A method for quality detection and risk assessment of aggregate for bridge concrete mixing

Abstract

The application provides a kind of bridge concrete mixing aggregate quality detection and risk assessment method, comprising the following steps: S1, multi-source fusion intelligent sampling;S2, multi-modal refinement detection;S3, accelerated activity simulation;S4, dynamic cloud platform risk assessment.The application realizes accurate sampling through multi-source information fusion, characterizes the total amount and form of iron elements through multi-modal detection, simulates activity through on-site deployable accelerated reaction device, realizes risk classification and continuous optimization through dynamic cloud platform, simultaneously provides graded detection mode to adapt to different engineering requirements, and finally realizes active prevention and control and intelligent management of bridge concrete aggregate quality.

Description

Technical Field

[0001] This invention belongs to the field of civil engineering material testing and quality and safety control technology, specifically a method for quality testing and risk assessment of aggregates used in bridge concrete mixing. Background Technology

[0002] Bridges are key nodes in transportation networks, playing a vital role in facilitating the passage of numerous vehicles and pedestrians. If problems arise that lead to performance degradation and subsequently cause serious accidents such as partial collapses, they will greatly threaten the driving safety of passing vehicles and the lives and property of pedestrians. Therefore, the assessment of their health status and risks is of particular importance.

[0003] Limestone aggregate, a key aggregate in bridge concrete, directly determines the mechanical properties and durability of the concrete structure. If the aggregate contains excessive iron, or if the iron exists in harmful forms such as pyrite (mainly FeS2) or magnetite (mainly Fe3O4), electrochemical oxidation and chemical reactions will occur in the alkaline (typically pH ≥ 12), oxygen-rich, and humid environment where the concrete is used. Specifically, in certain environments, the iron in the limestone aggregate is first oxidized to ferrous ions, and then converted into products such as ferric hydroxide and ferric sulfate. These products can expand several times in volume, generating enormous internal stress, leading to the spalling and cracking of the concrete protective layer, and even structural collapse. Simultaneously, rust stains will pollute the bridge's appearance, reducing its aesthetic appeal and public trust.

[0004] The existing technology uses GB / T 14685-2022 "Construction Gravel and Crushed Stone" as the core standard, which only provides routine specifications for aggregate soundness, sulfide and total sulfate content, and does not establish a special testing system for iron, which has the following significant technical defects: (1) Single detection dimension: Existing technologies generally use the potassium dichromate chemical titration method, which can only determine the total iron content and cannot distinguish the occurrence form of iron, thus leading to the misjudgment of "total content is qualified but actual harm is present". For example, stable hematite (main component is Fe2O3) has no significant harm even if the content is high, while low content pyrite (main component is FeS2) will cause serious swelling risk. In this case, the determination method that only measures the total iron content but cannot measure the pyrite content may lead to the problem that the total iron content is qualified but the actual use contains risks.

[0005] (2) Disconnection between environmental simulation and actual conditions: Existing technologies are mostly based on laboratory testing under normal temperature and pressure, which do not reproduce the “CO2 / O2 penetration + humidity fluctuation + stress coupling” environment of concrete in actual service. They cannot predict the long-term chemical activity of iron, and the test results deviate greatly from the actual engineering.

[0006] (3) Delayed risk warning: The existing method is "post-event traceability". Only when concrete has obvious defects such as rust spots and cracks can aggregate problems be investigated in reverse. It is impossible to actively control risks during the aggregate arrival stage, resulting in high rework costs and accumulated structural safety hazards.

[0007] (4) Poor sampling representativeness: Iron is unevenly distributed in gravel. Traditional random sampling methods are prone to missing high-iron enrichment areas. The sample cannot reflect the true quality of the batch, further amplifying the detection error.

[0008] Therefore, there is an urgent need to develop a full-chain detection method that takes into account "precise quantification, morphological identification, activity prediction, and intelligent early warning" to overcome the limitations of existing technologies and provide technical support for the quality control of bridge engineering. Summary of the Invention

[0009] In view of this, the purpose of this invention is to overcome the shortcomings of existing limestone crushed stone iron element detection methods, which are "single-dimensional, detached from reality, have delayed early warning, and have sampling bias," and to provide an intelligent detection and risk assessment method based on multimodal fusion and dynamic learning. Specifically, it provides a quality detection and risk assessment method for aggregates used in bridge concrete mixing.

[0010] To achieve the above objectives, the present invention proposes the following technical solution: Firstly, a method for quality testing and risk assessment of aggregates used in bridge concrete mixing is proposed, including the following steps: S1. Obtain full-area image data of the crushed stone batch using image recognition technology, obtain full-area elemental distribution data of the crushed stone batch using X-ray fluorescence analysis technology, perform spatial registration and data fusion processing on the full-area image data and the full-area elemental distribution data to obtain fusion results, locate iron enrichment hotspots in the crushed stone batch based on the fusion results, and obtain comprehensive samples by targeted sampling based on the iron enrichment hotspots. S2. After pretreatment of the composite sample, the total iron content of the composite sample is determined by inductively coupled plasma atomic absorption spectrometry or atomic absorption spectrometry; and X-ray diffraction phase analysis is performed to obtain information on the occurrence speciation of iron. S3. Prepare concrete specimens from the composite sample, place the specimens in an accelerated reaction device for reaction, and measure the expansion rate of the specimens after the reaction; wherein, the accelerated reaction device is used to simulate the service environment of concrete carbonation and oxidation. S4. Upload the detection data from S2 and S3 to the cloud platform, process the detection data using a machine learning model based on incremental learning, and output the comprehensive risk level of the crushed stone batch and corresponding engineering prevention and control suggestions.

[0011] Furthermore, S1 includes: The image recognition technology is used to identify and extract mineral particle regions with preset morphological characteristics in the crushed stone batch; Using the mineral grain region as the target region, the X-ray fluorescence analysis technique is employed to obtain the global elemental distribution data. The preset morphological features are associated with the typical morphology of iron-bearing ore particles, and the preset morphological features include at least one of color, structure, or shape.

[0012] Furthermore, S1 also includes: The convolutional neural network performs pixel-level semantic segmentation on the global image data, and then outputs the precise contour information of the mineral particle region in the crushed stone batch to complete the automatic identification of the mineral particle region. The convolutional neural network is an encoder-decoder structure, trained and optimized using rock thin-section images.

[0013] Furthermore, in step S1, the targeted sampling includes the following steps: S11. In accordance with the provisions of GB / T 14685-2022 standard, a first sample is obtained by randomly sampling from the crushed stone batch; S12. In each of the iron-rich hotspot areas located based on the data fusion results, an additional 3 to 5 suspected iron-bearing ore particles are collected as a second sample. S13. Combine the first sample and the second sample to obtain the composite sample.

