Multimodal ai analysis of construction and demolition waste recycling crushing and screening method

Abstract

The application provides a multi-modal AI analysis building waste recycling crushing and screening method, and belongs to the technical field of building waste treatment. The method comprises the following steps: conveying the material to be crushed to a primary crushing device through a conveyor, estimating the weight of magnetic metal by combining electromagnetic induction scanning, separating and detecting the actual weight after crushing, and calculating the weight difference ΔG; obtaining the particle size D1 / D2 and the particle proportion P before and after crushing based on visual recognition, and constructing a multi-parameter crushing effect model to evaluate the crushing degree; dynamically adjusting the parameters of the primary crushing device, and outputting qualified products through the cooperation of recycling crushing and screening. The application realizes accurate evaluation of crushing degree and intelligent adjustment of parameters through multi-modal AI analysis, reduces the metal residue rate to a low level, significantly reduces the unit energy consumption, and improves the qualified rate of finished products to a high level, which is suitable for large-scale building waste treatment scenes.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of construction waste treatment, in particular to a construction waste recycling crushing and screening method based on multi-modal AI analysis. BACKGROUND

[0002] The construction waste can be divided into concrete blocks, waste bricks and stones, waste plastics, waste metals, waste wood and the like.

[0003] At present, the construction waste is generally screened out by a screening device, and the excellent screening device can screen out high-purity concrete blocks and waste bricks and stones to produce coarse and fine aggregates, which can be used to produce concrete, mortar or prepare building material products such as blocks, wallboards, floor tiles and the like. Among them, the screened concrete blocks and waste bricks and stones need to be crushed by a crusher to meet the required particle size. Since the mixed metals in the high-purity concrete blocks and waste bricks and stones are mainly steel bars, the traditional crushing device can only observe the crushing effect of the crushed materials by manual observation, which makes the crushing device unable to be adjusted in time to make the concrete blocks and waste bricks and stones be fully crushed. SUMMARY

[0004] The present application relates to the technical field of construction waste treatment, in particular to a construction waste recycling crushing and screening method based on multi-modal AI analysis.

[0005] The technical solution adopted by the present application is as follows: the present application provides a construction waste recycling crushing and screening method based on multi-modal AI analysis, comprising the following steps:

[0006] Step S1: conveying the to-be-crushed material to a primary crushing device for crushing by a conveyor I, wherein the to-be-crushed material includes construction waste containing magnetic metals;

[0007] Step S2: based on multi-frequency electromagnetic signal analysis of an electromagnetic induction scanning device, the volume of the magnetic metals in the to-be-crushed material conveyed into the primary crushing device per unit time is obtained in advance, and the estimated weight is calculated by combining the density parameter of the magnetic metals;

[0008] Step S3: after the primary crushing, the material is separated by a magnetic separation device through a conveyor II, and the magnetic metals are separated to a collection device, and the collection device is integrated with a weight detection device to measure the actual collection weight in real time;

[0009] Step S4: calculating the weight difference per unit time, and inputting the weight difference into a crushing effect model to obtain the crushing sufficiency;

[0010] Step S5: based on the crushing sufficiency, dynamically adjusting the operating parameters of the primary crushing device;

[0011] Step S6: The primary crushed material is sent to the screening device by conveyor two, and the finished product is screened out;

[0012] Step S7: The retained material on the screening device is sent to the secondary crushing device by conveyor three for secondary crushing, and the secondary crushed material is returned to the downstream of the magnetic separation device on conveyor two;

[0013] Step S8: Steps S1-S7 are repeatedly executed to continuously output the finished product.

[0014] In some embodiments, in step S2, the following steps are included:

[0015] Step S21: The multi-channel array sensor emits multi-band electromagnetic signals at a preset frequency gradient, each band scans for a preset time, and the receiving end collects phase difference data points at a preset interval;

[0016] Step S22: The signal is filtered and smoothed, and is converted to a digital signal by analog-to-digital conversion;

[0017] Step S23: A three-dimensional grid data is constructed by a spatial interpolation method, a phase difference threshold is set to determine the magnetic metal region, and a voxel counting method is used to calculate the volume;

[0018] Step S24: The estimated weight is calculated by combining the standard density of the magnetic metal and the speed correction coefficient.

[0019] In some embodiments, in step S4, the establishment of the crushing effect model includes the following steps:

[0020] Step S41: A preset number of sample groups at a preset interval, the sample group contains five parameters: weight difference ΔG, average particle size of the material to be crushed , average particle size of the material after primary crushing , and primary crushing device working frequency f and working gap d; wherein, based on the visual recognition algorithm, respectively and are obtained, and the proportion of particles with a particle size greater than a preset particle size is calculated based on the visual recognition data statistics , combined with and ΔG to generate a crushing sufficiency label, including three levels of "excellent", "good", and "poor".

