Method and system for measuring rheological properties of steel fiber reinforced concrete

Abstract

The application relates to a steel fiber reinforced concrete rheological property measurement method and system, and solves the problem that in traditional rotary shear testing, fiber migration and rearrangement are caused by local forced shearing, measured values deviate from the initial state of the material, and internal flow field information cannot be synchronously obtained. The method comprises the following steps: after steel fiber reinforced concrete is loaded into a loading container, a whole deformation triggering mechanism is used to apply a whole action force to make the material produce whole deformation, non-contact scanning devices are used to synchronously obtain appearance morphology evolution data, a lateral non-destructive perspective scanning device is used to synchronously obtain internal flow field evolution data, the two types of data are spatiotemporally correlated and fused, and a physical information neural network model based on embedded rheological physical equation constraints is used to inversely calculate rheological property parameters. The application has the following effects: under the premise of keeping the initial distribution state of fibers, synchronous monitoring of macroscopic appearance morphology and internal flow field is realized, and the whole rheological property of steel fiber reinforced concrete can be truly reflected.

Description

Technical Field

[0001] This invention relates to the field of building material testing technology, and in particular to a method and system for measuring the rheological properties of steel fiber reinforced concrete. Background Technology

[0002] Steel fiber reinforced concrete is increasingly widely used in the field of construction engineering due to its excellent crack resistance and toughness. The rheological properties of steel fiber reinforced concrete, especially its yield stress and plastic viscosity, are important indicators for guiding pumping construction, pouring and vibration, and quality control of molding. Accurate measurement of these properties is of great significance for mix design optimization and setting of construction process parameters.

[0003] Currently, the measurement of the rheological properties of steel fiber reinforced concrete mainly uses a rotating shear rheometer. This type of instrument is based on the homogeneous fluid assumption, applying localized forced shear to the sample via a rotor, recording the relationship between torque and rotational speed, and combining this with a pre-set rheological model to calculate shear stress, shear rate, and rheological parameters.

[0004] However, steel fiber reinforced concrete is a heterogeneous multiphase material, and its macroscopic rheological behavior is closely related to the distribution, orientation, and contact state of the internal steel fibers. During the rotational shear test, the localized forced shear applied by the rotor may cause the steel fibers to migrate, rotate, aggregate, or reorient, causing the internal structure of the material to deviate from its pre-test state. Consequently, the yield stress and plastic viscosity measured are insufficient to fully characterize the rheological behavior of the material in an undisturbed state, thus affecting the accuracy of the mix design and construction process optimization of steel fiber reinforced concrete. Summary of the Invention

[0005] In order to achieve simultaneous monitoring of macroscopic appearance and internal flow field while maintaining the initial fiber distribution state, and to truly reflect the overall rheological properties of steel fiber reinforced concrete, this application provides a method and system for measuring the rheological properties of steel fiber reinforced concrete.

[0006] In a first aspect, this application provides a method for measuring the rheological properties of steel fiber reinforced concrete, comprising the following steps:

[0007] The pre-mixed steel fiber reinforced concrete is loaded into a pre-designated loading container;

[0008] An overall force is applied to the steel fiber reinforced concrete by a preset overall deformation triggering mechanism, causing the steel fiber reinforced concrete to undergo overall deformation; during the overall deformation process, natural shear is generated inside the steel fiber reinforced concrete, accompanied by the evolution of its appearance and internal flow field.

[0009] During the overall deformation process, the external morphology evolution data is acquired simultaneously using a pre-set non-contact scanning device, and the internal flow field evolution data is acquired simultaneously using a pre-set lateral non-destructive perspective scanning device.

[0010] The rheological parameters of steel fiber reinforced concrete are calculated by spatiotemporally fusing external morphological evolution data with internal flow field evolution data and inverting the data based on a pre-defined physical information neural network model with embedded pre-defined rheological physical equations.

[0011] Secondly, this application provides a system for measuring the rheological properties of steel fiber reinforced concrete, comprising:

[0012] The sample loading module is used to load pre-mixed steel fiber reinforced concrete into a pre-set loading container;

[0013] The deformation driving module is used to apply an overall force to the steel fiber concrete through a preset overall deformation triggering mechanism, so that the steel fiber concrete will undergo overall deformation; during the overall deformation process, natural shear is generated inside the steel fiber concrete, accompanied by the evolution of appearance and internal flow field.

[0014] The synchronous acquisition module is used to simultaneously acquire external morphology evolution data using a preset non-contact scanning device during the overall deformation process, and simultaneously acquire internal flow field evolution data using a preset lateral non-destructive perspective scanning device.

[0015] The fusion inversion module is used to perform spatiotemporal correlation fusion of external morphology evolution data and internal flow field evolution data, and to invert and calculate the rheological characteristic parameters of steel fiber reinforced concrete based on a preset physical information neural network model with embedded preset rheological physical equation constraints. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating the method for measuring the rheological properties of steel fiber reinforced concrete provided in an embodiment of this application.

[0017] Figure 2 This is a schematic cross-sectional view of the steel fiber reinforced concrete under free-flowing conditions, as provided in an embodiment of this application.

[0018] Figure 3 This is a schematic cross-sectional view of the steel fiber reinforced concrete suction and leakage process provided in an embodiment of this application.

[0019] Figure 4 This is a schematic diagram of the structure of steel fiber reinforced concrete under push-and-collapse as provided in the embodiments of this application. Detailed Implementation

[0020] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0021] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Features thus defined as "first" or "second" may explicitly or implicitly include one or more features. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0022] like Figure 1 As shown in the figure, this application provides a method for measuring the rheological properties of steel fiber reinforced concrete, including the following steps:

[0023] S1: Load the pre-mixed steel fiber reinforced concrete into the pre-set loading container.

[0024] In this embodiment, steel fiber reinforced concrete refers to a heterogeneous, high-viscosity, multiphase composite material made by mixing cement matrix, aggregates, steel fibers, and admixtures, with a fiber volume content of 0.01%–2%, and in a preferred embodiment, 0.5%–2%. The pre-set loading container refers to a transparent acrylic container used to hold the steel fiber reinforced concrete and withstand the overall force applied by the overall deformation triggering mechanism. The pre-set loading container is made of X-ray permeable acrylic material with a wall thickness of 5mm–10mm. This wall thickness range has been verified by X-ray transmittance to ensure that the transmittance is not less than 10% under tube voltage conditions of 120kV–200kV, in order to meet the signal-to-noise ratio requirements of industrial CT reconstruction. The pre-mixed steel fiber reinforced concrete is filled into the pre-set loading container in a uniform paving manner, with a filling volume of 80%–95% of the effective volume of the pre-set loading container. After filling, the upper surface of the steel fiber reinforced concrete is flat and there are no gaps between it and the side wall of the pre-set loading container.

[0025] The specific implementation process is as follows: First, the mixed and evenly distributed steel fiber reinforced concrete is transferred to a pre-designed loading container. It is then poured in layer by layer, with each layer controlled to a thickness of 5cm-10cm. After each layer, a manual tamping rod is used to gently compact the concrete evenly around the circumference of the container. The tamping rod is inserted to a depth of 1 / 2 to 2 / 3 of the current layer thickness, and each layer is tamped 3 to 5 times, avoiding prolonged mechanical vibration. This layering and tamping process is repeated until the filling height of the steel fiber reinforced concrete reaches 80% to 95% of the effective volume of the pre-designed loading container. In one embodiment, the filling height is 90% of the effective volume. Finally, a scraper is used to level the top surface, ensuring the upper surface of the steel fiber reinforced concrete is parallel to the top edge of the pre-designed loading container, completing the filling process. It is worth noting that the pre-set loading container must be made of an X-ray-penetrable material such as acrylic. If a metal or ordinary glass container is used, it will cause serious artifacts in the industrial computed tomography image sequence, making it impossible to accurately obtain the internal flow field evolution data. At the same time, it is strictly forbidden to transfer materials by rotary shearing or high-pressure pumping during the loading process, so as not to destroy the original random distribution state of the steel fibers before the test.

[0026] Understandably, step S1 involves layering and light manual compaction to load the pre-mixed steel fiber reinforced concrete into a pre-set loading container in a low-disturbance manner, so that the steel fiber reinforced concrete after loading remains in an initial state that is close to random distribution, and provides test samples with clear boundary conditions for subsequent overall deformation triggering and synchronous monitoring.

[0027] S2: The steel fiber reinforced concrete is subjected to an overall force through a preset overall deformation triggering mechanism, causing the steel fiber reinforced concrete to undergo overall deformation; during the overall deformation process, the steel fiber reinforced concrete undergoes natural shearing, accompanied by changes in appearance and internal flow field.

[0028] In this embodiment, the overall deformation triggering mechanism refers to a mechanical actuator installed at the bottom or side of the loading container to induce macroscopic overall displacement of the steel fiber reinforced concrete, including a bottom perforated baffle, a bottom vertical piston, or a lateral horizontal piston; the overall force refers to the volume force or surface force acting on the steel fiber reinforced concrete as a whole rather than locally, including gravity, piston suction force, or horizontal pushing force. The intensity and rate of application of this overall force are controlled by the overall deformation triggering mechanism and do not produce local forced shear; overall deformation refers to the macroscopic continuous deformation of the steel fiber reinforced concrete under the action of the overall force. During this overall deformation process, natural shear will occur inside the material, accompanied by macroscopic evolution of the appearance and microscopic evolution of the internal flow field. Among them, natural shear refers to the shear action spontaneously formed due to the velocity difference between the internal layers during the overall displacement of the material.

[0029] It should be noted that the specific method and steps for applying an overall force to the steel fiber reinforced concrete through a preset overall deformation triggering mechanism to induce overall deformation of the steel fiber reinforced concrete are detailed in steps S201 to S224.

[0030] like Figure 2 As shown in the embodiment of this application, the preset overall deformation triggering mechanism is a preset hole baffle plate disposed at the bottom of the preset loading container; Figure 2 The left side shows the loading state when the perforation baffle is closed, and the right side shows the process of steel fiber reinforced concrete freely leaking through the bottom opening after the perforation baffle is opened, forming a central sinkhole and an edge curve on the upper surface. An overall force is applied to the steel fiber reinforced concrete through a preset overall deformation triggering mechanism, including the following steps:

[0031] S201: Open the hole baffle to open the bottom opening of the preset loading container.

[0032] In this embodiment, the pre-set hole baffle refers to a flat or door-type sealing structure that can be quickly opened and installed at the bottom opening of the loading container. The opening response time of the hole baffle is no more than 0.1 seconds, and in a preferred embodiment, no more than 0.05 seconds. The hole shape is circular, rectangular, polygonal, or other irregular shapes. Different cross-sectional shapes of the bottom opening produce different boundary constraints and shear field distributions, which can be used as an auxiliary dimension for inverting the characteristic parameters of different rheological models. The hole area is 10%-30% of the bottom area of ​​the loading container. The bottom opening refers to the channel opening at the bottom of the loading container for releasing steel fiber reinforced concrete. The bottom opening is completely closed before the hole baffle is opened, and forms an open channel after opening.

[0033] The specific implementation process is as follows: First, ensure that the bottom perforated baffle of the loading container is completely closed. A sealing gasket is installed between the perforated baffle and the bottom of the loading container to prevent slurry leakage during the loading stage. After loading the pre-mixed steel fiber reinforced concrete into the loading container according to step S1, activate the rapid opening mechanism of the perforated baffle. Control the perforated baffle to fully open within no more than 0.05 seconds. An electromagnetic fast-release mechanism is preferred to achieve millisecond-level opening, ensuring that the disappearance of the lower surface support is approximately a step function, satisfying the instantaneous assumption of the initial condition for free-fall leakage, causing the bottom opening of the loading container to open suddenly. It is worth noting that the opening speed of the perforated baffle must be controlled within 0.05 seconds. If the opening is too slow, it will cause the material to be slowly extruded instead of falling freely as a whole, introducing non-gravity-driven factors that interfere with the test.

