A method and system for formulation optimization of rubber-construction waste mix

Abstract

The present application relates to the cross technical field of civil engineering material test, geotechnical engineering information technology and solid waste resource utilization, in particular to a formula optimization method and system of rubber-architectural solid waste mixture. The method comprises: obtaining digital particle archives of rubber particles and architectural solid waste; obtaining training samples through multifunctional linkage test and constructing performance prediction model; obtaining particle swarm optimization algorithm and improving the particle swarm optimization algorithm to obtain improved particle swarm optimization algorithm; based on the improved particle swarm optimization algorithm, using the digital particle archives and the performance prediction model to calculate the optimal formula. The present application solves the problems of low efficiency and unreliability of rubber-architectural solid waste mixture formula optimization in the prior art. Through multifunctional linkage test to obtain training samples and construct performance prediction model, and using improved particle swarm optimization algorithm for intelligent reverse optimization of engineering target, the present application improves the automation level, design reliability and efficiency of formula optimization.

Description

Technical Field

[0001] This invention relates to the interdisciplinary fields of civil engineering material testing, geotechnical engineering information technology, and solid waste resource utilization, specifically a method and system for optimizing the formulation of rubber-construction solid waste mixtures. Background Technology

[0002] In the fields of civil engineering and solid waste resource utilization, mixing waste tire rubber granules with construction solid waste (such as crushed concrete and bricks) to form lightweight fillers has become an important way to solve environmental problems and obtain engineering materials. The engineering performance of such rubber-construction solid waste mixtures is affected by a complex coupling of multiple factors such as mix proportion, particle morphology, interface characteristics, stress history, and ambient temperature. Therefore, formula optimization is crucial for engineering applications.

[0003] Currently, formulation optimization methods for such mixtures suffer from the following shortcomings: First, there is a lack of digital characterization of raw materials. Traditional methods only involve simple crushing and sieving of raw materials, without digitally characterizing the three-dimensional morphology and internal microstructure of particles. This makes it impossible to establish a quantitative correlation between the inherent properties of raw materials and macroscopic performance, resulting in a lack of precise input for formulation optimization. Second, there is insufficient in-situ multiphysics testing methods. Mechanical testing, microstructure observation, micro-damage monitoring, and environmental temperature control are usually performed by independent equipment. This makes it impossible to simultaneously acquire multi-scale data of stress, temperature, structural, and damage fields under the same sample and loading path, resulting in a lack of multiphysics features in training samples and limiting the prediction accuracy of subsequent machine learning models. Third, there is a lack of intelligent prediction and reverse design capabilities. Existing methods rely on trial-and-error forward verification, which is inefficient and cannot recommend the optimal ratio based on the target engineering performance. Furthermore, there is a lack of machine learning models based on multi-source data fusion, making it difficult to establish a high-precision nonlinear mapping relationship between formulation and performance.

[0004] Therefore, there is an urgent need for a method to optimize rubber-construction solid waste mixture formulations that can achieve digital characterization of raw materials, in-situ synchronous testing of multi-physics fields, fusion of multi-source data, and intelligent reverse design, so as to improve the efficiency and reliability of formulation design. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a method and system for optimizing the formulation of rubber-construction solid waste mixtures, solving the problems of low efficiency and unreliability in the optimization of rubber-construction solid waste mixture formulations in existing technologies.

[0006] To achieve the above objectives, one aspect of the present invention provides a method for optimizing the formulation of a rubber-construction solid waste mixture, the method comprising: acquiring digital particle files of rubber particles and construction solid waste; acquiring training samples and constructing a performance prediction model through a multi-functional linkage experiment; acquiring a particle swarm optimization algorithm and improving the particle swarm optimization algorithm to obtain an improved particle swarm optimization algorithm; and calculating the optimal formulation based on the improved particle swarm optimization algorithm using the digital particle files and the performance prediction model.

[0007] This invention acquires digital particle files of rubber granules and construction solid waste, accurately establishes the input basis of the inherent properties of raw materials, obtains training samples through multi-functional linkage experiments and constructs a performance prediction model, realizes multi-physics data fusion and high-precision performance mapping, improves the particle swarm optimization algorithm and calculates the optimal formula by combining digital files and performance prediction model, realizes intelligent reverse optimization for engineering goals, effectively avoids local optima, and improves the automation level, design reliability and efficiency of formula optimization.

[0008] Optionally, obtaining the digital particle profiles of rubber granules and construction solid waste includes: crushing and screening the rubber granules and construction solid waste to obtain solid waste particles; performing three-dimensional morphological scanning and internal structure scanning on the solid waste particles to obtain particle point cloud data and internal microstructure data; performing surface reconstruction using spherical harmonic functions based on the particle point cloud data to obtain morphological reconstruction coefficients; and binding the morphological reconstruction coefficients with the internal microstructure data to generate the digital particle profiles.

[0009] This invention obtains uniform solid waste particles by crushing and screening rubber granules and construction solid waste, ensuring the stability and accuracy of subsequent scanning. It performs three-dimensional morphological and internal structure scanning on the solid waste particles to acquire particle point cloud data and internal microstructure data, achieving comprehensive and accurate acquisition of particle features. Based on the particle point cloud data, surface reconstruction is performed using spherical harmonic functions to obtain morphological reconstruction coefficients, improving the accuracy and fidelity of the particle's three-dimensional morphology representation. The morphological reconstruction coefficients are then bound to the internal microstructure data to generate a digital particle archive, achieving a unified correlation between the particle's external morphology and internal structure, thus improving the scientific rigor and accuracy of the digital particle archive.

[0010] Optionally, the step of obtaining training samples and constructing a performance prediction model through multi-functional linkage experiments includes: acquiring multiple historical digital particle archives and preparing standard samples using the historical digital particle archives based on pre-set ratio parameters; subjecting the standard samples to triaxial loading and simultaneously performing microfocus X-ray CT scanning, distributed acoustic emission monitoring, and high and low temperature environmental control to obtain multi-physics field test data; obtaining a multi-scale feature dataset and macroscopic performance indicators based on the multi-physics field test data; associating and storing the ratio parameters, the multi-scale feature dataset, the digital particle archives, and the macroscopic performance indicators to construct a multi-scale performance database; and training a multi-task deep neural network based on the multi-scale performance database to obtain the performance prediction model.

[0011] This invention prepares standard samples by acquiring historical digital particle archives, conducts triaxial loading and multi-physics field synchronous monitoring to obtain experimental data, forms a multi-scale feature dataset and macroscopic performance indicators, constructs a multi-scale performance database and trains a multi-task deep neural network, realizes data closure and accurate model iteration, and improves the generalization ability and prediction accuracy of the performance prediction model.

[0012] Optionally, obtaining the multi-scale feature dataset and macroscopic performance indicators based on the multi-physics test data includes: calculating the peak deviatoric stress, secant modulus, and critical state friction angle based on the multi-physics test data to form macroscopic performance indicators; and obtaining the multi-scale feature dataset based on the multi-physics test data.

[0013] This invention calculates peak deviatoric stress, secant modulus, and critical friction angle based on multiphysics field test data to form macroscopic performance indicators. Simultaneously, it extracts multiscale feature datasets based on multiphysics field test data, achieving accurate matching between macroscopic performance indicators and multiscale feature data. This ensures complete data dimensions and comprehensive feature coverage, improving the utilization rate of test data and the effectiveness of features.

[0014] Optionally, obtaining the multi-scale feature dataset based on the multiphysics test data includes: extracting mechanical data, CT image data, acoustic emission data, and temperature data from the multiphysics test data; calculating the true volumetric strain based on the CT image data using digital volume image correlation methods; calculating the apparent volumetric strain based on the mechanical data; correcting the apparent volumetric strain based on the true volumetric strain to obtain the corrected volumetric strain; performing time-frequency analysis, nonlinear dynamic analysis, and high-order feature extraction on the acoustic emission data to obtain a damage feature vector; and fusing the mechanical data, the CT image data, the temperature data, the corrected volumetric strain, and the damage feature vector to obtain the multi-scale feature dataset.

[0015] This invention extracts mechanical, CT image, acoustic emission, and temperature data from multi-physics field test data, combines them with digital volume image correlation methods to accurately calculate the true volumetric strain and correct the apparent volumetric strain, thereby improving the accuracy of strain data. At the same time, it performs multi-dimensional analysis of acoustic emission data to obtain highly recognizable damage feature vectors, effectively integrating multi-source data to form a complete multi-scale feature dataset, thus improving data reliability.

[0016] Optionally, the corrected volumetric strain satisfies the following formula: , in, Loading time Corrected volumetric strain at that time Rubber content and ambient temperature The relevant correction factor, Loading time Apparent volumetric strain at time This is the weighting adjustment coefficient. Loading time The actual volumetric strain at that time, Loading time The axial strain value at that time.

[0017] This invention introduces a corrected volumetric strain formula, combines it with a correction coefficient related to rubber content and ambient temperature, integrates apparent volumetric strain with actual volumetric strain measured by CT, and dynamically adjusts the correction weight through an attenuation coefficient. This effectively corrects the measurement deviation of traditional apparent volumetric strain, accurately restores the true volumetric deformation of the mixture during loading, and improves the accuracy and reliability of volumetric strain data.

[0018] Optionally, the step of performing time-frequency analysis, nonlinear dynamics analysis, and high-order feature extraction on the acoustic emission data to obtain the damage feature vector includes: extracting amplitude, energy, rise time to amplitude ratio, and center frequency from the acoustic emission data; performing wavelet packet decomposition on the acoustic emission data and calculating the wavelet packet energy entropy; performing multifractal detrending fluctuation analysis on the acoustic emission data and calculating the multifractal spectrum width; performing phase space reconstruction on the acoustic emission data and calculating the maximum Lyapunov exponent; and combining the amplitude, energy, rise time to amplitude ratio, center frequency, wavelet packet energy entropy, multifractal spectrum width, and maximum Lyapunov exponent to obtain the damage feature vector.

[0019] This invention extracts fundamental time-domain features such as amplitude and energy from acoustic emission data, performs wavelet packet decomposition to calculate energy entropy to capture time-frequency domain damage information, conducts multifractal detrending fluctuation analysis to obtain spectral width to characterize the nonlinear evolution of damage, and calculates the maximum Lyapunov exponent through phase space reconstruction to quantify the chaotic characteristics of the system. By combining multidimensional features to form a highly recognizable damage feature vector, it comprehensively covers the multi-scale nonlinear features of damage evolution, thereby improving the characterization accuracy and identification reliability of the damage state of the mixture.