[0021] Step S42: A multi-input and multi-output neural network model is constructed, the input layer contains five features: ΔG, , , , , the hidden layer uses a nonlinear activation function, and the output layer uses a probability type activation function;

[0022] Step S43: training the model with a preset loss function, setting the learning rate and the number of iterations, and stopping training when the preset accuracy is reached;

[0023] Step S44: collecting new samples at a preset period, and starting incremental training when the error rate exceeds the preset threshold.

[0024] In some embodiments, the visual recognition algorithm in step S41 obtains the process includes:

[0025] A visual recognition system 1 is installed above the feeder conveyor of the primary crushing device, continuously collecting images of the material to be crushed, pre-processing and Mask R-CNN model segmentation of particles, extracting the equivalent circle diameter of each particle, and calculating .

[0026] In some embodiments, the visual recognition algorithm in step S41 obtains and the particle ratio P, the process includes:

[0027] A visual recognition system 2 is installed above the conveyor 2 at the discharge end of the primary crushing device, continuously collecting images of the material after primary crushing, pre-processing and Mask R-CNN model segmentation of particles, extracting the equivalent circle diameter of each particle, and calculating and P.

[0028] In some embodiments, in step S5, the following steps are included:

[0029] Step S51: determining the crushing sufficiency level according to the maximum probability value output by the crushing effect model;

[0030] Step S52: when the level is a preset level, adjusting the working frequency of the primary crushing device according to a preset formula, and setting the step size and upper limit value as preset values;

[0031] Step S53: when the working frequency is adjusted and the target level is still not reached within a preset period, adjusting the working gap of the primary crushing device, and setting a minimum working gap value;

[0032] Step S54: continuously monitoring for a preset period, maintaining the parameters if the standard is met, or repeating the adjustment otherwise.

[0033] In some embodiments, it further includes a dynamic optimization database construction and pre-adjustment step:

[0034] Step S01: constructing a dynamic optimization database, which stores historical processing data, including the characteristic parameters of the material to be crushed, the optimal operating parameter combination of the primary crushing device, and the corresponding crushing sufficiency;

[0035] ​Step S02: Real-time acquisition of the characteristic parameters of the current material to be crushed, and matching of historical data sets with a similarity reaching a preset threshold in the database through a data matching algorithm;

[0036] Step S03: Extracting the best operating parameters in the matched historical data set as the initial adjustment parameters of the current primary crushing device;

[0037] Step S04: If the data matching fails for a continuous preset number of groups, enable the real-time adjustment mode of the crushing effect model, and store the new parameter combination and processing result in the database.

[0038] In some embodiments, the dynamic optimization database is updated by clustering at a preset period, and a clustering algorithm is used to divide the historical data into a preset number of working condition clustering centers, each clustering center stores the optimal parameter template under the working condition, and when new data is added, the clustering center parameters are automatically updated.

[0039] The beneficial effects of the present application are as follows:

[0040] 1. In the present application, a neural network crushing effect model is constructed by fusing the weight difference, particle size before and after crushing, and particle proportion, realizing three-dimensional evaluation of "particle size compliance - crushing efficiency - metal separation effect", and the accuracy of crushing sufficiency determination is greatly improved, which is a significant progress compared with the traditional single parameter method, and avoids parameter adjustment errors caused by misjudgment.

[0041] 2. In the present application, the dynamic optimization database realizes initial parameter pre-adjustment, combined with the real-time closed-loop adjustment of the crushing effect model, the adaptation efficiency of the working frequency and gap of the primary crushing device is significantly improved, the unit product energy consumption is obviously reduced, and the manual intervention is greatly reduced.

[0042] 3. In the present application, the visual recognition system and the clustering update mechanism ensure that the device adapts to the particle size fluctuation and metal content change of construction waste, and still maintains stable operation when the material characteristics mutate

[0043] 4. In the present application, the recycling and efficient resource recovery of construction waste are realized: through secondary crushing and recycling process, the retained material is processed again, improving the resource utilization rate. BRIEF DESCRIPTION OF DRAWINGS

[0044] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiment or prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings obtained according to these drawings without creative labor are still within the scope of the present application.

[0045] Figure 1 Flow chart of the construction waste recycling crushing and screening method of the multi-modal AI analysis in the present application;

[0046] Figure 2 Flow chart of the method for calculating the estimated weight in the present application;

[0047] Figure 3 Flow chart of the method for establishing a crushing effect model in the present application;

[0048] Figure 4 Flow chart of the method for dynamically adjusting the primary crushing equipment in the present application;

[0049] Figure 5 Flow chart of the method for constructing a dynamic optimization database in the present application. DETAILED DESCRIPTION

[0050] The following description provides specific application scenarios and requirements of the present specification, with the purpose of enabling those skilled in the art to manufacture and use the contents of the present specification. Various local modifications to the disclosed embodiments will be apparent to those skilled in the art, and the general principles defined herein can be applied to other embodiments and applications without departing from the spirit and scope of the present specification. Therefore, the present specification is not limited to the embodiments shown, but is consistent with the widest scope of the claims.