[0034] Understandably, step S201 opens the bottom opening suddenly by quickly opening the hole baffle, providing an instantaneous release condition for the subsequent downward flow of steel fiber reinforced concrete under gravity, ensuring that the overall force is only constant gravity and avoiding the introduction of non-gravity factors.

[0035] S202: The open bottom opening causes the steel fiber reinforced concrete to lose its lower surface support at the bottom opening.

[0036] In this embodiment, the lower surface support refers to the vertical support force provided by the bottom hole baffle of the loading container to the lower surface of the steel fiber concrete. When the hole baffle is closed, the lower surface support balances the gravity load of the steel fiber concrete, keeping the material in a static filling state.

[0037] Specific implementation process: When the perforated baffle in step S201 is fully opened, the bottom opening of the loading container is in an open state. The lower surface support originally provided by the perforated baffle suddenly disappears, and the contact force between the lower surface of the steel fiber concrete and the perforated baffle drops to zero. The steel fiber concrete loses its bottom constraint in the vertical direction and is only subject to side wall friction and gravity.

[0038] It is worth noting that the disappearance of the lower surface support must be instantaneous. If a slow opening method is adopted, the lower surface support will gradually decrease, and the material will creep instead of flowing downward, resulting in a change in the overall deformation mode.

[0039] Understandably, step S202 breaks the static equilibrium of the material within the loading container by removing the lower surface support of the steel fiber reinforced concrete, creating the necessary conditions for gravity-driven downward flow and ensuring that subsequent deformation is triggered by pure gravity.

[0040] S203: The steel fiber reinforced concrete at the bottom opening flows downward through the bottom opening under the action of gravity; during the downward flow, the steel fiber reinforced concrete undergoes overall deformation.

[0041] In this embodiment, the gravitational force refers to the downward vertical volume force exerted by Earth's gravity on the steel fiber reinforced concrete. This gravitational force is a constant value, equal to the mass of the steel fiber reinforced concrete multiplied by the gravitational acceleration, i.e. ,in, Gravity is expressed in Newtons (N). The mass of steel fiber reinforced concrete is expressed in kilograms (kg). This represents the acceleration due to gravity, measured in meters per second squared, with the symbol m / s². 2 Take 9.8 m / s 2 Downward flow refers to the movement of steel fiber reinforced concrete as a whole, displacing downward under the constant force of gravity after losing the support of the lower surface. During this downward flow, the different layers inside the material have different speeds due to their own weight, which in turn induces natural shear.

[0042] Specific implementation process: When the steel fiber reinforced concrete loses its lower surface support in step S202, the material is subjected to gravity. Under the influence of the flow, the material begins to move downwards as a whole. During this downward flow, the steel fiber reinforced concrete maintains continuity, with the upper material moving downwards synchronously with the lower material, forming a macroscopic downward flow. Due to frictional resistance between the material and the sidewalls of the container, the material near the sidewalls flows downwards at a slightly lower velocity than the central region. This velocity difference creates a natural shear layer within the material. The central region of the upper surface of the steel fiber reinforced concrete gradually collapses, forming a central sinkhole. The depth of the sinkhole increases with the duration of downward flow, and the edge region forms a specific edge curve morphology. This edge curve can be captured in three dimensions using a non-contact scanning device. Simultaneously, natural shear occurs within the material due to the difference in downward flow velocities between layers, causing relative displacement between the steel fibers and aggregates, resulting in an internal flow field evolution. This internal flow field evolution can be monitored in situ using a lateral non-destructive radiographic scanning device. The final result is a downward-flowing overall deformation model used for subsequent synchronous monitoring of the external morphology and internal flow field.

[0043] It is worth noting that during the downward flow process, the sidewalls of the loading container should be kept smooth with a friction coefficient not exceeding 0.3 to reduce the interference of sidewall friction on the overall deformation mode and allow gravity to become the dominant driving force. In one embodiment, an adjustable inclined flow channel is also provided below the bottom opening of the loading container. The steel fiber reinforced concrete flowing out of the bottom opening flows along the inclined flow channel. By changing the inclination angle of the inclined flow channel, the component of gravity along the flow direction can be adjusted, thereby finely changing the driving stress for testing the critical yield stress. Furthermore, when the fiber volume content of the steel fiber reinforced concrete approaches the upper limit of 2%, the internal cohesion of the material increases significantly, which may lead to incomplete downward flow or the formation of suspension; in this case, supplementary measurements should be taken in conjunction with a suction drop test.

[0044] Understandably, step S203 achieves the downward flow of steel fiber reinforced concrete through gravity. By using constant gravity as the overall force, nonlinear interference caused by variable speed or variable force is avoided, providing stable mechanical conditions for inverting the foundation rheological parameters through morphological features under constant load.

[0045] like Figure 3 As shown, in some embodiments of this application, the preset overall deformation triggering mechanism is a preset vertical piston disposed at the bottom of the preset loading container; Figure 3 The left side shows the initial state of the vertical piston installed at the bottom opening of the loading container, and the right side shows the process of the vertical piston moving downward to create suction force and forming a concave shape on the upper surface of the steel fiber reinforced concrete. An overall force is applied to the steel fiber reinforced concrete through a preset overall deformation triggering mechanism, including the following steps:

[0046] S211: Controls the vertical piston to move downwards at a preset speed.

[0047] In this embodiment, the preset vertical piston refers to a cylindrical piston body installed at the bottom opening of the loading container and capable of moving downward in the vertical direction at a set speed. The diameter of the vertical piston matches the inner diameter of the loading container, with a fitting clearance of 1mm-3mm. The vertical piston is driven by a servo motor ball screw or a hydraulic servo drive. The preset speed refers to the set linear speed at which the vertical piston moves downward. The adjustable range of the preset speed is 1mm / s-100mm / s, and a constant speed mode or a variable speed mode can be selected according to the test requirements.

[0048] The specific implementation process is as follows: First, the vertical piston is reset to the upper edge of the bottom opening of the loading container, so that the top surface of the piston is flush with the bottom surface of the container; after the pre-mixed steel fiber concrete is loaded into the loading container according to step S1, the downward movement speed of the vertical piston is set by the servo controller; the vertical piston drive mechanism is started to control the vertical piston to move downward at a set preset speed.

[0049] It is worth noting that the preset lower limit of the vertical piston speed is 1 mm / s, which corresponds to a sampling time of no more than 100 seconds when the stroke is 100 mm. In a preferred embodiment, a lower speed of 1 mm / s is used in conjunction with a 50 mm stroke, corresponding to a sampling time of 50 seconds, in order to balance sampling efficiency and morphology resolution. The upper limit is 100 mm / s. If this limit is exceeded, the inertial effect will be significant, and non-rheological impact disturbances will be generated inside the material, resulting in distortion of rheological parameter inversion.

[0050] Understandably, step S211, by controlling the vertical piston to move downward at a preset speed, provides a precise displacement input for the subsequent formation of controllable suction force, thereby achieving active control of the overall deformation rate and making it possible to test rheological properties under different equivalent shear rates.

[0051] S212: The downward movement of the vertical piston creates a suction force at the bottom of the preset loading container that corresponds to the preset speed.

[0052] In this embodiment, the suction force refers to the negative pressure attraction formed at the bottom of the loading container when the vertical piston moves downward. This suction force acts as an overall force to induce the upper steel fiber concrete to flow downward. The magnitude of the suction force is related to the moving speed of the vertical piston, the cross-sectional area of ​​the piston, and the viscosity of the steel fiber concrete. The suction force corresponding to the preset speed refers to the suction intensity with a defined numerical range determined by the preset speed.

[0053] Specific implementation process: When the vertical piston in step S211 moves downward at a preset speed, the space between the top surface of the piston and the lower surface of the steel fiber reinforced concrete increases, forming a local low-pressure zone. This low-pressure zone generates a downward suction force on the upper steel fiber reinforced concrete. The direction of the suction force is vertically downward, acting on the entire lower surface of the steel fiber reinforced concrete. It is an overall force rather than a local shear force. The magnitude of the suction force increases with the increase of the piston speed. When the preset speed is 1 mm / s, the empirically estimated suction force is about 0.1 kPa-1 kPa. When the preset speed is 100 mm / s, the empirically estimated suction force is about 10 kPa-50 kPa. The above values ​​are empirical ranges under typical mix proportions. The actual suction force is affected by the nonlinear coupling between the yield stress and plastic viscosity of the steel fiber reinforced concrete and needs to be calibrated according to the specific mix proportion.

[0054] It is worth noting that the formation of suction force requires a good seal between the piston and the container wall, with a clearance of 1mm-3mm. If the clearance is too large, the suction force will leak, and the upper material cannot be effectively induced to flow. In one embodiment, the downward movement of the vertical piston is not limited to a constant preset speed. It can also adopt variable speed modes such as step speed change, continuous acceleration or deceleration, or reciprocating oscillation to stimulate different flow responses of the material under different working conditions, which can be used to evaluate the thixotropy and dynamic viscosity of steel fiber reinforced concrete.

[0055] Understandably, step S212 generates a suction force corresponding to a preset speed by moving the vertical piston downward, converting the mechanical displacement into a pressure difference driving force acting on the entire material, thus achieving non-contact application of the overall force and avoiding local shear damage between the piston and the material.

[0056] S213: Suction force induces steel fiber reinforced concrete to flow downwards; during the flow, the steel fiber reinforced concrete undergoes overall deformation.

[0057] In this embodiment of the application, induced flow refers to the macroscopic continuous downward flow of steel fiber reinforced concrete under the action of the overall suction force. During the induced flow process, natural shearing occurs inside the material, accompanied by the evolution of appearance and internal flow field. The flow direction is consistent with the direction of suction force, both being vertically downward.

[0058] The specific implementation process is as follows: When the suction force generated in step S212 acts on the lower surface of the steel fiber reinforced concrete, the upper steel fiber reinforced concrete flows downward as a whole under the combined action of suction force and gravity. During the flow, the material maintains continuity, with the flow velocity in the central region being greater than that in the edge region. This velocity difference forms a natural shear layer within the material. The steel fibers and aggregates undergo relative displacement during the flow, but the fiber distribution remains random and does not produce directional alignment. The upper surface of the steel fiber reinforced concrete develops a concave morphology, and the curvature and depth of the concavity vary with the suction velocity. When the preset velocity is 1 mm / s, the concavity depth is approximately 5 mm–20 mm; when the preset velocity is 100 mm / s, the concavity depth can reach 50 mm–150 mm, with the specific values ​​depending on the yield stress and plastic viscosity of the material. An internal flow field evolution corresponding to natural shear is generated within the material, and this internal flow field evolution can be monitored in situ using a lateral non-destructive radiographic scanning device. Finally, the overall deformation of the material under suction and fall is obtained for testing the rheological properties at different equivalent shear rates.

[0059] It is worth noting that the continuity of material flow should be monitored during the induced flow process. If breaks or voids occur, it indicates that the suction force is too large or the material viscosity is too high, and the preset speed should be reduced or the material mix ratio adjusted. Different suction speeds correspond to different internal shear rates. By comparing the morphology data at different speeds, the dynamic rheological parameters of the material can be calculated. This test mode perfectly replaces the variable-speed rotational shearing process of traditional rheometers.

[0060] Understandably, step S213 induces the steel fiber reinforced concrete to flow downwards through suction, causing overall deformation of the material under a controllable overall force. This provides an experimental basis for inverting dynamic rheological parameters through morphological differences at different speeds. By inducing overall deformation during the suction-driven flow, the steel fiber reinforced concrete forms a quantifiable concave morphology and internal flow field at different preset speeds, providing multiple sets of experimental data for inverting dynamic rheological parameters such as plastic viscosity through variable-speed testing.

