Sorting and intelligent resource scheduling method for crushed materials of waste electric appliances

Abstract

The invention provides a sorting and intelligent resource scheduling method for crushed materials of waste electric appliances. The method comprises the following steps: S1, establishing a full-process equipment rate correlation model: defining a coupling relationship among a feeding rate (Vf), a first-stage crusher rotor rotating speed (Vp1), a second-stage crusher rotor rotating speed (Vp2), a vortex separator magnetic roller rotating speed (Vs) and a winnower fan frequency (Va); wherein Vp1 and Vf are associated through an adjustable proportionality coefficient K1, and K1 is dynamically set according to the identified refrigerator material characteristics; the Vp2 is associated with the real-time average particle size (D50) of the material after primary crushing; vs is associated with the expected recovery rate (eta) of the target non-ferrous metal through a pre-trained nonlinear model, and the model is established based on sorting experiment data of copper and aluminum particles with different characters.

Description

Technical Field

[0001] This invention belongs to the field of intelligent scheduling, and specifically relates to a method for sorting and intelligent resource scheduling of crushed waste electrical appliances. Background Technology

[0002] Currently, in the field of refrigerator waste recycling production lines, existing rate scheduling methods face a series of severe challenges caused by the inherent uncertainties of waste products. These challenges seriously restrict the improvement of processing efficiency and resource recycling benefits. The shortcomings of current technologies are concentrated in several aspects:

[0003] Traditional scheduling optimization theories are difficult to apply directly to remanufacturing systems because the recycling of used refrigerators involves significant uncertainties in terms of quantity, time, quality, and subsequent dismantling sequence and depth, making it difficult for production lines to operate stably. The quality uncertainty of recycled parts directly leads to randomization of remanufacturing processing paths and process times, greatly increasing the difficulty of production scheduling, and traditional static scheduling methods cannot effectively adapt to such dynamic changes.

[0004] At the level of specific recycling processes, such as the eddy current separation stage for recovering non-ferrous metals from crushed materials, the separation efficiency is affected by multiple factors, including material shape, feeding speed, magnetic roller speed, and baffle angle. The separation efficiency varies greatly for copper and aluminum particles of different shapes. Traditional production lines lack the intelligent scheduling capability to dynamically adjust these parameters based on the characteristics of the incoming materials, resulting in low resource recovery rates. Fourth, the dismantling process of retired products also faces uncertainties in terms of source, type, structure, and residual performance value. This makes the stability of the reverse production scheduling of dismantling highly susceptible to disturbances from uncertainties such as the type of incoming materials, dismantling sequence, and dismantling method. Traditional scheduling methods relying on manual experience are insufficient to achieve balanced equipment utilization and maximize dismantling efficiency.

[0005] Furthermore, in production environments that combine manufacturing and remanufacturing, traditional scheduling methods often struggle to simultaneously coordinate the scheduling of new and remanufactured products when dealing with random recovery rates and multi-product economic batch scheduling problems. This can easily lead to infeasible solutions and limited optimization effects, making it impossible to minimize the total unit cost.

[0006] Existing technologies do not adequately consider the complexity of remanufacturing workshop scheduling in closed-loop supply chains. Particularly when the remanufacturing processing time of recycled parts is uncertain, traditional scheduling methods struggle to optimize effectively. Workshop scheduling suffers from the blindness of randomly swapping neighboring processes, making it difficult to maximize overall system profit while minimizing the maximum completion time. These factors collectively result in significant shortcomings in the efficiency, adaptability, and resource recovery benefits of existing refrigerator breakage and recycling production lines. Summary of the Invention

[0007] This invention proposes a sorting and intelligent resource scheduling method for crushed waste electrical appliances, which solves the problems of inconsistent equipment speeds, low sorting efficiency, unstable overall resource recovery rate, and inability to adaptively adjust according to material changes in existing refrigerator crushing and recycling production lines due to uncertain material input.

