waste materials. The discrete element method is used to simulate the dynamic contact process between the discrete element model of construction waste materials and the digital model of equipment structure, and the local stress distribution and contact time data are obtained; Construct a joint simulation environment including material input module, crushing dynamics module, screening dynamics module, wear calculation module, equipment response module, parameter mapping module and data output module to obtain an interactive process simulation system; By using the control variable method to design multiple groups of different operating parameter combinations, and performing simulation calculations on the interactive process simulation system, a simulation data set containing equipment operating parameters, material characteristic parameters and corresponding equipment wear degree data is obtained.

[0030] In this embodiment, a finite element analysis system is used to establish digital models of a crusher and a vibrating screen. By analyzing the structural features of the equipment, including the rotor, hammer head, liner of the crusher, and the screen and supporting structure of the vibrating screen, a three-dimensional model of the equipment is drawn using CAD software, and then the model is imported into the finite element analysis system for meshing. The equipment structure is discretized into a finite number of units, and a mathematical model of the equipment in the process of force and deformation is established through the connection relationship between nodes and units. The finer the mesh, the higher the accuracy of the simulation results, but the amount of calculation will also increase. For example, for the hammer head area of ​​the crusher, due to the high impact force and friction, a high-precision hexahedral mesh is used to make the stress and strain distribution in this area more accurate. After the digital model of the equipment structure is established, a model of equipment wear progression is constructed based on Archard wear theory and cumulative damage theory. Archard wear theory points out that the amount of wear Contact pressure , Slide distance And material hardness The calculation formula is proportional to the reciprocal of:

[0031] in, is the wear coefficient, which indicates the wear sensitivity of the material under specific working conditions; It is the stress condition on the equipment surface; is the relative sliding distance; is the hardness of the material; is the amount of wear. In practical applications, the wear coefficient is obtained through laboratory wear tests or material databases. , combined with the local pressure calculated in the finite element analysis and sliding distance , predicting the degree of wear of key equipment components under different operating conditions. At the same time, the cumulative damage theory is used to evaluate the fatigue damage of equipment materials under various stress cycles, and its damage accumulation model is expressed as:

[0032] in, Indicates the degree of cumulative damage. It is in The actual number of cycles under the stress amplitude is is the failure life of the material under this stress condition. When it reaches 1, the equipment component is considered to have suffered fatigue failure. By combining the wear model with the cumulative damage model, the wear progression of the equipment during long-term operation can be predicted. At the same time, in order to accurately simulate the interaction between construction waste and equipment, a discrete element model of construction waste materials is constructed. The material properties of construction waste include density distribution , particle size distribution , Form Factor , elastic modulus and friction coefficient These parameters are obtained through experimental testing or sample analysis, such as particle size distribution obtained through sieving tests. The cumulative distribution function of , elastic modulus determined by triaxial compression test , the static friction coefficient of the material is obtained through contact friction test . Based on these parameters, in the discrete element software, a discrete element model of the material is constructed through the particle generation tool. Each particle is given specific physical properties and mechanical behavior models, such as the coefficient of restitution and rolling friction coefficient when particles collide. These models can simulate the actual movement state of the material inside the equipment. After the digital model of the equipment structure and the discrete element model of the construction waste material are constructed, the dynamic contact process between the construction waste material and the equipment is simulated using the coupling technology of the discrete element method and the finite element method. During the simulation process, when each particle comes into contact with the surface of the equipment, the system calculates the local stress distribution at the contact point. and contact time , the local stress calculation formula is:

[0033] in, is the contact force, is the contact area. Through dynamic simulation, the stress distribution diagram and contact time data of the equipment surface at different times and positions are obtained, and a joint simulation environment containing multiple functional modules is constructed, including material input module, crushing dynamics module, screening dynamics module, wear calculation module, equipment response module, parameter mapping module and data output module. For example, the material input module automatically generates input particle flow according to the characteristic parameters of construction waste materials, the crushing dynamics module calculates the crushing efficiency by analyzing the impact process between the material and the crusher hammer, the screening dynamics module simulates the movement trajectory and screening efficiency of the material in the vibrating screen, and the wear calculation module calculates the wear amount of the equipment by combining the stress distribution and sliding distance data. The equipment response module analyzes the stress and deformation of the equipment under load changes, the parameter mapping module maps the equipment operation status with the wear characteristics, and the data output module outputs the simulation results as a standardized data set, including the equipment operation parameters. , Material characteristic parameters And the corresponding degree of equipment wear :

[0034] in, For the equipment wear prediction model, multiple groups of different operating parameter combinations are designed by the control variable method, such as adjusting the feed rate of the equipment Crusher speed , screening frequency and screening amplitude , and simulate the interactive process simulation system to evaluate the degree of equipment wear under each working condition one by one , and finally form a system containing equipment operating parameters , Material characteristic parameters Corresponding equipment wear data The simulation data set.