[0061] like Figure 4 As shown, in some embodiments of this application, the preset overall deformation triggering mechanism is a preset horizontal piston disposed inside a preset loading container; the preset loading container is a horizontally placed cylinder with a preset opening at one end and a preset test platform disposed outside the preset opening. Figure 4 The diagram shows a side view and a top view of the lateral collapse that occurs after a horizontal piston pushes the steel fiber reinforced concrete from a pre-set opening onto a test platform. The top view shows the lateral collapse diffusion profile and flow strip distribution. A pre-set overall deformation triggering mechanism applies an overall force to the steel fiber reinforced concrete, including the following steps:

[0062] S221: Controls the horizontal piston to move in the direction of the preset opening.

[0063] In this embodiment, the preset horizontal piston refers to a cylindrical piston body installed inside a horizontally placed barrel that can move forward at a set speed in the horizontal direction. The diameter of the horizontal piston matches the inner diameter of the barrel, and the fitting clearance is 1mm-3mm. The preset opening refers to an opening provided at one end of the horizontally placed barrel. The preset opening is circular, rectangular, polygonal, or other irregular shapes, and the opening area is 80%-95% of the cross-sectional area of ​​the barrel. The preset opening direction refers to the axial movement direction of the horizontal piston toward the preset opening.

[0064] The specific implementation process is as follows: First, the horizontal piston is reset to the rear limit position of the material cylinder. The pre-mixed steel fiber concrete is then loaded into the horizontal material cylinder according to step S1, with the loading volume being 80%-90% of the effective volume of the material cylinder. The forward movement speed of the horizontal piston is set through the servo controller. The horizontal piston drive mechanism is started to control the horizontal piston to move towards the preset opening direction.

[0065] It is understandable that step S221, by controlling the horizontal piston to move in the direction of the preset opening, provides a controllable horizontal driving force for subsequently pushing the steel fiber reinforced concrete onto the test platform, thereby achieving the precise application of the overall horizontal force.

[0066] S222: A horizontal piston pushes steel fiber reinforced concrete from a pre-set opening onto the test platform.

[0067] In this embodiment, "pushing" refers to the horizontal overall pressure applied by the horizontal piston to the steel fiber reinforced concrete, which causes the steel fiber reinforced concrete to shift in the direction of the preset opening. The preset test platform refers to a horizontally placed acrylic plate with a surface roughness Ra of no more than 3.2 μm. In a preferred embodiment, the original surface of the acrylic sheet or the surface of the acrylic sheet is polished to a Ra of no more than 1.6 μm with fine sandpaper to achieve a balance between reducing friction and controlling costs. The platform area is no less than 5 times the cross-sectional area of ​​the barrel. This multiple is determined based on the empirical value of the typical yield stress range of 50 Pa to 500 Pa to ensure that the side collapse process is not affected by the platform boundary constraints.

[0068] The specific implementation process is as follows: When the horizontal piston in step S221 moves towards the preset opening direction, the horizontal piston applies a horizontal pushing force to the steel fiber concrete. This pushing force acts as an overall force, causing the steel fiber concrete to move towards the preset opening direction as a whole. Under the action of the pushing force, the steel fiber concrete is pushed out from the preset opening and enters the space above the test platform. During the pushing process, natural horizontal shear is generated inside the material, and the steel fibers and aggregates undergo relative displacement, but the fiber distribution remains random.

[0069] It is worth noting that the magnitude of the pushing force should be controlled during the pushing process so that the material is pushed out in a continuous whole form. If the pushing force is too large, it may cause the material to break or splash, while if the pushing force is too small, it will not be able to form an effective side collapse morphology.

[0070] Understandably, step S222 pushes the steel fiber reinforced concrete onto the test platform using a horizontal piston, causing the material to undergo overall displacement under horizontal force, thus providing initial positional conditions for subsequent lateral collapse after the loss of sidewall restraint.

[0071] S223: After the steel fiber reinforced concrete is pushed out of the preset opening, it loses the side wall constraint of the barrel.

[0072] In the embodiments of this application, the sidewall constraint refers to the radial restraining force provided by the sidewall of the horizontal cylinder to the steel fiber concrete. This sidewall constraint prevents the material from spreading radially during the pushing process, so that the material maintains the same cross-sectional shape as the cylinder. The loss of sidewall constraint means that after the steel fiber concrete leaves the preset opening, the radial restraining force suddenly disappears, and the material can spread freely in all directions.

[0073] Specific implementation process: When the steel fiber reinforced concrete in step S222 is pushed out of the preset opening, the material leaves the area surrounded by the side wall of the horizontal cylinder, and the radial constraint originally provided by the side wall of the cylinder suddenly disappears; the material loses its boundary restriction in the horizontal direction and is only constrained by the friction of the bottom surface of the test platform; the disappearance of the side wall constraint causes the material to change from a constrained columnar shape to a free-diffusion flow shape.

[0074] It is worth noting that the area of ​​the preset opening should be 80%-95% of the cross-sectional area of ​​the barrel. If the opening is too small, the material will be subjected to local shearing at the edge of the opening during extrusion, which will destroy the original fiber distribution. If the opening is too large, the sidewall constraint will disappear too early and an effective pushing section cannot be formed.

[0075] Understandably, step S223 breaks the radial boundary restriction of the steel fiber reinforced concrete by removing the sidewall constraint of the cylinder, thus creating free diffusion conditions for lateral collapse under gravity.

[0076] S224: Steel fiber reinforced concrete that has lost its sidewall restraints will collapse laterally under the action of gravity; during the collapse, the steel fiber reinforced concrete will undergo overall deformation.

[0077] In the embodiments of this application, lateral collapse refers to the composite deformation process in which steel fiber reinforced concrete, after losing the restraint of the side walls, freely diffuses in all directions under the action of gravity and collapses vertically. During this lateral collapse process, the material generates a composite shear of horizontal flow shear and gravity collapse.

[0078] The specific implementation process is as follows: When the steel fiber reinforced concrete in step S223 loses its sidewall constraint, the material undergoes lateral collapse under gravity. During the collapse, the material freely diffuses in all directions, forming a lateral collapse profile. Simultaneously, the material collapses vertically, with the collapse depth in the central region being greater than that in the edge region, forming a specific three-dimensional lateral collapse morphology. A composite shearing effect of horizontal flow shear and gravity collapse occurs within the material, causing relative displacement between the steel fibers and aggregates in both horizontal and vertical directions. During the collapse, the extruded material forms a specific lateral collapse profile and flow stripe distribution on the test platform. The lateral collapse profile can be acquired in three dimensions using a non-contact scanning device, and the flow stripe distribution can be obtained through top-view image recognition. An internal flow field evolution corresponding to natural shear is generated within the material, which can be monitored in situ using a lateral non-destructive scanning device. Finally, the overall deformation of the lateral collapse under horizontal pushing conditions is obtained for testing its rheological properties.

[0079] It is worth noting that the area of ​​the test platform must be no less than five times the cross-sectional area of ​​the barrel to ensure that the lateral collapse process is not affected by the platform boundary constraints and to truly reflect the free flow characteristics of the material; the surface roughness Ra of the platform should not exceed 3.2 μm to reduce the interference of platform friction on the lateral collapse morphology. Lateral collapse under gravity causes steel fiber reinforced concrete to form a free-diffusion three-dimensional lateral collapse morphology after losing sidewall constraints, providing macroscopic deformation samples for inverting rheological properties through the correspondence between the lateral collapse profile and the pushing velocity. Changes in the pushing velocity lead to different degrees of shear and lateral collapse morphology. Based on the correspondence between the lateral collapse morphology and the pushing velocity, the rheological properties of the material can be calculated. This test mode can comprehensively evaluate the rheological behavior of materials under combined horizontal flow and gravity collapse conditions.

[0080] In summary, step S2 uses a perforated baffle at the bottom of the loading container, a vertical piston, or a horizontal piston inside the barrel as the overall deformation triggering mechanism. The overall force, such as gravity, suction force corresponding to the preset speed, or horizontal pushing force, induces the overall deformation of the steel fiber reinforced concrete. This maintains the original random distribution state of the steel fibers under non-destructive rheological testing conditions, allowing the material to undergo natural shearing accompanied by changes in appearance and internal flow field, thus truly reflecting the intrinsic rheological properties of the material.

[0081] S3: During the overall deformation process, the external morphology evolution data is acquired simultaneously using a preset non-contact scanning device, and the internal flow field evolution data is acquired simultaneously using a preset lateral non-destructive perspective scanning device.

[0082] In this application embodiment, the non-contact scanning device refers to a device that collects real-time data on the three-dimensional geometric changes of the steel fiber reinforced concrete surface without physical contact with it. This includes non-contact three-dimensional reconstruction devices such as 3D laser scanners, structured light scanners, binocular stereo vision cameras, infrared depth cameras, or multi-view high-frequency camera arrays. The appearance morphology evolution data refers to spatiotemporal sequence data continuously collected during the overall deformation process, reflecting the three-dimensional morphological changes of the steel fiber reinforced concrete surface. This data exists in the form of a three-dimensional contour point cloud sequence. The lateral non-destructive imaging scanning device refers to an industrial computed tomography (CT) scanner installed on the side of a loading container or cylinder, which uses X-rays to penetrate the container wall and perform tomographic scanning of the material's interior. The highest spatial resolution of this industrial CT scanner is no greater than 0.1 mm / voxel. In one embodiment, the lateral non-destructive fluoroscopic scanning device can also be a magnetic resonance imaging device or a neutron beam imaging device, as long as it can penetrate the wall of the loading container and identify the steel fibers and aggregates; the internal flow field evolution data refers to the spatiotemporal sequence data reflecting the relative motion of each phase component inside the steel fiber reinforced concrete, which is continuously collected during the overall deformation process. This data exists in the form of an industrial computed tomography (CT) image sequence; synchronous acquisition refers to using the start time of the overall deformation as the synchronization reference, simultaneously activating the non-contact scanning device and the lateral non-destructive fluoroscopic scanning device, so that the three-dimensional contour point cloud sequence and the CT tomographic image sequence form a data pair at the same moment according to a preset time synchronization relationship; when the frame rates of the two are different, nearest neighbor matching, downsampling, or time interpolation is used to form a corresponding relationship, and the accuracy of this time synchronization relationship is no greater than 0.05 seconds. It should be noted that the specific implementation process of step S3 is elaborated in detail by steps S301-S304, and the specific implementation steps include:

[0083] S301: Using the start time of the overall deformation as the synchronization reference, simultaneously start the preset non-contact scanning device and the preset lateral non-destructive radiographic scanning device.

[0084] In this embodiment of the application, the synchronization reference refers to the reference time used to unify the zero point of the time of the non-contact scanning device and the lateral non-destructive fluoroscopy scanning device. The synchronization reference is the starting time of the overall deformation, that is, the moment the hole baffle opens, the moment the vertical piston starts, or the moment the horizontal piston starts. Simultaneous start means that at the synchronization reference time, a synchronization trigger signal is sent to the non-contact scanning device and the lateral non-destructive fluoroscopy scanning device through a synchronization trigger or network time protocol, so that the data acquisition program of the two devices starts running at the same time.

[0085] The specific implementation process is as follows: First, determine the starting moment of the overall deformation. For the downward flow test, the starting moment is the instant when the orifice baffle is fully opened; for the suction and fall test, the starting moment is the instant when the vertical piston begins to move downward; for the push and collapse test, the starting moment is the instant when the horizontal piston begins to move forward. At the determined synchronization reference moment, a hardware trigger signal is sent to the non-contact scanning device and the lateral non-destructive radiographic scanning device through a synchronization trigger, or a software synchronization command based on the network time protocol is sent to the two devices through the local area network, so that the data acquisition programs of the two devices start running at the same millisecond level.