[0014] Furthermore, in step S2, the detection limit of the inductively coupled plasma atomic absorption spectrometry is ≤0.001%, and the detection limit of the atomic absorption spectrometry is ≤0.005%. The composite sample was scanned using an X-ray diffraction device at a 2θ angle range of 5° to 80° and a step size of 0.02° to obtain a diffraction pattern. The diffraction pattern was fully fitted using the Rietveld refinement method to semi-quantitatively determine the content of iron-bearing mineral phases in the composite sample, and the proportion of iron in harmful forms was assessed based on the content. The iron-bearing mineral phases include at least pyrite and magnetite.

[0015] Furthermore, in step S3, the concrete specimen preparation method includes the following steps: The composite sample, reference cement, and standard sand were mixed in a mass ratio of 6:4:10, and water was added at a water-cement ratio of 0.50. The mixture was then molded into prism specimens of 40mm x 40mm x 160mm. The specimen was cured under standard curing conditions of 20±1 ℃ and relative humidity ≥95% for 24 h before demolding.

[0016] Furthermore, in step S3, a mixed gas containing CO2 and O2 is continuously introduced into the accelerating reaction device, and the temperature is maintained at 55±2 ℃ and the relative humidity is ≥95%. The mixed gas has a CO2 concentration of 5±1% to simulate the carbonation environment of concrete and an O2 concentration of 20±2% to simulate the oxidation environment. The reaction cycle is at least 14 days.

[0017] Furthermore, the accelerated reaction device includes a portable reaction chamber or a large fixed multi-field coupling reaction chamber; The portable reaction chamber has a volume of <1 m³. 3 It includes at least a reaction chamber module, an environmental control module, and a data acquisition module; The reaction chamber module is used to provide a sealed space for the specimen to react; The environmental control module is used to adjust the reaction environment parameters inside the reaction chamber module; The data acquisition module is used to monitor and acquire reaction data inside the reaction chamber module; The large fixed multi-field coupling reaction chamber has a volume ≥10 m³. 3 It can also apply an additional 0-5 MPa compressive stress to the specimen placed therein to simulate the service stress state of concrete.

[0018] Furthermore, in step S3, the step of determining the expansion rate of the specimen includes: Before and after the reaction, the reference length of the specimen was measured using a dial gauge or a laser displacement sensor, and the expansion rate of the specimen was calculated based on the change in the reference length. The dial indicator has a measurement accuracy of not less than 0.001 mm, and the laser displacement sensor has a measurement accuracy of not less than 0.002 mm.

[0019] Furthermore, S4 includes: The machine learning model is constructed based on the incremental random forest algorithm, and a preset amount of historical engineering case data and basic experimental data are used as the initial training set to train the machine learning model. The cloud platform receives and monitors the detection data. For every preset number of newly added valid detection and verification data, the incremental learning process of the machine learning model is automatically triggered. Through the incremental learning process, the feature weights and discrimination thresholds of the machine learning model are continuously and dynamically optimized until the accuracy of the machine learning model in predicting the comprehensive risk level is consistent with the prediction accuracy.

[0020] The beneficial effects of this invention are: This invention achieves precise sampling through multi-source information fusion, characterizes the total amount and form of iron through multimodal detection, simulates activity through a field-deployable accelerated reaction device, and realizes risk classification and continuous optimization through a dynamic cloud platform. It also provides graded detection modes to adapt to different engineering needs, ultimately achieving proactive prevention and intelligent management of bridge concrete aggregate quality. Specifically, it includes: (1) Significantly improved detection accuracy: This invention uses image recognition and XRF technology to perform multi-source fusion sampling, which solves the problem of insufficient representativeness of traditional sampling and improves the sample qualification rate, i.e. the matching degree with the actual quality of the batch, to more than 95%. Furthermore, multimodal detection is used to simultaneously characterize the total amount and form of iron in aggregates, accurately identify the content of harmful forms of iron, and avoid misjudgment of "total amount qualified but form harmful". The detection dimensions cover the chemical nature and risk source of iron.

[0021] (2) Strong engineering practicality: The present invention uses a portable reaction chamber to realize on-site activity simulation, shortening the detection cycle from 3 months in the traditional laboratory to 14 days, and the deployment cost is significantly reduced compared with the large equipment used in the prior art, which is suitable for the rapid quality control needs of the construction site; and adopts a graded detection mode, which can be flexibly selected according to the importance of the project and the budget, taking into account both detection accuracy and economy.

[0022] (3) Foresight of risk warning: This invention shifts the detection direction in the prior art from "post-event traceability" to "pre-event warning". Targeted detection can be carried out at the aggregate entry stage and its long-term activity risk can be predicted, avoiding concrete cracking and rework due to aggregate problems, and effectively reducing rework costs.

[0023] (4) Sustainable technological evolution: The present invention adopts a cloud platform with dynamic incremental learning to continuously optimize the model as data accumulates, and the prediction accuracy increases from the initial 85% to more than 98%, forming a technological barrier. Subsequent addition of engineering data can further enhance the model's advantages, making it difficult to be replaced. It has a wide range of applications, especially for specific engineering fields, to optimize the model and form a targeted engineering prevention and risk prediction system.

[0024] (5) Clear engineering guidance: This invention directly outputs the risk level and specific prevention and control measures without the need for secondary interpretation by professionals. It can be directly used for engineering decision-making, promote the transformation of test results into "quality control actions", and can be widely applied to engineering prevention and control projects. Corresponding measures can be directly adopted according to specific circumstances, simplifying the engineering guidance process.

[0025] It should be understood that all combinations of the foregoing concepts and the additional concepts described in more detail below can be considered part of the inventive subject matter of this disclosure, provided that such concepts do not contradict each other. Attached Figure Description

[0026] The accompanying drawings are not drawn to scale. In the drawings, each identical or nearly identical component shown in the various figures may be denoted by the same reference numeral. For clarity, not every component is labeled in each figure. Embodiments of various aspects of the invention will now be described by way of example and with reference to the accompanying drawings, wherein: Figure 1 This is a schematic diagram of the overall process and graded testing mode selection of the quality testing and risk assessment method for aggregates used in bridge concrete mixing provided by the present invention. Figure 2 This is a schematic diagram of the multi-source fusion intelligent sampling system in S1 of the present invention; Figure 3 This is a schematic diagram of the modular structure of the portable reaction chamber in S3 of the present invention; Figure 4 This is a schematic diagram of the working mechanism of the dynamic incremental learning cloud platform in S4 of the present invention; Figure 5 This is the performance optimization curve of the dynamic incremental learning model in S4 of this invention. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the described embodiments of the present invention without creative effort are within the scope of protection of the present invention. Unless otherwise defined, the technical or scientific terms used herein should have the ordinary meaning understood by those skilled in the art.

[0028] The terms "first," "second," and similar words used in the specification and claims of this patent application do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Similarly, unless the context clearly indicates otherwise, the singular forms of "an," "a," or "the," etc., do not indicate a quantity limitation, but rather indicate the presence of at least one. Terms such as "comprising" or "including" mean that the element or object preceding "comprising" encompasses the features, integrals, steps, operations, elements, and / or components listed following "comprising" or "including," and do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components, and / or collections thereof.