[0035] Optionally, the measurement subsystem 13 is specifically used for: Select multiple target devices as monitoring objects on the actual construction waste sorting line, configure sensors according to the same installation scheme as the sensor array, and obtain the actual monitoring system; Set multiple groups of different typical working condition parameters for the actual measurement monitoring system, and continuously run and regularly record the equipment surface morphology data and sensor data under different working conditions to obtain the actual measured raw data; Extract time domain features from measured original data to obtain a time domain feature set, and perform fast Fourier transform and wavelet packet decomposition on the measured original data to obtain a frequency domain and time-frequency domain feature set; The three-dimensional point cloud data obtained by the deep vision camera module is used to calculate the surface roughness index, the main curvature distribution, the area and depth distribution of the wear area, and obtain the depth feature set; The most representative features are selected from the time domain feature set, frequency domain and time-frequency domain feature set, and depth feature set. The wear degree of the equipment is calibrated by combining visual inspection with laser scanning data to obtain a labeled measured wear data set.

[0036] In this embodiment, multiple target devices are selected as monitoring objects on the actual construction waste sorting line, including crushers, vibrating screens, conveyor belts and other key equipment. These devices are susceptible to the impact and friction of materials and mechanical fatigue caused by long-term operation during the sorting process, so the wear problem is particularly prominent. Sensors are configured on these target devices according to the same installation scheme as the sensor array to form a measurement monitoring system. The sensors include pressure sensors, three-axis acceleration sensors, displacement sensors and depth vision camera modules. The pressure sensor is installed on the inner wall of the crusher working surface to measure the changes in the force on the inner wall when the equipment is crushing materials, and obtain pressure data. ,in Indicates pressure, The three-axis acceleration sensor is installed at the key connection of the equipment to capture the vibration data during the operation of the equipment. , the acceleration is , , The components in the directions are , , ; The displacement sensor is mainly installed on the edge of the vibrating screen to monitor the movement trajectory of the screen in real time and obtain displacement data The depth vision camera module is fixed at a suitable position outside the device, and obtains 3D point cloud data on the device surface through non-contact laser scanning or structured light technology. , used to analyze the changes in the surface morphology of the equipment. After the actual measurement monitoring system is built, by setting multiple sets of different typical operating parameters, the equipment is continuously operated under various actual operating conditions to comprehensively evaluate the wear of the equipment under different working conditions. Typical operating parameters include feed rate , Crusher speed , screening frequency , screening amplitude Through the combined changes of these parameters, the operating status of the equipment under low load, high load and extreme working conditions is simulated. Under these different working conditions, the actual measurement monitoring system continuously collects data through the preset sampling frequency, and regularly records the equipment surface morphology data and sensor data to obtain the actual measured raw data, including pressure , acceleration , displacement and 3D point cloud Data. Extract time domain features from the measured raw data to obtain a time domain feature set. Time domain features are characteristic parameters of the signal in the time domain, including the mean , Standard Deviation , Peak and RMS value . These characteristics are calculated using the following formula:

[0037]

[0038] in, Indicates the number of data sampling points, It's in time For the acceleration signal , respectively calculate , , The time domain characteristics of the equipment reflect the vibration intensity and stability of the equipment in different directions during operation. At the same time, the measured raw data is subjected to fast Fourier transform and wavelet packet decomposition to obtain the frequency domain and time-frequency domain feature sets. The frequency domain characteristics mainly include the center frequency of the signal spectrum. , main frequency components and bandwidth :

[0039] in, is the spectrum of the signal, is the frequency. In frequency domain analysis, for example, a specific frequency is found in the spectrum of a vibration signal. The peak at is significantly higher than other frequencies, indicating that some parts of the equipment have resonated, which is a typical feature of mechanical wear or looseness. Through wavelet packet decomposition, the signal is decomposed into sub-signals with different frequency bandwidths. By analyzing the distribution of these sub-signals on the time axis, the time-frequency domain characteristics are obtained, which helps to capture the frequency response characteristics of the equipment under dynamic load. 3D point cloud data obtained using the deep vision camera module , by calculating the surface roughness index , principal curvature distribution , the area of ​​the wear zone and depth distribution , and obtain a deep feature set. For example, surface roughness It is expressed as:

[0040] in, It is the point cloud data The height value of the point, is the average surface height, is the number of data points. The area of ​​the worn area The projected area of ​​the damaged surface in the three-dimensional model is calculated, and the depth distribution By analyzing the depth of the depression in the wear area and combining it with the fine topological data obtained by laser scanning, the quantitative analysis of the changes in the surface morphology of the equipment can be achieved. The most representative features are selected from the time domain feature set, frequency domain and time-frequency domain feature set, and depth feature set through feature selection methods (such as principal component analysis or mutual information method). Combining visual inspection with laser scanning data, the degree of equipment wear is calibrated to obtain a labeled measured wear data set.