[0086] It is worth noting that the delay of the synchronization trigger signal should not exceed 0.01 seconds to ensure that the actual start-up time difference between the two devices is controlled within an acceptable range; if software synchronization is used, it should be ensured that the two devices are connected to the same local area network switch and the network delay is not greater than 1 millisecond.

[0087] It is understandable that step S301 provides a precise time zero point for establishing the time correspondence between the three-dimensional contour point cloud sequence and the industrial computed tomography image sequence by using the start time of the overall deformation as a synchronization reference and starting the two scanning devices simultaneously, thus ensuring that the macroscopic appearance monitoring and microscopic internal monitoring are strictly aligned in time.

[0088] S302: Throughout the entire evolution stage of the overall deformation, the three-dimensional contour point cloud sequence of the steel fiber reinforced concrete surface is continuously acquired through a preset non-contact scanning device as appearance evolution data.

[0089] In this embodiment of the application, the three-dimensional contour point cloud sequence refers to the set of three-dimensional coordinate points describing the surface of steel fiber reinforced concrete that are continuously collected in chronological order during the overall deformation process. Each frame in the three-dimensional contour point cloud sequence contains the three-dimensional coordinate information of each sampling point on the material surface. The appearance morphology evolution data refers to the spatiotemporal sequence data reflecting the three-dimensional morphological changes of the surface of steel fiber reinforced concrete that are continuously collected during the overall deformation process. The evolution stage refers to the entire process from the start of the synchronization reference time to the point where the overall deformation is basically stable or the preset collection time is reached.

[0090] Specific implementation process: Throughout the entire evolution stage of overall deformation, a non-contact scanning device continuously acquires a three-dimensional contour point cloud sequence of the steel fiber reinforced concrete surface at a frequency of no less than 10 frames / second, serving as the appearance morphology evolution data. Each frame in the three-dimensional contour point cloud sequence contains the three-dimensional coordinate information of each sampling point on the material surface, with a sampling point density of no less than 100 points / cm². 2For the downward flow test, the sampling duration is 5 to 30 seconds until the material flows downward steadily or reaches the bottom of the container; for the suction drop test, the sampling duration is related to the stroke of the vertical piston, usually 10 to 60 seconds; for the push-and-collapse test, the sampling duration is 20 to 120 seconds until the collapse process is basically stable.

[0091] It is worth noting that the sampling frequency of the non-contact scanning device should be no less than the frame rate of industrial computed tomography (CT) scans, and in a preferred embodiment, no less than twice that frame rate. When using a fast electron beam CT route, a multi-view high-frequency camera array can be used to meet this matching requirement. When the frame rate of industrial CT scans is higher than the achievable frame rate of the non-contact scanning device, the point cloud data is interpolated according to the preset time synchronization relationship in step S304 to ensure the spatiotemporal correspondence accuracy between macroscopic appearance monitoring and microscopic internal monitoring. The distance between the scanning device and the material surface should be maintained within the range of 0.3m to 1.5m to ensure measurement accuracy and field of view coverage.

[0092] Understandably, step S302, by continuously acquiring a three-dimensional contour point cloud sequence throughout the entire evolution stage of the overall deformation, fully records the dynamic evolution process of the steel fiber reinforced concrete surface morphology from the initial state to the final state, providing the original data basis for subsequent extraction of macroscopic morphological feature variables such as collapse depth, edge curvature, and maximum wave height.

[0093] S303: Simultaneously, a CT tomographic image sequence of steel fiber reinforced concrete is continuously acquired through a preset lateral non-destructive fluoroscopic scanning device as data characterizing the evolution of the internal flow field; wherein, each frame in the three-dimensional contour point cloud sequence corresponds to the frame at the corresponding time in the CT tomographic image sequence according to a preset time synchronization relationship; when the frame rates of the two are different, nearest neighbor matching, downsampling or time interpolation is used to form a correspondence.

[0094] In this embodiment, the industrial computed tomography image sequence refers to a collection of tomographic projection images or three-dimensional reconstruction data that are continuously acquired in chronological order during the overall deformation process and reflect the internal structural state of steel fiber reinforced concrete; the internal flow field evolution data refers to the spatiotemporal sequence data that are continuously acquired during the overall deformation process and reflect the relative motion of the various phase components inside the steel fiber reinforced concrete; and lateral non-destructive radiography refers to using X-rays to penetrate the container wall from the side of the loading container or cylinder to perform tomographic scanning of the material's interior without damaging the material's internal structure.

[0095] Specific implementation process: Throughout the entire evolution stage of overall deformation, a sequence of industrial computed tomography (CT) images of steel fiber reinforced concrete is continuously acquired using a lateral non-destructive radiographic scanning device as data for the internal flow field evolution. To address the motion artifact problem during dynamic deformation, one of the following two optional technical approaches is adopted:

[0096] Route 1, namely the single-plane high-speed tomography monitoring route: A high-frame-rate X-ray flat panel detector is used in conjunction with a fixed X-ray source to perform single-plane continuous perspective imaging on the outlet section of the barrel in the push-and-collapse test or the middle section of the container in the downward flow test, with a frame rate of not less than 60fps, to directly obtain a two-dimensional dynamic distribution image of the fibers in the monitoring plane; the fiber motion in the monitoring plane is regarded as a representative cross-sectional flow field of the three-dimensional internal flow field, which is used to calculate the fiber network connectivity of the cross-section.

[0097] Route 2, namely the fast electron beam CT route: adopts an electron beam fast CT system, which generates rotating X-rays by bombarding a ring target with an electromagnetically deflected electron beam. It can complete the projection acquisition of single-frame three-dimensional volume data in milliseconds without mechanical rotation, and the reconstruction time step is no more than 0.1 seconds, thereby controlling motion artifacts within an acceptable range.

[0098] It is important to distinguish that the acquisition frequency of X-ray projection data can be 0.5 seconds. -1 up to 5s -1 This corresponds to one projection every 0.2–2 seconds; when using a fast electron beam CT route, the acquisition frequency can be increased to 30 seconds. -1 up to 60s -1 Each frame of CT tomographic image is reconstructed from X-ray projection data acquired within a preset reconstruction time window: In the fast electron beam CT route, approximately 2-30 projections can be acquired within the reconstruction time window of 0.05-0.5 seconds, at which point sparse perspective reconstruction algorithms such as compressed sensing or algebraic iteration are used to complete the tomographic reconstruction; in the static or quasi-static scanning stages before and after testing, at least 180 projections can be acquired using conventional mechanical rotation CT to obtain high signal-to-noise ratio reconstructed images; in the single-plane high-speed tomographic monitoring route, tomographic reconstruction is not required, and single-plane fluoroscopic images are directly and continuously acquired at a high frame rate of not less than 60fps. In one embodiment, the preset reconstruction time window can be 0.05 to 0.5 seconds, and the preset observation time step can be 0.05 to 0.2 seconds. For rapid deformation processes, the preset reconstruction time window and preset observation time step can be reduced; for slow deformation processes, the preset reconstruction time window can be increased to improve the reconstruction signal-to-noise ratio. Each frame in the CT tomographic image sequence is a two-dimensional perspective projection image of the monitoring plane in Route 1, and a three-dimensional volume data obtained through rapid reconstruction in Route 2. The image grayscale resolution is not less than 16 bits. The scanning device is arranged on the side of the loading container or cylinder, with the X-ray source and detector located on both sides of the container to ensure that the X-ray penetration path is perpendicular to the main direction of material flow. For downward flow tests and suction drop tests, the scanning plane is located at the middle height of the container. For pushing and side collapse tests, the scanning plane is located near the cylinder outlet.

[0099] It is worth noting that the selection of the X-ray projection data acquisition frequency is determined based on the expected duration of the overall deformation. When using a fast electron beam CT route, a 60s frequency is used when the overall deformation duration is less than 5 seconds. -1 High-speed mode captures rapidly evolving details; a 30-second interval is used when the duration exceeds 5 seconds. -1 The standard mode balances data volume and resolution; when using the standard mechanical rotation CT path, the acquisition frequency is 0.5s. -1 up to 5s -1 It is suitable for static or quasi-static scanning before and after testing; the container must be made of X-ray permeable material such as acrylic, with a wall thickness of 5mm-10mm to ensure effective X-ray penetration.

[0100] Understandably, step S303, by synchronously and continuously acquiring industrial computed tomography image sequences, fully records the relative motion trajectories of steel fibers, aggregates, and slurry inside the steel fiber reinforced concrete during the overall deformation process, providing a raw perspective data basis for subsequent extraction of fiber network connectivity and tracking of internal flow field evolution.

[0101] The 3D contour point cloud sequence and the CT tomographic image sequence correspond to each other according to a preset time synchronization relationship. When the two have the same time resolution, they can be directly matched one by one according to the timestamp; when the two have different time resolutions, the correspondence can be established by nearest neighbor matching, downsampling, or linear interpolation.

[0102] For any target time The point cloud frame and CT tomographic image frame with the smallest time difference are selected to form a candidate correspondence; the preset synchronization threshold is set to no more than 0.05 seconds; when the time difference is greater than this threshold, linear time interpolation is used first to fill in the missing data, and the interpolation is based on the spatiotemporal coordinates of two adjacent frames; if the number of consecutive missing frames exceeds 3 frames, the time period is marked as a data missing area and will not participate in the subsequent inversion.

[0103] Through the above steps, dual-source observation data on the synchronous evolution of macroscopic appearance and internal fiber structure over time can be obtained.

[0104] In summary, step S3 employs non-contact scanning devices such as 3D laser scanners, structured light scanners, binocular stereo vision cameras, infrared depth cameras, or multi-view high-frequency camera arrays, as well as an industrial computed tomography (CT) scanner installed on the side of the loading container. Both devices are activated simultaneously with the overall deformation initiation time as the synchronization reference. Throughout the entire evolution of the overall deformation, 3D contour point cloud sequences and industrial CT image sequences are continuously acquired to achieve spatiotemporal synchronous monitoring of macroscopic appearance and microscopic internal flow field, establishing a dual-source spatiotemporal monitoring dataset with strict time correspondence.

[0105] S4: Spatiotemporally correlate and fuse the external morphology evolution data with the internal flow field evolution data, and calculate the rheological characteristic parameters of steel fiber reinforced concrete based on the preset physical information neural network model with embedded preset rheological physical equation constraints.

[0106] In this embodiment, spatiotemporal correlation fusion refers to the process of spatially registering, extracting features, and storing the evolution data of the external morphology and the internal flow field at the same moment to form a unified dataset that can simultaneously describe the geometric morphology of the material surface and the internal structural state; dynamic boundary conditions refer to the boundary constraints that change with time in fluid dynamics calculations, and in this application, specifically refer to the free surface boundary contour extracted from the free surface geometric model; fiber network connectivity refers to the dimensionless parameter characterizing the integrity of the internal skeleton of steel fiber reinforced concrete; macroscopic morphological feature variables refer to the set of parameters extracted from the free surface geometric model for quantifying the evolution state of the external morphology; and preset rheological physical equation constraints refer to the constitutive equations describing the relationship between shear stress and shear rate established based on the Modified Bingham model, the Herschel-Bulkley model, or the conventional power function model, which are embedded as soft constraints in the loss function of the physical information neural network model.

[0107] Understandably, step S4 unifies macroscopic appearance data and microscopic internal data into the same spatiotemporal framework through spatiotemporal correlation fusion, and performs rheological parameter inversion based on a physical information neural network model constrained by rheological physical equations. This solves the problem that traditional rheometers cannot directly calibrate steel fiber reinforced concrete due to shear damage to fiber distribution, and realizes intelligent inversion of rheological parameter inversion based on dual verification of macroscopic morphology and microscopic flow field.

[0108] It should be noted that the specific implementation process of step S4 is detailed by steps S401 to S406, steps S4061 to S4066, and steps S411 to S416 respectively; after the inversion calculation is completed, steps S4067 to S4074 are executed to verify the final state and correct the physical information neural network model through solidification and accompanying backpropagation.