[0020] Optionally, the step of calculating the optimal formula based on the improved particle swarm optimization algorithm using the digital particle profile and the performance prediction model includes: setting an initial population; calculating the predicted performance index of individuals in the initial population using the performance prediction model based on the initial population and the digital particle profile; obtaining the target engineering performance index; calculating the objective function value of the individual using the target engineering performance index and the predicted performance index; and iterating using the improved particle swarm optimization algorithm based on the objective function value to obtain the optimal formula.

[0021] This invention sets up an initial population and combines it with digital particle profiles. It uses a performance prediction model to calculate predicted performance indicators, matches the target engineering performance indicators to accurately calculate individual objective function values, and uses an improved particle swarm optimization algorithm for efficient iteration based on the objective function values. This achieves intelligent reverse precision solution of the formula, effectively avoiding the defects of traditional algorithms such as local optima and low optimization efficiency, and improving the accuracy, convergence speed and engineering applicability of formula optimization.

[0022] Optionally, the improved particle swarm optimization algorithm satisfies the following formula: , in, For the first During the nth iteration The speed of each particle For inertial weights, For the first During the nth iteration The speed of each particle All are learning factors. It is a random number. For the first The individual best position in the history of each particle. The optimal position globally. For the first The optimal neighborhood position of each particle. For the first During the nth iteration The position of each particle. For the first During the nth iteration The position of each particle.

[0023] This invention introduces an improved particle swarm optimization algorithm formula that includes the optimal neighborhood position. Based on the traditional individual optimality and global optimality, it adds a neighborhood optimality guiding term and combines inertial weights to balance global exploration and local development capabilities. This effectively avoids the defects of premature convergence and getting trapped in local optima, improves the convergence speed, solution accuracy and global search capability of recipe optimization, and enhances the reliability and efficiency of recipe optimization.

[0024] In another aspect, a system for optimizing the formulation of a rubber-construction waste mixture is provided, comprising: a processor, an input device, an output device, and a memory, wherein the processor, the input device, the output device, and the memory are interconnected, wherein the memory is used to store a computer program, the computer program including program instructions, and the processor is configured to invoke the program instructions to execute a method for optimizing the formulation of a rubber-construction waste mixture as described in any of the preceding aspects of the present invention.

[0025] The present invention provides a rubber-construction solid waste mixture formulation optimization system, which has a compact structure, stable performance, high integration and simple composition. It can stably execute the rubber-construction solid waste mixture formulation optimization method provided in the preceding aspect of the present invention, further improving the overall applicability and practical application capability of the present invention. Attached Figure Description

[0026] Figure 1 This is a flowchart of a method for optimizing the formulation of a rubber-construction solid waste mixture according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of the intelligent testing and digital twin system for solid waste-based civil engineering materials according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the raw material processing and digital characterization module according to an embodiment of the present invention; Figure 4 This is a schematic diagram of the intelligent sample preparation and assembly module according to an embodiment of the present invention; Figure 5 This is a schematic diagram of the structure of the triaxial-CT-acoustic emission-temperature control multifunctional linkage test module according to an embodiment of the present invention; Figure 6 This is a schematic diagram of the multi-scale data synchronous acquisition and intelligent analysis module according to an embodiment of the present invention; Figure 7 This is a schematic diagram of the machine learning analysis module and the central control and database platform structure according to an embodiment of the present invention; Figure 8 This is a schematic diagram of a rubber-construction solid waste mixture formulation optimization system according to an embodiment of the present invention. Detailed Implementation

[0027] Specific embodiments of the present invention will now be described in detail. It should be noted that the embodiments described herein are for illustrative purposes only and are not intended to limit the invention. In the following description, numerous specific details are set forth in order to provide a thorough understanding of the invention. However, it will be apparent to those skilled in the art that these specific details are not necessary to practice the invention. In other instances, well-known circuits, software, or methods have not been specifically described to avoid obscuring the invention.

[0028] Throughout this specification, references to "an embodiment," "an embodiment," "an example," or "an example" mean that a particular feature, structure, or characteristic described in connection with that embodiment or example is included in at least one embodiment of the invention. Therefore, the phrases "in an embodiment," "in an embodiment," "an example," or "an example" appearing in various places throughout the specification do not necessarily refer to the same embodiment or example. Furthermore, specific features, structures, or characteristics can be combined in one or more embodiments or examples in any suitable combination and / or sub-combination. Moreover, those skilled in the art will understand that the illustrations provided herein are for illustrative purposes and are not necessarily drawn to scale.

[0029] First, we construct an intelligent testing and digital twin system for solid waste-based civil engineering materials based on a cyber-physics system architecture, such as... Figure 2 As shown, the intelligent testing and digital twin system for solid waste-based civil engineering materials includes a raw material processing and digital characterization module 1, an intelligent sample preparation and assembly module 2, a triaxial-CT-acoustic emission-temperature control multi-functional linkage testing module 3, a multi-scale data synchronous acquisition and intelligent analysis module 4, a machine learning analysis module 5, and a central control and database platform 6. It aims to break down the technical barriers of isolated equipment and fragmented data in traditional civil engineering material testing, constructing a closed-loop network for deep interaction from the physical testing space to the digital twin space. In terms of logical topology, the system is constructed from bottom to top as a perception and execution layer, a data fusion and transmission layer, and an intelligent decision-making and twin layer. The sensing and execution layer is equipped with high-precision preparation machinery, a multiphysics loading framework, and a cross-scale sensor array, used to transform physical materials into standard samples and acquire real-time physical responses under complex working conditions; the data fusion and transmission layer relies on a hardware-level clock synchronization network and industrial Ethernet, configured to align, clean, and reduce the features of massive multi-source heterogeneous raw data on a unified time axis; the intelligent decision-making and twin layer relies on machine learning algorithms and multi-scale databases to construct a digital twin of the material.

[0030] like Figure 3As shown, the raw material processing and digital characterization module includes an automated crushing and gradation unit 11, a multi-dimensional morphology and structure scanning unit 12, and a digital filing and mathematical modeling unit 13. These units are connected in sequence to convert solid waste raw materials into digital particle files.

[0031] like Figure 4 As shown, the intelligent sample preparation and assembly module includes a precision metering and mixing casting unit 21, a servo vibration and compaction molding unit 22, a standard environment constant temperature and humidity curing unit 23, and a multi-dimensional sensor automated patching and assembly unit 24, which are used to complete sample molding, curing and sensor assembly.

[0032] like Figure 5 As shown, the triaxial-CT-acoustic emission-temperature control multi-functional linkage test module includes a triaxial loading subsystem 31, a microfocus X-ray CT scanning subsystem 32, a distributed acoustic emission monitoring subsystem 33, a high and low temperature environment control subsystem 34, a sample mounting base 311, a permeable stone 312, a loading piston rod 313, an X-ray source 321, a flat panel detector 322, a broadband acoustic emission sensor 331, and a thermocouple temperature probe 332. All subsystems are coaxially integrated to achieve synchronous observation of stress field, temperature field, structural field, and damage field.

[0033] like Figure 6 As shown, the multi-scale data synchronous acquisition and intelligent analysis module includes an FPGA synchronous acquisition unit 41, a multi-source data fusion processor 42, an acoustic emission feature analysis unit 43, and a CT image post-processing unit 44, which are used to complete data synchronization, correction, feature extraction and fusion.

[0034] like Figure 7 The diagram shows the structure of the machine learning analysis module and the central control and database platform. It mainly illustrates the core components of intelligent prediction and reverse optimization, including a multi-task deep neural network unit 51, a damage pattern intelligent classification unit 52, a reverse design optimization unit 53, and a central control and database platform 6, which are used to realize performance prediction, damage identification, and formulation optimization.

[0035] From the perspective of the underlying mathematical logic of system operation, the overall architecture of this invention is configured to execute a complex nonlinear mapping and feedback process. Let the material formulation and process parameter space be... The external environment and loading path space are The final performance output space of the system is The overall architecture of the system is based on physical experiment operators. With digital proxy model operators The deep fusion is used to achieve target mapping, and its system-level mapping formula is expressed as: , in, This represents the overall performance state space output by the system. For spatiotemporal fusion operators targeting multi-source heterogeneous data from physics experiments, For deep learning-based performance prediction operators, These are the network weight parameters. This is the confidence level dynamic adjustment coefficient. For system observation and modeling residual terms, For loading time. Based on this architecture, the system is configured not only to output test results positively, but also to minimize... The deviation from the target performance is used to solve in reverse and output the optimal parameter space. In terms of specific physical structure and hardware / software implementation, the system highly integrates six core modules that are sequentially connected. Specifically, the system includes a raw material processing and digital characterization module 1, an intelligent sample preparation and assembly module 2, a triaxial-CT-acoustic emission-temperature control multi-functional linkage test module 3, a multi-scale data synchronous acquisition and intelligent analysis module 4, a machine learning analysis module 5, and a central control and database platform 6. The raw material processing and digital characterization module 1 serves as the digital starting point for physical input, configured to convert solid waste into a digital particle file with a precise mathematical description and transmit it to the next level. The intelligent sample preparation and assembly module 2 is connected to the raw material processing and digital characterization module 1, serving as a standardized building block for physical entities. It eliminates the discreteness of manual sample preparation through closed-loop feedback control to ensure parameter space. Precise execution; the triaxial-CT-acoustic emission-temperature control multi-functional linkage test module 3 serves as the core of multi-physics field coupling sensing and the operator for executing physical tests. The key hardware is configured to achieve simultaneous observation of stress field, temperature field, internal structure field, and micro-damage field; the multi-scale data synchronous acquisition and intelligent analysis module 4 is communicatively connected to the above-mentioned experimental module 3, undertakes the data fusion layer function, and is responsible for executing the spatiotemporal fusion operator. This enables microsecond-level data alignment and high-order physical feature extraction; the machine learning analysis module 5, as the system's digital decision-making center, is responsible for executing the digital proxy model operators. The system is configured to achieve macroscopic performance prediction and reverse intelligent optimization of material formulations through graph neural networks and multi-task learning. The central control and database platform 6 establishes bidirectional communication connections with each of the above modules, serving as the system's neural center and data storage carrier. It is configured to coordinate the control timing of all modules, manage massive amounts of time-series and spatial data, and generate a three-dimensional visualized digital twin interactive interface. All the above modules work collaboratively to constitute the highly automated and intelligent complete testing and analysis system of this invention.