[0041] Optionally, the training subsystem 14 further includes: A design unit for designing an adaptive domain adversarial neural network structure including a feature extractor, a domain classifier, and a wear predictor, wherein the feature extractor is composed of multiple fully connected layers, batch normalization layers, and activation functions; The adversarial unit is used to define a multi-task loss function including wear prediction loss and domain adversarial loss, and implement adversarial training through the gradient reversal layer to obtain the training optimization target; A partitioning unit, used for partitioning the simulation data set into a first training set and a validation set, and partitioning the labeled measured wear data set into a second training set and a test set; The joint unit is used to use the first training set to perform network pre-training, and then perform joint training in combination with the second training set, and adopt a dynamic weight adjustment strategy to gradually increase the domain adversarial loss weight to obtain a device wear degree prediction model.

[0042] In this embodiment, a network structure including a feature extractor, a domain classifier, and a wear predictor is designed. Through a multi-task loss function and adversarial training method, accurate prediction of the degree of equipment wear and seamless migration between simulation data and measured data are achieved. The whole process includes the collaborative work of the design unit, adversarial unit, partitioning unit, and joint unit to ensure that the model can not only accurately predict the equipment wear state, but also adapt to the distribution differences between different data domains. In the design unit, a feature extractor is constructed. The feature extractor consists of multiple fully connected layers, batch normalization layers, and activation functions. Its goal is to convert the input multidimensional feature vector Convert to latent feature representation ,in is the dimension of the input features, is the extracted potential feature dimension. The feature extraction process is expressed as: ; in, It is The output features of the layer, is the weight matrix of the fully connected layer, is the bias vector, is the activation function (e.g. ReLU function ), Indicates The batch normalization layer calculates the mean of the feature distribution at each layer. and standard deviation , and then normalize the features: ; This processing method can accelerate the convergence of the model and avoid the problem of gradient disappearance or gradient explosion. The potential features extracted by the feature extractor is input to both the domain classifier and the wear predictor. The role of the domain classifier is to discriminate the input features The domain classifier is composed of two to three fully connected layers and a sigmoid activation function, and outputs the domain prediction result. ,in It indicates that it comes from simulation data (source domain), It indicates that it comes from the measured data (target domain). The domain classification loss is calculated using the binary cross entropy loss function:

[0043] in, is the actual domain label (0 or 1), is the prediction result of the domain classifier. Through the gradient reversal layer, the gradient sign is reversed during back propagation, making the features learned by the feature extractor more domain invariant, achieving the effect of domain adversarial training. The wear predictor is used to predict the degree of wear of the equipment, and the prediction result is expressed as , the actual wear label is , the wear prediction loss is calculated by the mean square error loss: ; in, is the sample size, and Respectively The adversarial unit defines a loss that includes the actual wear value and the predicted wear value of the sample. Domain adversarial loss Multi-task loss function , by dynamically adjusting the loss weight Implementing adversarial training:

[0044] in, is the domain adversarial loss weight coefficient, initially Small enough to ensure that the model prioritizes learning the wear prediction task, and gradually increases in the later stages of training. , so that the model pays more attention to domain adaptability. During the training process, the gradient descent method (such as Adam optimizer) is used to calculate the loss function of the model parameters. Gradient Perform back propagation and parameter update: ; in, is the learning rate, which controls the step size of model parameter update. In the partition unit, the simulation data set Divide into the first data set and the second dataset , the measured wear dataset with labels Divide into the third data set And the fourth data set During the data division process, the consistency of data distribution is ensured, for example, the data set is divided in a ratio of 80%:20% to avoid overfitting and data leakage problems. Pre-train the model with the goal of minimizing the wear prediction loss Without introducing domain adversarial loss, the initial stage When the model is applied to the second dataset When the loss on t becomes stable, it means that the model has learned enough wear prediction features. Perform joint training, dynamically increase the domain adversarial loss weight For example, the value of is increased linearly:

[0045] in, is the current training progress (the ratio of training rounds to total rounds), is the adjustment coefficient. The dynamic adjustment strategy ensures that the model focuses on the wear prediction task in the early stage and enhances the domain adaptation ability in the later stage. When the joint training is completed, on the fourth dataset The wear prediction accuracy of the model evaluated on and domain classification confusion , the wear prediction accuracy calculation formula is: ; in, is the number of samples. When the model reaches the domain classification confusion state with a small wear prediction error, it is considered that the model has successfully achieved the cross-domain migration of simulation data and measured data, and finally the equipment wear degree prediction model is obtained.

[0046] Optionally, the combined unit is specifically used for: Initialize the parameters of the adaptive domain adversarial neural network to obtain the initial network parameters; The first training set is input into the adaptive domain adversarial neural network in batches, and only the feature extractor and wear predictor parameters are updated until the verification loss is stable, thereby obtaining a pre-trained model; Build a data loader, mark the first training set as source domain data, mark the second training set as target domain data, and sample them in the same proportion to form a mixed batch to obtain a domain adversarial training data stream; For each batch of data in the domain adversarial training data stream, features are extracted through the feature extractor of the adaptive domain adversarial neural network, and the extracted features are simultaneously input into the domain classifier and the wear predictor to obtain domain prediction and wear prediction results; The domain adversarial loss weight coefficient is calculated according to the current training progress, and the total loss is calculated by combining the wear prediction loss and the domain adversarial loss. The network parameters are updated through the gradient reversal layer to obtain the dynamically adjusted network weights. After every 50 rounds of training, the learning rate is decayed to the original 0.1, and the wear prediction error and domain classification confusion matrix are calculated on the test set. When the prediction error is less than the threshold and the domain classification accuracy is close to the random guessing level, the cross-domain migration model is obtained.