[0029] Unless otherwise stated, the abbreviations used in this invention have the following meanings: In this invention, CNN refers to Convolutional Neural Network; XRF refers to X-ray Fluorescence Spectrometer; ICP-OES refers to Inductively Coupled Plasma Emission Spectrometer; AAS refers to Atomic Absorption Spectrometer; XRD refers to X-ray Diffractometer; and SEM-EDS refers to Scanning Electron Microscope-X-ray Energy Dispersive Spectrometer.

[0030] This invention discloses a method for quality testing and risk assessment of aggregates used in bridge concrete mixing, as shown in the attached figure. Figure 1 It includes the following steps: S1. Multi-source fusion intelligent sampling, see appendix for details. Figure 2 : (1) Full-area scanning and data acquisition: using image recognition technology to obtain the volume of the pile body ≥10 m³ 3 The whole-area image data of limestone crushed stone batches were obtained, and the whole-area elemental distribution data of the crushed stone batches were obtained using X-ray fluorescence analysis technology.

[0031] Furthermore, the above-mentioned image recognition technology is used to automatically identify and extract mineral particle areas with preset morphological characteristics in the crushed stone batch. Using the mineral particle areas as target areas, X-ray fluorescence analysis technology is used to obtain global elemental distribution data.

[0032] Among them, the preset morphological characteristics are associated with the typical morphology of iron-bearing ore particles, including color, structure or shape.

[0033] The aforementioned pre-defined morphological characteristics include particles that appear yellowish-brown or black under an optical microscope.

[0034] Preferably, the image recognition technology is at least one of hyperspectral imaging technology, microscopic imaging technology, and digital imaging processing technology, and the abnormal morphological regions are obtained after recognition.

[0035] Specifically, the image recognition technology is a digital imaging processing technology that performs recognition by acquiring visible light images; it can also be extended to include hyperspectral imaging technology or microscopic imaging technology. In the following embodiments of the present invention, visible light-based digital imaging processing technology is used.

[0036] Further preferably, a high-resolution industrial camera or handheld imaging device is used to perform a full-range visible light scan of the crushed stone batch, wherein the industrial camera has a resolution of ≥20 megapixels, set according to manufacturer standards such as the Basler ace 2 series specifications. The aforementioned high-resolution industrial camera and handheld imaging device both refer to devices capable of acquiring digital images in the visible light band.

[0037] Preferably, X-ray fluorescence analysis is performed using an X-ray fluorescence spectrometer with a detection accuracy of ≤10 ppm, set according to manufacturer standards such as the Olympus Vanta series specifications; the X-ray fluorescence spectrometer can be handheld or stationary.

[0038] Specifically, a grid scanning method was adopted, with a scanning interval of ≤5 cm. The scanning interval was set to be more than 3 times the maximum particle size in GB / T 14685-2022 to ensure representative coverage. The measurement time for each grid point was no less than 10 seconds to ensure data stability. After analysis, the high iron value area was obtained.

[0039] (2) Data fusion and hotspot location: The whole-domain image data and the whole-domain element distribution data are fused to obtain the fusion result, and the iron element enrichment hotspot area in the crushed stone batch is located based on the fusion result.

[0040] Furthermore, the data fusion module spatially matches the morphologically abnormal regions obtained through image recognition and the high-value iron regions obtained through X-ray fluorescence analysis to locate iron enrichment hotspots. The iron content threshold of the iron enrichment hotspots can be defined according to actual needs.

[0041] The data fusion module is a processing unit for spatial registration and fusion of multi-source data, which can be used with existing general image processing and data fusion software or custom-developed.

[0042] Preferably, the iron enrichment hotspot region is defined as a continuous region with an iron content ≥ 0.5%.

[0043] Specifically, the iron distribution information obtained by X-ray fluorescence analysis is spatially registered and superimposed with the texture, color, or feature region information obtained by image recognition technology.

[0044] In areas of suspected iron-bearing ore particles with abnormal color identified by image recognition technology, X-ray fluorescence spectrometry is used simultaneously to perform grid scanning on the batch material to generate an iron element distribution cloud map. If the iron element content in this area exceeds a set threshold, it is determined to be an iron element enrichment hotspot area.

[0045] Furthermore, the aforementioned image recognition technology performs pixel-level semantic segmentation on the global image data based on a convolutional neural network, thereby outputting precise contour information of the mineral particle area in the crushed stone batch to complete the automatic identification of the mineral particle area.

[0046] Specifically, an image segmentation algorithm based on convolutional neural networks (such as U-Net and DeepLab v3+) is used to automatically identify mineral particle regions in crushed stone batches. The convolutional neural network is an encoder-decoder structure used to perform pixel-level semantic segmentation and output precise contour information of the mineral particle regions. The convolutional neural network is trained on a large number of rock thin section images (e.g., more than 10,000), achieving an average intersection-over-union (mIoU) ratio of over 90% and an accuracy of over 95% on the validation set.

[0047] (3) Based on the distribution of iron enrichment hotspots, a comprehensive sample was obtained by targeted sampling. The comprehensive sample is a representative original crushed stone extracted from the crushed stone batch.

[0048] Furthermore, the aforementioned targeted sampling includes the following steps: S11. In accordance with the provisions of GB / T 14685-2022 standard, a first sample is obtained by randomly sampling from the crushed stone batch; S12. In each iron enrichment hotspot area that has been located, additionally collect 3 to 5 abnormal mineral particles as a second sample. S13. Combine the first sample and the second sample to obtain a composite sample. The total mass of the composite sample is 50±5 kg.

[0049] This quality setting is based on the provisions of GB / T 14685-2022 for the sampling quantity of crushed stone with a maximum nominal particle size of 31.5mm, and takes into account the sample requirements of subsequent multi-item tests, so as to avoid excessive sample quantity while ensuring representativeness.

[0050] Preferably, the above-mentioned grasping method includes robotic arm or manual sampling.

[0051] S2, Multimodal Refined Detection: (1) The composite sample obtained above is pretreated.

[0052] The above pretreatment requires crushing the composite sample to a particle size ≤ 5 mm, grinding it to a fineness ≥ 200 mesh, and reducing it to the amount required for detection.

[0053] (2) The total iron content of the comprehensive sample was determined by inductively coupled plasma spectroscopy or atomic absorption spectroscopy; and X-ray diffraction phase analysis was performed to obtain information on the occurrence of iron.

[0054] Furthermore, the detection limit of the inductively coupled plasma spectrometry method is ≤0.001%, set according to GB / T 36590-2018 standard; the detection limit of the atomic absorption spectrometry method is ≤0.005%, set according to GB / T 7728-2021 standard.