[0008] The technical solution of this invention is implemented as follows: a method for sorting and intelligent resource scheduling of crushed waste electrical appliances, the method comprising the following steps:

[0009] S1. Establish a full-process equipment rate correlation model: Define the coupling relationship between feed rate (Vf), primary crusher rotor speed (Vp1), secondary crusher rotor speed (Vp2), eddy current separator magnetic roller speed (Vs), and air separator fan frequency (Va); among them, Vp1 and Vf are correlated through an adjustable proportional coefficient K1, which is dynamically set according to the identified refrigerator material characteristics; Vp2 is correlated with the real-time average particle size (D50) of the material after primary crushing; Vs is correlated with the expected recovery rate (η) of the target non-ferrous metal through a pre-trained nonlinear model, which is established based on the sorting experimental data of copper and aluminum particles with different properties;

[0010] S2. Multi-source real-time sensing and material characteristic identification: The material composition of the refrigerator to be processed is estimated through the visual recognition system at the feed end; D50 is obtained in real time through the online particle size analyzer at the discharge port of the primary crusher; the type, shape and proportion of non-ferrous metals are identified through the component detection unit at the feed port of the eddy current separator.

[0011] S3. Multi-objective dynamic optimization decision: The central control unit receives the real-time data from step S2, and constructs a multi-objective optimization function with the goals of maximizing the overall output value and balancing the equipment load, and with system power consumption as a constraint. The function is solved online using a non-dominated sorting genetic algorithm with an elite retention strategy, and a set of optimal equipment rate parameter combinations is selected from the Pareto solution set according to a preset priority. The priority rule is to prioritize the scheme with a more balanced load when the output value difference is less than a set range.

[0012] S4. Rate Execution and Feedforward Compensation: The optimal rate parameters selected in step S3 are sent to the actuators of each device, and Vf, Vp1, Vp2, Vs and Va are adjusted synchronously. At the same time, based on the identification results of the next refrigerator to be processed, the changes in material characteristics are predicted, and the key parameters in the correlation model are adjusted by feedforward compensation one control cycle in advance to suppress the system disturbance caused by the feed switching.

[0013] S5. Closed-loop feedback and model self-learning: Monitor the actual recovery rate (η') of non-ferrous metals at the discharge port of the eddy current separator and compare it with the expected recovery rate η; if the error continues to exceed the preset threshold, the self-learning mechanism is triggered, and the mapping relationship between Vs and η in step S1 is incrementally updated using the latest production data, so that the model can continuously adapt to the long-term changes in material properties.

[0014] Existing refrigerator crushing and recycling production lines suffer from systemic defects in rate scheduling. The core problem lies in the lack of intelligent, adaptive, and collaborative control capabilities when faced with highly uncertain and dynamically changing incoming material conditions. Specific technical challenges primarily stem from the unpredictability of the materials. Waste refrigerators exhibit significant differences in model, age, and structural components, resulting in strong fluctuations in the composition, particle size, and shape distribution of the crushed material. Traditional control systems, employing fixed parameters or simple adjustments based on limited feedback, cannot anticipate these changes. This leads to a mismatch between the processing capacity of upstream crushing equipment and downstream sorting equipment. For example, when the discharge particle size of the primary crusher suddenly coarsens, the secondary crusher may fail to accelerate in time, leading to a risk of blockage. Or, when a large amount of flake aluminum accumulates at the inlet of the eddy current separator, the fixed magnetic roller speed cannot achieve optimal sorting efficiency, resulting in the loss of valuable metals.

[0015] Existing technologies lack a global optimization perspective. Each piece of equipment is often controlled independently or only has a simple interlocking start-stop relationship. They fail to treat maximizing the output value of the entire production line, optimizing energy consumption, and balancing equipment load as a unified, multi-objective optimization problem. Their scheduling decisions are usually local and short-sighted, and they cannot pursue the best economic benefits while ensuring the stable operation of the system. For example, they may overload the eddy current separator in order to achieve a short-term high throughput, which will eventually lead to a shortened equipment life and a decrease in sorting purity.

[0016] Existing control strategies are reactive rather than predictive, and can only be adjusted after materials have entered downstream equipment and caused deviations in process parameters. This "locking the stable door after the horse has bolted" approach inevitably involves efficiency losses and product quality fluctuations during the adjustment process, and cannot achieve a smooth transition of production status.