[0047] In this embodiment, the neural network is initialized to obtain initial network parameters. In the model initialization stage, all weight parameters in the feature extractor, domain classifier and wear predictor are and bias parameters Both use Xavier initialization or He initialization methods. These two initialization methods can keep the initial values ​​of the parameters within a reasonable distribution range and effectively avoid the problem of gradient vanishing or gradient exploding in the early training process of the model. For example, for the weight matrix of the fully connected layer ,in and are the number of input and output nodes respectively. The Xavier initialization method randomly samples the weights in the interval:

[0048] in, Indicates that in the interval This initialization method ensures that the variance of signals between different layers remains stable, which helps the model converge quickly. After completing the model parameter initialization, the first training set The batches are fed into the adaptive domain adversarial neural network for pre-training. During the pre-training phase, only the parameters of the feature extractor and the wear predictor are updated, while the parameters of the domain classifier are kept frozen, which means that in the multi-task loss function, the domain adversarial loss weight coefficient Set to 0 to make the model focus on equipment wear prediction tasks. Wear prediction loss By calculating the mean square error, the formula is:

[0049] in, is the number of batch samples, For the The actual wear degree of each sample, is the wear value predicted by the model. In each batch, the model calculates the predicted value through forward propagation and calculates the gradient of the loss function with respect to the model parameters through back propagation. , and then update the network parameters of the feature extractor and wear predictor through the gradient descent method (such as Adam optimizer):

[0050] in, is the learning rate, which controls the step size of the model parameter update. During the pre-training process, the model loss on the validation set gradually decreases. When it converges to a stable value, the model is considered to have learned enough equipment wear prediction features and a pre-trained model is obtained. After the model pre-training is completed, a data loader is built to load the first training set Marked as source domain data (simulated data), the second training set Marked as target domain data (measured data), by setting the same sampling strategy, the source domain and target domain data are mixed to generate a batch data stream, which is called domain adversarial training data stream. In the mixed data batch, each batch of data contains both source domain data and target domain data. Also includes target domain data , to ensure that the model can simultaneously learn the wear prediction task in the source domain and the domain adaptation task in the target domain during the training process. In the domain adversarial training stage, for each mixed batch of data, the input features are extracted by the feature extractor High-dimensional feature representation , the feature extraction process is expressed as:

[0051] in, is the weight matrix of the feature extractor, is the bias, is the activation function (such as ReLU function). The extracted features It is input into the domain classifier and wear predictor at the same time. The domain classifier outputs the domain prediction result ,in Represents source domain data Represents the target domain data. The domain classification loss is calculated by the cross entropy loss function: ; in, is the actual domain label (0 or 1), which is passed through the gradient reversal layer to convert the gradient Take the opposite approach to achieve the effect of adversarial training. During the training process, according to the current training progress Compute domain adversarial loss weight coefficient Dynamic changes of adopting linear increase strategy, for example:

[0052] in, ,along with increase, The smooth transition from close to 0 to close to 1 allows the model to focus on the wear prediction task in the early stage and enhance domain adaptation ability in the later stage. It is composed of the weighted wear prediction loss and domain adversarial loss:

[0053] Each time the model parameters are updated, the total loss function is minimized by the gradient descent method to achieve dynamic adjustment of the network weights. The decay is reduced to the original 0.1 to avoid oscillation in the later stage of model training and improve the accuracy of model convergence. During model training, the wear prediction error is calculated on the test set. And domain classification confusion matrix Wear prediction error Expressed as the deviation between the predicted value and the true value:

[0054] in, is the number of test set samples, and Respectively The actual and predicted wear values ​​of samples. Less than a preset threshold (e.g. 0.05) and the domain classification confusion matrix The domain classification accuracy in is close to the random guessing level (50%), indicating that the model has achieved the alignment of the feature distribution of the simulation data and the measured data, achieving the expected effect of domain adversarial training, and finally obtaining a cross-domain migration model.

[0055] Optionally, the output subsystem 15 further includes: A filtering unit is used to collect point cloud data of the working surface of the construction waste sorting device through a deep vision camera module, and to perform adaptive voxel filtering and normal vector estimation on the collected point cloud data to obtain a regularized voxel grid and local geometric features; A feature extraction unit is used to extract features from the regularized voxel grid and local geometric features to obtain a high-dimensional feature vector of the point cloud; A dimensionality reduction processing unit is used to input the high-dimensional feature vector of the point cloud into the variational autoencoder for dimensionality reduction processing to obtain the wear fault feature; A clustering unit is used to cluster the wear patterns of the wear failure characteristics based on the prediction results of the equipment wear degree prediction model to obtain the wear failure risk assessment results; The construction unit is used to construct the Takagi-Sugeno fuzzy model to describe the nonlinear dynamic characteristics of the construction waste sorting device, and to design a multi-rate sampling framework to group the sensor acquisition frequency. The feedback control gain is determined by solving the linear matrix inequality optimization problem to obtain the basic strategy for parameter optimization. The solving unit is used to solve the optimal operating parameter combination based on the wear failure risk assessment results and the basic parameter optimization strategy, combined with the Bayesian optimization system, to dynamically adjust the feed rate, crushing speed, screening frequency and screening amplitude of the construction waste sorting device to obtain a fault closed-loop response plan.