[0051] In the description of the present application, it should be noted that, unless otherwise explicitly specified and limited, the orientation or positional relationship indicated by the terms "longitudinal", "transverse", "radial", "length", "width", "thickness", "upper", "lower", "left", "right", "front", "rear", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc. is based on the orientation or positional relationship shown in the drawings, and these terms are mainly used to better describe the present application and its embodiments, and are not used to limit the indicated devices, elements or components to have a specific orientation, or to be constructed and operated in a specific orientation.

[0052] It should be noted that the terms "first", "second", and similar words do not indicate any order, quantity or importance, but are only used to distinguish different components, and should not be understood as limiting the embodiments of the present application.

[0053] It should be noted that the terms "mounting", "setting", "provided with", "connection", "connected" should be broadly understood. For example, it can be fixedly connected, detachably connected, or integrally constructed; it can be mechanically connected, or electrically connected; it can be directly connected, or indirectly connected through an intermediate medium, or it can be the internal communication between two devices, elements or components.

[0054] It should be noted that the terms "in some embodiments," "exemplarily," and "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design described in this application as "in some embodiments," "exemplarily," or "for example" should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of terms such as "in some embodiments," "exemplarily," and "for example" is intended to present related concepts in a specific manner, meaning that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of the above terms in various places in the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. Those skilled in the art will explicitly and implicitly understand that the embodiments described herein can be combined with other embodiments.

[0055] Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances.

[0056] Regarding the accompanying drawings of this application, it should be clearly understood that the drawings are for illustrative and descriptive purposes only and are not intended to limit the scope of this specification. It should also be understood that the drawings are not necessarily drawn to scale.

[0057] like Figures 1 to 5 As shown in the figure, this specification provides a method for recycling, crushing, and screening construction waste using multimodal AI analysis, including the following steps:

[0058] Step S1: The material to be crushed is conveyed to the primary crushing equipment via conveyor one for crushing. The material to be crushed includes construction waste containing magnetic metals. It can be understood that the construction waste containing magnetic metals includes concrete blocks, bricks and mixed metal components. The high-purity material to be crushed is screened out by professional screening equipment, among which the mixed metal components are mainly steel bars.

[0059] Step S2: Based on the multi-band electromagnetic signal analysis of the electromagnetic induction scanning device, the volume of magnetic metal in the material to be crushed fed into the primary crushing equipment per unit time is obtained in advance, and the estimated weight is calculated by combining the density parameter of the magnetic metal; specifically, step S2 includes the following steps:

[0060] Step S21: The multi-channel array sensor emits multi-band electromagnetic signals at a preset frequency gradient, such as 1kHz, 3kHz, 5kHz, 8kHz, and 10MHz. Each band is continuously scanned for a preset time, which is 0.5-1s. The receiving end collects 500-1000 phase difference data points at a preset interval of 1μs.

[0061] Generally, the multi-channel array sensor is automatically preheated after starting to ensure the stability of the sensor signal. The multi-channel array sensor and the conveyor are vertically arranged, and the distance between them is adjusted according to the size of the material to be crushed.

[0062] Step S22: filtering and mean smoothing of the signal, and analog-to-digital conversion to a digital signal; specifically, the original signal is sequentially subjected to a Butterworth low-pass filter and mean smoothing with a window size N=5 to remove environmental noise interference, and is converted to a digital signal through 16-bit analog-to-digital conversion to improve data reliability.

[0063] Step S23: constructing a three-dimensional grid data through a spatial interpolation method, setting a phase difference threshold to determine a magnetic metal region, and calculating a volume through voxelization counting; specifically, a 1mm precision three-dimensional grid data is constructed through Kriging interpolation, a 5° phase difference threshold is set to determine a magnetic metal region, and a voxelization counting method is used to calculate the volume of the magnetic metal with a volume of 1mm³ per voxel to realize accurate volume quantification.

[0064] Step S24: calculating the estimated weight in combination with the standard density, correction coefficient and speed correction coefficient of the magnetic metal.

[0065] The calculation formula of the estimated weight is: ;

[0066] wherein, is the volume of the magnetic metal;

[0067] is the standard density, and since the main metal is steel bar, the density of the steel bar is 7.85g / cm³.

[0068] is the correction coefficient, for example, a corrosion correction coefficient of 0.95-1.0; it can be understood that the correction coefficient is not necessarily required.

[0069] is the speed correction coefficient, which is the ratio of the actual conveying speed of the conveyor to the rated speed, and the actual conveying speed is monitored in real time by installing a speed sensor such as an optical encoder or a Hall sensor on the non-load-bearing section of the driving drum or the conveyor belt of the conveyor to monitor the running speed of the conveyor belt.