[0036] Please see Figure 1 In order to solve the problems in the prior art, in an alternative implementation, such as Figure 1 The method for optimizing the formulation of a rubber-construction solid waste mixture, as shown, includes the following steps: Step S1: Obtain digital particle profiles of rubber granules and construction solid waste.

[0037] The process of obtaining digital particle profiles for rubber granules and construction solid waste includes the following sub-steps: Step S11: The rubber particles and the construction solid waste are crushed and screened to obtain solid waste particles.

[0038] In this embodiment, the raw material processing and digital characterization module 1 serves as the digital starting point and data source for the physical input of this system. It is configured to transform multi-source, heterogeneous, and morphologically diverse solid waste into digital particle files with precise mathematical descriptions, thereby providing high-fidelity basic data support for subsequent physical sampling and digital twin modeling. Specifically, the raw material processing and digital characterization module 1 further includes an automated crushing and grading unit 11, a multi-dimensional morphology and structure scanning unit 12, and a digital filing and mathematical modeling unit 13, all connected sequentially via a material transfer channel and a data communication bus. The automated crushing and grading unit 11 is equipped with a jaw crusher, a multi-layer servo vibrating screen, and a precision weighing sensor. In a specific embodiment, for large pieces of construction solid waste (such as waste concrete, waste bricks, etc.), the jaw crusher first pre-crushes them to a maximum particle size of less than 50mm. The reason for this parameter setting is that the preparation of standard cylindrical specimens in civil engineering (such as those with a diameter of 150mm and a height of 300mm) requires that the maximum aggregate particle size not exceed one-third of the minimum specimen size, in order to eliminate testing errors caused by size effects. Subsequently, the crushed material enters the multi-layer servo vibrating screen. Based on a preset continuous or discontinuous gradation target curve, the system precisely grades the material into a fine aggregate range of 0-5mm, and multiple coarse aggregate ranges such as 5-20mm and 20-40mm. This unit uses a closed-loop control system to monitor the material quality of each particle size range in real time and mix them proportionally. This effectively eliminates the initial random errors introduced by factors such as environmental humidity and human operation during traditional manual material preparation, ensuring a high degree of consistency in the physical input parameter space.

[0039] Step S12: Perform three-dimensional morphological scanning and internal structure scanning on the solid waste particles to obtain particle point cloud data and internal microstructure data.

[0040] In this embodiment, after standardized crushing and gradation of the physical morphology, the material is automatically conveyed to the multi-dimensional morphology and structure scanning unit 12 located downstream of the automated crushing and gradation unit 11. The multi-dimensional morphology and structure scanning unit 12 is equipped with a high-precision laser profilometer and an X-ray micro-CT scanner, configured to perform comprehensive, cross-scale non-destructive scanning of representative solid waste particles after gradation. The high-precision laser profilometer is used to quickly acquire high-resolution three-dimensional point cloud data of the particle surface, i.e., particle point cloud data, to quantify the high surface roughness and angularity unique to solid waste aggregates; while the X-ray micro-CT scanner is used to penetrate the particle surface to acquire the initial microcrack topology and density gradient data of the attached mortar, i.e., internal mesoscopic structure data. The above dual-source scanning data are registered and fused in a unified spatial coordinate system to form an original dataset containing all-dimensional physical characteristics of the particles.

[0041] Step S13: Surface reconstruction is performed using spherical harmonic functions based on the particle point cloud data to obtain morphological reconstruction coefficients.

[0042] In this embodiment, to achieve accurate mapping and dimensionality reduction of physical particles to a digital twin space, the digital archiving and mathematical modeling unit 13 establishes a high-speed communication connection with the multidimensional morphology and structure scanning unit 12, configured to receive the original dataset and perform mathematical reconstruction on it. Preferably, the digital archiving and mathematical modeling unit 13 has a built-in surface reconstruction algorithm based on spherical harmonics, configured to transform the highly irregular surface morphology of solid waste particles into a linear combination of orthogonal basis functions. Its spatial representation mathematical reconstruction process is configured to execute the following formula: , in, Represents the polar angle of the particle surface in spherical coordinates. With azimuth radial distance in the direction, The maximum expansion order of the spherical harmonic function (its value is adaptively determined by the system based on the particle surface roughness, and is typically used for construction solid waste aggregates). to To balance computational efficiency and reconstruction accuracy. For the first Rank Standard spherical harmonic function of order 1 The morphological reconstruction coefficients correspond to the spherical harmonic function.

[0043] Step S14: Bind the morphological reconstruction coefficients to the internal microstructure data to generate the digital particle profile.

[0044] In this embodiment, the morphological reconstruction coefficients and internal microstructure data are bound one-to-one to generate a unique corresponding digital particle profile. This digital particle profile is then encrypted and transmitted to the central control and database platform for persistent storage. The establishment of this profile provides precise physical material preparation and spatial stacking basis for the intelligent sample preparation and assembly module. On the other hand, it also provides high-fidelity geometric boundary conditions and input parameters for the machine learning analysis module when constructing micromechanical proxy models (such as discrete element DEM simulation), thereby completely opening up the homogeneous transformation channel between "physical entity" and "digital model" at the forefront of the system.

[0045] Step S2: Obtain training samples and construct a performance prediction model through multi-functional linkage experiments.

[0046] Multi-functional linkage test refers to a multi-physics field coupling test that simultaneously performs triaxial mechanical loading, microfocus X-ray CT scanning, distributed acoustic emission monitoring, and high and low temperature environment control on a standard sample.

[0047] The process of obtaining training samples and constructing a performance prediction model through multi-functional linkage experiments includes the following sub-steps: Step S21: Obtain multiple historical digitized particle files, and prepare standard samples using the historical digitized particle files based on pre-set mixing parameters.

[0048] In this embodiment, the historical digital particle archive refers to the set of digital particle archives generated by the preliminary raw material processing and digital characterization module and stored in the central control and database platform. The precise metering and mixing casting unit in the intelligent sample preparation and assembly module calls up one or more historical digital particle archives according to the proportioning parameters (e.g., water-cement ratio of 0.45, solid waste aggregate replacement rate of 50%) issued by the central control platform, automatically calculates the target dosage of each component, and utilizes a high-precision fluid dosing pump, planetary mixer, and multi-degree-of-freedom... A casting robotic arm sequentially adds solid waste aggregates of various particle sizes, cement-based binders, water, and admixtures to a mixer using a step-by-step mixing process: first, 60 seconds of dry mixing to ensure uniform dispersion of the solid waste aggregate and binder powders; then, water and water-reducing agent are added for 120 seconds of wet mixing, forming a uniform interface transition layer on the aggregate surface. The resulting slurry is precisely injected by the casting robotic arm into a standard civil engineering cylindrical mold (Φ150mm×300mm) or a cubic mold (150mm×150mm×150mm). The cast samples are then conveyed to a servo vibration and compaction unit equipped with an electromagnetic servo vibration table with adjustable frequency and amplitude. The system performs compaction according to preset preparation process parameters (including a vibration frequency of 50Hz, an amplitude of 0.5mm, and a duration of 30 to 45 seconds) to achieve an optimal balance between eliminating internal air bubbles and preventing segregation of the solid waste aggregate. After molding, the samples are transferred to a standard environmental constant temperature and humidity curing unit. This unit is a closed environmental control box, in which the internal temperature is strictly kept constant at 20±2℃ and the relative humidity is maintained above 95%. After 24 hours of initial curing in this environment, the samples are demolded by an automatic demolding mechanism and continue to be cured to the preset target age of 3 days, 7 days or 28 days. These curing conditions are also part of the preparation process parameters to ensure the full hydration reaction inside the sample and the stable development of its strength. Once the sample reaches the target age, it is automatically transferred to the automated multi-dimensional sensor patching and assembly unit for final preparation before testing. This unit is equipped with a machine vision positioning system and a precision patching robot. It automatically grinds, applies adhesive, and attaches strain gauges or fiber Bragg grating sensors at specific geometric locations on the sample surface (such as the symmetrical sides of the center of a cylindrical sample). Simultaneously, it couples acoustic emission probes to the ends or sides of the sample to capture microcrack initiation and propagation signals in real time during subsequent loading. After assembly, a laser marking machine generates a unique RFID electronic tag or QR code on the sample surface. This tag is deeply data-bound to the historical digital particle archive, the mixing parameters used in this step, and the sensor spatial coordinates, and simultaneously uploaded to a central database, achieving traceability throughout the physical sample's entire lifecycle. This completes the preparation of the standard sample.

[0049] Step S22: Triaxial loading is applied to the standard sample, and microfocus X-ray CT scanning, distributed acoustic emission monitoring, and high and low temperature environment control are performed simultaneously to obtain multi-physics field test data.

[0050] In this embodiment, a standard sample is installed in a triaxial CT acoustic emission temperature control multi-functional linkage test module. Through unified scheduling by a central control and database platform, triaxial mechanical loading is applied simultaneously, microfocus X-ray CT in-situ scanning is performed, distributed acoustic emission signals are acquired, and high and low temperature environments are controlled. Thus, under the same sample, the same loading path, and the same temperature conditions, mechanical data, CT image data, acoustic emission data, and temperature data are acquired synchronously to form a multi-physics field test dataset. The specific implementation is as follows: The triaxial-CT-acoustic emission-temperature control multi-functional linkage test module 3, as the core innovative component of the intelligent preparation and performance prediction system of this invention, adopts a multi-physics field integrated coaxial architecture. It deeply integrates and precisely couples the traditional independent and difficult-to-coordinate triaxial mechanical loading test, micro-focus X-ray CT microstructure observation, distributed acoustic emission micro-damage dynamic monitoring, and high and low temperature environment simulation functions. It completely solves the technical bottlenecks of traditional test platforms being scattered, sample states being unreproducible, internal damage evolution being unable to be captured in real time in situ, and multi-field data being unable to be acquired synchronously from the perspectives of hardware structure, spatial layout, temporal coordination, and data interaction. It realizes full-dimensional, high-precision, synchronous coupled observation of stress field, temperature field, microstructure field, and micro-damage evolution field under the same sample, the same loading path, and the same environmental conditions. It provides core hardware support and full-link in-situ data for multi-scale data fusion, intelligent damage diagnosis, accurate prediction of material properties, and reverse design of mix proportions.