[0055] The composite sample was scanned using an X-ray diffraction device at a 2θ angle range of 5° to 80° with a step size of 0.02° to obtain a diffraction pattern. The diffraction pattern was then fitted to the full spectrum using the Rietveld refinement method to semi-quantitatively determine the content of iron-bearing mineral phases in the composite sample, and the proportion of iron in harmful forms was assessed based on the content.

[0056] The Rietveld refinement method described above is a standard method for quantitative analysis of crystal structure through full-spectrum fitting, which is an existing technology and will not be elaborated further here.

[0057] The iron-bearing mineral phases include at least pyrite and magnetite.

[0058] Specifically, semi-quantitative analysis refers to quantitative results with an accuracy between 1% and 5%, which is accurate and appropriate for mineral phase analysis.

[0059] The aforementioned harmful forms refer to iron minerals that undergo electrochemical oxidation and chemical reactions in the alkaline (usually pH≥12), oxygen-rich, and humid environment where concrete is in service. For example, pyrite can cause serious expansion risks, and magnetite may undergo mineral transformation, affecting the stability of concrete structures.

[0060] Specifically, the total iron content of the composite sample is determined by inductively coupled plasma atomic absorption spectrometry (ICP-AES) or atomic absorption spectrometry as a quantitative benchmark. X-ray diffraction (XRD) phase analysis is used to obtain information on the occurrence and chemical forms of iron. The analytical parameters of the XRD apparatus are set to cover the main diffraction peaks of all relevant phases, including pyrite and magnetite, ensuring sufficient resolution and precision.

[0061] In some alternative implementations, the pretreated composite sample can also be microscopically characterized, i.e., a scanning electron microscope-X-ray energy dispersive spectroscopy (SEM-XEDS) can be used to observe the microscopic distribution of iron in the gravel and analyze the binding state between the minerals and the matrix, providing microscopic evidence for activity assessment.

[0062] Specifically, the resolution of the aforementioned scanning electron microscope-X-ray energy dispersive spectrometer is ≤10 nm, as set according to manufacturer standards such as the Zeiss Gemini 300 specification.

[0063] S3, Accelerated Activity Simulation: (1) Specimen preparation: The composite sample is made into concrete specimens.

[0064] Furthermore, the method for preparing concrete specimens includes the following steps: Mix the composite sample, reference cement, and standard sand in a mass ratio of 6:4:10, add water at a water-cement ratio of 0.50, stir, and then form a prism specimen of 40mm x 40mm x 160mm. The prism specimen was cured for 24 hours under standard curing conditions of 20±1 ℃ and relative humidity ≥95% before demolding.

[0065] Specifically, the aforementioned benchmark cement meets the requirements of GB 8076-2008, and the standard sand meets the requirements of ISO 679:2009.

[0066] Specifically, a water-cement ratio of 0.50 means that the mass ratio of water to cement in concrete is 1:2, that is, the mass ratio of water to the mixture of composite sample, reference cement and standard sand is 1:2; the prism specimen is 40mm x 40mm x 160mm. This size and proportion refer to GB / T 17671-2021 Cement Mortar Strength Test Method, and are adjusted according to the aggregate testing requirements to ensure the homogeneity of the specimen and the sensitivity of the test.

[0067] (2) Environmental simulation reaction: The specimens prepared above were placed in an accelerated reaction device to simulate the service environment of concrete carbonation and oxidation.

[0068] Furthermore, a mixed gas containing CO2 and O2 is continuously introduced into the accelerated reaction device, and the temperature is maintained at 55±2℃ and the relative humidity is ≥95%. The CO2 concentration in the mixed gas is 5±1%, and the O2 concentration is 20±2%. The total flow rate of the mixed gas is set according to the volume of the accelerated reaction device to ensure that the gas in the device is renewed at least twice per hour.

[0069] The reaction cycle is at least 14 days. This cycle is determined based on previous tests and can effectively distinguish the risk levels of different reactive aggregates while ensuring a safety margin. It can also be extended to 21 days or 28 days depending on the urgency of the project to obtain more conservative results.

[0070] Preferably, the accelerated reaction device includes a portable reaction chamber or a large fixed multi-field coupling reaction chamber.

[0071] Portable reaction chamber volume <1 m 3 It includes at least a reaction chamber module, an environmental control module, and a data acquisition module.

[0072] The reaction chamber module provides a sealed space for the specimen to react; the environmental control module adjusts the reaction environment parameters inside the reaction chamber module; and the data acquisition module monitors and acquires the reaction data inside the reaction chamber module.

[0073] Specifically, the environmental control module is responsible for regulating the humidity, temperature, and concentration of the introduced gas within the cavity, while the data acquisition module includes temperature and humidity sensors to detect the temperature and humidity data within the cavity. This modular design allows for assembly on-site within one hour, making it easy to use.

[0074] See attached document Figure 3 The environmental control module includes an intelligent controller, a temperature and humidity control unit, and a gas mixing and delivery unit. The reaction chamber module includes an outer shell and insulation layer, a sample rack, an observation window and an airtight quick-opening door, and an interface panel. The data acquisition module includes an embedded data logger, a sensor module, and a communication unit.

[0075] The intelligent controller uses a touchscreen human-machine interface. The temperature and humidity control unit has a temperature control range of 20~80 ℃ and a relative humidity control range of ≥95%. The gas mixing and delivery unit mixes CO2 and O2 with N2 as a balance gas and delivers it into the reaction chamber. The environmental control module is connected to the reaction chamber module via a quick-connect interface and controls its environmental parameters.

[0076] Inside the reaction chamber module, the insulation layer is located inside the outer shell, and the sample holder is located inside the chamber, accommodating 6-8 sets of samples. The interface panel includes quick-connect interfaces for gas and electrical circuits. The reaction chamber module connects to the data acquisition module via a quick-connect interface and returns environmental signals.

[0077] The data acquisition module and sensor module include, but are not limited to, temperature sensors, humidity sensors, pressure sensors and gas concentration sensors, which are responsible for detecting the temperature, humidity, pressure and corresponding gas concentration data inside the cavity. The communication unit can adopt Wi-Fi, 4G, 5G and Ethernet.

[0078] The data acquisition module outputs environmental data logs and reaction cycle reports from the portable reaction chamber via wired or wireless transmission.

[0079] Large fixed multi-field coupling reaction chamber with a volume ≥10 m³ 3 It can also apply an additional 0-5 MPa compressive stress to the specimen placed inside to simulate the service stress state of concrete.

[0080] The reaction chamber volume setting can eliminate the size effect, accommodate larger and more realistic specimens, and ensure environmental uniformity. The setting range of the additional compressive stress refers to the stress level setting of typical concrete members in GB 50010-2010 "Code for Design of Concrete Structures", which covers the stress level inside typical concrete structures and basically encompasses the common stress environment of concrete under service conditions.

[0081] (3) Determination of activity index: After the reaction is completed, at least the expansion rate of the specimen should be determined.