[0017] The system's long-term adaptability is insufficient. Slow drift in equipment performance, tool wear, and seasonal changes in material properties can all cause the initially set control model to gradually become ineffective. Traditional methods rely on manual experience for periodic, offline parameter tuning, resulting in slow response and limited accuracy. The technical challenge addressed by this invention is to systematically address these challenges by constructing an intelligent scheduling system that integrates real-time sensing, multi-objective dynamic optimization, feedforward compensation, and closed-loop self-learning. It accurately senses and predicts material characteristics through a multi-source sensor network, providing a data foundation for dynamic decision-making; it achieves coordinated matching of equipment speeds by establishing a rate correlation model between equipment and using advanced algorithms for online global optimization; it significantly reduces the disturbance to the system caused by material switching by introducing feedforward control based on next material prediction; and finally, through the model's self-learning mechanism, the system can continuously evolve, maintaining high efficiency and stability in long-term operation, thus fundamentally overcoming the inherent technical bottlenecks of traditional static, passive, and isolated control modes.

[0018] In a preferred embodiment, the visual recognition system performs three-dimensional contour scanning and model recognition, and queries a preset material database to obtain the estimated material composition of the refrigerator to be processed; the composition detection unit identifies and classifies the physical morphology of non-ferrous metal particles at the feed inlet of the eddy current separator in real time through spectral analysis and high-speed image acquisition.

[0019] As a preferred implementation, when constructing the multi-objective optimization function, the output value per unit time of the production line is used as the core economic indicator, the load balance of key equipment is used as the stability indicator, and the total power consumption is limited to a safe threshold as a constraint.

[0020] As a preferred implementation, the feedforward compensation adjustment specifically involves: predicting the characteristics of the next batch of refrigerator material to be processed, and adjusting the coordinated working points of the downstream key sorting equipment and the upstream crushing equipment in advance to mitigate the impact of drastic fluctuations in the composition of the incoming material on the sorting effect.

[0021] As a preferred implementation, after the model self-learning mechanism is triggered, it collects recent production data to form an incremental dataset, fine-tunes the parameters of the nonlinear model, and verifies the model before it is put into online application to ensure the improvement of its prediction accuracy.

[0022] After adopting the above technical solution, the beneficial effects of the present invention are as follows: the most core benefit is that it simultaneously improves the overall production efficiency and resource recovery rate. Through precise real-time perception and dynamic optimization, the system can always keep each piece of equipment running at the working point that best matches the material it is currently processing, ensuring that the crushing efficiency and sorting accuracy are always maintained at a high level, thereby directly increasing the output of valuable metals per unit time and reducing the valuable components remaining in the waste, thus improving the level of comprehensive resource utilization.

[0023] This method greatly enhances the stability and robustness of the production system. Faced with complex and ever-changing upstream materials, the system no longer responds passively but can adjust key parameters in advance through a feedforward compensation mechanism. This effectively buffers the impact of sudden changes in material characteristics on the process flow, avoids abnormal operating conditions such as equipment overload and blockage, and ensures continuous and stable production. At the same time, the consideration of equipment load balancing in multi-objective optimization avoids long-term overload operation of a single piece of equipment, which helps to extend the service life of core equipment.

[0024] This method enables intelligent and adaptive production processes. The system no longer relies on manual intervention and experience-based judgment by operators, reducing the demand for manpower and operational uncertainty. In particular, the built-in model self-learning function allows the control system to continuously improve and optimize itself as production data accumulates. Even in the face of long-term changes in material composition or equipment performance degradation, it can automatically maintain a high level of control quality and has the ability to cope with future uncertainties.

[0025] In terms of energy consumption and economic benefits, this method avoids unnecessary energy waste through system power consumption constraints and global optimization. For example, it automatically reduces the energy consumption of the crusher when the material is light and optimizes the energy consumption of the air separator when the separation pressure is low, thereby reducing the comprehensive energy consumption cost per unit product and improving the overall economy of the production line.

[0026] This invention upgrades traditional rigid production lines into intelligent, flexible, and adaptive production systems, resulting in profound positive impacts on multiple dimensions, including efficiency, stability, intelligence level, and economic benefits. Attached Figure Description

[0027] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0028] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0029] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0030] Example:

[0031] like Figure 1 As shown, a method for sorting and intelligent resource scheduling of crushed waste electrical appliances is disclosed, the method comprising the following steps:

[0032] S1. Establish a full-process equipment rate correlation model: Define the coupling relationship between feed rate (Vf), primary crusher rotor speed (Vp1), secondary crusher rotor speed (Vp2), eddy current separator magnetic roller speed (Vs), and air separator fan frequency (Va); among them, Vp1 and Vf are correlated through an adjustable proportional coefficient K1, which is dynamically set according to the identified refrigerator material characteristics; Vp2 is correlated with the real-time average particle size (D50) of the material after primary crushing; Vs is correlated with the expected recovery rate (η) of the target non-ferrous metal through a pre-trained nonlinear model, which is established based on the sorting experimental data of copper and aluminum particles with different properties;

[0033] S2. Multi-source real-time sensing and material characteristic identification: The material composition of the refrigerator to be processed is estimated through the visual recognition system at the feed end; D50 is obtained in real time through the online particle size analyzer at the discharge port of the primary crusher; the type, shape and proportion of non-ferrous metals are identified through the component detection unit at the feed port of the eddy current separator.

[0034] S3. Multi-objective dynamic optimization decision: The central control unit receives the real-time data from step S2, and constructs a multi-objective optimization function with the goals of maximizing the overall output value and balancing the equipment load, and with system power consumption as a constraint. The function is solved online using a non-dominated sorting genetic algorithm with an elite retention strategy, and a set of optimal equipment rate parameter combinations is selected from the Pareto solution set according to a preset priority. The priority rule is to prioritize the scheme with a more balanced load when the output value difference is less than a set range.

[0035] S4. Rate Execution and Feedforward Compensation: The optimal rate parameters selected in step S3 are sent to the actuators of each device, and Vf, Vp1, Vp2, Vs and Va are adjusted synchronously. At the same time, based on the identification results of the next refrigerator to be processed, the changes in material characteristics are predicted, and the key parameters in the correlation model are adjusted by feedforward compensation one control cycle in advance to suppress the system disturbance caused by the feed switching.

[0036] S5. Closed-loop feedback and model self-learning: Monitor the actual recovery rate (η') of non-ferrous metals at the discharge port of the eddy current separator and compare it with the expected recovery rate η; if the error continues to exceed the preset threshold, the self-learning mechanism is triggered, and the mapping relationship between Vs and η in step S1 is incrementally updated using the latest production data, so that the model can continuously adapt to the long-term changes in material properties.

[0037] The technical solution protected in this application, in the specific implementation scenario of a refrigerator crushing and recycling production line, works on the basis of a closed-loop control system that integrates real-time perception, intelligent decision-making, forward-looking control, and self-optimization, thereby improving the overall efficiency of the production line through multi-level technical collaboration.

[0038] During implementation, when the waste refrigerator enters the starting end of the processing line through the feeding mechanism, the vision recognition system installed above the feeding conveyor is immediately activated to perform three-dimensional scanning and image acquisition of the refrigerator shell. Through the embedded deep learning algorithm, the refrigerator model is identified and the pre-stored material database is queried to estimate the overall material composition of the refrigerator, including the distribution of metal parts, plastic type and approximate weight. This information provides important pre-processing parameters for the subsequent crushing process.

[0039] At the same time, a rate correlation model based on physical laws and process knowledge has been established for the key equipment on the production line. This model dynamically correlates core parameters such as feed rate, primary crusher rotor speed, secondary crusher rotor speed, eddy current separator magnetic roller speed, and air separator fan frequency, forming a parameter system that is mutually restrictive and works in synergy.

[0040] As the refrigerator enters the primary crusher, the crushed material is transported to the next process via a discharge conveyor. During this process, an online particle size analyzer installed at the discharge port of the primary crusher continuously monitors the particle size distribution of the crushed material, calculates and outputs the average particle size value in real time. This real-time data is directly fed back to the control system to dynamically adjust the rotor speed of the secondary crusher, ensuring that the feed particle size for the secondary crushing remains within the ideal range. After the material passes through the secondary crushing stage, it enters the sorting stage. At the inlet of the eddy current separator, a component detection unit consisting of a near-infrared spectrometer and a high-speed camera begins operation. The near-infrared spectrometer analyzes the polymer components in the material, while the high-speed camera captures the visual characteristics of the material in flight. Image recognition algorithms distinguish the shape characteristics and approximate proportions of non-ferrous metal particles such as copper and aluminum. This real-time sensed data is combined with a pre-established nonlinear model to dynamically optimize the magnetic roller speed setpoint of the eddy current separator.