[0051] This module uses a high-rigidity integrated frame as the load-bearing foundation, and integrates the triaxial loading subsystem 31, microfocus X-ray CT scanning subsystem 32, distributed acoustic emission monitoring subsystem 33, and high and low temperature environment control subsystem 34 in a coaxial design. This ensures that the mechanical loading center axis, CT scanning center axis, acoustic emission array monitoring center, and temperature control flow field action center are strictly aligned. Under the unified scheduling of the central control and database platform 6, each subsystem achieves fully automatic linkage operation according to the preset loading system, temperature conditions, and data acquisition sequence, ensuring that the test process is stable and controllable, with no blind spots in observation and no data distortion.

[0052] The triaxial loading subsystem 31 employs a servo motor-driven dual-column high-rigidity loading frame, integrating axial loading, confining pressure closed-loop control, and dual-mode adjustment of displacement and stress. It provides stable, uniform, and reproducible three-dimensional mechanical boundary conditions for rubber-construction solid waste mixture samples, ensuring continuous controllability of the sample's stress state under CT scanning, acoustic emission monitoring, and high / low temperature environments, thus avoiding image artifacts, signal interference, and data deviations caused by loading fluctuations. This subsystem possesses a wide loading range and high-precision control accuracy, meeting the mechanical property testing needs of mixtures in different engineering scenarios and providing reliable assurance for macroscopic mechanical response acquisition.

[0053] To ensure the specimen remains under stable stress during the CT scan phase, the loading system employs a load-holding-scan-trigger linkage control strategy. Upon entering the CT scan cycle, the system can control axial load fluctuations to a minimum within milliseconds, ensuring no significant displacement or deformation during imaging. During loading, the specimen's three-dimensional stress state satisfies the mechanical equilibrium control equations. By solving the correspondence between the stress tensor and boundary loads in real time, a precise characterization of the mechanical state is achieved. , In the formula: These are the components of the stress tensor; These are volume force components; For the sample calculation domain, This is the calculation region for the sample.

[0054] The real-time coordinated matching relationship between axial load and confining pressure can be expressed as: , In the formula: This represents the real-time maximum principal stress. For real-time confining pressure; This is the real-time axial load; The initial cross-sectional area of ​​the sample; This is the confining pressure conversion factor; This refers to the real-time oil pressure of the hydraulic system.

[0055] Microfocus X-ray CT Scanning Subsystem: The microfocus X-ray CT scanning subsystem 32 adopts a symmetrical layout, with the X-ray source 321 and the flat panel detector 322 placed on opposite sides of the pressure chamber. Radial distance adjustment and coaxial alignment of the optical path are achieved using a precision linear slide. Without disassembling the sample, interrupting loading, or changing the ambient temperature, in-situ three-dimensional imaging and dynamic tracking of the particle arrangement, pore structure, particle breakage, and interface contact state within the mixture sample are performed, fully recording the evolution of the microstructure during the loading process. This subsystem possesses advantages such as high resolution, rapid imaging, and low artifacts, and can clearly distinguish between construction solid waste particles, rubber particles, and the three-phase porous medium, providing a high-quality image data source for quantitative analysis of the microstructure.

[0056] To quantify the impact of mechanical disturbances on image quality, this invention introduces a CT motion artifact coefficient. As a core evaluation index for module integration stability, the imaging accuracy is ensured by constraining the artifact coefficient, and its expression is: , In the formula: For the first Sub-CT reconstructed image matrix; For the first Sub-CT reconstructed image matrix; This represents the total number of scans. Let Frobenius be the matrix norm.

[0057] To achieve precise matching between the mechanical space and the CT imaging space and eliminate coordinate deviations during multi-field data fusion, the system obtains the spatial transformation matrix from the mechanical coordinate system to the CT coordinate system through standard phantom calibration. To achieve sub-pixel level spatial registration, the coordinate transformation relationship is as follows: , In the formula: Coordinates in the mechanical coordinate system; Coordinates in the CT coordinate system; This is the coordinate translation compensation vector.

[0058] Distributed Acoustic Emission Monitoring Subsystem: The distributed acoustic emission monitoring subsystem 33 is used to capture micro-damage elastic wave signals generated by rubber-construction solid waste mixtures during stress deformation in real time. This includes typical damage events such as crushing and fracture of rigid construction solid waste particles, tensile tearing of rubber particles, and interparticle interface friction slippage. Through multi-channel array deployment and continuous full-waveform acquisition, it fully preserves the time-domain, frequency-domain, and amplitude characteristics of the damage signals, enabling real-time detection, feature extraction, and spatial localization of damage events. This provides original high-quality signals for subsequent intelligent classification of damage modes and quantitative analysis of damage evolution. This subsystem uses 8-channel broadband acoustic emission sensors, evenly and spirally arranged along the outer wall of the pressure chamber. The sensor operating frequency covers 20kHz-1MHz, and the system sampling rate is up to 10MHz. It supports continuous full-waveform acquisition, fully preserving high-frequency transient damage information and avoiding the loss of critical signals. Simultaneously, it employs a dual coupling structure of waveguide rod and high-temperature coupling agent to reduce the impact of high and low temperature environments on signal transmission efficiency, ensuring stable and reliable signal acquisition over a wide temperature range. Acoustic emission signals propagate as elastic waves within the mixture. Based on the time difference of arrival of signals from multi-channel sensors, a three-dimensional spatial damage source localization equation can be constructed, enabling accurate inversion of the damage location. , In the formula: The spatial coordinates of the damage source; For the first Coordinates of the acoustic emission sensor; The propagation speed of longitudinal waves; For the signal to arrive at the Channel timing; This is the initial moment when the damage occurs.

[0059] To eliminate interference from sensor gain differences, baseline drift, and environmental noise across channels, the system performs standardization preprocessing on the raw signals to ensure that the signals from each channel are at a uniform dimension, facilitating subsequent multi-channel collaborative analysis. The signal standardization formula is as follows: , In the formula: For the first Original acoustic emission waveform of the channel; For the first Channel signal mean; For the first Channel signal standard deviation; This is the waveform after standardization.

[0060] By conducting collaborative analysis of multi-channel acoustic emission signals, multi-dimensional features such as cumulative energy, event rate, amplitude distribution, spectral entropy, and spatial distribution density of damage events can be extracted to construct high-dimensional damage feature vectors, providing an input basis for deep learning damage classification models.

[0061] High and Low Temperature Environment Control Subsystem: The high and low temperature environment control subsystem 34 adopts a combined temperature control mode of semiconductor refrigeration and resistance heating. It fully encloses the triaxial pressure chamber through a jacketed sealed flow channel, achieving rapid adjustment, uniform distribution, and long-term stability of the sample's ambient temperature. It can simulate complex working conditions in actual engineering, such as low temperatures in cold regions, high temperatures in summer, and alternating day and night temperatures, accurately revealing the temperature regulation mechanism of the mechanical properties, damage evolution, and microstructural changes of rubber-construction solid waste mixtures. This subsystem features a wide temperature range, high precision, and fast response. The temperature control range is -30℃ to 80℃, with a temperature control accuracy better than ±0.5℃ and a temperature stability of ±0.3℃. The heating and cooling rates are continuously adjustable within the range of 0.5-5℃ / min, enabling rapid attainment of the target temperature and ensuring uniform temperature throughout the sample, avoiding uneven stress, abnormal damage evolution, and data distortion caused by localized temperature differences.

[0062] To ensure temperature field uniformity, the system deploys high-precision thermocouples at three key cross-sections (top, middle, and bottom) of the sample for real-time monitoring. Temperature deviations are compensated through closed-loop feedback adjustment, ensuring the overall temperature uniformity of the sample meets the following constraints: , In the formula: For the first Real-time temperature at the measuring point; The average temperature over the entire sample area.

[0063] The dynamic adjustment process of the temperature control system exhibits inertial lag characteristics, and its dynamic response can be described using a first-order inertial element. The temperature control model is as follows: , In the formula: This refers to the time constant of the temperature control system. For system gain; This is the output control quantity for temperature control; The initial ambient temperature; This represents the real-time temperature of the sample.

[0064] Multi-physics field synchronous linkage control mechanism: To achieve efficient coordination of triaxial loading, CT scanning, acoustic emission monitoring, and high / low temperature control, this module employs the IEEE 1588v2 precision clock synchronization protocol and FPGA hardware synchronization technology to construct a globally unified time reference. This enables microsecond-level synchronous acquisition of data from four sources: mechanics, CT, acoustic emission, and temperature, ensuring strict alignment of multi-source data on the time axis and providing a timing foundation for subsequent multi-scale data fusion. The maximum time deviation between each acquisition channel meets the following constraints: , In the formula: For the first Channel and the Inter-channel time deviation.

[0065] The linkage control follows strict timing logic. The CT scan is triggered only after the loading system enters the stable holding phase, ensuring that there are no motion artifacts in the imaging. The timing constraints are as follows: , In the formula: This refers to the execution time of the CT scan. To ensure the system is stable during loading; This represents the total duration of a single scan. Through the integrated multi-physics field, high-precision collaborative control, and full-dimensional in-situ observation described above, the triaxial-CT-acoustic emission-temperature control multi-functional linkage test module 3 overcomes the inherent defects of traditional testing technologies, such as single dimension, spatiotemporal asynchrony, and data fragmentation. It constructs a synchronous observation system integrating stress, temperature, structure, and damage, providing a core hardware platform and high-quality data support across the entire chain for the study of multi-scale performance evolution laws, intelligent performance prediction, and reverse design of rubber-construction solid waste mixtures.