[0109] S401: The three-dimensional contour point cloud sequence and the CT tomographic image sequence are matched one by one according to the preset time synchronization relationship to form a data pair at the same time.

[0110] In the embodiments of this application, the time synchronization relationship refers to the time correspondence rule between each frame in the three-dimensional contour point cloud sequence and the corresponding frame in the industrial computed tomography image sequence, and the accuracy of the time synchronization relationship is no greater than 0.05 seconds; the data pair refers to the associated data unit composed of a frame of three-dimensional contour point cloud data and a frame of industrial computed tomography image data at the same time.

[0111] The specific implementation process is as follows: First, the acquisition timestamps of the 3D contour point cloud sequence and the industrial computed tomography (CT) image sequence are read. The timestamp accuracy of the 3D contour point cloud sequence is at the millisecond level, and the timestamp accuracy of the industrial CT image sequence is at the frame level. For each frame of the industrial CT image, the frame with the smallest absolute time difference in the 3D contour point cloud sequence is found as the corresponding frame. If the minimum time difference is not greater than 0.05 seconds, the two frames are combined into a data pair. If the minimum time difference is greater than 0.05 seconds, the time is marked as missing data and linear interpolation is used to fill in the missing data. All frames are traversed to form a sequence of data pairs arranged in chronological order.

[0112] It is worth noting that timestamp reading should be based on the unified clock source of the hardware clock or synchronous trigger of each device to avoid time drift caused by using their own independent system clocks; if the two devices have different acquisition frame rates, the device with the lower frame rate should be used as the benchmark, and the data with the higher frame rate should be downsampled or averaged.

[0113] Understandably, step S401 establishes a precise temporal mapping between macroscopic appearance data and microscopic internal data by forming data pairs one by one according to the time synchronization relationship, providing a time-aligned data foundation for subsequent joint analysis.

[0114] S402: Use a preset surface reconstruction algorithm to reconstruct the surface of the three-dimensional contour point cloud in the data pair, and generate a free surface geometric model of steel fiber reinforced concrete at the current moment.

[0115] In this embodiment of the application, the surface reconstruction algorithm refers to a mathematical algorithm that converts a discrete three-dimensional contour point cloud into a continuous triangular mesh or a parametric surface, including the Poisson surface reconstruction algorithm or the Delaunay triangulation algorithm; the free surface geometric model refers to a mathematical model generated by the surface reconstruction algorithm that describes the continuous three-dimensional shape of the outer surface of steel fiber reinforced concrete at a certain moment; the three-dimensional contour point cloud refers to a set of discrete three-dimensional coordinate points acquired by a non-contact scanning device.

[0116] Specific implementation process: For each data pair formed in step S401, extract the three-dimensional contour point cloud data; before reconstruction, estimate the normal vector of the three-dimensional contour point cloud. In one implementation, principal component analysis based on local plane fitting is used to estimate the normal vector of each sampling point, with 10-30 neighborhood points to meet the input requirements of the Poisson surface reconstruction algorithm for point clouds with normal vectors; use the Poisson surface reconstruction algorithm or the Delaunay triangulation algorithm to reconstruct the surface of the three-dimensional contour point cloud, generating a free surface geometric model of steel fiber reinforced concrete at the current moment, wherein the mesh resolution of the surface reconstruction is set to 0.5mm-1mm; the octree depth of the Poisson surface reconstruction algorithm is set to 8-10 to adapt to the mesh resolution of 0.5mm-1mm; the triangulation criterion of the Delaunay triangulation algorithm is maximum and minimum angle optimization to avoid the occurrence of narrow triangles; the reconstructed free surface geometric model is stored in the form of triangular mesh or parametric surface, including vertex coordinates, normal vectors and topological connection relationships.

[0117] It is worth noting that the grid resolution of the surface reconstruction should match the sampling point density of the non-contact scanning device. If the grid resolution is too high, the computational load will increase and noise may be introduced; if it is too low, detailed features will be lost. Before reconstruction, outlier removal and smoothing filtering should be performed on the point cloud data, with the outlier removal threshold set to three times the average spacing of the point cloud. It is understandable that step S402 uses a surface reconstruction algorithm to convert the discrete 3D contour point cloud into a continuous free surface geometric model, providing a structured 3D geometric representation for subsequent extraction of macroscopic morphological feature variables and dynamic boundary conditions.

[0118] S403: Obtain macroscopic morphological feature variables from the free surface geometric model using a preset morphological feature extraction algorithm.

[0119] In this embodiment of the application, the morphological feature extraction algorithm refers to an algorithm that quantitatively extracts macroscopic morphological feature variables such as collapse depth, edge curvature, diffusion area, maximum wave height and fluctuation time from a free surface geometric model; macroscopic morphological feature variables refer to a set of parameters used to quantify the evolution state of the appearance morphology of steel fiber reinforced concrete.

[0120] The specific implementation process is as follows: From the free surface geometric model generated in step S402, macroscopic morphological feature variables are obtained using a preset morphological feature extraction algorithm; the collapse depth is obtained by calculating the maximum vertical distance of each vertex on the free surface geometric model relative to the initial plane; the edge curvature is obtained by extracting the boundary contour line of the free surface geometric model and calculating the reciprocal of the radius of curvature of each point on the boundary contour line; the diffusion area is obtained by calculating the projected area of ​​the free surface geometric model on the horizontal plane; the maximum wave height is obtained by calculating the maximum vertical displacement of all vertices on the free surface geometric model relative to the corresponding position points in the initial state; and the fluctuation time is obtained by tracking the time interval required for the maximum wave height to reach the final stable value from the initial value.

[0121] It is worth noting that morphological feature extraction should be performed immediately after surface reconstruction to avoid data accumulation; the calculation of each feature variable should be based on the same coordinate system and the same reference plane to ensure comparability between different time steps. Understandably, step S403 extracts macroscopic morphological feature variables from the free surface geometric model, transforming the three-dimensional appearance morphology into quantifiable numerical indicators, providing directly input macroscopic feature data for subsequent rheological parameter inversion.

[0122] S404: Based on a preset boundary extraction threshold, extract the boundary contour of the free surface from the free surface geometry model, and use it as the dynamic boundary condition for the fluid dynamics calculation at the current moment.

[0123] In this embodiment, the boundary extraction threshold refers to the geometric discrimination threshold used to distinguish the fluid free surface from the background or container wall in the free surface geometry model. This boundary extraction threshold is set to an angle greater than 85° between the surface normal vector and the gravity direction, or a surface curvature greater than 0.5 mm. -1 Boundary profile refers to the closed curve formed by the vertices that satisfy the boundary extraction threshold condition in the free surface geometric model.

[0124] Specific implementation process: Based on a preset boundary extraction threshold, the boundary contour of the free surface is extracted from the free surface geometric model generated in step S402. This boundary contour is extracted by determining whether the angle between the normal vector of each vertex of the surface and the direction of gravity is greater than 85° or whether the surface curvature is greater than 0.5mm. -1 The vertices that meet the conditions form a free surface boundary; the extracted boundary contour node coordinates are arranged in clockwise or counterclockwise order to form a closed boundary contour line, which serves as the dynamic boundary condition for the fluid dynamics calculation at the current moment; the dynamic boundary condition is stored in the form of a node coordinate list, which includes the three-dimensional coordinates of each node and the connection relationship between adjacent nodes.

[0125] It is worth noting that the curvature threshold in the boundary extraction threshold is 0.5mm. -1This threshold is applicable to steel fiber reinforced concrete with general aggregate sizes. If the aggregate size is greater than 10mm, this threshold should be adjusted to 0.2mm. -1 —0.3mm -1 To avoid aggregate edges being misjudged as free surface boundaries; the threshold of 85° for the angle between the normal vector and the direction of gravity is suitable for most collapse and side collapse morphologies. If the material has extremely poor fluidity, resulting in a nearly vertical surface, this threshold can be adjusted to 80°-85°.

[0126] Understandably, step S404 extracts dynamic boundary conditions from the free surface geometric model, transforming the measured appearance into variable boundary inputs for fluid dynamics calculations, thus providing time-evolving geometric constraints for subsequent forward modeling and boundary updates.

[0127] S405: Based on CT tomographic image sequences, a preset fiber recognition algorithm is used to extract the spatial location and motion features of steel fibers, and the fiber network connectivity, which characterizes the integrity of the internal skeleton, is calculated according to a preset connectivity calculation formula.

[0128] In this application embodiment, the fiber recognition algorithm refers to an image processing algorithm that identifies and extracts the spatial location of steel fibers based on the grayscale and morphological features of industrial computed tomography (CT) images. This includes a tubular structure enhancement filtering algorithm based on the Hessian matrix or a fiber segmentation algorithm based on deep learning. The tubular structure enhancement filtering algorithm based on the Hessian matrix refers to an image filtering algorithm that enhances tubular or linear structures by calculating the eigenvalues ​​of the Hessian matrix at each point in the image. In this application, it specifically refers to an enhancement filtering algorithm used to highlight the high grayscale structure of steel fibers, thereby improving the contrast between the fibers and the matrix. Its input is an industrial CT image, and its output is the enhanced fiber. Structural response map; The deep learning-based fiber segmentation algorithm refers to a semantic segmentation algorithm that uses a convolutional neural network to perform pixel-level classification of industrial computed tomography (CT) images. In this application, it specifically refers to a segmentation algorithm used to identify the pixel region of steel fibers, which is used to achieve sub-pixel level fiber boundary extraction. Its input is an industrial CT image and its output is a fiber mask image. In one embodiment, the segmentation algorithm uses a U-Net network, in which the encoder extracts multi-scale features through 4 downsampling operations and the decoder restores resolution through 4 upsampling operations. Feature maps are stitched together using skip connections between the same layers. The convolutional kernel size is 3×3, and the training samples are no less than 500 CT images with manually labeled fiber regions.

[0129] The specific implementation process is as follows: Based on CT tomographic image sequences, a preset fiber recognition algorithm is used to extract the spatial location and motion features of steel fibers. This fiber recognition algorithm first performs Gaussian filtering on the industrial computed tomographic scan images for noise reduction, with the standard deviation of the Gaussian kernel set to 1-2 pixels; then, a tubular structure enhancement filter based on the Hessian matrix is ​​used to highlight the fibrous structure, with the eigenvalue threshold of the Hessian matrix set to 0.1-0.5; finally, the centerline coordinates and direction vectors of each fiber are obtained through threshold segmentation and skeleton extraction. The threshold segmentation adopts the Otsu adaptive thresholding method, where the Otsu adaptive thresholding method refers to a statistical method that automatically determines the image segmentation threshold by maximizing the inter-class variance. In this application, it specifically refers to the adaptive threshold determination algorithm used to distinguish fiber pixels from background pixels, which is used to achieve automatic extraction of fiber regions without manual intervention. Its input is the enhanced fiber structure response map and its output is a binary fiber mask.

[0130] The fiber network connectivity, representing the integrity of the internal skeleton, is calculated according to a preset connectivity calculation formula. The design principle of this formula is that the integrity of the fiber network skeleton depends on both the proportion of fibers participating in connectivity and the average length contribution of connected paths; the product of these two factors comprehensively represents the integrity of the internal skeleton. The maximum connected component is obtained using a disjoint-set data structure algorithm or a breadth-first search algorithm: traversing the undirected graph. For all nodes, nodes connected by edges are grouped into the same connected component, and the connected component with the largest number of nodes is taken as the largest connected component. Therefore, the formula for calculating connectivity is derived and expressed mathematically as follows:

[0131] .