[0041] The central control unit, as the intelligent core of the entire system, gathers all sensing data and runs a multi-objective optimization algorithm. This algorithm simultaneously considers two objectives: maximizing the overall production value and balancing equipment load. Under the safety constraint of total system power consumption, it uses a non-dominated sorting genetic algorithm with an elite retention strategy to solve the problem in real time, generating multiple Pareto optimal solutions. Then, according to preset decision rules, it selects the most suitable combination of equipment parameters. When the predicted production values ​​of different parameter combinations are similar, the system prioritizes the scheme with more balanced equipment load to ensure production stability. During the parameter execution phase, the control system sends the optimized rate commands to the actuators of each device, synchronously adjusting the feed rate, crusher speed, separator magnetic roller speed, and fan frequency to achieve coordinated operation of the entire line.

[0042] More importantly, the system has a forward-looking control capability. When the vision recognition system has completed the feature recognition of the next refrigerator to be processed, the control system will predict the changing trend of material characteristics in advance, and make feedforward compensation adjustments to the key parameters in the correlation model based on this prediction. For example, when it is detected that the next refrigerator contains more thick-walled metal parts, the speed reference value of the crusher will be appropriately increased in advance. This feedforward mechanism effectively suppresses system fluctuations caused by material switching.

[0043] At the end of the sorting process, a metal composition detection sensor is installed at the discharge port of the eddy current separator to continuously monitor the actual recovery rate of the sorted material and compare this measured value with the expected value of the control system. When the deviation continues to exceed the allowable range, the system automatically triggers the model self-learning mechanism to incrementally update the control model of the eddy current separator using the latest production data. Through parameter fine-tuning, the model can better adapt to the slow changes in material characteristics during long-term operation. This self-optimization capability ensures the continuous improvement of system performance.

[0044] The entire workflow starts with material identification, goes through parameter optimization, instruction execution, feedforward compensation and feedback learning, forming a complete intelligent control closed loop. This enables the refrigerator crushing and recycling production line to autonomously adapt to changing incoming material conditions and maintain a highly efficient, stable and optimized operating state, effectively improving resource recycling efficiency and the level of production line intelligence.

[0045] In the actual operation of a refrigerator recycling production line, the implementation of the visual recognition system forms the front-end perception foundation of the entire adaptive control system. When a used refrigerator enters the pre-processing area via a chain conveyor, the 3D visual recognition system installed above the feed inlet immediately starts working. This system consists of multiple industrial-grade stereo cameras and structured light projectors, capable of simultaneously performing a panoramic scan of the moving refrigerator from different angles. During the scanning process, the system accurately reconstructs the refrigerator's external contour using point cloud data processing technology, accurately capturing its length, width, height, door structural features, and key geometric parameters such as the compressor compartment position. This 3D morphological data, combined with synchronously acquired 2D surface texture images, is fed into an image recognition neural network trained on a large number of samples. This network can quickly match the corresponding product model based on visual features such as the refrigerator's brand logo, door handle design, and heat dissipation grille style. After obtaining accurate model information, the system immediately queries a pre-set material database, which compiles detailed material lists for mainstream refrigerator models on the market, including key information such as the thickness of the cabinet steel plate, the type of insulation material, the layout of the refrigeration pipes, and the compressor specifications. Based on this data, the system can estimate the approximate material composition of the refrigerator being processed, such as the ratio of galvanized steel sheet to ABS plastic in the outer shell, and the estimated content of internal copper pipes and aluminum fins. This estimated data provides an important basis for the subsequent pre-adjustment of crushing process parameters, enabling the system to prepare corresponding processing strategies in advance for refrigerators with different structures.

[0046] When the material undergoes multi-stage crushing and reaches the sorting section, the component detection unit installed at the feed inlet of the eddy current separator plays a crucial role. This detection unit is a comprehensive monitoring platform integrating multiple sensing technologies, with its core including a high-resolution near-infrared spectrometer and a high-speed industrial camera system. As the crushed mixture passes through the detection area in a uniform thin layer, the near-infrared spectrometer emits light waves of a specific wavelength at extremely high frequencies onto the material surface and identifies the molecular vibrational characteristics of different materials by receiving the reflected spectral signals. This spectral analysis can effectively distinguish common plastic types such as polyethylene, polypropylene, and polyurethane foam, and can also identify the presence of metals such as copper and aluminum through characteristic absorption peaks. Simultaneously, a high-speed camera equipped with a special lighting system continuously captures the falling material flow at an extremely high acquisition rate, capturing the flight attitude and surface morphology of each particle. These high-definition images are transmitted in real time to the image processing unit, where specialized machine learning algorithms analyze the particle contour features, surface reflectivity, and trajectory. The trained algorithm can accurately distinguish different forms of non-ferrous metal particles, such as sheet-like copper, wire-like copper, block-like aluminum, and foil-like aluminum. By integrating spectral features and visual morphological information, the system can not only identify the types of non-ferrous metals, but also accurately classify their physical forms. This real-time detection data provides a direct basis for subsequent optimization of the operating parameters of the eddy current separator.