[0082] Furthermore, the steps for determining the expansion rate of the specimen include: Before and after the reaction, the reference length of the specimen was measured using a dial indicator or a laser displacement sensor, and the expansion rate of the specimen was calculated based on the change in the reference length. The measurement accuracy of the dial indicator was not less than 0.001 mm, and was set according to the JJG345-2013 specification. The measurement accuracy of the laser displacement sensor was not less than 0.002 mm, and was set according to the manufacturer's standard such as the KEYENCELK-H series specifications.

[0083] Wherein, the reference length of the specimen before the reaction is L0, and the reference length of the specimen after the reaction is L1, the unit of reference length is mm, and the formula for calculating the specimen expansion rate E is: .

[0084] Since the expansion of the specimen during the accelerated reaction is on the order of a few hundredths of a millimeter, the aforementioned high-precision measuring device ensures that its changes can be accurately captured and a reliable risk assessment can be conducted.

[0085] In some optional embodiments, an electronic balance can be used to determine the rate of change in specimen mass, and an X-ray diffractometer can be used to analyze the composition of surface precipitates to verify the reaction products; wherein the accuracy of the electronic balance is 0.01g. Before and after the reaction, the mass of the specimen is determined using an electronic balance, and the rate of change in specimen mass is calculated based on the change in mass.

[0086] Wherein, the mass of the specimen before the reaction is M0, the mass of the specimen after the reaction is M1, the unit of mass is g, and the formula for calculating the mass change rate ΔM is: .

[0087] S4. Risk assessment of dynamic cloud platform, please refer to the appendix for details. Figure 4 : (1) Data upload: Upload the detection data of S2 and S3 above to the cloud platform.

[0088] Furthermore, the total iron content and the proportion of harmful iron forms measured in S2, as well as the expansion rate and mass change rate measured in S3, are uploaded to the cloud-based risk assessment platform via an encrypted network.

[0089] (2) Model calculation: The detection data is processed by the machine learning model based on incremental learning set in the cloud platform.

[0090] Furthermore, a machine learning model is constructed based on the incremental random forest algorithm, and a preset number of historical engineering case data and basic experimental data are used as the initial training set to train the machine learning model. The cloud platform receives and monitors the detection data, and the incremental learning process of the machine learning model is automatically triggered every time a preset number of valid detection and verification data are added. Through the incremental learning process, the feature weights and discrimination thresholds of the machine learning model are continuously and dynamically optimized until the accuracy of the machine learning model in predicting the comprehensive risk level is consistent with the prediction accuracy.

[0091] Specifically, the aforementioned machine learning model is a model built based on the incremental random forest algorithm.

[0092] The incremental random forest algorithm is an ensemble learning algorithm that can gradually absorb new data and update the model without retraining all the data. It is an existing technology and will not be elaborated here.

[0093] The model input features include total iron content, pyrite content, magnetite content, expansion rate, and mass change rate. Ten-fold cross-validation is used for model training and validation to ensure its generalization ability.

[0094] The initial training set for the machine learning model includes at least 1,000 sets of historical engineering case data and 500 sets of basic experimental data. Data structure examples include: total iron content, content of harmful forms of iron (such as the proportion of pyrite and magnetite), expansion rate, mass change rate, and final engineering risk level label.

[0095] The cloud platform is configured to automatically trigger the incremental learning process of the machine learning model to optimize its internal feature weights and discrimination thresholds whenever a preset number of new valid detection and verification data are accumulated.

[0096] The preset number of valid detection and verification data can be set to 10, 50 or 100 sets.

[0097] See attached document Figure 5This is a schematic diagram illustrating the increase in prediction accuracy of a machine learning model through incremental learning as the cumulative amount of input data increases. When the cumulative amount of data is 0-1000 sets, the model is defined as being in the initial learning stage, with a prediction accuracy of less than 60%. When the cumulative amount of data is 1000-2000 sets, the model is defined as being in the rapid improvement stage, with the prediction accuracy remaining at around 80%. When the cumulative amount of data is 2000-5000 sets, the model is defined as being in the stable optimization stage, with the prediction accuracy continuously increasing from 84% to around 98%.

[0098] The results demonstrate that through the incremental learning process, the accuracy of the machine learning model in predicting the overall risk level continuously improves and optimizes as the cloud platform's operating time increases and the data scale expands.

[0099] (3) Output of results: The model outputs the comprehensive risk level of the crushed stone batch based on the input data.

[0100] Furthermore, the aforementioned comprehensive risk level is at least partially based on the specimen expansion rate, and the comprehensive risk level is divided into several risk levels, with corresponding engineering control recommendations output.

[0101] The aforementioned engineering risk control recommendations include not only aggregate usage recommendations, such as recommended use, downgraded use, prohibited use, or recommendations for blending with low-risk aggregates, but also other engineering risk control recommendations, such as engineering design applications. The results and outputs applied to engineering decisions form new labeled data pairs, which are automatically fed back and incorporated into the training set for data accumulation. Every 100 newly added valid data pairs automatically trigger the incremental learning process of the machine learning model to optimize its internal feature weights and discrimination thresholds, thereby continuously improving the prediction accuracy of the comprehensive risk level.

[0102] Specifically, the overall risk level is classified according to the specimen expansion rate as follows: Low risk level: inflation rate ≤ 0.1%; Medium risk level: 0.1% < inflation rate ≤ 0.3%; High risk level: Inflation rate > 0.3%.

[0103] In some optional embodiments, the overall risk level is determined based on the specimen expansion rate, total iron content, and content of harmful iron elements. The overall risk level is classified as follows: Low-risk level: Expansion rate ≤ 0.1%, total iron content ≤ 0.5%, and content of harmful iron elements ≤ 5%; Medium risk level: 0.1% < expansion rate ≤ 0.3%, or 0.5% < total iron content ≤ 1.0%, or 5% < content of harmful iron elements ≤ 15%; High-risk level: Expansion rate > 0.3%, or total iron content > 1.0%, or content of harmful iron elements > 15%.

[0104] The content of harmful forms of iron includes the proportion of pyrite and / or magnetite, which can be selected according to the actual situation of the aggregate.

[0105] The risk levels mentioned above are related to the engineering applicability of crushed stone materials. Low risk level means recommended use, which can be used directly in engineering projects; medium risk level means restricted use, which needs to be verified by compounding with admixtures; high risk level means prohibited use.

[0106] The above method includes several selectable detection modes, and the detection modes include at least: First mode: S1, S2, S3 and S4 are executed sequentially; wherein, in S2, at least inductively coupled plasma spectroscopy, X-ray diffraction phase analysis and scanning electron microscopy-X-ray energy dispersive spectroscopy are used to obtain the occurrence morphology information of iron element; the accelerated reaction device in S3 adopts a large fixed multi-field coupled reaction chamber.

[0107] Second mode: S1, S2, S3 and S4 are executed sequentially; wherein, in S2, at least atomic absorption spectrometry and X-ray diffraction phase analysis are used to obtain the occurrence form information of iron element; in S3, the accelerated reaction device is a portable reaction chamber.