[0056] In this embodiment, the filtering unit collects point cloud data of the working surface of the construction waste sorting device through a depth vision camera module (such as a laser radar or a structured light 3D scanner). These point cloud data are stored in the form of three-dimensional coordinates, and each point contains spatial coordinates. ,in , , Respectively indicate points Position in three-dimensional space. Adaptive voxel filtering is performed on the original point cloud data. By dividing the point cloud into cubes of uniform size (voxels), and simplifying multiple points in each voxel into one representative point, this process greatly reduces the amount of point cloud data while retaining spatial structure information. Assuming the voxel size is , the voxel center position is , then the point after voxel filtering The calculation formula is:

[0057] in, is the number of points in a voxel. After filtering, a regularized voxel grid is obtained to improve the efficiency of point cloud processing. After completing the voxel filtering, the filtering unit estimates the normal vector of the point cloud data to obtain the local geometric features of each point. Normal vector Reflects the directional characteristics of the point cloud surface in three-dimensional space. The normal vector is calculated by the neighborhood of each point in the point cloud (using the k-nearest neighbor algorithm, assuming the neighborhood size is ) The eigenvalue decomposition of is:

[0058] in, is the centroid of the neighborhood points, and the eigenvector corresponding to the minimum eigenvalue of the covariance matrix is ​​the normal vector , and obtains geometric features such as local curvature, flatness, and edge characteristics through normal vector estimation. The feature extraction unit extracts high-dimensional features from the regularized voxel grid and local geometric features. Using a three-dimensional point cloud feature descriptor (such as FPFH or SHOT features), the spatial geometric information and normal vector information of the point cloud are converted into a high-dimensional feature vector. ,in is the feature dimension. For example, by calculating the angle distribution and distance distribution of the normal vector between each point in the point cloud and the midpoint in its neighborhood, a multi-dimensional feature representation describing the surface morphology and wear characteristics of the equipment is obtained. After obtaining the high-dimensional feature vector, the dimensionality reduction processing unit uses a variational autoencoder to reduce the dimensionality of the high-dimensional feature vector of the point cloud. The variational autoencoder converts the high-dimensional input feature vector into a Compressing to a low-dimensional latent space ,in The encoder takes as input the features Mapping to mean vector and the standard deviation vector :

[0059] in, and is the weight matrix, and is the bias vector. The latent representation is sampled by reparameterization technique:

[0060] The decoder transforms the latent representation Remap back to the original feature space and calculate the reconstruction loss (mean square error) and regularized loss (KL divergence) , the total loss function for:

[0061] in, is the weight coefficient of the regularization term. By minimizing the total loss function, feature dimensionality reduction is achieved to obtain the wear fault feature z. The clustering unit clusters the wear fault features into wear patterns based on the prediction results of the equipment wear degree prediction model. The K-means or DBSCAN clustering algorithm is used to cluster the low-dimensional features into The wear patterns are classified into different categories, such as light wear, uniform wear, local wear and abnormal wear, to obtain the wear failure risk assessment results. After realizing the wear pattern recognition, the construction unit describes the nonlinear dynamic characteristics of the construction waste sorting device by constructing the Takagi-Sugeno (TS) fuzzy model. The TS fuzzy model consists of multiple linear sub-models. Composition, through fuzzy rules For inference, the model output is the weighted average of all sub-models:

[0062] in, is the fuzzy weight, Yes Input In the To improve sampling efficiency, a multi-rate sampling framework is constructed to group sensor acquisition frequencies, dynamically adjust the sampling rate according to the device status, and determine the feedback control gain by solving the linear matrix inequality optimization problem. , forming a basic strategy for parameter optimization. The solution unit combines the wear failure risk assessment results and the basic strategy for parameter optimization with the Bayesian optimization system to solve the optimal operating parameter combination. Bayesian optimization builds an agent model Predicting different operating parameters Under the device performance, choose to maximize the acquisition function Parameter combinations:

[0063] Dynamically adjust the feed rate of the construction waste sorting device through repeated iterations , Crushing speed , screening frequency and screening amplitude , and finally output a closed-loop fault response plan to achieve automatic control of the equipment under complex working conditions and improve the stability and safety of equipment operation.