[0047] In the control system of a refrigerator crushing and recycling production line, the construction of a multi-objective optimization function is the core of the entire intelligent scheduling decision-making process. After the central control unit receives real-time data from various sensing nodes, it first needs to establish economic indicators that comprehensively reflect the production line's operating status. The output value per unit time, as the core economic indicator, is calculated considering not only the real-time market prices of various recyclable components in the current material but also the material composition prediction data provided by the visual recognition system and the actual sorting effect feedback from the component detection unit. The system dynamically evaluates the real-time value output of recyclables such as copper, aluminum, and plastics of different qualities through a built-in value calculation model, while also considering equipment operating costs and energy consumption expenditures, forming a comprehensive economic benefit evaluation indicator. This indicator accurately reflects the production line's profitability under the current parameter settings, providing a clear economic guide for the optimization algorithm. In actual operation, when processing high-value refrigerators rich in copper tubes and aluminum fins, the system tends to select parameter combinations that maximize the recovery of these valuable metals; while when processing low-value refrigerators with predominantly plastic shells, it focuses more on energy consumption control and production cost optimization.

[0048] Alongside economic efficiency indicators, stability indicators of the production system are considered, primarily reflected through the load balance of key equipment. In actual production environments, the load status of core equipment such as primary crushers, secondary crushers, eddy current separators, and air separators directly affects the continuous operation capability of the entire production line. The control system calculates the real-time load rate of each piece of equipment by monitoring operating parameters such as motor current, bearing temperature, and vibration amplitude. The load balance index reflects the overall system's coordination by assessing the dispersion of the load rates of these key equipment. When some equipment is under high load for an extended period while other equipment is under light load, the system's balance index deteriorates. In this case, the optimization algorithm tends to adjust the operating parameters between equipment, appropriately shifting the load from high-load equipment to low-load equipment. For example, when the system detects that the primary crusher is overloaded while the secondary crusher still has redundant capacity, it may appropriately reduce the feed rate or adjust the crusher speed ratio, sacrificing a small amount of processing capacity to achieve a more balanced distribution of equipment load, thereby avoiding cascading shutdowns caused by overload of a single piece of equipment.

[0049] While pursuing economic benefits and operational stability, the system must also ensure that total power consumption remains within a safe threshold range. This constraint is set based on the carrying capacity of the production line's power supply system and the rated power parameters of each piece of equipment. The control system continuously monitors the instantaneous and cumulative energy consumption of the entire production line by collecting real-time power data from the main circuit. When the optimization algorithm is searching for the optimal parameter combination, any solution that causes total power consumption to exceed the safe threshold is automatically eliminated. In practical applications, this constraint mechanism effectively prevents grid overload caused by multiple high-power devices operating at peak speeds simultaneously. For example, during periods of high summer temperatures, when ambient temperature causes a decrease in equipment heat dissipation efficiency, the system automatically lowers the power consumption threshold and adjusts the equipment operating parameters accordingly to ensure production safety. This intelligent decision-making system based on multi-objective optimization enables the production line to find the optimal balance between economic benefits, operational stability, and energy consumption, achieving truly intelligent production management.

[0050] In the actual operation of a refrigerator crushing and recycling production line, a feedforward compensation adjustment mechanism is a crucial technical means to address fluctuations in incoming materials. This proactive control strategy relies on the accurate identification of subsequent refrigerators and the prediction of their material properties by a visual recognition system. Once the system obtains detailed characteristics of the next refrigerator through 3D scanning and model recognition, it immediately initiates the feedforward compensation analysis process. The control unit compares the identified refrigerator model with information in the material database to obtain typical material composition characteristics of that model, including the distribution of metal components, the proportion of different types of plastic materials, and the installation locations of special components. Based on this predictive information, the system simulates the material state of the refrigerator after crushing and anticipates potential impacts on downstream sorting equipment. For example, when the next refrigerator is identified as a commercial display case made of thick-walled steel plates and dense copper tubing, the system predicts that crushing it will generate a large amount of heavy metal fragments and coarse-diameter copper materials, which will significantly affect the sorting effect of the eddy current separator.