[0070] The actual speed of the conveyor will fluctuate due to working conditions, such as automatic deceleration when the material is too thick and acceleration when the load is empty, which deviates from the rated speed, resulting in changes in the amount of material passing through the electromagnetic scanning area per unit time. If not corrected:

[0071] When the actual speed is greater than the rated speed, the volume of the magnetic metal V per unit time is underestimated (more material passes through in the same time, and the total amount of metal increases), and direct calculation will result in being too small;

[0072] When the actual speed < rated speed, V is overestimated, It is too large.

[0073] By After correction, "volume per unit time" can be unified to the time reference of rated speed, ensuring Can truly reflect the total amount of magnetic metal in unit time, and the difference between the subsequent actual collection weight (Weight per unit time) Calculation is accurate, providing reliable data for broken fullness evaluation.

[0074] The time coefficient makes the estimated weight directly comparable to the actual collected weight, providing a reliable data reference for subsequent broken fullness evaluation and equipment adjustment.

[0075] Step S3: After primary crushing, the material passes through the magnetic separation device through the conveyor II, preferably the magnetic separation device uses the existing technology self-unloading electromagnetic iron remover, separates the magnetic metal to the collection device, and the collection device integrates the weight detection device to measure the actual collection weight in real time ;

[0076] Step S4: Calculate the weight difference per unit time, input the weight difference into the crushing effect model to obtain the crushing fullness; Specifically, in step S4, the establishment of the crushing effect model includes the following steps:

[0077] Step S41: Pre-set a pre-set number of sample groups at a pre-set interval, and the sample group contains five parameters: weight difference ΔG, average particle size of the material to be crushed , average particle size of the material after primary crushing , and working frequency f and working gap d of the primary crushing equipment; wherein, and are obtained based on visual recognition algorithm respectively, and the proportion of particles greater than the pre-set particle size is calculated through visual recognition data statistics , combined with and ΔG to generate a crushing fullness label, including "excellent", "good", and "poor".

[0078] Exemplarily, the generation rule of the crushing fullness label is:

[0079] When the following conditions are all met: P≤5%, ΔD≥100mm and ΔG≤3000kg / h, the label is "excellent";

[0080] When any of the following conditions is met: 5% < P ≤ 15%, 50mm ≤ ΔD < 100mm or 3000kg / h < ΔG ≤ 6000kg / h, the label is "good";

[0081] When any of the following conditions is met: P > 15%, ΔD < 50mm or ΔG > 6000kg / h, the label is "poor".

[0082] Step S42: Construct a multi-input multi-output neural network model, the input layer contains ΔG, , , , Five features, the hidden layer uses a nonlinear activation function, and the output layer uses a probability type activation function; specifically, a 3-layer fully connected neural network is used, the input layer contains ΔG, , , , Five features, the first hidden layer has 12 neurons, the ReLU activation function, prevents gradient disappearance, the second hidden layer has 8 neurons, the LeakyReLU activation function, the slope is 0.1, and the output layer has 3 neurons, the Softmax activation function, and outputs the "excellent - good - poor" probability.

[0083] Step S43: Train the model with a preset loss function, set the learning rate and the number of iterations, and stop training when the preset accuracy is reached; for example, a cross-entropy loss function is used, the learning rate is set to 0.001-0.01, the number of iterations is set to 5000-10000 times, and the training is stopped when the validation set accuracy is ≥ 90%, ensuring the generalization ability of the model.

[0084] Step S44: Collect new samples at a preset period, and start incremental training when the error rate exceeds the preset threshold. For example, 10 groups of new samples are collected every 24 hours, and if the prediction error rate is > 10%, incremental training is started (fixed 90% weight, fine-tuned 10% weight), and the model adaptability is maintained.

[0085] The process of acquiring by the visual recognition algorithm in step S41 includes:

[0086] A visual recognition system is installed above the feeder conveyor of the primary crushing device, and the images of the material to be crushed are continuously collected, preprocessed, and segmented by the Mask R-CNN model, the equivalent circle diameter of each particle is extracted, and is calculated.

[0087] The process of acquiring and the particle proportion P, the process includes:

[0088] A visual recognition system two is installed above the conveyor two at the discharge end of the primary crushing device, continuously collects material images, segments particles through pre-processing and Mask R-CNN model, extracts the equivalent circle diameter of each particle, and calculates and P.

[0089] Preferably, a vibration conveyor is additionally arranged upstream of the conveyors corresponding to the visual recognition system one and the visual recognition system two, double excitation motors are symmetrically installed at the bottom of the conveyor, horizontal excitation force is generated by synchronous reverse rotation of the motors, so that the material makes periodic jumping motion on the surface of the belt, and the layered state is broken; secondly, the surface of the belt is made of high-friction coefficient rubber material and is pressed into a diamond pattern, which reduces material sliding and further disperses the layers through the friction difference during vibration. In this way, the detected substances are transported along the corresponding conveyors in a flat state as much as possible, thereby improving the accuracy of visual detection.