[0066] Multi-scale data synchronous acquisition and intelligent analysis module: The multi-scale data synchronous acquisition and intelligent analysis module 4 is bidirectionally connected to the triaxial-CT-acoustic emission-temperature control multi-functional linkage test module 3. As the core of the system's data fusion processing and intelligent feature extraction unit, it is used to uniformly receive, hardware synchronize, spatiotemporally align, suppress noise, purify features, and standardize the fusion output of multi-source heterogeneous data from the triaxial loading subsystem 31, microfocus X-ray CT scanning subsystem 32, distributed acoustic emission monitoring subsystem 33, and high and low temperature environment control subsystem 34. This completely solves the technical defects of traditional tests, such as misaligned timing, chaotic format, missing features, and difficulty in correlation of mechanical, structural, damage, and temperature data. This module relies on the FPGA synchronous acquisition unit 41, the multi-source data fusion processor 42, the acoustic emission feature analysis unit, and the CT image post-processing unit to achieve integrated synchronous acquisition and intelligent analysis of macro-fine-micro multi-scale data. It provides a high-fidelity, strongly correlated, and interpretable input feature set for the machine learning analysis module 5, providing data support for material performance prediction, damage pattern recognition, and multi-scale evolution law analysis.

[0067] Multi-physics data hardware synchronous acquisition and spatiotemporal unification: The multi-scale data synchronous acquisition and intelligent analysis module 4, with the FPGA synchronous acquisition unit 41 as its core, adopts the IEEE 1588v2 precision clock synchronization protocol to provide a unified global timestamp for mechanical data, acoustic emission data, CT image data, and temperature data. This enables microsecond-level synchronous acquisition of multi-source data under the same time reference, ensuring accurate correspondence between stress field, temperature field, microstructure field, and microdamage field data. For the dynamic characteristics of different physical signals, the system adopts a channel-based independent sampling strategy: the sampling frequency of macroscopic mechanical signals is set to 1kHz to accurately record the complete evolution process of stress-strain and confining pressure-volume deformation; acoustic emission signals are sampled at a high frequency of 10MHz, supporting continuous acquisition of the entire waveform and fully preserving high-frequency details of transient damage such as particle breakage, rubber tearing, and interface slippage; CT projection images are acquired discontinuously at preset loading trigger points to capture microstructure features under critical states; and temperature signals are recorded in real time at a sampling frequency of 10Hz to ensure continuous and traceable temperature field changes.

[0068] The multi-channel data synchronization accuracy meets the global constraint condition, and the maximum time deviation between any two signals is no greater than 10μs. Its mathematical constraint expression is: , In the formula: , These are the timestamps of any two acquisition channels; .

[0069] To ensure the fidelity of the original signal, the FPGA synchronous acquisition unit 41 performs programmable amplification, anti-aliasing filtering, and baseline drift correction on the input signal. The amplitude-frequency characteristics of the filtering system satisfy the following: , In the formula: This is the filtered transfer function; This is the system cutoff angular frequency; ω is the angular frequency of the input signal.

[0070] Step S23: Based on the multi-physics field test data, obtain the multi-scale feature dataset and macroscopic performance indicators.

[0071] The specific steps for obtaining the multi-scale feature dataset and macroscopic performance indicators based on the multi-physics field experimental data include the following: Step S231: Calculate the peak deviatoric stress, secant modulus, and critical state friction angle based on the multiphysics test data to form macroscopic performance indicators.

[0072] In this embodiment, axial load, axial displacement, confining pressure, and volumetric strain are extracted from the multiphysics test data to obtain mechanical data. Axial principal stress is calculated based on the axial load and the initial cross-sectional area of ​​the specimen, and axial strain is calculated based on the axial displacement and the initial height of the specimen. A deviatoric stress-axial strain curve is plotted. Peak deviatoric stress is calculated based on the difference between the peak axial principal stress and the corresponding confining pressure. The secant slope corresponding to 50% of the peak deviatoric stress is extracted from the deviatoric stress-axial strain curve to determine the secant modulus. Parallel triaxial loading tests are conducted using multiple levels of different confining pressures. The Mohr-Coulomb critical strength envelope is fitted using the failure principal stress corresponding to each level of confining pressure to obtain the critical state friction angle. Macroscopic performance indicators are formed based on the peak deviatoric stress, the secant modulus, and the critical state friction angle.

[0073] Step S232: Obtain a multi-scale feature dataset based on the multi-physics field test data.

[0074] The multi-scale feature dataset obtained from the multi-physics experimental data includes: Step S2321: Extract mechanical data, CT image data, acoustic emission data, and temperature data from the multiphysics test data.

[0075] In this embodiment, data is extracted from multiphysics test data according to the characteristics of different physical signals: macroscopic mechanical data (including real-time axial strain, real-time radial strain, axial displacement, confining pressure, volumetric strain, etc.) is acquired at a sampling frequency of 1 kHz to record the entire stress-strain process; CT image data is extracted at preset loading trigger points to capture microstructural features; acoustic emission data is extracted at a high sampling rate of 10 MHz to extract the full waveform signal to retain high-frequency information of transient damage; and temperature data is extracted at a sampling frequency of 10 Hz to record temperature field changes. These data will be used for subsequent fusion correction and feature extraction, respectively.

[0076] Step S2322: Calculate the true volumetric strain based on the CT image data using a digital volume image correlation method.

[0077] In this embodiment, to address the issues of non-uniform deformation of rubber-construction solid waste mixture and distortion in external sensor volumetric strain measurement, the multi-source data fusion processor 42 combines the three-dimensional structural image output by the micro-focus X-ray CT scanning subsystem 32 with the digital volume image correlation (DVC) method to calculate the true displacement field and volumetric strain inside the sample. Based on this, the external mechanical sensor data is adaptively corrected to construct a high-precision multi-source fusion volumetric strain model.

[0078] First, the true volumetric strain of the sample is calculated based on the CT 3D reconstruction results: , In the formula: This is the initial volume of the sample; The full-field displacement vector obtained from DVC calculation; The calculation region for the sample; Spatial coordinates; This refers to the loading time.

[0079] Step S2323: Calculate the apparent volumetric strain based on the mechanical data.

[0080] In this embodiment, the mechanical data includes real-time axial strain and real-time radial strain, and apparent volumetric strain. : , In the formula: For real-time axial strain; For real-time radial strain.

[0081] Step S2324: Correct the apparent volumetric strain according to the actual volumetric strain to obtain the corrected volumetric strain.

[0082] The corrected volumetric strain satisfies the following formula: , in, Loading time Corrected volumetric strain at that time Rubber content and ambient temperature The relevant correction factor, Loading time Apparent volumetric strain at time This is the weighting adjustment coefficient. Loading time The actual volumetric strain at that time, Loading time The axial strain value at that time.

[0083] It should be noted that the exponent term The strain-time coupled attenuation factor means that not only does a larger axial strain lead to a faster attenuation of the correction deviation, but also a longer loading duration results in a faster attenuation. For rubber-construction waste mixtures exhibiting significant viscoelastic behavior, introducing the time dimension of attenuation is reasonable and necessary. (If 't' is deleted and replaced with...) If the attenuation factor no longer includes a time variable, it means that under the same axial strain level, whether the loading is 1 second or 1000 seconds, the correction weight is exactly the same, and the ability to characterize the time accumulation effect will be lost.

[0084] Where the correction coefficient Quadratic polynomial fitting was used: , In the formula: , , , , , These are the calibration coefficients for the experiment. After fusion correction, the volumetric strain measurement error can be controlled within 0.05%, significantly improving the accuracy of deformation monitoring.

[0085] The weighting adjustment coefficient and The method was determined through a combination of experimental calibration and optimization. Specifically, at least five groups covering different rubber content were first selected. and ambient temperature For the standard specimen under combined working conditions, sampling times were synchronously collected at a uniform sampling frequency throughout the entire triaxial loading process. The apparent volumetric strain below True volume strain and axial strain ,in, , This represents the total number of synchronous sampling points corresponding to a single set of experiments. Based on the synchronous sampling results, a calibration sample set is established: , Based on this, using true volume strain As a reference value, the corrected volumetric strain With true volume strain Minimizing the deviation between the parameters is the optimization objective. An objective function is constructed for a single set of working conditions: , in, Calculated by the following formula: , Substituting the corrected volumetric strain expression into the objective function, we can obtain the following about... and The nonlinear optimization problem is preferably solved using the nonlinear least squares method, with the objective function... Minimum value is used as the optimization criterion. To ensure the stability of the solution process, the following is set: The initial value is , The initial value is and to and Apply a nonnegativity constraint, that is: , During the iterative solution process, the parameters are considered convergent when the objective function values ​​corresponding to two consecutive iterations satisfy the following equation: , in, For the first The objective function value of the next iteration. For the first The objective function value of the next iteration. If the convergence condition is not met after reaching the preset maximum number of iterations, the current iteration result is output as the approximate optimal solution under the corresponding working condition.

[0086] Furthermore, in order to improve the and To assess applicability under varying rubber content and ambient temperature conditions, the objective functions for all calibration conditions are jointly optimized to construct a comprehensive objective function: , in, To calibrate the total number of operating condition groups, For the first The objective function under the combined working conditions, For the first The weighting coefficients corresponding to the group of working conditions, and satisfying the following: , Preferably, the weighting coefficient The objective function is determined based on the proportion of sampling points for each working condition to the total number of sampling points, or pre-set based on the importance of each working condition in actual engineering applications. By performing a minimization solution, uniform parameters applicable to all calibration conditions can be obtained. and .

[0087] After obtaining the above and Then, the corrected volumetric strain is substituted into the corrected volumetric strain formula to verify the correction effect on the verification samples that were not calibrated. Specifically, the corrected volumetric strain of each verification sample at each sampling time is calculated. and the actual volumetric strain at the corresponding moment Error comparison is performed, and a verification error index is calculated. Preferably, the mean relative error or root mean square error is used as the verification index; when the verification error is not greater than a preset threshold, the calibrated result is considered valid. and It meets the usage requirements.

[0088] in, Used to characterize the contribution weight of the difference between true and apparent volumetric strain in the correction result. This is used to characterize the decay rate of the difference term as it changes over the loading process. As the loading process progresses, through... and The coordinated adjustment of the two technologies enables the dynamic fusion of real volumetric strain information obtained from CT measurements with apparent volumetric strain information obtained from external measurements, thereby improving the authenticity, continuity and stability of volumetric strain measurement results.

[0089] Step S2325: Perform time-frequency analysis, nonlinear dynamics analysis, and high-order feature extraction on the acoustic emission data to obtain the damage feature vector.