[0051] Based on this prediction, the control system adjusts the operating parameters of the upstream crushing equipment in advance. For conditions involving the processing of heavy metal materials, the system appropriately increases the speed setting of the primary crusher to ensure effective crushing of thicker metal shells; simultaneously, it adjusts the rotor clearance of the secondary crusher to better suit the processing of high-hardness metal materials. These adjustments aim to ensure that the crushed material achieves an ideal particle size distribution suitable for sorting, creating favorable pretreatment conditions for downstream sorting processes. More importantly, the system synchronously adjusts the expected operating state of the downstream sorting equipment, especially making preliminary settings for key parameters of the eddy current separator. When a significant change in the metal content and characteristics of the subsequent incoming material is predicted, the system recalibrates the operating point of the separator in advance, including adjusting parameters such as magnetic roller speed, electrode spacing, and belt speed, ensuring that it is already in optimal operating condition when the material arrives at the sorting zone.

[0052] The core value of this feedforward compensation mechanism lies in breaking the lag pattern of "detection-response" in traditional control systems and establishing an advanced "prediction-preparation" control logic. In actual production scenarios, when a household refrigerator, primarily made of plastic, is followed closely by a commercial refrigerator densely packed with metal, without feedforward compensation, the sorting system will only detect the decrease in sorting efficiency when the heavy metal material arrives, and then begin adjusting parameters. This process results in the loss of a significant amount of valuable metal. However, with feedforward compensation, the system begins adjusting the operating parameters of the entire production line before the commercial refrigerator even enters the crusher. This ensures that when the material flow characteristics change abruptly, each piece of equipment is already prepared, effectively mitigating the impact of fluctuations in incoming material composition on the sorting effect. This control method greatly improves the production line's ability to cope with complex incoming material conditions, ensuring the stability of recycling efficiency and product quality.

[0053] In the long-term operation of the refrigerator crushing and recycling production line, the model self-learning mechanism is a crucial guarantee for the continuous optimization of system performance. This mechanism begins with the detection of self-learning trigger conditions. The control system continuously compares the actual recovery rate monitored at the eddy current separator outlet with the expected recovery rate predicted by the model. When the deviation between the two consistently exceeds the allowable range, the self-learning process is automatically triggered. This triggering mechanism is designed to fully consider the complexity of the industrial production environment, avoiding false triggers caused by instantaneous fluctuations and ensuring that model updates are only initiated when a systematic deviation actually occurs. After triggering, the system immediately begins collecting complete operational data from the most recent production cycle. This data includes the visual recognition results of each refrigerator, equipment parameters during the crushing process, the operating status of the sorting stage, and the final product quality inspection results, forming a timely and representative incremental dataset.

[0054] After data acquisition, the system enters the model parameter fine-tuning stage. This stage first preprocesses the collected incremental data, including outlier removal, data standardization, and feature engineering, to ensure the quality and consistency of the input data. Then, an incremental learning algorithm is used to optimize the parameters of the eddy current sorting nonlinear model established in step S1. During this process, the system retains the core features of the original model, only adjusting parameters that show deviations. This gradual optimization approach ensures model stability while effectively adapting to the slow changes in production processes. The algorithm used in the optimization process specifically considers the real-time requirements of industrial applications, enabling rapid calculations while maintaining accuracy, avoiding disruption to normal production. The entire parameter adjustment process is cross-validated based on historical data to ensure that each parameter modification brings a real improvement in the model's predictive ability.

[0055] Model validation is a crucial step in the self-learning process. After parameter fine-tuning, the system does not immediately deploy the new model online but first conducts thorough testing in a validation environment. The validation process uses recently collected production data as a test set to comprehensively evaluate the updated model's prediction accuracy and generalization ability under various operating conditions. The system simulates different incoming material conditions and equipment states to examine the new model's performance in these scenarios, paying particular attention to its stability under boundary conditions. Only when the new model demonstrates performance superior to or at least equal to the original model in all test scenarios is it approved for online application. During model switching, the system employs a smooth transition strategy, gradually transferring control from the old model to the new model while closely monitoring changes in key process parameters to ensure a smooth transition in production. This rigorous self-learning mechanism ensures that the control system can continuously adapt to the impact of long-term factors such as equipment wear and changes in material properties, always maintaining optimal operating conditions.