[0064] Optionally, the dimension reduction processing unit is specifically used for: Construct a variational autoencoder consisting of an encoder and a decoder, where the encoder consists of three fully connected layers with 256, 128, and 64 nodes, respectively. Each fully connected layer is followed by a batch normalization layer and a ReLU activation function. The high-dimensional feature vector of the point cloud is input into the variational autoencoder for feature processing to obtain the encoded output features, which are mapped into mean vectors and variance vectors respectively, and sampled through the reparameterization technology to obtain the potential representation after dimensionality reduction; Apply orthogonal constraints to the latent representation after dimensionality reduction, and obtain orthogonal latent features by calculating the difference between the latent variable covariance matrix and the identity matrix and minimizing its norm; Map the orthogonal latent features back to the original feature space to obtain the reconstructed feature vector, calculate the mean square error between the decoder network and the reconstructed feature vector as the reconstruction loss, and calculate the KL divergence between the latent variable distribution and the standard normal distribution as the regularization loss. The reconstruction loss and regularization loss are combined to form the total loss function; Iterative optimization is performed based on the total loss function to output wear fault characteristics.

[0065] In this embodiment, a variational autoencoder model consisting of an encoder and a decoder is constructed. The variational autoencoder is a generative model that reduces the dimensionality by learning the latent space representation of the input data, and reconstructs the latent representation back to the original data space through the decoder. The encoder part consists of multiple fully connected layers. The designed encoder contains three fully connected layers, consisting of 256, 128 and 64 nodes respectively. Each fully connected layer is followed by a batch normalization layer and a ReLU activation function, which helps to speed up the training process of the model and prevent overfitting. Assume that the input high-dimensional feature vector is ,in is the dimension of the features, and the encoder maps these features into the latent space, generating two vectors: the mean vector and variance vector , these two vectors are used to define the distribution of the latent space. The first fully connected layer takes the input features Mapped to an intermediate space of size 256, we get ,in is the weight matrix of the first layer, is the bias term. Through the batch normalization layer and the ReLU activation function Process and obtain BatchNorm The second and third layers are processed in a similar way to obtain the intermediate features and , the output is mapped to a mean vector through two independent fully connected layers and variance vector :

[0066] In getting the mean and variance Then, the latent variables are transformed into Sampling from the distribution of the latent space. By introducing noise variables , so that the latent variable becomes differentiable, allowing back-propagation optimization. Latent variables Calculated by the following formula:

[0067] This approach brings the control of mean and variance into the latent space and provides higher flexibility and robustness for the subsequent generation process. In order to enhance the structure of the latent space and avoid high correlation between latent representations, orthogonal constraints are imposed on the latent representations after dimensionality reduction to ensure that there is no strong correlation between different dimensions of the latent space. For the potential features after dimensionality reduction, calculate the covariance matrix of the latent variables :

[0068] in, is the sample size, is the mean vector of all potential representations. To impose orthogonality constraints, we minimize the covariance matrix and the identity matrix The difference between them, the objective function is expressed as:

[0069] in, is the Frobenius norm, which represents the sum of the squares of the matrix elements. This constraint ensures that the dimensions of the latent space are orthogonal to each other, resulting in more structured latent features. The decoder maps it back to the original feature space. The decoder structure is a mirror image of the encoder, consisting of multiple fully connected layers. Assume that the output of the decoder is the reconstructed feature vector , the goal of the decoder is to minimize the reconstruction error, that is, the original feature and reconstruction features The mean square error is used as the reconstruction loss :

[0070] Calculate the regularization loss of the latent variable by calculating the distribution of the latent variable and the standard normal distribution The KL divergence measures the difference between two probability distributions and is expressed as:

[0071] in, and are the elements of the mean vector and variance vector respectively. The role of KL divergence is to push the distribution of the latent space closer to the standard normal distribution, so that the model can learn smooth and meaningful latent features. The final total loss function is It is the weighted sum of reconstruction loss and regularization loss, which is optimized using the gradient descent algorithm during training:

[0072] in, and They are the weight coefficients of KL divergence and orthogonality constraint, which are used to adjust the contribution of each loss to the total loss.

[0073] In the embodiment of the present application, a multi-dimensional sensor array of pressure sensors, triaxial acceleration sensors, displacement sensors and deep vision camera modules is configured on the working surface of the crusher and the edge of the vibrating screen, and multi-source data fusion is performed by combining wavelet transform noise reduction, time synchronization calibration and Kalman filter algorithm to achieve high-precision monitoring of the wear state of the equipment and capture complex wear characteristics that cannot be recognized by traditional single sensors. The simulation model constructed based on finite element analysis, Archard wear theory and discrete element method is combined with small sample measured data to form a mechanism-data hybrid driven fault prediction method, which effectively solves the problem of poor generalization ability of pure data-driven methods for small sample data, and avoids the prediction bias caused by over-simplification of pure mechanism models. By designing an adaptive domain adversarial neural network including feature extractors, domain classifiers and wear predictors, effective knowledge transfer from the simulation data domain to the measured data domain is achieved, and a large amount of simulation data is fully utilized to improve the training effect of the model, while maintaining adaptability to actual working conditions, significantly improving the generalization ability of the fault prediction model. The deep vision camera module is used to collect point cloud data, and the three-dimensional convolutional neural network and variational autoencoder are combined for feature extraction and dimensionality reduction. The orthogonal constraint and KL divergence optimization are used to generate low-dimensional wear fault features with high discrimination. The Takagi-Sugeno fuzzy model is used to describe the nonlinear dynamic characteristics of the sorting system. The sensor acquisition frequency is grouped in the multi-rate sampling framework. The feedback control gain is determined by solving the linear matrix inequality optimization problem, which effectively copes with the uncertainty caused by sensor failure and time delay, and improves the stability and reliability of the system under complex working conditions. Based on the wear fault risk assessment results and the basic strategy of parameter optimization, the Bayesian optimization system is combined to solve the optimal operating parameter combination, and the dynamic adjustment of the feed rate, crushing speed, screening frequency and screening amplitude of the sorting device is realized, forming a complete fault closed-loop response plan, which effectively extends the service life of the equipment while ensuring the sorting efficiency.