[0132] in: This represents the total number of fibers identified in the 3D reconstruction at the current moment. When using a single-plane high-speed tomography monitoring route, it represents the total number of fibers identified in the 2D image of the monitoring plane. The number of fibers in the largest connected domain within the 3D reconstruction body or monitoring plane at the current moment; The average length of the fiber centerline within the largest connected region; This refers to the nominal length of a single steel fiber. This represents the connectivity coverage ratio; This is a length contribution factor. When... When, define ;when At that time, take To maintain the connectivity normalization upper limit at 1.

[0133] The derivation of this formula is based on the fact that the fiber network is represented as an undirected graph. ,in For a set of fiber nodes, For the set of fiber contact relationships, an edge is established when the minimum distance between the centerlines of two fibers is less than a preset contact discrimination distance. This contact discrimination distance is preset to be no greater than 1.5 times the diameter of a single steel fiber, and in one embodiment, it is taken as 0.05mm-0.5mm. The integrity of the fiber network skeleton is affected by both the proportion of connected fibers and the contribution of the length of connected fibers: if Larger but Smaller size indicates that although there are many fibers, the connecting chains are relatively short; if Larger but A smaller value indicates the presence of only a few long fiber chains. Neither of these scenarios can form an effective load-bearing skeleton; therefore, the product of the two values ​​is used to characterize fiber network connectivity. This normalization is based on the following physical fact: the saturation enhancement effect of fiber network connectivity is fully reflected at the single fiber length scale; longer connected chains, while increasing the network scale, do not change the relative contribution rate of local skeleton enhancement. Therefore, the nominal length of a single fiber is used as the normalization benchmark to ensure... This ensures that subsequent dynamic constitutive constraints have stable physical boundaries.

[0134] It is worth noting that the accuracy of the fiber identification algorithm is affected by the resolution of the industrial computed tomography scan image. When the scanning resolution is not less than 0.1 mm / voxel, steel fibers with a diameter of not less than 0.3 mm can be identified. If the fiber diameter is less than 0.3 mm, the scanning resolution needs to be increased to 0.05 mm / voxel or a fiber segmentation algorithm based on deep learning needs to be used.

[0135] Understandably, step S405 uses a fiber identification algorithm and connectivity calculation formula to convert the industrial computed tomography image sequence into quantified fiber network connectivity, providing microstructural feature inputs for the dynamic constitutive constraints in the subsequent physical information neural network model.

[0136] S406: Link and store dynamic boundary conditions, fiber network connectivity, and macroscopic morphological feature variables to generate a spatiotemporal correlation fusion dataset.

[0137] In this embodiment of the application, associated storage refers to the process of binding and writing dynamic boundary conditions, fiber network connectivity and macroscopic morphological feature variables at the same time in a structured manner into a data file; the spatiotemporal associated fusion dataset refers to a structured data set organized in time series, containing dynamic boundary conditions, fiber network connectivity and macroscopic morphological feature variables at each time. Each time step record of the dataset contains a timestamp, coordinates of dynamic boundary condition nodes, fiber network connectivity values ​​and macroscopic morphological feature variable values.

[0138] The specific implementation process is as follows: The dynamic boundary conditions extracted in step S404, the fiber network connectivity calculated in step S405, and the macroscopic morphological feature variables obtained in step S403 are associated and stored according to time series to generate a spatiotemporal associated fusion dataset. The data storage format adopts structured text format or database format. Each time step record includes a timestamp, a list of dynamic boundary condition node coordinates, fiber network connectivity value, collapse depth value, edge curvature value, diffusion area value, maximum wave height value, and fluctuation time value. The timestamps are uniformly set to millisecond precision to ensure consistency with the original acquisition timestamps. After the dataset is generated, integrity verification is performed to check whether each time step record contains all fields. Missing fields are filled in using interpolation or marking methods. The free surface geometric model is stored and retrieved independently according to time index.

[0139] It is worth noting that the storage capacity of the spatiotemporal correlation fusion dataset increases with the number of time steps and spatial resolution. When the acquisition time is 60 seconds and the observation time step is 0.05 seconds, that is, 20 time steps per second, the dataset has approximately 1200 records and the storage space is approximately 50MB-200MB. It is recommended to adopt a compressed storage or block storage strategy to save storage space and improve reading efficiency.

[0140] Understandably, step S406 achieves the fusion of a unified spatiotemporal framework for dual-source heterogeneous data by associating and storing dynamic boundary conditions, fiber network connectivity, and macroscopic morphological feature variables, thus providing a structured and correlated input data foundation for the subsequent inversion calculation of the physical information neural network model.

[0141] It should be noted that the subsequent processing and updating of dynamic boundary conditions, fiber network connectivity, and macroscopic morphological feature variables are detailed in steps S4061 to S4066. S4061: With a preset time step as the period, the free surface geometric model corresponding to the current moment is obtained sequentially in chronological order, and the corresponding dynamic boundary conditions are extracted from the spatiotemporal correlation fusion dataset.

[0142] In this embodiment, the preset time step refers to the time discrete interval used in the dynamic boundary update and topography prediction correction process, and the preset time step is denoted as . The time step is set to 0.05 seconds to 0.2 seconds, consistent with the preset observation time step range in step S303, and not greater than the time interval between two adjacent frames of the CT tomographic image sequence; the time step within the computational fluid dynamics (CFD) forward model can be no greater than... numerical substep And satisfy the Courant stability condition:

[0143] .

[0144] in, For Courant number, To calculate the maximum velocity within the domain, For local mesh feature size, A preset stability upper limit is set. In one implementation, The value ranges from 0.5 to 1.0. The current time refers to the point in time being processed during the iterative calculation process. This current time is selected sequentially from the spatiotemporal correlation fusion dataset. The free surface geometric model refers to the mathematical model generated by the surface reconstruction algorithm that describes the continuous three-dimensional morphology of the outer surface of steel fiber reinforced concrete at a certain time.

[0145] Specific implementation process: First, use a preset observation time step. The preset observation time step is set to 0.05 seconds, and the current moment is extracted sequentially from the spatiotemporal correlation fusion dataset generated in step S406. Free surface geometric model and its corresponding dynamic boundary conditions During extraction, the current time is located based on the timestamp index. The corresponding record is retrieved, along with the list of dynamic boundary condition node coordinates and the values ​​of macroscopic topographic feature variables. If the current time... If there is no exact matching record in the dataset, linear interpolation is used to interpolate the free surface geometry model and dynamic boundary conditions at the current time from the records of adjacent time steps.

[0146] It is worth noting that the preset observation time step The choice requires a trade-off between computational accuracy and efficiency, especially when using a fast electron beam CT route with a projection acquisition frequency of 30s. -1 hour, It is recommended to use a time interval of 0.05 seconds to 0.1 seconds; when the projection acquisition frequency is 60 seconds. -1 hour, A value of 0.05 seconds is recommended. The time interval should not exceed the time between two adjacent frames to avoid skipping measured data points and causing inaccurate forward modeling predictions. The numerical substep size within the CFD forward modeling process. When using explicit time-progression schemes, the aforementioned Courant stability conditions should be met; when using first-order implicit Euler schemes or second-order Crank-Nicolson schemes, although theoretically unconditionally stable, it is still recommended to... Limitation to making Within a certain range, to balance computational accuracy and convergence speed, the time resolution of free surface tracking is ensured to be no less than the time resolution of appearance morphology data acquisition.

[0147] Understandably, step S4061 extracts the free surface geometric model and dynamic boundary conditions at the current moment by using a preset observation time step as the period, providing a time-series boundary input for the subsequent forward modeling prediction driven by the rheological parameter set, ensuring that the forward modeling calculation and the measured data closely follow each other on the time axis.

[0148] S4062: Obtain the rheological parameter set corresponding to the current time step; wherein, in the first time step, the rheological parameter set is the preset initial rheological parameter set; in subsequent time steps, the rheological parameter set is the rheological parameter set obtained after optimization in the previous time step.

[0149] In this embodiment of the application, the rheological parameter set refers to the set of physical parameters used to describe the rheological properties of steel fiber reinforced concrete, including yield stress. First plastic viscosity Second viscosity coefficient Flow Index Contribution coefficient of fiber network The initial rheological parameter set refers to the initial estimated values ​​of rheological parameters used at the first time step. This initial rheological parameter set is determined based on the mix design experience of the steel fiber reinforced concrete to be tested. In one embodiment, the initial rheological parameter set can also be directly estimated by a pre-established empirical identification model based on the measured macroscopic morphological characteristics during the initial stage of overall deformation. This empirical identification model uses the maximum wave height as the basis for estimation. With fluctuation time The independent variable is an exponential function:

[0150] .

[0151] .

[0152] .

[0153] in, The maximum wave height on the free surface during the overall deformation process is expressed in mm. The fluctuation time required for the maximum wave height to reach its final stable value from the initial value, expressed in seconds; , , , , , , , , These are empirical coefficients, calibrated using least-squares regression by substituting calibration materials or historical experimental data. The reason for using an exponential function is that it possesses monotonicity and asymptotic saturation characteristics, consistent with the physical behavior of rheological parameter estimation tending to a constant value when morphological characteristics stabilize. Furthermore, it has fewer parameters and is easier to regress on small samples. The specific values ​​of the empirical coefficients are related to the unit system, apparatus size, and material system used in calibration; in practical applications, recalibration should be performed for the current apparatus and material system. When the estimated value of the empirical identification model exceeds the following initial value range or is negative, it is truncated to the boundary of that initial value range. The initial value range is 50 Pa to 500 Pa. The initial value range is 1 Pa·s - 10 Pa·s. The initial value range is 0.1 Pa·s n -1 Pa·s n , The initial value range is 0.5-2.0. The initial value range is 0.01-0.5; the rheological parameter set obtained after optimization of the previous time step refers to the rheological parameter values ​​updated after forward modeling prediction and morphological error correction of the previous time step.

[0154] Specific implementation process: Obtain the rheological parameter set corresponding to the current time step. In the first time step At that time, rheological parameter set Using the aforementioned preset initial rheological parameter set; in subsequent time steps At that time, rheological parameter set The rheological parameter set obtained after optimization at the previous time step superscript This represents the optimized value. The optimization process uses gradient descent or the Levenberg-Marquardt algorithm to minimize the morphological error. The rheological parameter set is iteratively adjusted, and the adjustment step size of each rheological parameter is set independently according to its dimensions and order of magnitude: the yield stress and viscosity coefficient use a relative step size of 1%-10% of the current value or an absolute step size, such as 1 Pa and 0.1 Pa·s; the flow index and fiber network contribution coefficient use a relative step size of 1%-5% of the current value. Alternatively, an adaptive optimizer such as Adam can be used to automatically adjust the learning rate until the morphological error converges or the maximum number of iterations is reached.

[0155] It is worth noting that the accuracy of the initial rheological parameter set selection directly affects the convergence speed of forward modeling. If the initial value deviates from the true value by more than an order of magnitude, it is recommended to use a physical information neural network model for pre-estimation to obtain an initial rheological parameter set that is closer to the true value. During the optimization of the rheological parameter set, the upper and lower bounds of each parameter should be constrained to prevent physically unreasonable negative values ​​or extremely large values.

[0156] Understandably, step S4062 obtains the rheological parameter set corresponding to the current time step, providing constitutive parameter inputs describing the rheological properties of the material for subsequent forward modeling of the fluid dynamics equations, and realizing the dynamic update of the rheological parameters from initial estimation to time-step optimization.

[0157] S4063: Based on the dynamic boundary conditions and rheological parameter set at the current moment, the free surface morphology at the next moment is predicted by forward modeling the preset fluid dynamics equations, and this morphology is defined as the predicted free surface morphology at the next moment.

[0158] In this embodiment of the application, forward modeling of fluid dynamics equations refers to the process of calculating the theoretical free surface morphology at the next moment by solving the fluid dynamics equations based on the dynamic boundary conditions and rheological parameter set at the current moment; the predicted free surface morphology refers to the geometric shape of the theoretical free surface at the next moment obtained by forward modeling of fluid dynamics equations.