[0090] Specifically, the visual recognition algorithm includes:

[0091] Step D1: configure a double industrial camera and light source system, use a 100mmx100mm checkerboard calibration board to shoot multiple images at different angles, complete camera internal and external parameter calibration using OpenCV library, and realize double camera synchronous triggering based on corresponding conveyor encoder signals;

[0092] Step D2: when the material enters the camera field of view, trigger the camera exposure and continuously collect images through the encoder pulse signal, further, the images can be sequentially processed for grayscale, noise reduction filtering and threshold segmentation to generate a binary image to improve image quality.

[0093] Step D3: segment the overlapping particles using morphological operations or deep learning models on the pre-processed images, filter out invalid targets, extract the contours of valid particles and calculate the equivalent circle diameter; specifically, the following steps are included:

[0094] Step D31, overlapping particle segmentation:

[0095] Mild overlap refers to 2-3 particles sticking together: perform a 3x3 structure element erosion operation (2 iterations) on the binary image to break the thin connection between particles; calculate the distance of each pixel to the background by distance transform, find the maximum distance point as the particle center; use watershed transform to segment the sticking area.

[0096] Severe overlap, refers to ≥4 particles accumulation: based on Mask R-CNN deep learning model segmentation, model backbone uses ResNet-50, training data set contains 10000 labeled images, particle categories include concrete block, brick, metal block, boundary box labeling accuracy ±2mm; model is trained on a graphics card, loss function is reduced to 0.02, and pixel-level mask is output.

[0097] Step D32: invalid target filtering, calculate the area and aspect ratio of connected domain by OpenCV function, filter noise points with area <100-150 pixels and non-particle targets with aspect ratio >5-10, such as conveyor belt edges.

[0098] Step D33: feature extraction, Douglas-Peucker algorithm is used to extract particle simplified contour, pixel area is obtained by counting pixels within the contour, and equivalent circle diameter is calculated.

[0099] Step D4: adopt 3σ rule to eliminate outliers, calculate average particle size by weighting actual particle area, and transmit results to control system in real time, ensure timestamp error ≤10ms with weight difference, working frequency and working gap, support real-time analysis of crushing effect model.

[0100] Working frequency It refers to the number of periodic movements of the core working component of the primary crushing equipment per unit time, which directly determines the impact kinetic energy during crushing. For rotary type equipment such as hammer crusher, "working frequency" is "rotational speed", the number of hammer head rotations per minute. For swing type equipment such as jaw crusher, "working frequency" is "swing frequency", the number of swing movements of the moving jaw per minute.

[0101] Working gap It refers to the minimum distance between the core working components in the primary crushing equipment, which directly determines the stress space of the material in the crushing chamber and the discharge particle size, and is the core mechanical parameter to control the crushing effect.

[0102] Generally, the primary crushing equipment is preferably the working frequency and working gap adjustable crusher in the prior art, the working frequency and working gap of the primary crushing equipment are obtained through the control system of the primary crushing equipment. Modern crushers have hydraulic drive and sensor technology: the hydraulic device can accurately control the gap, and realize remote adjustment through linkage with PLC system, and the sensor monitors the gap value in real time, such as laser ranging sensor, which feeds back to the control system to form a closed-loop regulation.

[0103] The minimum discharge gap between the fixed jaw and the movable jaw in the jaw crusher, i.e., the minimum passage width of the material after crushing. The jaw crusher adjusts the position of the wedge through a hydraulic system to change the distance between the movable jaw and the fixed jaw, thereby achieving stepless adjustment.

[0104] The minimum distance between the hammer head (or the impact plate) and the liner in the hammer crusher (impact crusher), which determines the strength of the impact and grinding of the material. The gap between the impact plate and the rotor in the impact crusher can be adjusted by a hydraulic device, and the distance between the hammer head and the liner in the hammer crusher is adjusted.

[0105] Step S5: dynamically adjusting the operating parameters of the primary crushing equipment based on the crushing degree of fineness;

[0106] Specifically, in step S5, the following steps are included:

[0107] Step S51: extracting the crushing degree of fineness corresponding to the maximum probability of the crushing effect model output ("excellent", "good", "poor"); for example, the model inputs the real-time collected ΔG, 、 、 、 parameters, and outputs the probabilities of "excellent", "good", and "poor", such as a probability of "excellent" of 60%, a probability of "good" of 30%, and a probability of "poor" of 10%. The grade corresponding to the maximum probability is taken as the evaluation result of the current crushing effect, which is used to guide the parameter adjustment of the primary crushing equipment.