[0074] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned system embodiments and will not be repeated here.

[0075] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including a number of instructions to enable an electronic device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the system described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc., various media that can store program codes.

[0076] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. An intelligent construction waste sorting and recycling system, characterized in that: The system comprises: A collection subsystem, which is used to install a sensor array on the crusher working surface and the edge of the vibrating screen of the construction waste sorting device, and collect a characteristic data set reflecting the wear status of the equipment; A simulation subsystem, used for establishing a simulation model of the construction waste sorting device according to the characteristic data set, and generating a simulation data set containing a combination of multiple working condition parameters based on the simulation model; The measurement subsystem is used to collect small sample measured data under actual construction waste sorting conditions and calibrate the wear level to obtain a labeled measured wear data set; A training subsystem, used for inputting the simulation data set and the measured wear data set with labels into an adaptive domain adversarial neural network for training, so as to obtain a prediction model for the degree of equipment wear; The output subsystem is used to identify wear fault characteristics through the equipment wear degree prediction model, and based on the wear fault characteristics, optimize the operating parameters of the construction waste sorting device and output a fault closed-loop response plan.

2. The intelligent construction waste sorting and recycling system according to claim 1 is characterized in that: The acquisition subsystem is specifically used for: A pressure sensor, a three-axis acceleration sensor, a displacement sensor and a depth vision camera module are installed on the inner wall of the crusher working surface and the edge of the vibrating screen of the construction waste sorting device to obtain a multi-dimensional sensor array; Setting differentiated sampling frequencies for the multidimensional sensor array to collect pressure data, vibration data, displacement data and three-dimensional point cloud data respectively to obtain raw data of equipment wear; Performing wavelet transform noise reduction processing on the raw data of equipment wear to obtain a noise-reduced data segment set, and performing time synchronization calibration on the noise-reduced data segment set to obtain a standardized data set; Time domain features, frequency domain features and time-frequency domain features are extracted from the standardized data set to obtain a multidimensional feature vector, and a Kalman filter algorithm is used to perform multi-source data fusion on the multidimensional feature vector to obtain a feature data set reflecting the current wear state of the equipment.

3. The intelligent construction waste sorting and recycling system according to claim 1 is characterized in that: The simulation subsystem is specifically used for: Based on the characteristic data set, a finite element analysis system is used to establish a digital model of the equipment structure of the crusher and the vibrating screen; According to Archard wear theory and cumulative damage theory, the equipment wear progression model of wear coefficient and equipment material hardness is established; The density distribution, particle size distribution, shape factor, elastic modulus and friction coefficient of different types of construction waste are parameterized to obtain a discrete element model of construction waste materials. Using the discrete element method to simulate the dynamic contact process of the discrete element model of the construction waste material and the digital model of the equipment structure, to obtain local stress distribution and contact time data; Construct a joint simulation environment including material input module, crushing dynamics module, screening dynamics module, wear calculation module, equipment response module, parameter mapping module and data output module to obtain an interactive process simulation system; A plurality of different operating parameter combinations are designed by the control variable method, and the interactive process simulation system is simulated and calculated to obtain a simulation data set including equipment operating parameters, material characteristic parameters and corresponding equipment wear degree data.

4. The intelligent construction waste sorting and recycling system according to claim 1 is characterized in that: The measurement subsystem is specifically used for: Select multiple target devices as monitoring objects on the construction waste sorting line actually in operation, configure sensors according to the same installation scheme as the sensor array, and obtain a measured monitoring system; Setting a plurality of different groups of typical working condition parameters for the actual measurement monitoring system, and continuously operating and regularly recording equipment surface morphology data and sensor data under different working conditions to obtain actual measurement raw data; Extracting time domain features from the measured original data to obtain a time domain feature set, and performing fast Fourier transform and wavelet packet decomposition on the measured original data to obtain a frequency domain and time-frequency domain feature set; The three-dimensional point cloud data obtained by the deep vision camera module is used to calculate the surface roughness index, the main curvature distribution, the area and depth distribution of the wear area, and obtain the depth feature set; The most representative features are selected from the time domain feature set, the frequency domain and time-frequency domain feature set, and the depth feature set, and the wear degree of the equipment is calibrated by combining visual inspection with laser scanning data to obtain a labeled measured wear data set.