[0159] Forward modeling uses the following set of fluid dynamic equations:

[0160] ;

[0161] ;

[0162] in: This is a velocity vector, with units of m / s; Time, in seconds; This is the equivalent density of steel fiber reinforced concrete, in kg / m³. 3 The equivalent density is determined by dividing the sample mass by the measured volume of the charge. For conventional steel fiber reinforced concrete, its typical value range is 2200 kg / m³. 3 -2500kg / m 3 ; Pressure, unit is Pa; This is the deviatoric stress tensor, with units of Pa. This is the gravitational acceleration vector, with units of m / s². 2 .

[0163] The modified Bingham constitutive equation is:

[0164] .

[0165] in, Shear stress in the Modified Bingham model, in Pa; The yield stress is expressed in Pa. The first plastic viscosity is expressed in Pa·s. This is the second viscosity coefficient, measured in Pa·s. n ; Shear rate, in seconds. -1 ; The flow index is dimensionless and is typically taken as 0.8-1.2 for steel fiber reinforced concrete.

[0166] The effective shear stress after considering the contribution of the fiber network is: ;

[0167] in, To account for the effective shear stress after considering the influence of the fiber network, The contribution coefficient of the fiber network. For fiber network connectivity. The deviatoric stress tensor is expressed in generalized Newtonian fluid form as: ,in For the strain rate tensor, For effective viscosity, The equivalent scalar of the strain rate tensor is the Mises equivalent shear rate, where For strain rate tensor The component; under the generalized Newtonian fluid approximation, take This extends the Modified Bingham constitutive equation to three-dimensional flow. To prevent small values ​​from being regularized by division by zero in the low-shear region, take... The magnitude of this regularization is... It automatically degenerates into the true constitutive model, introducing only a negligible viscous approximation in the quasi-rigid region, without affecting the overall accuracy of rheological parameter inversion.

[0168] Specific implementation process: Based on the dynamic boundary conditions extracted in step S4061 at the current moment. and the rheological parameter set obtained in step S4062 The next moment is predicted through forward modeling of the preset fluid dynamics equations. The free surface morphology is determined, and the forward modeling prediction uses the finite volume method or finite element method to solve the above fluid dynamics equations; the computational domain is defined by the dynamic boundary conditions at the current moment. For fixed boundaries, using rheological parameter sets In , , , , Using constitutive parameters, the velocity and pressure fields within the fluid domain are solved. Then, the free surface position at the next moment is calculated using a free surface tracking algorithm. In one implementation, the free surface tracking employs the fluid volume fraction (VOF) method, determining the free surface by solving the transport equations for the fluid volume fraction within each computational grid and reconstructing the volume fraction isosurfaces. In another implementation, the level set method is used, tracking the free surface through the zero isosurface of the signed distance function; thus, the theoretically predicted free surface morphology is obtained. .

[0169] It is worth noting that the computational grid resolution for forward modeling should be set to 0.5mm–2mm to balance computational accuracy and efficiency; the time-progression scheme should use a first-order implicit Euler scheme or a second-order Crank-Nicolson scheme to ensure numerical stability; if numerical oscillations or divergences occur in the forward modeling results, the resolution should be reduced. Or increase the grid density.

[0170] Understandably, step S4063 uses forward modeling of fluid dynamics equations to predict the free surface morphology at the next moment, transforming the current boundary conditions and rheological parameters into the theoretical future morphology state, thus providing a predictive benchmark for subsequent comparison of the error with the measured morphology.

[0171] S4064: Obtain the free surface geometric model corresponding to the next moment and define the geometric model as the measured free surface morphology at the next moment.

[0172] In this embodiment, the measured free surface morphology refers to the free surface geometric model actually measured at the next time step, obtained from the sequence of free surface geometric models stored independently by time index; the next time step refers to the current time step. Plus the preset observation time step The subsequent time point, i.e. .

[0173] Specific implementation process: From the sequence of free surface geometric models stored independently by time index, find the one with a timestamp equal to or closest to... Record the time interval, extract the free surface geometric model from the record, and define the free surface geometric model as the measured free surface morphology at the next time interval. If an exact match exists For records at specific times, the data is extracted directly; if no exact matching record exists, linear interpolation or nearest neighbor interpolation is used to reconstruct the measured free surface topography at that time from records at adjacent times.

[0174] It is worth noting that the extraction accuracy of the measured morphology of a free surface depends on the temporal resolution of the spatiotemporally correlated fusion dataset, which is determined by the preset observation time step. Decision; when the frame rate of the non-contact scanning device is higher than the frame rate of the CT tomographic image sequence, if If the interval is set to 0.05 seconds, the point cloud data at some points in time need to be obtained through interpolation, and the interpolation error should be controlled within 0.1 mm. It is understandable that step S4064, by obtaining the measured free surface morphology at the next time step, provides a real observation benchmark for comparing the predicted morphology with the actual measured morphology, making the optimization and correction of rheological parameters based on evidence.

[0175] S4065: Calculate the morphology error between the predicted morphology of the free surface and the measured morphology of the free surface.

[0176] In the embodiments of this application, the topography error refers to the geometric deviation between the predicted topography of the free surface and the measured topography of the free surface.

[0177] for The root mean square morphology error for each matched sampling point is:

[0178] ;

[0179] The normalized morphology error is:

[0180] ;

[0181] in, To predict the morphology Vertical displacement of each sampling point This represents the vertical displacement of the sampling points corresponding to the measured topography. For reference length, the characteristic dimensions of the loading container, the maximum collapse depth, or the characteristic dimensions of the test platform can be used.

[0182] For morphologies with significant local mutations, let... for Predicted morphology of free surface at any time The point set, for Measured morphology of free surface at any time If the set of points is given, then the Hausdorff distance is:

[0183] .

[0184] The normalized Hausdorff error is .

[0185] The specific implementation process is as follows: The shape error between the predicted free surface shape obtained in step S4063 and the measured free surface shape obtained in step S4064 is calculated. The shape error is calculated using the root mean square error. The matching of corresponding sampling points adopts the nearest neighbor matching algorithm. The nearest neighbor matching algorithm refers to the search algorithm that searches for the matching point with the smallest Euclidean distance to the target point in the point set. In this application, it specifically refers to the point matching algorithm used to establish the correspondence between the sampling points of the predicted shape and the measured shape. It is used to realize the point-by-point deviation calculation between the two shapes. Its input is the predicted shape point set and the measured shape point set, and its output is the matching point index pair. The search radius is set to 2mm-5mm.

[0186] It should be noted that the Hausdorff distance is chosen as an auxiliary measure of topography error in this application because: when there are extreme topography features such as local collapse pits or protrusions on the free surface of steel fiber reinforced concrete, the root mean square error may underestimate local deviations due to the averaging effect. The Hausdorff distance, however, is sensitive to the maximum deviation, ensuring that local topography prediction failures are not overlooked during the inversion optimization process. It is applicable in scenarios where there are significant local abrupt changes or voids in the topography, and where the maximum prediction deviation needs to be strictly monitored. When the overall topography is smooth and local abrupt changes are not significant, only the root mean square error can be used as a measure of topography error. For topography features with significant horizontal diffusion, such as those observed in side collapse tests, the horizontal diffusion radius error is supplemented as an auxiliary measure.

[0187] It is worth noting that the calculation of topographic error should be carried out in the same coordinate system. The origin and coordinate axis direction of the predicted topographic and the measured topographic must be consistent. Otherwise, coordinate registration is required first. If the topographic error is greater than the preset topographic error threshold, it indicates that the current rheological parameter set is seriously deviating from the true value. The rheological parameter set should be re-initialized or the optimization step size should be reduced.

[0188] Understandably, step S4065 quantifies the degree of deviation between forward modeling prediction and actual observation by calculating the topographic error, providing a numerical objective function for the subsequent optimization and correction of the rheological parameter set.

[0189] S4066: The dynamic boundary conditions are forcibly updated based on the measured free surface topography at the next moment, and the updated dynamic boundary conditions at the next moment are used as the calculation input at the next moment after that.

[0190] In this embodiment, the updated dynamic boundary condition refers to the boundary condition used as the calculation input for the next time step, which is forcibly corrected by the measured topography of the free surface; forced update means that in the iterative calculation, the boundary extracted from the measured topography is directly used as the boundary condition for the next time step instead of the boundary result predicted by forward modeling, in order to suppress error accumulation; the next time step after that refers to the current time step. Adding the time point after doubling the preset observation time step, i.e. .

[0191] Specific implementation process: The dynamic boundary conditions are forcibly updated using the measured free surface topography obtained in step S4064. Boundary contours are extracted from the measured free surface topography, using the same extraction method as in step S404, with boundary extraction threshold judgment; the updated dynamic boundary conditions are then obtained. Update the dynamic boundary conditions As the next moment The computational input for the forward modeling prediction of the fluid dynamics equations; through this forced update mechanism, the forward modeling prediction is always based on the latest measured boundary, rather than on the boundary of the previous forward modeling prediction, thereby avoiding the accumulation and amplification of boundary errors in the time progression process.

[0192] It is worth noting that the forced update mechanism requires the spatiotemporal correlation fusion dataset to contain complete measured topography at each time step. If the measured data at a certain time step is missing, the interpolated topography is used for updating, and the data confidence is marked as low at that time step. The updated dynamic boundary conditions should be smoothed to remove boundary spikes caused by measurement noise. The smoothing filter window is set to 3-5 nodes.

[0193] Understandably, step S4066 achieves dynamic adaptive correction of the boundary by forcibly updating the dynamic boundary conditions with the measured morphology, so that the forward modeling calculation is always anchored to the real observation data, effectively suppressing the accumulation and divergence of rheological parameter inversion errors on the time axis.

[0194] Based on the updated dynamic boundary conditions obtained in step S4066 and the rheological parameter set optimized at this time step Among them, those with superscript The rheological parameter set represents the parameter set optimized by morphological error. Step S411 is executed. Steps S4061 to S4066 constitute a coarse-grained rheological parameter pre-estimation module based on forward modeling of fluid dynamics. Its function is to rapidly converge to the feasible region of the rheological parameters within a wide initial value range using the strict constraints of the physical conservation equations, and to use the optimized rheological parameter set as the initial rheological parameter estimate values ​​for the physical information neural network model; simultaneously, the dynamic boundary conditions, forcibly corrected by measured morphology, are used as the macroscopic geometric input of PINN. Steps S411 to S416 constitute a fine-grained rheological parameter inversion module based on physical information neural network. Its function is to refine the rheological parameters near the initial values ​​provided by the CFD pre-estimation by fusing the microscopic features of the fiber network with the nonlinear fitting capability of the neural network, and output the final inversion result. The two modules are in a cascade structure: the CFD pre-estimation solves the problem of PINN's sensitivity to initial values, and the PINN refinement compensates for the CFD model's insufficient representation of the microscopic structure of the fiber network.

[0195] S411: Obtain the fiber network connectivity and macroscopic morphology characteristic variables corresponding to the next time step, and obtain the updated dynamic boundary conditions and morphology error for the next time step.

[0196] In this embodiment, the updated dynamic boundary condition refers to the boundary condition used for calculation input at the current moment after being forcibly corrected by step S4066; the morphology error refers to the geometric deviation between the predicted morphology of the free surface and the measured morphology of the free surface, which is calculated by step S4065; the macromorphological feature variables refer to the set of parameters extracted from the measured morphology of the free surface for quantifying the evolution state of the appearance morphology, including at least one of collapse depth, edge curvature, diffusion area, maximum wave height and fluctuation time.

[0197] Specific implementation process: Obtain the fiber network connectivity from step S405. Obtain the updated dynamic boundary conditions from step S4066. Obtain the normalized morphology error from step S4065. Macroscopic morphological feature variables are extracted from the measured free surface morphology obtained in step S4064; data cleaning and normalization are performed on each variable to remove outliers and map each variable to the 0-1 interval.