[0108] Step S52: when the grade is a preset grade, adjusting the working frequency of the primary crushing equipment according to a preset formula, and the adjustment step and the upper limit value are preset values; specifically, when the grade is "good" or "poor", adjusting the working frequency according to the preset formula: wherein, is the historical maximum difference, is the current working frequency, the adjustment step ≤ 50-100 r / min, and the upper limit value is set to avoid overloading of the equipment and the lower limit value is set to ensure the crushing force.

[0109] Step S53: when the working frequency is still not up to the target grade within a preset period after the adjustment, adjusting the working gap of the primary crushing equipment, and setting a minimum working gap value; specifically, if the working frequency is still "poor" after 3 periods of adjustment, adjusting the working gap: , is the adjustment step, which is generally set according to the requirements, the minimum working gap ≥ 5 mm, which is generally set according to the requirements, and high-precision control is achieved through a hydraulic device.

[0110] During the gap adjustment, the working frequency is maintained at the current value to avoid system instability caused by simultaneous changes in the two parameters.

[0111] Step S54: Continuously monitor for 3 preset periods. If the standard is met, maintain the parameters. Otherwise, repeat the adjustment, but reduce the adjustment step by half to avoid excessive adjustment leading to a sharp increase in energy consumption.

[0112] Through the above logic, step S5 realizes intelligent and fine adjustment of the parameters of the primary crushing device, taking into account crushing efficiency, particle size standard and energy consumption control, providing core closed-loop control capability for construction waste recycling crushing and screening.

[0113] Step S6: The material crushed by the primary crushing device is sent to the screening device by conveyor two. Preferably, the screening device is a double-layer vibrating screen, and the finished material is screened out. Through different aperture screens, such as the upper layer screen 30-50mm and the lower layer screen 5-10mm, the finished material is screened out through the two layers of screens.

[0114] Step S7: The retained material on the screening device (upper layer screen and lower layer screen) is sent to the secondary crushing device by conveyor three for secondary crushing, and the secondary crushed material is returned to the downstream of the magnetic separation device on conveyor two;

[0115] Step S8: Steps S1-S7 are executed in a cycle to continuously output finished material and realize full-process closed-loop processing.

[0116] At this point, the main crushing and screening cycle realizes real-time monitoring and dynamic adjustment of the construction waste crushing and screening process through multi-source data acquisition and AI analysis.

[0117] It also includes dynamic optimization database construction and pre-adjustment steps:

[0118] Step S01: Construct a dynamic optimization database, which stores historical processing data, including the characteristic parameters of the material to be crushed, the optimal operating parameter combination of the primary crushing device, and the corresponding crushing degree of completion. The characteristic parameters of the material to be crushed include the average particle size and the content of magnetic metal , and the optimal operating parameter combination of the primary crushing device includes the working frequency and working gap of the primary crushing device.

[0119] Step S02: Real-time acquisition of the characteristic parameters of the current material to be crushed, matching of historical data sets with a similarity of a preset threshold in the database through a data matching algorithm. Specifically, match the historical data sets with a similarity of ≥90% in the database through the K-Nearest Neighbor algorithm. The similarity calculation formula is:

[0120] ;

[0121] Wherein, , , The weight coefficients of the material characteristic parameters and the equipment operation parameters are respectively, by adjusting the proportion of the two, the influence of different parameters on the similarity matching can be highlighted, so as to more accurately screen out the historical data similar to the current working condition, and provide a reliable basis for the pre-adjustment of the equipment parameters.

[0122] wherein, : the average particle size of the current material to be crushed;

[0123] : the average particle size of the material to be crushed in a historical data in the database;

[0124] : the maximum value of the average particle size of all historical data in the database, used for normalization processing to eliminate the influence of dimension.

[0125] : the weight proportion of magnetic metal in the current material to be crushed;

[0126] ; wherein, the total weight of the material to be crushed is obtained by adding a belt scale device, an electromagnetic induction scanning device is installed upstream of the conveyor I, a high-precision belt scale such as a strain gauge type dynamic weighing device is installed, the weight of the material per unit length on the conveyor I can be collected in real time, and the total weight can be calculated in combination with the running speed of the conveyor I.

[0127] : the weight proportion of magnetic metal in a historical data in the database;

[0128] : the maximum value of the weight proportion of magnetic metal of all historical data in the database, used for normalization processing.

[0129] Step S03: extracting the best operation parameters in the historical data group that matches successfully as the initial adjustment parameters of the current primary crushing equipment;

[0130] Step S04: if the continuous preset number of data matching fails, the real-time adjustment mode of the crushing effect model is enabled, and the new parameter combination and the processing result are stored in the database.

[0131] With the accumulation of data, directly storing all original data will lead to: when matching, the calculation amount is too large, the K - nearest neighbor algorithm needs to compare all data, the time consumption increases, and the redundant data occupies the storage space.