[0108] The third mode involves sequentially executing S1 and S3; in S1, only X-ray fluorescence analysis is used for rapid screening of iron content; in S3, a portable reaction chamber is used for the accelerated reaction. The measured expansion rate data is then input into a local risk assessment module or a cloud platform to determine the risk level.

[0109] The first mode is a comprehensive inspection mode, which is suitable for key projects such as extra-large span bridges and cross-sea bridges; the second mode is a core inspection mode, which is suitable for conventional highway or railway bridge projects; and the third mode is an economic screening mode, which is suitable for small bridges or rapid sampling inspection of aggregates upon arrival at the site.

[0110] Different modes can be selected for testing based on actual needs and working conditions.

[0111] The following detailed description, with reference to specific embodiments, further illustrates the intelligent detection and risk assessment method for iron content and activity in limestone crushed stone used in bridge concrete disclosed in this invention. Unless otherwise specified, the materials, equipment, and instruments used in the embodiments are all commercially available; specific product models and other information are as follows: The handheld imaging device, model DS-2TD1217B-3 / PA, was purchased from Hangzhou Hikvision Digital Technology Co., Ltd. The X-ray fluorescence spectrometer was purchased from Olympus Corporation of Japan, model Vanta C series; The inductively coupled plasma atomic emission spectrometer was purchased from Thermo Fisher Scientific, USA, and its model is iCAPPRO. The atomic absorption spectrometer was purchased from Beijing Purkinje General Instrument Co., Ltd., model TAS-990. The X-ray diffractometer, model D8 ADVANCE, was purchased from Bruker AXS GmbH, Germany. The scanning electron microscope-X-ray energy dispersive spectrometer was purchased from Hitachi High Technology Corporation of Japan, model SU5000. The dial indicator was purchased from TESA in Switzerland, model TESATAST 6. The laser displacement sensor was purchased from Keyence Corporation of Japan, model LK-G500.

[0112] Example 1: Core Detection Mode (Applicable to conventional highway or railway bridge engineering) The core detection mode was used to test the limestone gravel of a secondary highway bridge: Material information: Limestone crushed stone for a secondary highway bridge, with a stockpile volume of approximately 50 m³ and a particle size of 5~31.5 mm.

[0113] Equipment used: handheld high-definition imager (Hikvision DS-2TD1217B-3 / PA), handheld XRF analyzer (Olympus Vanta C series), atomic absorption spectrometer (Purkinje General TAS-990), small benchtop XRD instrument (Bruker D8 ADVANCE), portable reaction chamber (volume 0.8 m³), ​​electronic balance (accuracy 0.01 g), dial gauge (TESA TESATAST6), cloud-based risk assessment platform (800 sets of initial data already loaded).

[0114] Implementation steps: S1, Multi-source fusion intelligent sampling

[0115] (1) Use a handheld high-definition imager to scan the entire area of ​​the material pile. Specifically, scan 10 areas along the top and sides of the material pile. Use an image segmentation algorithm based on convolutional neural network (CNN) to automatically identify 12 yellowish-brown areas suspected to contain iron ore particles, among which the preset morphological feature is color abnormality.

[0116] (2) Using a handheld XRF analyzer, scan the stockpile along a 5cm×5cm grid to generate an iron distribution cloud map, and locate three hot spots with an iron content ≥0.5%, each with an area of ​​0.2 m². 2 0.3 m 2 0.15 m2 .

[0117] (3) Randomly sample 40 kg according to GB / T 14685-2022 standard, and then take 5 additional abnormal mineral particles from each hot spot area to form a 50 kg comprehensive sample (total mass 50±5 kg).

[0118] S2, Multimodal Refined Detection (1) Sample pretreatment: The composite sample was crushed to a particle size ≤ 5 mm, ground to 200 mesh, and then divided into two portions of 10 g each.

[0119] (2) AAS detection: The total iron content was determined to be 0.65% by atomic absorption spectrometry (AAS) (detection limit ≤ 0.005%, in accordance with GB / T 7728-2021).

[0120] (3) XRD detection: A small benchtop XRD instrument was used to scan in the 2θ angle range of 5°~80° with a step size of 0.02°. The full spectrum was fitted by the Rietveld refinement method. The semi-quantitative calculation showed that the proportion of pyrite (FeS2) was 15%, and magnetite was not detected.

[0121] S3, Accelerated Activity Simulation (1) Specimen preparation: Take the comprehensive sample, mix the comprehensive sample, reference cement and standard sand in a mass ratio of 6:4:10, add water at a water-cement ratio of 0.50 and stir to form 3 sets of prism specimens of 40mm×40mm×160mm. Place them under standard curing conditions of 20±1℃ and relative humidity ≥95% for 24 hours before demolding.

[0122] (2) Environmental simulation reaction: The specimens prepared above were placed in a portable reaction chamber, and the chamber environment was set to a temperature of 55℃ and a relative humidity of 96%. A mixed gas containing 5% CO2 and 20% O2 (with N2 as the equilibrium gas) was introduced, and the reaction cycle was 14 days.

[0123] (3) Determination of activity index: Before and after the reaction, the mass change rate of the specimen was measured by an electronic balance; the reference length of the specimen was measured by a dial gauge, and the expansion rate of the specimen was calculated based on the change of the reference length.

[0124] The measured mass change rates of the three groups of specimens were +0.30%, +0.33%, and +0.29%, respectively, and the average mass change rate of the three groups of specimens was 0.31%.

[0125] The volume expansion rates of the three groups of specimens were measured to be 0.20%, 0.23%, and 0.21%, respectively, with an average volume expansion rate of 0.21%.

[0126] S4, Dynamic Cloud Platform Risk Assessment (1) Upload data: The above-measured characteristic data, such as total iron content of 0.65%, pyrite content of 15%, volume expansion rate of 0.21%, and mass change rate of 0.31%, are uploaded to the cloud platform through an encrypted network.

[0127] (2) Model calculation: The platform processes data based on the incremental random forest model (current version V2.1, which has been optimized by 50 sets of new data). After inputting data, the model matches historical cases (total iron 0.6%~0.7%, pyrite 12%~18%, expansion rate 0.2%~0.25%) to determine the risk level.

[0128] (3) Results output: The risk level is determined to be "medium risk". The engineering recommendation is "mix with slag powder at a ratio of 1:0.15, measure the crack resistance of the mixed concrete, and it can be used after meeting the standard".

[0129] Result verification: This batch of crushed stone, following the "medium-risk recommendation," was compounded with S95 grade slag powder at a mass ratio of 1:0.15. Concrete test blocks were prepared according to the "Technical Specification for Crack Prevention of Concrete Structures in Highway Engineering" and subjected to a 28-day restricted shrinkage cracking test. The total crack area per unit area was ≤100 mm² / m², and the crack width was ≤0.1 mm, meeting the crack resistance requirements for bridge concrete. In actual use in bridge cap beam construction, a visual inspection after 6 months showed no rust spots or cracks, verifying the effectiveness of this method.