[0090] The process of performing time-frequency analysis, nonlinear dynamics analysis, and high-order feature extraction on the acoustic emission data to obtain the damage feature vector specifically includes the following sub-steps: Step S23251: Extract amplitude, energy, rise time to amplitude ratio, and center frequency from the acoustic emission data.

[0091] In this embodiment, the amplitude of each acoustic emission event is obtained through a peak detection algorithm; the energy is calculated by square integral of the waveform signal; the time from when the signal first exceeds the threshold to when it reaches the peak value is measured, and the ratio of this time to the amplitude is calculated to obtain the rise time to amplitude ratio; the waveform spectrum is analyzed using Fast Fourier Transform, and the frequency corresponding to the maximum amplitude is extracted as the center frequency. The above four features constitute the basic waveform part of the damage feature vector.

[0092] Step S23252: Perform wavelet packet decomposition on the acoustic emission data and calculate the wavelet packet energy entropy.

[0093] In this embodiment, the signal is first decomposed into 5-level wavelet packets, and the wavelet packet energy entropy is calculated to describe the signal energy distribution complexity: , In the formula: The wavelet packet energy entropy, For the first Energy of each wavelet packet node; This is the normalized energy ratio; For the first Energy of each wavelet packet node.

[0094] Step S23253: Perform multifractal detrending fluctuation analysis on the acoustic emission data and calculate the multifractal spectrum width.

[0095] In this embodiment, multifractal detrended fluctuation analysis (MF-DFA) is used to calculate the multifractal spectrum width of the signal. This reflects the nonlinear characteristics of the damage signal: , In the formula: , These are the maximum and minimum values ​​of the singularity index, respectively.

[0096] Step S23254: Perform phase space reconstruction on the acoustic emission data and calculate the maximum Lyapunov exponent.

[0097] In this embodiment, the maximum Lyapunov exponent is calculated. Characterizing the chaotic properties of damage evolution: , In the formula: For the first phase space Group of adjacent trajectory vectors at discrete time steps The difference vector, For the first phase space Group of adjacent trajectory vectors at discrete time steps The difference vector; For the number of iterations, For discrete time steps.

[0098] Step S23255: Combine the amplitude, the energy, the rise time to amplitude ratio, the center frequency, the wavelet packet energy entropy, the multifractal spectrum width, and the maximum Lyapunov exponent to obtain the damage feature vector.

[0099] In this embodiment, the amplitude ,energy Rise time to amplitude ratio Center frequency Wavelet packet energy entropy Multifractal spectral width Maximum Lyapunov index Combined, a 7-dimensional damage feature vector is formed. : , This vector can effectively distinguish three typical damage modes: rigid particle breakage, rubber particle tearing, and particle-particle interface slippage.

[0100] Step S2326: The mechanical data, the CT image data, the temperature data, the corrected volumetric strain, and the damage feature vector are fused to obtain a multi-scale feature dataset.

[0101] In this embodiment, the CT image post-processing unit 44 and the multi-source data fusion processor 42 perform unified format conversion, outlier removal, missing value interpolation and normalization on the processed mechanical data, CT image data, temperature data, corrected volumetric strain and damage feature vectors to form a standardized dataset.

[0102] Step S24: Associate and store the ratio parameters, the multi-scale feature dataset, the digital particle profile, and the macroscopic performance indicators to construct a multi-scale performance database.

[0103] In this embodiment, the proportioning parameters, digitized particle profiles, multi-scale feature datasets, and macroscopic performance indicators are all uploaded to the central control and database platform. First, the features in each dimension of the multi-scale feature dataset are subjected to min-max normalization: ,in This is the original data. , These are the minimum and maximum values ​​of the feature, respectively. The data is normalized. Then, using the test number, timestamp, and sample ID as primary keys, all the above data are linked by a multi-dimensional index: proportioning parameters, morphological and physical property information from the digital particle archive, normalized multi-scale feature datasets, and macroscopic performance indicators are stored respectively in a time-series database (storing time-series data such as mechanics and temperature), a relational database (storing structured parameters and indicators), and an object storage system (storing unstructured data such as acoustic emission waveforms and CT images). This constructs a highly correlated multi-scale performance database, providing complete, reusable, and traceable data support for subsequent machine learning analysis modules, realizing fully automated processing from raw signals to engineering features and then to model input.

[0104] Step S25: Train a multi-task deep neural network based on the multi-scale performance database to obtain the performance prediction model.

[0105] In this embodiment, the machine learning analysis module 5 is bidirectionally connected to the multi-scale data synchronous acquisition and intelligent analysis module 4 and the central control and database platform 6, serving as the intelligent decision-making core and performance prediction center of the system of this invention. This module receives standardized and fused multi-scale mechanical, damage, structural, and temperature characteristic data. Through deep neural networks and intelligent optimization algorithms, it establishes a high-precision nonlinear mapping relationship between material formulation, preparation process, external working conditions, macroscopic mechanical properties, and damage evolution laws, realizing multi-task performance forward prediction and material formulation reverse design for engineering objectives. This module incorporates a multi-task deep neural network unit 51, a damage mode intelligent classification unit 52, and a reverse design optimization unit 53, relying on a multi-scale performance database to achieve online model training, real-time inference, and continuous iterative updates.

[0106] Multi-task deep neural network model construction: The multi-task deep neural network unit 51 adopts a multi-task learning (MTL) architecture based on the fusion of bidirectional long short-term memory network (BiLSTM) and attention mechanism. During the training phase, the complete normalized feature vector output by the multi-scale data synchronous acquisition and intelligent analysis module 4 is used as input. This input includes material ratio, gradation characteristics, initial state, loading conditions, and damage evolution characteristics (such as accumulated acoustic emission energy) acquired in real time during the experiment. Particle breakage incident rate (etc.), totaling 15 key parameters, forming the training input vector: , The network output layer is configured with three parallel task branches, each predicting the peak deviatoric stress. Secant modulus Critical friction angle The shared feature extraction layer consists of three BiLSTM layers and one attention mechanism layer, used to automatically mine deep correlation features hidden in time series data. The expression for calculating the attention weights is: , In the formula: For BiLSTM Always hide your status; , , These are the parameters for network learning. For the first Time-feature weights The sequence length is given.

[0107] The model as a whole employs a weighted multi-task loss function for joint optimization: , In the formula: For the first The mean squared error loss for each task; This refers to the task weighting coefficient; for Regularization coefficient; These are all the learnable parameters of the network.

[0108] Through the training process described above, the model not only establishes a mapping between formulation, process, environment, and macroscopic performance, but also implicitly learns the correlation between static inputs and dynamic damage evolution characteristics. Therefore, in practical deployments for formulation reverse engineering or rapid performance evaluation, the model can rely solely on static design variables (rubber content). Gradation unevenness coefficient Curvature coefficient Dry density Initial porosity Ambient temperature Confining pressure Loading speed (etc.) as input, no need to provide , Real-time monitoring data is collected. At this point, the model utilizes the temporal feature mapping relationships learned during training to still output high-precision macroscopic performance predictions. Experiments show that the prediction accuracy of the model with simplified input decreases by no more than 3% compared to the complete input, fully meeting the needs of engineering applications.

[0109] Intelligent Damage Pattern Classification Based on Deep Learning: The intelligent damage pattern classification unit 52 uses the 7-dimensional acoustic emission high-order feature vector output by the multi-scale data synchronous acquisition and intelligent analysis module 4 as input to construct a deep convolutional neural network (CNN) classification model, realizing real-time automatic identification and quantitative statistics of three typical damage patterns: rigid building solid waste particle breakage, rubber particle tearing, and particle-particle interface slippage. The classification model uses a Softmax classifier to output the probability distribution of each type of damage. , In the formula: It is a 7-dimensional damage feature vector; This is the high-dimensional feature mapping function for the convolutional layer; , For output layer weights and biases; For input features belonging to the first The probability of damage.

[0110] Through training and validation with multiple batches of experimental samples, the classification model achieves an accuracy rate of no less than 94.2% on the independent test set, which can meet the real-time online diagnosis needs of damage patterns during the experiment.

[0111] Through the above process, training samples (with macroscopic performance indicators as output labels) are constructed using a multi-scale performance database. A multi-task deep neural network based on BiLSTM and attention mechanisms is then trained, and the network parameters are optimized using a weighted multi-task loss function. Finally, a performance prediction model capable of simultaneously predicting the three mechanical indicators is obtained. This model can be deployed in a machine learning analysis module for subsequent reverse engineering of formulations and rapid performance evaluation.

[0112] Step S3: Obtain the particle swarm optimization algorithm and improve the particle swarm optimization algorithm to obtain the improved particle swarm optimization algorithm.

[0113] The improved particle swarm optimization algorithm satisfies the following formula: , in, For the first During the nth iteration The speed of each particle For inertial weights, For the first During the nth iteration The speed of each particle All are learning factors. It is a random number. For the first The individual best position in the history of each particle. To be the globally optimal position For the first The optimal neighborhood position of each particle. For the first During the nth iteration The position of each particle. For the first During the nth iteration The position of each particle.

[0114] The optimal neighborhood position It refers to the first The optimal position of a particle within its preset neighborhood, corresponding to the historical optimal position of the individual particle; specifically, based on the pre-defined neighborhood topology, the optimal position relative to the first particle is determined. The particle is identified by its at least one neighboring particle, and the individual historical best position of that at least one neighboring particle is compared with that of the first particle. The individual historical best positions of each particle are considered as candidate positions. The optimal objective function value is used as the criterion to select the nth particle from these candidate positions. The optimal neighborhood position of each particle The inertial weight The speed update mechanism derived from the particle swarm optimization algorithm is used to characterize the first... The particle pair The degree of inheritance of its own motion state in each iteration is used to balance global search and local development; preferably, the inertia weight To pre-set parameters or make dynamic adjustments based on the iteration process, the particles can maintain a strong search capability in the early stage of the iteration and improve convergence accuracy in the later stage of the iteration, thereby improving the stability and reliability of the optimal formula solution.

[0115] Step S4: Based on the improved particle swarm optimization algorithm, calculate the optimal formula using the digital particle profile and the performance prediction model.

[0116] The calculation of the optimal formulation based on the improved particle swarm optimization algorithm, using the digital particle profile and the performance prediction model, specifically includes the following sub-steps: Step S41: Set an initial population, and calculate the predicted performance index of individuals in the initial population using the performance prediction model based on the initial population and the digitized particle profile.