[0056] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for sorting and intelligent resource scheduling of crushed waste electrical appliances, characterized in that, The method includes the following steps: S1. Establish a full-process equipment rate correlation model: Define the coupling relationship between feed rate Vf, primary crusher rotor speed Vp1, secondary crusher rotor speed Vp2, eddy current separator magnetic roller speed Vs, and air separator fan frequency Va; among them, Vp1 and Vf are correlated through an adjustable proportional coefficient K1, which is dynamically set according to the identified refrigerator material characteristics; Vp2 is correlated with the real-time average particle size D50 of the material after primary crushing; Vs is correlated with the expected recovery rate η of the target non-ferrous metal through a pre-trained nonlinear model, which is established based on the sorting experimental data of copper and aluminum particles with different properties; S2. Multi-source real-time sensing and material characteristic identification: The material composition of the refrigerator to be processed is estimated through the visual recognition system at the feed end; D50 is obtained in real time through the online particle size analyzer at the discharge port of the primary crusher; the type, shape and proportion of non-ferrous metals are identified through the component detection unit at the feed port of the eddy current separator. S3. Multi-objective dynamic optimization decision: The central control unit receives the real-time data from step S2, and constructs a multi-objective optimization function with the goals of maximizing the overall output value and balancing the equipment load, and with system power consumption as a constraint. The function is solved online using a non-dominated sorting genetic algorithm with an elite retention strategy, and a set of optimal equipment rate parameter combinations is selected from the Pareto solution set according to a preset priority. The priority rule is to prioritize the scheme with a more balanced load when the output value difference is less than a set range. S4. Rate Execution and Feedforward Compensation: The optimal rate parameters selected in step S3 are sent to the actuators of each device, and Vf, Vp1, Vp2, Vs and Va are adjusted synchronously. At the same time, based on the identification results of the next refrigerator to be processed, the changes in material characteristics are predicted, and the key parameters in the correlation model are adjusted by feedforward compensation one control cycle in advance to suppress the system disturbance caused by the feed switching. S5. Closed-loop feedback and model self-learning: Monitor the actual recovery rate η' of non-ferrous metals at the discharge port of the eddy current separator and compare it with the expected recovery rate η; if the error continues to exceed the preset threshold, the self-learning mechanism is triggered, and the mapping relationship between Vs and η in step S1 is incrementally updated using the latest production data, so that the model can continuously adapt to the long-term changes in material properties.

2. The method for sorting and intelligent resource scheduling of waste electrical appliance crushed materials as described in claim 1, characterized in that: The visual recognition system performs three-dimensional contour scanning and model recognition, and queries a pre-set material database to obtain the estimated material composition of the refrigerator to be processed; the composition detection unit identifies and classifies the physical morphology of non-ferrous metal particles at the feed inlet of the eddy current separator in real time through spectral analysis and high-speed image acquisition.

3. The method for sorting and intelligent resource scheduling of waste electrical appliance crushed materials as described in claim 1, characterized in that: When constructing the multi-objective optimization function, the output value per unit time of the production line is used as the core economic indicator, the load balance of key equipment is used as the stability indicator, and the total power consumption is limited to a safe threshold as a constraint.

4. The method for sorting and intelligent resource scheduling of crushed waste electrical appliances as described in claim 1, characterized in that: The feedforward compensation adjustment specifically refers to: predicting the characteristics of the next batch of refrigerator materials to be processed, and adjusting the coordinated working points of the downstream key sorting equipment and the upstream crushing equipment in advance to mitigate the impact of drastic fluctuations in the composition of the incoming material on the sorting effect.

5. The method for sorting and intelligent resource scheduling of waste electrical appliance crushed materials as described in claim 1, characterized in that: After being triggered, the model's self-learning mechanism collects recent production data to form an incremental dataset, fine-tunes the parameters of the nonlinear model, and verifies the model before it is put into online application to ensure the improvement of its prediction accuracy.

Images