[0132] Therefore, the dynamic optimization database is updated by clustering at a preset period, the historical data is divided into a preset number of working condition clustering centers by using a clustering algorithm, each clustering center stores the optimal parameter template under the working condition, and when new data is added, the clustering center parameters are automatically updated.

[0133] The clustering update is to divide the data into several categories according to the "working condition similarity" by using a clustering algorithm (such as K-means), and the "center" (cluster center) of each category is the "typical representative" of the data in the category. Each cluster center is not only a "typical value of working condition", but also stores the optimal parameter template of the working condition, that is, the working frequency f and the working gap d of the "crushing sufficiency for optimization" in the working condition. When matching the new working condition, it is not necessary to compare all the original data, but only to compare several cluster centers, so that the most similar working condition and the corresponding optimal parameters can be quickly found, and the matching efficiency is greatly improved.

[0134] The cluster center and the parameter template are not "outdated" due to old data, and can always reflect the latest production rule, ensuring the accuracy of the pre-adjustment.

[0135] It can be understood that the above-mentioned multi-modal AI analysis building waste recycling crushing and screening method can be applied to the following building waste recycling crushing and screening system, which comprises a conveyor I, a conveyor II, a conveyor III, a primary crushing device, a secondary crushing device, an electromagnetic induction scanning device, a visual recognition system I, a visual recognition system II, a weight detection device, a magnetic separation device, a screening device, a detection system and a control system.

[0136] The output end of the conveyor I is connected to the feeding end of the primary crushing device, the discharging end of the primary crushing device is connected to the input end of the conveyor II, the magnetic separation device is installed above the conveyor II, the output end of the conveyor II is connected to the feeding end of the screening device, the input end of the conveyor III is connected to the retained material outlet of the screening device, and the output end is connected to the feeding end of the secondary crushing device. The discharging end of the secondary crushing device is connected to the downstream of the magnetic separation device of the conveyor II through a chute.

[0137] The detection system comprises an electromagnetic induction scanning device, a visual recognition system I, a visual recognition system II and a weight detection device. The electromagnetic induction scanning device and the visual recognition system I are installed above the conveyor I along the conveying direction. The visual recognition system II is installed above the conveyor II near the primary crushing device. The weight detection device is integrated at the bottom of the collection device of the magnetic separation device.

[0138] The control system is electrically connected with the primary crushing device, the secondary crushing device and the detection system respectively, for receiving detection data and adjusting the operation parameters of the devices.

[0139] In view of the foregoing, it will be seen that the foregoing detailed description of the application is only illustrative in nature and not limiting. Although specific reference can be made to the examples contained herein, this has been done for illustrative purposes only, and is not intended to limit the application of the application in any way. Other ways of implementing the application will be apparent to those skilled in the art in view of the foregoing detailed description of the application. Accordingly, the true spirit and scope of the application is indicated by the appended claims, rather than by the foregoing description; and changes can be made to the details of the application without departing from the application as defined by the claims. Although there has been described above what are considered to be preferred embodiments of the present application, various modifications, alterations, and improvements will readily occur to those skilled in the art. It is the intention, therefore, to include all such modifications, alterations, and improvements within the scope of the present application. In particular, it is intended that only reasonable variants that can be appreciated by those skilled in the art and with the benefit of this disclosure fall within the scope of the application. Accordingly, the application is not to be restricted, except as by the appended claims and their equivalents.

[0140] Furthermore, it is to be understood that the application can be carried out by specifically designated apparatus, and that apparatus may, by virtue of certain design choices, constitute a combination of several features. However, this does not mean that the combination of these features is necessary, and that a person skilled in the art, upon reading the application, could not identify a part of the apparatus as a separate embodiment. That is, the embodiments of the application can also be understood as an integration of several sub-embodiments. The content of each sub-embodiment is valid when there are less than all the features of a single previously disclosed embodiment.

[0141] Finally, it should be understood that the embodiments of the application disclosed herein are illustrative of the principles of the present application. Other modifications that fall within the spirit and scope of the application will be apparent to those skilled in the art. Accordingly, the application is not to be restricted, except as by the appended claims and their equivalents.