[0130] Example 2: Comprehensive Inspection Mode (Applicable to key projects such as extra-long span bridges and cross-sea bridges) A comprehensive testing approach was adopted to test the limestone gravel of a cross-sea bridge. Material information: Limestone crushed stone specifically for the anchorage area of ​​a cross-sea bridge tower, mined from a specific high-quality mining site, with a stockpile volume of approximately 200 m³. 3 The particle size ranges from 5 to 26.5 mm.

[0131] Equipment used: Handheld high-definition imager (Hikvision DS-2TD1217B-3 / PA), handheld XRF analyzer (Olympus Vanta C series), inductively coupled plasma atomic emission spectrometer (Thermo Fisher iCAP PRO), X-ray diffractometer (Brook D8 ADVANCE), scanning electron microscope-X-ray energy dispersive spectrometer (Hitachi SU5000), large fixed multi-field coupling reaction chamber (volume 12 m³). 3 It can apply 0~5 MPa compressive stress), electronic balance (accuracy 0.01 g), dial indicator (TESATE SATAST 6), and cloud-based risk assessment platform (800 sets of initial data have been loaded).

[0132] Mode Features Description: This embodiment adopts the most comprehensive detection mode, which provides the highest reliability risk assessment for critical structural components by integrating micromorphological analysis (SEM-EDS) and multi-field coupled stress simulation.

[0133] Implementation steps: S1, Multi-source fusion intelligent sampling The sampling process is the same as in Example 1. A 50 kg composite sample is obtained from the batch material through image recognition and XRF fusion positioning.

[0134] S2, Multimodal Refined Detection (1) Sample pretreatment is the same as in Example 1.

[0135] (2) Determination of total iron content: The total iron content was determined to be 0.45% by inductively coupled plasma optical emission spectrometry (ICP-OES) (detection limit ≤ 0.001%, which is more accurate than that of Example 1).

[0136] (3) Morphological and microscopic analysis: XRD analysis revealed characteristic peaks of pyrite (FeS2), and Rietveld's refined calculations showed that it accounted for 8% of the total iron.

[0137] SEM-EDS analysis: Microscopic observation of pyrite-bearing mineral particles revealed that they were sparsely embedded in the limestone matrix in a star-like pattern, with no obvious interconnected fissures, indicating good microstructural stability.

[0138] S3, Accelerated Activity Simulation (1) The specimen preparation is the same as in Example 1.

[0139] (2) Environmental simulation reaction: A large fixed multi-field coupling reaction chamber was used to apply a constant compressive stress of 2 MPa to the specimen (simulating the continuous compression state of the cable tower concrete). Other environmental conditions (temperature 55±2℃, RH≥95%, mixed gas containing 5% CO2 and 20% O2) were the same as in Example 1, and the reaction period was 14 days.

[0140] (3) Activity index determination: The reference length of the specimen was measured with a dial indicator before and after the reaction, and the average expansion rate was calculated to be 0.08%.

[0141] S4, Dynamic Cloud Platform Risk Assessment (1) Data upload: Upload comprehensive data such as total iron content of 0.45%, pyrite content of 8%, SEM-EDS micromorphological description, and stress expansion rate of 0.08% to the cloud platform.

[0142] (2) Model calculation: The platform model integrates all data, especially the good micro-morphology and low stress expansion rate, and determines that its long-term risk is controllable.

[0143] (3) Results output: The overall risk level is determined to be "low risk", and the engineering recommendation is "all indicators are excellent, and it is recommended to use C50 and above grade concrete directly for the tower of the cross-sea bridge".

[0144] Result verification: This batch of crushed stone was used directly in the concrete construction of the cable tower as recommended. It reached the required strength 28 days after pouring, and no abnormal strain was observed inside the structure through internal sensor monitoring, which verified the reliability of the comprehensive testing mode for the safety assessment of key engineering materials.

[0145] Example 3: Economic Screening Mode (Applicable to rapid sampling inspection of small bridges or aggregates upon arrival) An economic screening model was used to test limestone gravel on a rural road. Material information: A rural road renovation project uses limestone crushed stone. The material source is complex, consisting of a mixture supplied by multiple small mining sites. The stockpile volume is approximately 30 m³. 3 The particle size ranges from 5 to 31.5 mm.

[0146] Equipment used: Handheld XRF analyzer (Olympus Vanta C series), portable reaction chamber (0.8m³). 3 ), dial gauge (TESA TESATAST 6), and local risk assessment module.

[0147] The local risk assessment module employs a simplified risk assessment system based on an inflation rate threshold, implemented using an embedded system (such as a Raspberry Pi) or an industrial controller.

[0148] Mode Features Description: This embodiment is an economical screening mode that prioritizes detection efficiency. To save time and costs to the greatest extent, image recognition technology is not used for morphology-assisted localization in the S1 sampling stage; only XRF is used for rapid element screening. Multimodal refined detection steps are not used in the S2 stage, and complex morphological analysis is omitted. Finally, a local simplified model is used to complete the initial risk assessment in the shortest time and at the lowest cost. It is suitable for engineering scenarios that are cost-sensitive or have a relatively high risk tolerance.

[0149] Implementation steps: S1, Rapid Sampling and Screening A handheld XRF analyzer was used to perform a rapid grid scan of the stockpile, with a scan interval of 10 cm, which is sparser than the core detection mode to improve speed. Areas with an iron content ≥0.8% were identified as abnormal areas. Based on this result, samples were taken from the abnormal areas and other parts of the batch to obtain a 50 kg composite sample.

[0150] S3, Accelerated Activity Simulation (1) The specimen preparation is the same as in Example 1.

[0151] (2) Environmental simulation reaction: Same as in Example 1, using a portable reaction chamber, with a reaction cycle of 14 days.

[0152] (3) Activity index determination: The reference length of the specimen was measured with a dial indicator before and after the reaction, and the average expansion rate was calculated to be 0.35%.

[0153] Rapid Risk Assessment: Input the inflation rate data (0.35%) into the local risk assessment module for offline calculation. This module makes its judgment based on a preset single inflation rate threshold (≤0.1% is low risk, 0.1%~0.3% is medium risk, and >0.3% is high risk).

[0154] Output result: The risk level is determined to be "high risk", and the engineering recommendation is "The expansion rate of this batch of aggregate exceeds the standard. It is prohibited to use it directly in the main structure. It is recommended to downgrade it for use in the subgrade or to dispose of it as waste material".

[0155] Results verification and explanation: The rapid screening results were consistent with the subsequent core pattern test results for this batch of material (total iron content 1.2%, pyrite content 18%), both of which were determined to be high-risk.