[0117] In this embodiment, the reverse design optimization unit first sets up an initial population containing multiple candidate formulation individuals, each individual being determined by the design variables to be optimized (including rubber content). Gradation unevenness coefficient Curvature coefficient Dry density The process involves several steps: first, the formulation parameters of each individual in the population are integrated with the inherent properties of the raw materials in the digital particle archive (such as particle morphology reconstruction coefficient, internal porosity, etc.), and then input into the step performance prediction model. The model outputs the predicted performance indicators corresponding to that individual in parallel, including peak deviatoric stress, secant modulus, and critical state friction angle.

[0118] Step S42: Obtain the target engineering performance index, and use the target engineering performance index and the predicted performance index to calculate the objective function value of the individual.

[0119] In this embodiment, the reverse design optimization unit first obtains the engineering target performance indicators, including the target peak deviatoric stress, the target secant modulus, and the target critical state friction angle; then, for each individual in the population, it substitutes its predicted performance indicators and the corresponding target performance indicators into the weighted relative error sum of squares function to calculate the target function value for that individual. ,in Let be the design variables to be optimized, and represent the rubber content, respectively. Gradation unevenness coefficient Curvature coefficient Dry density ; For the first Predicted values ​​of the following performance indicators (in order: peak deviatoric stress, secant modulus, and critical friction angle). For the corresponding engineering target performance indicators, These are the weighting coefficients. Simultaneously, the design variables must satisfy the boundary constraints. ,in and These are the lower and upper bound vectors for each design variable; in addition, inequality constraints must also be satisfied. ,in Indicates the first Constraint functions include process feasibility, material cost, and construction requirements. The smaller the objective function value, the closer the corresponding formulation is to the engineering goal. This value will be used as the basis for evaluating the quality of individuals in subsequent particle swarm iterations.

[0120] Step S43: Based on the objective function value, the improved particle swarm optimization algorithm is used to iterate and obtain the optimal formula.

[0121] In this embodiment, the reverse design optimization unit uses an improved particle swarm optimization algorithm that introduces a neighborhood optimum term for iterative optimization: based on the objective function value of each individual, the unit updates the individual historical optimal position, global optimal position, and neighborhood optimal position of each particle, and adjusts the particle's flight direction and step size according to the speed and position update rules of the improved particle swarm optimization algorithm; the unit repeatedly executes performance prediction, objective function value calculation, optimal position update, and particle state update until the preset iteration termination condition is met (such as reaching the maximum number of iterations or the objective function value converges). At this point, the proportioning parameters corresponding to the global optimal position are output (including rubber content, gradation non-uniformity coefficient, curvature coefficient, and dry density) to form the optimal formulation scheme.

[0122] Online Model Updates and Continuous Learning: To ensure the long-term accuracy and generalization ability of the model, the machine learning analysis module has the capability of online incremental learning and adaptive updates. After each new set of experiments is completed, the central control and database platform automatically imports the newly added multi-scale data into the training set. The system adopts the Elastic Weight Consolidation (EWC) strategy to fine-tune and update the neural network. While retaining historical knowledge and avoiding catastrophic forgetting, it quickly absorbs the inherent laws of new materials, new ratios, and new working conditions. This allows the model's prediction accuracy, classification accuracy, and reverse design reliability to continuously improve with the accumulation of database data, always maintaining its adaptive adaptability to complex engineering environments.

[0123] After the model is updated, the system automatically stores the new version weights, evaluation indicators, and prediction errors into the model library, forming a complete closed loop of experimental collection, data fusion, intelligent prediction, reverse design, physical verification, and model iteration. This provides continuously iterative intelligent algorithm support for the intelligent R&D, engineering application, and standardized promotion of rubber-construction solid waste mixtures throughout their entire life cycle.

[0124] Central Control and Database Platform: After the machine learning analysis module completes performance prediction, damage classification, and reverse design optimization, all system data, hardware instructions, algorithm tasks, and visual interactions are uniformly integrated into the central control and database platform for closed-loop management. This platform serves as the nerve center, data foundation, and interaction entry point of the entire system, closely receiving the operational status, multi-source data, algorithm results, and control requirements of all upstream modules. It adopts an EtherCAT industrial Ethernet real-time control architecture, a hybrid architecture database storage system, and WebGL 3D digital twin rendering technology. This enables six core functions: system-wide hardware collaborative scheduling, multi-module temporal logic control, multi-scale integrated data storage, digital twin visual interaction, full lifecycle management of machine learning models, and closed-loop iterative testing. It completely solves the common industry problems of traditional solid waste mixture testing platforms, such as dispersed equipment, asynchronous instructions, fragmented data, weak human-machine interaction, and poor traceability. This ensures long-term stable operation of the system under high-precision, high-reliability, and unattended conditions, providing unified, secure, and scalable top-level support for the intelligent R&D and engineering application of rubber-construction solid waste mixtures. The implementation process of this platform strictly follows the technical roadmap of real-time collaborative control, integrated data management, digital twin visualization, model and task scheduling, and closed-loop iteration. The specific implementation steps are as follows: Real-time collaborative control and global hardware scheduling: The central control and database platform adopts the EtherCAT industrial Ethernet real-time bus to build a distributed collaborative control system with a control cycle of 1ms. It can uniformly issue commands, collect status, interlock protection and schedule timing for the raw material processing and digital characterization module 1, intelligent sample preparation and assembly module 2, triaxial-CT-acoustic emission-temperature control multi-functional linkage test module 3, multi-scale data synchronous acquisition and intelligent analysis module 4, and machine learning analysis module 5, to ensure that the entire system's actions are precise, synchronized and conflict-free.

[0125] The platform has a built-in preset test process engine that can automatically execute a fully unattended operation, from crushing and screening, sample preparation and casting, vibration compaction, standard curing, sensor assembly, triaxial loading, temperature control, CT scanning, acoustic emission acquisition, data fusion, model inference to reverse output. (The last sentence appears to be incomplete and possibly refers to timing errors in the execution of various modules.) satisfy: , In the formula: This refers to the actual execution time of the module; This refers to the moment when the platform issues instructions.

[0126] The platform possesses comprehensive multi-level safety protection logic: when the axial load exceeds the threshold, the confining pressure is abnormal, the temperature exceeds the limit, the CT source is not ready, the acoustic emission signal is distorted, or communication is interrupted, the system immediately triggers graded protection actions. For example, when the sample loading stress exceeds the set value by 120%, the platform immediately triggers an emergency shutdown, cutting off the loading power, shutting off the X-ray source, maintaining stable confining pressure, and activating audible and visual alarms to prevent equipment damage and data loss; if the communication interruption exceeds 500ms, the system automatically saves the current data and enters a safe state to prevent test failure. Without this protection mechanism, serious consequences such as sample breakage, pressure chamber oil leakage, false CT triggering, and data loss are likely to occur.

[0127] Integrated storage and standardized management of multi-source heterogeneous data: The platform constructs a hybrid data architecture of time-series database + relational database + object storage, realizing unified classification, indexing, storage and fast retrieval of mechanical time-series data, acoustic emission waveform data, CT three-dimensional image data, temperature field data, formula parameters, sample preparation process, model weights and prediction results, fundamentally eliminating "data silos".

[0128] 1. Time Series Database (InfluxDB): Stores 1kHz mechanical data, 10MHz acoustic emission waveforms, and 10Hz temperature time series data, supporting high-frequency writing and millisecond-level queries; 2. Relational database (MySQL): Stores structured information such as sample ID, formulation, rubber content, gradation, curing age, loading path, temperature control curve, and test personnel; 3. Object Storage System (MinIO): Stores large-capacity unstructured data such as CT 3D reconstruction images, high-resolution slices, digital twin models, and test reports.

[0129] All data is linked using a globally unique sample ID and timestamp as an index. Before being stored, the data is automatically cleaned, validated, outlier removed, and missing value filled, ensuring high data efficiency. , In the formula, The number of valid data entries. This represents the total number of data entries collected. For data efficiency.

[0130] Example: For a sample numbered SAMPLE-2026-0318-007, its raw material particle information, stirring parameters, vibration compaction parameters, curing temperature and humidity, triaxial loading curve, CT sequence images, acoustic emission waveform, predicted intensity, and reverse formulation can be retrieved with a single click via ID, with a response time of ≤200ms. Without a unified data architecture, similar data is scattered across multiple devices, resulting in query times of several hours and a high risk of data loss or mismatch.

[0131] Digital Twin 3D Visualization and Human-Computer Interaction: Based on WebGL real-time graphics rendering technology, the platform constructs a 3D digital twin interface for the entire process of the test sample and experiment. It dynamically displays macroscopic stress-strain curves, volumetric strain evolution, microscopic CT slices, spatiotemporal distribution of acoustic emission damage, temperature field changes, machine learning prediction curves, damage identification cloud maps, and reverse design schemes through multi-view linkage, realizing integrated visualization of "physical field + data field + model field".

[0132] Visualized interface rendering frame rate satisfy: , Ensure a smooth and lag-free process throughout the loading, damage, and structural evolution stages. Example: During the experiment, operators can simultaneously observe: ① real-time stress-strain curves; ② detailed CT scans of the structure at the current moment; ③ three-dimensional spatial distribution of acoustic emission damage sources; ④ global temperature cloud map of the sample; ⑤ real-time intensity prediction by the model. If the visualization frame rate is less than 30fps, screen stuttering, delayed damage location, and curve jumps will occur, leading to misjudgments.

[0133] The platform supports mouse-interactive slicing, panning observation, parameter annotation, curve fitting, result export, and automatic report generation. It can directly output standard format charts and data files that meet the requirements of engineering acceptance and academic research.

[0134] Machine Learning Model Management and Intelligent Multi-Task Scheduling: The platform provides full lifecycle management for machine learning models, including model loading, inference scheduling, version management, accuracy evaluation, incremental updates, and rollback mechanisms. It uniformly schedules multi-task deep neural network prediction, damage pattern classification, and reverse engineering optimization algorithms, and sets strict priorities. , In the formula: To control task priorities in real time, Prioritize data analysis tasks. Prioritize data storage tasks. Prioritize model update tasks. Real-time control tasks have the highest priority, followed by online analysis, data storage, and model updates, ensuring the system remains stable and uncontrolled under high concurrency.