Claims

1. A multi-modal AI analyzed construction and demolition waste recycling crushing and screening method, characterized by, The process includes the following steps: Step S1: The material to be crushed is conveyed to the primary crushing equipment via conveyor one. The material to be crushed includes construction waste containing magnetic metal. Step S2: Based on the multi-band electromagnetic signal analysis of the electromagnetic induction scanning device, the volume of magnetic metal in the material to be crushed conveyed into the primary crushing equipment per unit time is obtained in advance, and the estimated weight is calculated in combination with the density parameter of the magnetic metal. Step S3: After primary crushing, the material is conveyed to the magnetic separation device via conveyor two to separate the magnetic metal to the collection device. The collection device integrates a weight detection device to measure the actual collected weight in real time. Step S4: The weight difference per unit time is calculated, and the weight difference is input into the crushing effect model to obtain the degree of crushing. Step S5: Based on the degree of crushing, the operating parameters of the primary crushing equipment are dynamically adjusted. Step S6: The material crushed by primary crushing is sent to the screening equipment via conveyor two to screen out the finished material. Step S7: The material retained on the screening equipment is sent to the secondary crushing equipment via conveyor three for secondary crushing. The material that is crushed in the secondary crushing is returned to the downstream of the magnetic separation device on conveyor two. Step S8: Steps S1-S7 are executed repeatedly to continuously output the finished material. In step S4, the establishment of the crushing effect model includes the following steps: Step S41: Set a preset number of sample groups at preset intervals, wherein the sample groups contain five parameters: weight difference ΔG, average particle size of the material to be crushed, etc. Average particle size of material after primary crushing And the operating frequency f and the working interval d of the primary crushing equipment; among which, based on the visual recognition algorithm, respectively obtain and and through The visual recognition data shows the percentage of particles larger than the preset particle size. , combined Generate a fragmentation sufficiency label with ΔG, including three levels: excellent, good, and poor; Step S42: Construct a multi-input multi-output neural network model, with the input layer containing ΔG, , , , Five features: the hidden layer uses a non-linear activation function, and the output layer uses a probabilistic activation function; Step S43: train the model with a preset loss function, set the learning rate and number of iterations, and stop training when the preset accuracy is reached; Step S44: collect new samples at a preset period, and start incremental training when the error rate exceeds a preset threshold.

2. The method of multi-modal Al-analyzed construction and demolition waste recycling crushing and screening as claimed in claim 1, wherein, Step S2 includes the following steps: Step S21: The multi-channel array sensor emits multi-band electromagnetic signals according to a preset frequency gradient, and each band is continuously scanned for a preset time. The receiving end collects phase difference data points at preset intervals. Step S22: Filter and mean smooth the signal, and convert it into a digital signal via analog-to-digital conversion; Step S23: Construct three-dimensional mesh data using spatial interpolation, set a phase difference threshold to determine the magnetic metal region, and calculate the volume using voxelization counting method; Step S24: Calculate the estimated weight by combining the standard density of the magnetic metal and the velocity correction factor.

3. The method for recycling, crushing, and screening construction waste using multimodal AI analysis according to claim 1, characterized in that, The visual recognition algorithm in step S41 acquires The process of the above-mentioned includes: A visual recognition system 1 is installed above the feed end conveyor of the primary crushing device, continuously collects images of the material to be crushed, separates the particles through pre-processing and Mask R-CNN model, extracts the equivalent circle diameter of each particle, and calculates .

4. The method for recycling, crushing, and screening construction waste using multimodal AI analysis according to claim 1, characterized in that, The visual recognition algorithm in step S41 acquires and the proportion of particles P includes: A visual recognition system two is installed above the conveyor two at the discharge end of the primary crushing device, continuously collects images of the material after primary crushing, separates the particles through pre-processing and Mask R-CNN model, extracts the equivalent circle diameter of each particle, and calculates and P.

5. The method for construction waste recycling crushing and screening with multi-modal AI analytics of claim 1, wherein, Step S5 includes the following steps: Step S51: Determine the crushing sufficiency level based on the maximum probability value output by the crushing effect model; Step S52: When the level is the preset level, adjust the working frequency of the primary crushing equipment according to the preset formula, and adjust the step size and upper limit value to the preset value; Step S53: If the target level is not reached within the preset cycle after the working frequency is adjusted, adjust the working gap of the primary crushing equipment and set the minimum working gap value. Step S54: Continuously monitor the preset period. If the target is met, maintain the parameter; otherwise, repeat the adjustment.

6. The method of multi-modal Al analytics for construction and demolition waste recycling crushing and screening as claimed in claim 1, wherein, It also includes dynamically optimizing the database construction and pre-tuning steps: Step S01: Construct a dynamic optimization database, which stores historical processing data, including characteristic parameters of the material to be crushed, the optimal combination of operating parameters of the primary crushing equipment, and the corresponding crushing sufficiency. Step S02: Collect the characteristic parameters of the material to be crushed in real time, and match historical data groups with similarity reaching a preset threshold in the database using a data matching algorithm; Step S03: Extract the best operating parameters from the successfully matched historical data set and use them as the initial adjustment parameters for the current primary crushing equipment; Step S04: If the matching of the preset number of data sets fails, the real-time adjustment mode of the crushing effect model is activated, and the new parameter combination and processing results are stored in the database.

7. The method for recycling, crushing, and screening construction waste using multimodal AI analysis according to claim 6, characterized in that, The dynamic optimization database performs clustering updates at preset intervals. The clustering algorithm divides historical data into a preset number of working condition cluster centers. Each cluster center stores the optimal parameter template for that working condition. When new data is added, the cluster center parameters are automatically updated.

Images