5. The intelligent construction waste sorting and recycling system according to claim 1 is characterized in that: The training subsystem also includes: A design unit, for designing an adaptive domain adversarial neural network structure including a feature extractor, a domain classifier, and a wear predictor, wherein the feature extractor is composed of a plurality of fully connected layers, a batch normalization layer, and an activation function; The adversarial unit is used to define a multi-task loss function including wear prediction loss and domain adversarial loss, and implement adversarial training through the gradient reversal layer to obtain the training optimization target; A division unit, used for dividing the simulation data set into a first training set and a validation set, and dividing the labeled measured wear data set into a second training set and a test set; The joint unit is used to use the first training set to perform network pre-training, and then perform joint training in combination with the second training set, and adopt a dynamic weight adjustment strategy to gradually increase the domain adversarial loss weight to obtain a device wear degree prediction model.

6. The intelligent construction waste sorting and recycling system according to claim 5 is characterized in that: The combined unit is specifically used for: Initializing parameters of the adaptive domain adversarial neural network to obtain initial network parameters; Inputting the first training set into the adaptive domain adversarial neural network in batches, and only updating the feature extractor and wear predictor parameters until the verification loss is stable, thereby obtaining a pre-trained model; Construct a data loader, mark the first training set as source domain data, mark the second training set as target domain data, and sample them in the same proportion to form a mixed batch to obtain a domain adversarial training data stream; For each batch of data in the domain adversarial training data stream, extract features through the feature extractor of the adaptive domain adversarial neural network, and input the extracted features into the domain classifier and the wear predictor at the same time to obtain domain prediction and wear prediction results; The domain adversarial loss weight coefficient is calculated according to the current training progress, and the total loss is calculated by combining the wear prediction loss and the domain adversarial loss. The network parameters are updated through the gradient reversal layer to obtain the dynamically adjusted network weights. The learning rate is decayed to the original 0.1 after every 50 rounds of training, and the wear prediction error and domain classification confusion matrix are calculated on the test set. When the prediction error is less than the threshold and the domain classification accuracy is close to the random guessing level, the cross-domain migration model is obtained.

7. The intelligent construction waste sorting and recycling system according to claim 1 is characterized in that: The output subsystem also includes: A filtering unit is used to collect point cloud data of the working surface of the construction waste sorting device through a deep vision camera module, and to perform adaptive voxel filtering and normal vector estimation on the collected point cloud data to obtain a regularized voxel grid and local geometric features; A feature extraction unit, used to extract features from the regularized voxel grid and local geometric features to obtain a high-dimensional feature vector of the point cloud; A dimensionality reduction processing unit, used for inputting the high-dimensional feature vector of the point cloud into a variational autoencoder for dimensionality reduction processing to obtain a wear fault feature; A clustering unit, configured to cluster the wear patterns of the wear failure characteristics based on the prediction results of the equipment wear degree prediction model, and obtain a wear failure risk assessment result; A construction unit is used to construct a Takagi-Sugeno fuzzy model to describe the nonlinear dynamic characteristics of the construction waste sorting device, and to design a multi-rate sampling framework to group the sensor acquisition frequency, determine the feedback control gain by solving a linear matrix inequality optimization problem, and obtain a basic strategy for parameter optimization; A solving unit is used to solve the optimal operating parameter combination based on the wear failure risk assessment result and the parameter optimization basic strategy in combination with a Bayesian optimization system, dynamically adjust the feed rate, crushing speed, screening frequency and screening amplitude of the construction waste sorting device, and obtain a fault closed-loop response plan.

8. The intelligent construction waste sorting and recycling system according to claim 7 is characterized in that: The dimension reduction processing unit is specifically used for: Construct a variational autoencoder consisting of an encoder and a decoder, where the encoder consists of three fully connected layers with 256, 128, and 64 nodes, respectively. Each fully connected layer is followed by a batch normalization layer and a ReLU activation function. Inputting the high-dimensional feature vector of the point cloud into the variational autoencoder for feature processing to obtain encoded output features, mapping the encoded output features into a mean vector and a variance vector respectively, and sampling through a reparameterization technique to obtain a potential representation after dimensionality reduction; Applying orthogonal constraints to the latent representation after dimensionality reduction, and obtaining orthogonal latent features by calculating the difference between the latent variable covariance matrix and the identity matrix and minimizing the norm thereof; Mapping the orthogonal latent features back to the original feature space to obtain a reconstructed feature vector, calculating the mean square error between the decoder network and the reconstructed feature vector as a reconstruction loss, and calculating the KL divergence between the latent variable distribution and the standard normal distribution as a regularization loss, and combining the reconstruction loss and the regularization loss to form a total loss function; Iterative optimization is performed based on the total loss function to output wear fault characteristics.

Images