[0156] The construction company adopted the suggestion to downgrade the batch of aggregates for use in roadbed filling, thus avoiding potential quality hazards in the main structure. This proves that the economic screening model can effectively intercept high-risk batches of aggregates at the lowest cost and fastest speed. It is particularly suitable for the initial screening of aggregates from complex sources with varying quality control levels, serving as the first efficient line of defense for "early warning."

[0157] While the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the invention. Those skilled in the art can make various modifications and refinements without departing from the spirit and scope of the invention. Therefore, the scope of protection of the present invention shall be determined by the claims.

[0158] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can be appropriately combined to form other embodiments that can be understood by those skilled in the art. Technical details not described in detail in this invention can all be implemented using any existing technology in the art. In particular, all technical features not described in detail in this invention can be implemented using any existing technology.

Claims

1. A method for quality inspection and risk assessment of aggregates used in bridge concrete mixing, characterized in that, Includes the following steps: S1. Obtain full-area image data of the crushed stone batch using image recognition technology, obtain full-area elemental distribution data of the crushed stone batch using X-ray fluorescence analysis technology, perform spatial registration and data fusion processing on the full-area image data and the full-area elemental distribution data to obtain fusion results, locate iron enrichment hotspots in the crushed stone batch based on the fusion results, and obtain comprehensive samples by targeted sampling based on the iron enrichment hotspots. S2. After pretreatment of the composite sample, the total iron content of the composite sample is determined by inductively coupled plasma atomic absorption spectrometry or atomic absorption spectrometry. X-ray diffraction phase analysis was performed to obtain information on the occurrence morphology of iron. S3. Prepare concrete specimens from the composite sample, place the specimens in an accelerated reaction device for reaction, and measure the expansion rate of the specimens after the reaction; wherein, the accelerated reaction device is used to simulate the service environment of concrete carbonation and oxidation. S4. Upload the detection data from S2 and S3 to the cloud platform, process the detection data using a machine learning model based on incremental learning, and output the comprehensive risk level of the crushed stone batch and corresponding engineering prevention and control suggestions. The feature dimensions of the model input include total iron content, pyrite content, magnetite content, expansion rate, and mass change rate.

2. The method for quality inspection and risk assessment of aggregates used in bridge concrete mixing according to claim 1, characterized in that, S1 includes: The image recognition technology is used to identify and extract mineral particle regions with preset morphological characteristics in the crushed stone batch; Using the mineral grain region as the target region, the X-ray fluorescence analysis technique is employed to obtain the global elemental distribution data. The preset morphological features are associated with the typical morphology of iron-bearing ore particles, and the preset morphological features include at least one of color, structure, or shape.

3. The method for quality testing and risk assessment of aggregates used in bridge concrete mixing according to claim 2, characterized in that, S1 further includes: The convolutional neural network performs pixel-level semantic segmentation on the global image data, and then outputs the precise contour information of the mineral particle region in the crushed stone batch to complete the automatic identification of the mineral particle region. The convolutional neural network is an encoder-decoder structure, trained and optimized using rock thin-section images.

4. The method for quality inspection and risk assessment of aggregates used in bridge concrete mixing according to claim 1, characterized in that, In step S1, the targeted sampling includes the following steps: S11. In accordance with the provisions of GB / T 14685-2022 standard, a first sample is obtained by randomly sampling from the crushed stone batch; S12. In each of the iron-rich hotspot areas located based on the data fusion results, an additional 3 to 5 suspected iron-bearing ore particles are collected as a second sample. S13. Combine the first sample and the second sample to obtain the composite sample.

5. The method for quality inspection and risk assessment of aggregates used in bridge concrete mixing according to claim 1, characterized in that, In S2, The detection limit of the inductively coupled plasma atomic absorption spectrometry is ≤0.001%, and the detection limit of the atomic absorption spectrometry is ≤0.005%. The composite sample was scanned using an X-ray diffraction device at a 2θ angle range of 5° to 80° and a step size of 0.02° to obtain a diffraction pattern. The diffraction pattern was fully fitted using the Rietveld refinement method to semi-quantitatively determine the content of iron-bearing mineral phases in the composite sample, and the proportion of iron in harmful forms was assessed based on the content. The iron-bearing mineral phase includes at least pyrite and magnetite.

6. The method for quality inspection and risk assessment of aggregates used in bridge concrete mixing according to claim 1, characterized in that, In step S3, the concrete specimen preparation method includes the following steps: The composite sample, reference cement, and standard sand were mixed in a mass ratio of 6:4:10, and water was added at a water-cement ratio of 0.

50. The mixture was then molded into prism specimens of 40mm x 40mm x 160mm. The specimen was cured under standard curing conditions of 20±1 ℃ and relative humidity ≥95% for 24 h before demolding.

7. The method for quality inspection and risk assessment of aggregates used in bridge concrete mixing according to claim 1, characterized in that, In S3, The accelerated reaction device continuously introduces a mixed gas containing CO2 and O2, and maintains a temperature of 55±2 ℃ and a relative humidity of ≥95%. The mixed gas has a CO2 concentration of 5±1% to simulate the carbonation environment of concrete and an O2 concentration of 20±2% to simulate the oxidation environment. The reaction cycle is at least 14 days.

8. The method for quality inspection and risk assessment of aggregates used in bridge concrete mixing according to claim 7, characterized in that, The accelerated reaction device includes a portable reaction chamber or a large fixed multi-field coupling reaction chamber; The portable reaction chamber has a volume of <1 m³. 3 It includes at least a reaction chamber module, an environmental control module, and a data acquisition module; The reaction chamber module is used to provide a sealed space for the specimen to react; The environmental control module is used to adjust the reaction environment parameters inside the reaction chamber module; The data acquisition module is used to monitor and acquire reaction data inside the reaction chamber module; The large fixed multi-field coupling reaction chamber has a volume ≥10 m³. 3 It can also apply an additional 0-5 MPa compressive stress to the specimen placed therein to simulate the service stress state of concrete.

9. The method for quality inspection and risk assessment of aggregates used in bridge concrete mixing according to claim 1, characterized in that, In step S3, the step of determining the expansion rate of the specimen includes: Before and after the reaction, the reference length of the specimen was measured using a dial gauge or a laser displacement sensor, and the expansion rate of the specimen was calculated based on the change in the reference length. The dial indicator has a measurement accuracy of not less than 0.001 mm, and the laser displacement sensor has a measurement accuracy of not less than 0.002 mm.

10. The method for quality inspection and risk assessment of aggregates used in bridge concrete mixing according to claim 1, characterized in that, S4 includes: The machine learning model is constructed based on the incremental random forest algorithm, and a preset amount of historical engineering case data and basic experimental data are used as the initial training set to train the machine learning model. The cloud platform receives and monitors the detection data. For every preset number of newly added valid detection and verification data, the incremental learning process of the machine learning model is automatically triggered. Through the incremental learning process, the feature weights and discrimination thresholds of the machine learning model are continuously and dynamically optimized until the accuracy of the machine learning model in predicting the comprehensive risk level is consistent with the prediction accuracy.

Images