[0135] Example: When the system simultaneously performs "triaxial loading + CT scan + acoustic emission acquisition + real-time performance prediction + reverse recipe calculation," the platform prioritizes the real-time performance of loading and acquisition, while placing model inference and data storage in the background for asynchronous execution to avoid control delays caused by computing power contention. An unreasonable scheduling mechanism can easily lead to control delays, missed sampling points, prediction stuttering, or even system crashes.

[0136] After each set of experiments is completed, the platform automatically imports new data into the training set, triggering elastic weight consolidation (EWC) incremental updates. This improves model accuracy while preserving historical knowledge, forming a complete intelligent closed loop of experimentation, data, model, prediction, design, verification, and iteration.

[0137] System logs, traceability, and closed-loop iteration support: The platform automatically records the entire process operation logs, equipment status logs, data acquisition logs, algorithm operation logs, and abnormal alarm logs, with a retention period of no less than 3 years, achieving 100% traceability of the entire chain from raw materials to sample preparation, testing, analysis, prediction, and design.

[0138] Example: If the strength of a certain group of samples is abnormally low, the following can be quickly traced through logs: ① Whether the raw material gradation is qualified; ② Whether the stirring time is up to standard; ③ Whether the compaction density is sufficient; ④ Whether the curing temperature and humidity fluctuate; ⑤ Whether the loading rate is accurate; ⑥ Whether the sensor drifts; ⑦ Whether the model prediction is normal, thus enabling rapid problem localization. Without a complete log traceability mechanism, the cause of the anomaly cannot be investigated, and the test data is unreliable and unusable.

[0139] Under the closed-loop iterative mechanism, the system can automatically complete a round of model optimization after completing 10 sets of experiments. The prediction accuracy continues to improve with the increase of sample size. Long-term operation can realize the transformation from "relying on experiments" to "fewer experiments, higher prediction, and stronger design", shortening the material research and development cycle from the traditional 72 hours to less than 24 hours and improving design efficiency by more than 10 times.

[0140] Once the central control and database platform has completed all data archiving, model updates, visualization, and experimental loop, the entire process of the intelligent preparation and performance prediction system for rubber-construction solid waste mixtures based on multi-scale data fusion and machine learning, as described in this invention, has been fully executed. The system then enters standby mode and can directly start the next set of experiments or carry out customized design of engineering materials.

[0141] Summary of technical effects (including a complete calculation example of substituting solid waste aggregate with a particle size of 5-10mm into the formula) In summary, this invention, through digital characterization of raw materials, intelligent sample preparation, multi-physics field linkage testing, multi-scale data fusion, and machine learning intelligent prediction, can achieve a stable configuration of 80%–100% aggregate content (as a percentage of the total aggregate mass) and 5%–50% rubber particle content (as a percentage of the total solid mixture mass) of construction solid waste. While achieving a high proportion of resource utilization of bulk solid waste, it ensures that the mechanical properties, deformation characteristics, and engineering stability of the materials meet the requirements of engineering projects such as roadbed backfilling, high-fill load reduction, and tunnel subbase. Compared with traditional testing methods, this system improves testing accuracy by more than 5 times, R&D efficiency by more than 10 times, shortens the testing cycle from 72 hours to less than 24 hours, achieves a sample strength variation coefficient of less than 3%, controls the volumetric strain measurement error within 0.05%, has a performance prediction accuracy of no less than 92%, a damage identification accuracy of 94.2%, and possesses wide temperature range engineering adaptability from -30℃ to 80℃. When all modules work together and parameters are controlled reasonably, the system can stably output high-confidence test data and optimal mix ratios, realizing a paradigm upgrade from "experience-based trial and error" to "intelligent design and accurate prediction". If there are deviations in raw material gradation, sample preparation parameters, loading sequence or temperature control accuracy, the system will automatically identify, compensate in real time and provide a correction plan, which will not cause test failure or data scrapping.

[0142] like Figure 8 As shown, in another aspect, the present invention also provides a formulation optimization system for rubber-construction solid waste mixtures, comprising: a processor, an input device, an output device, and a memory, wherein the processor, the input device, the output device, and the memory are interconnected, wherein the memory is used to store a computer program, the computer program including program instructions, and the processor is configured to call the program instructions to execute relevant steps in a relevant embodiment of the formulation optimization method for rubber-construction solid waste mixtures of the present invention.

[0143] This invention provides a formulation optimization system for rubber-construction solid waste mixtures. The functional components can be integrated into a single processing unit, exist as individual physical entities, or be integrated into a single unit. The integrated components can be implemented in hardware or software.

[0144] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered within the scope of the claims and specification of the present invention.

Claims

1. A method for optimizing the formulation of a rubber-construction solid waste mixture, characterized in that, The method includes: Obtain digital particle profiles of rubber granules and construction solid waste; Training samples were obtained and a performance prediction model was constructed through multi-functional joint experiments; Obtain the particle swarm optimization algorithm, and improve the particle swarm optimization algorithm to obtain the improved particle swarm optimization algorithm; Based on the improved particle swarm optimization algorithm, the optimal formulation is calculated using the digital particle profile and the performance prediction model.

2. The method for optimizing the formulation of a rubber-construction solid waste mixture according to claim 1, characterized in that, The acquisition of digital particle files for rubber granules and construction solid waste includes: The rubber granules and the construction solid waste are crushed and screened to obtain solid waste granules. Three-dimensional morphological scanning and internal structure scanning were performed on the solid waste particles to obtain particle point cloud data and internal microstructure data. Based on the particle point cloud data, surface reconstruction is performed using spherical harmonic functions to obtain morphological reconstruction coefficients; The morphological reconstruction coefficients are bound to the internal microstructure data to generate the digital particle profile.

3. The method for optimizing the formulation of a rubber-construction solid waste mixture according to claim 1, characterized in that, The process of obtaining training samples and constructing a performance prediction model through multi-functional linkage experiments includes: Multiple historical digitized particle archives are acquired, and standard samples are prepared using the historical digitized particle archives based on pre-set mixing parameters; The standard sample was subjected to triaxial loading, and microfocus X-ray CT scanning, distributed acoustic emission monitoring and high and low temperature environment control were performed simultaneously to obtain multi-physics field test data; Based on the multiphysics field experimental data, a multi-scale feature dataset and macroscopic performance indicators were obtained; The ratio parameters, the multi-scale feature dataset, the digital particle archive, and the macroscopic performance indicators are associated and stored to construct a multi-scale performance database. The performance prediction model is obtained by training a multi-task deep neural network based on the multi-scale performance database.

4. The method for optimizing the formulation of a rubber-construction solid waste mixture according to claim 3, characterized in that, The multi-scale feature dataset and macroscopic performance indicators obtained based on the multiphysics experimental data include: Based on the multiphysics test data, the peak deviatoric stress, secant modulus, and critical state friction angle are calculated to form macroscopic performance indicators. A multi-scale feature dataset was obtained based on the multiphysics field test data.

5. The method for optimizing the formulation of a rubber-construction solid waste mixture according to claim 4, characterized in that, The multi-scale feature dataset obtained from the multiphysics experimental data includes: Mechanical data, CT image data, acoustic emission data, and temperature data are extracted from the multiphysics field test data; The true volumetric strain is calculated based on the CT image data using digital volume image correlation methods. Calculate the apparent volumetric strain based on the mechanical data; The apparent volumetric strain is corrected based on the actual volumetric strain to obtain the corrected volumetric strain; Time-frequency analysis, nonlinear dynamics analysis, and high-order feature extraction are performed on the acoustic emission data to obtain a damage feature vector; The mechanical data, the CT image data, the temperature data, the corrected volumetric strain, and the damage feature vector are fused to obtain a multi-scale feature dataset.

6. The method for optimizing the formulation of a rubber-construction solid waste mixture according to claim 5, characterized in that, The corrected volumetric strain satisfies the following formula: ， in, Loading time Corrected volumetric strain at that time Rubber content and ambient temperature The relevant correction factor, Loading time Apparent volumetric strain at time This is the weighting adjustment coefficient. Loading time The actual volumetric strain at that time, Loading time The axial strain value at that time.

7. The method for optimizing the formulation of a rubber-construction solid waste mixture according to claim 5, characterized in that, The damage feature vector obtained by performing time-frequency analysis, nonlinear dynamics analysis, and high-order feature extraction on the acoustic emission data includes: Extract amplitude, energy, rise time to amplitude ratio, and center frequency from the acoustic emission data; The acoustic emission data is decomposed into wavelet packets, and the wavelet packet energy entropy is calculated. Perform multifractal detrending fluctuation analysis on the acoustic emission data and calculate the multifractal spectrum width; The acoustic emission data are reconstructed in phase space, and the maximum Lyapunov exponent is calculated. The damage feature vector is obtained by combining the amplitude, the energy, the rise time to amplitude ratio, the center frequency, the wavelet packet energy entropy, the multifractal spectrum width, and the maximum Lyapunov exponent.

8. The method for optimizing the formulation of a rubber-construction solid waste mixture according to claim 1, characterized in that, The calculation of the optimal formulation based on the improved particle swarm optimization algorithm, using the digital particle profile and the performance prediction model, includes: An initial population is set up, and the predicted performance index of individuals in the initial population is calculated using the performance prediction model based on the initial population and the digitized particle profile. Obtain the target engineering performance index, and use the target engineering performance index and the predicted performance index to calculate the objective function value of the individual; The optimal formula is obtained by iteratively applying the improved particle swarm optimization algorithm based on the objective function value.

9. The method for optimizing the formulation of a rubber-construction solid waste mixture according to claim 1, characterized in that, The improved particle swarm optimization algorithm satisfies the following formula: ， in, For the first During the nth iteration The speed of each particle For inertial weights, For the first During the nth iteration The speed of each particle All are learning factors. It is a random number. For the first The individual best position in the history of each particle. To be the globally optimal position For the first The optimal neighborhood position of each particle. For the first During the nth iteration The position of each particle. For the first During the nth iteration The position of each particle.

10. A formulation optimization system for rubber-construction solid waste mixtures, characterized in that, include: The system includes a processor, an input device, an output device, and a memory, all interconnected, wherein the memory stores a computer program comprising program instructions, and the processor is configured to invoke the program instructions to execute a method for optimizing the formulation of a rubber-construction solid waste mixture as described in any one of claims 1 to 9.

Images