[0198] The normalization formula is:

[0199] ;

[0200] in, The normalized value is dimensionless. The original value; The minimum value in the training dataset; The maximum value in the training dataset; To prevent small amounts from being divided by zero.

[0201] It is worth noting that the outlier threshold for data cleaning is set to 3 times the standard deviation, meaning that variable values ​​with an absolute value greater than 3 times the standard deviation are considered outliers and removed; normalization is determined based on the training dataset during the training phase. and The same set of extreme values ​​is used in the reasoning stage, but for values ​​exceeding... The current sample feature value is truncated to the [0,1] interval, and the feature is marked as extrapolated to indicate that the reliability of the inversion result may be reduced, so as to ensure the consistency of data distribution.

[0202] Understandably, step S411 provides complete and standardized input data for the physical information neural network model by acquiring and preprocessing fiber network connectivity, dynamic boundary conditions, morphological errors, and macroscopic morphological feature variables, thus ensuring the quality and consistency of the model input.

[0203] S412: Use the updated dynamic boundary conditions and macroscopic morphological feature variables as inputs to the preset physical information neural network model.

[0204] In this application embodiment, the physical information neural network model refers to a deep learning architecture that embeds physical laws into the neural network loss function in the form of soft constraints. In this application, the model specifically refers to a physical information neural network model used to fuse macroscopic morphological features and microscopic fiber network features and invert rheological property parameters.

[0205] The physical information neural network model includes a boundary encoding branch, a macroscopic morphology encoding branch, a fiber network constraint branch, a physical constraint fusion layer, a rheological parameter output layer, and a stress decoding subnetwork. The boundary encoding branch receives updated dynamic boundary conditions and extracts the geometrical variation features of the free surface. The macroscopic morphology encoding branch receives macroscopic morphology feature vectors and extracts the overall material flow response features. The fiber network constraint branch receives fiber network connectivity and its time series, providing internal structural variables for dynamic constitutive constraints. The physical constraint fusion layer fuses the features from each branch and calculates the residuals of the rheological physical equations. The rheological parameter output layer outputs a set of rheological parameters. The stress decoding subnetwork takes the fused features from the physical constraint fusion layer as input, passes through two fully connected layers with 32 and 16 neurons respectively, and outputs the equivalent shear stress predicted by PINN using the ReLU activation function. In one implementation, the boundary encoding branch employs a multilayer perceptron or a point cloud encoding network. If the number of boundary nodes is fixed, the boundary node coordinates can be arranged into a fixed-length vector and then input into the multilayer perceptron; if the number of boundary nodes is not fixed, resampling can be performed first, or a point cloud feature extraction structure can be used. The boundary encoding branch may include three fully connected layers with 128, 128, and 64 neurons respectively, and the activation function is ReLU.

[0206] The macroscopic shape encoding branch employs a multilayer perceptron, with the input being a normalized macroscopic shape feature vector. This branch can include three fully connected layers with 64, 64, and 32 neurons respectively, using ReLU as the activation function. The output of this branch is the macroscopic shape encoding vector.

[0207] The fiber network constraint branch is used to process the fiber network connectivity sequence. If the fiber network connectivity is input as a scalar time series, a one-dimensional temporal convolutional network, a Long Short-Term Memory (LSTM) network, or a temporal multilayer perceptron can be used; if the fiber network connectivity is input as a spatial field, a ConvLSTM can be used. This branch outputs the fiber network state encoding vector, and simultaneously stores the current time-stamped state encoding vector. It is directly provided to dynamic constitutive constraints.

[0208] The physical constraint fusion layer concatenates the boundary encoding vector, macroscopic topography encoding vector, and fiber network state encoding vector, and forms a fused feature through one or more fully connected layers.

[0209] Rheological parameter output layer output: .

[0210] Among them, those with superscript The variable represents the PINN output value. To ensure that the output parameters satisfy physical nonnegativity, the yield stress and viscosity coefficient can be output using Softplus:

[0211] ,in, The rheological parameters are a linear weighted sum of the neurons corresponding to the output layer, and are dimensionless.

[0212] The flow index and fiber network contribution coefficient can be obtained using a scaled sigmoid output: ; .in, , These are the lower and upper limits of the liquidity index, respectively, and in one implementation, they are taken as 0.5 and 2.0. The upper limit of the fiber network contribution coefficient is set to 1.0 in one implementation. , These are the linear weighted sums of the corresponding neurons in the output layer. For the Sigmoid function: .

[0213] Specific implementation process: The updated dynamic boundary conditions obtained in step S411 are... The macroscopic morphological feature variables are used as inputs to the physical information neural network model. These inputs are normalized and then fed into the boundary encoding branch for macroscopic feature encoding. In one implementation, macroscopic operating parameters, the preset speed of the vertical or horizontal piston, and the initial volume of the steel fiber reinforced concrete are also used as auxiliary inputs, concatenated with the macroscopic morphological feature variables and input together, so that the physical information neural network model can explicitly perceive the driving conditions. The boundary encoding branch contains several fully connected hidden layers, and the activation function is the ReLU function, where ReLU refers to the linear rectified activation function, and its mathematical expression is: In this application, the activation function used to introduce nonlinear mapping is specifically used to enhance the feature representation ability of the network. Its input is a linear weighted sum of neurons, and its output is a nonlinear activation value. The dropout ratio is set to 0.2, where dropout refers to a regularization technique that randomly discards some neuron connections during neural network training. In this application, it specifically refers to a random deactivation mechanism used to prevent overfitting and enhance the model's generalization ability. Its input is the hidden layer output, and its output is the activation value after random masking. The output of the boundary encoding is a feature vector, which contains a high-order abstract representation of dynamic boundary conditions and macroscopic morphological features. The weight initialization of the fully connected layer adopts the He initialization method, where the He initialization method refers to the weight initialization method designed for the ReLU activation function. Its mathematical expression is that the weights are initialized from a mean of 0 and a variance of 2 / Sampling from a normal distribution, The input number of neurons is dimensionless. In this application, it specifically refers to the weight initialization strategy used to adapt to the ReLU activation function to avoid gradient vanishing or exploding. Its input is the layer input dimension and its output is the initial weight matrix.

[0214] It is worth noting that the input dimension of the boundary coding branch depends on the number of dynamic boundary condition nodes and the number of macroscopic morphological feature variables. When the number of dynamic boundary condition nodes is greater than 1000, principal component analysis should be performed first to reduce the dimension. Principal component analysis refers to a statistical method that converts high-dimensional variables into linearly independent low-dimensional principal components through orthogonal transformation. In this application, it specifically refers to a dimension reduction algorithm used to reduce the input dimension of dynamic boundary conditions. It is used to compress the input size and retain the main feature information. Its input is the coordinate vector of high-dimensional boundary nodes and its output is the low-dimensional principal component feature vector. The principal components that retain 95% of the variance information are used as input.

[0215] Understandably, step S412, by sending the updated dynamic boundary conditions and macroscopic morphological feature variables into the boundary encoding branch, realizes the high-order feature extraction and encoding of macroscopic appearance information, providing a macroscopic dimension feature representation for subsequent fusion with microscopic features.

[0216] S413: Use fiber network connectivity as the independent variable of dynamic constitutive constraints in the pre-defined physical information neural network model.

[0217] In the embodiments of this application, dynamic constitutive constraints refer to constraints that use fiber network connectivity as an independent variable to reflect the enhancing effect of fiber network on matrix rheological properties; independent variables refer to input variables that change independently in the mathematical model and affect the dependent variable.

[0218] Specific implementation process: The fiber network connectivity obtained in step S411 is... As the independent variable of the dynamic constitutive constraint in the physical information neural network model, this independent variable, after feature expansion, is fed into the fiber network constraint branch for micro-spatiotemporal feature encoding. Feature expansion refers to expanding the scalar form of fiber network connectivity into a vector form that matches the spatiotemporal dimension of the CT tomographic image sequence. The dimension of the expanded feature vector is the number of time steps multiplied by the number of spatial locations multiplied by 1. The fiber network constraint branch contains several temporal coding layers, and the activation function is the tanh function, where tanh refers to the hyperbolic tangent activation function, and its mathematical expression is:

[0219] .

[0220] in, This represents a gated linear weighted sum, i.e., the input variable is dimensionless; represents a natural constant, dimensionless. In this application, it specifically refers to the activation function used for state updates in time-series coding networks, controlling the range of information flow. Its input is a linear weighted sum of neurons, and its output is a hyperbolic tangent mapping value. The gated activation function is the sigmoid function, where the sigmoid function refers to the sigmoid activation function, and its mathematical expression is: In this application, "activation function" specifically refers to the activation function used to control the opening and closing of the gating of the temporal coding network, which is used to implement the gating logic of the forget gate, input gate and output gate. Its input is the gating linear weighted sum and its output is the gating weight between 0 and 1. The output of the micro-spatiotemporal feature encoding is a spatiotemporal feature vector, which contains dynamic information about the evolution of the fiber network connectivity over time.

[0221] It is worth noting that the input to the fiber network constraint branch should be a CT tomographic image sequence or a fiber network connectivity sequence after feature expansion. If the input is a single scalar value, it needs to be expanded into a sequence form by copying it in the time dimension first. The batch size of the temporal coding layer is set to 4-16 to balance memory usage and gradient stability.

[0222] Understandably, step S413, by feeding the fiber network connectivity as the independent variable of the dynamic constitutive constraint into the fiber network constraint branch, realizes the spatiotemporal evolution encoding of micro-fiber network features, providing a dynamic feature representation of the micro-dimensional dimension for subsequent fusion with macro-features.

[0223] S414: Use shape error as an additional weighting term in the loss function of the preset physical information neural network model.

[0224] In this embodiment, the additional weighting term refers to the shape error weighting term introduced in the loss function of the physical information neural network model in addition to the data-driven error term and the physical constraint term; the loss function refers to the scalar function used to measure the difference between the neural network prediction value and the true value, and guides the optimization direction of the network weights; the shape error refers to the geometric deviation between the predicted shape of the free surface and the measured shape of the free surface, which is calculated by step S4065.

[0225] PINN training data includes the following four categories:

[0226] 1. Synthetic samples generated from numerical simulations of fluid dynamics;

[0227] 2. Measured samples formed from historical tests of steel fiber reinforced concrete with known mix proportions;

[0228] 3. A control sample is formed from fiber-free reference mortar and alternative calibration materials, such as high-concentration corn starch paste or thick starch paste. Since the alternative calibration materials do not contain steel fibers, their rheological parameters can be directly measured using a traditional rotational rheometer, and used as the data-driven mean square error term. The source of the tags;

[0229] 4. Online observation samples generated in the current test batch.

[0230] The training process includes the following steps: data cleaning and normalization, training set partitioning, forward computation, total loss calculation, backpropagation, Adam parameter update, early stopping on the validation set, evaluation of the test set, and fine-tuning of the current batch. The training, validation, and test sets can be partitioned in a ratio of 7:1.5:1.5 or 8:1:1.

[0231] The loss function of the physical information neural network model is defined separately for the offline training stage and the online inversion stage:

[0232] During the offline training phase, historically labeled samples and synthetic samples are used, and the total loss function is:

[0233] .

[0234] in Calculated from historical calibration samples with known true values ​​of rheological parameters, it is used to guide the network to learn the mapping relationship between macroscopic morphology and rheological parameters; and This is a dimensionless weighting coefficient. The selection should make Typical values ​​in the early stages of training and Being on the same order of magnitude, such as to The loss can be automatically calibrated through loss statistics during the pre-training phase; during the online inversion phase, and It is fixed as the optimal value after pre-training convergence.

[0235] When testing currently unknown samples during the online inversion phase, historical labels are unavailable. The total loss function degenerates into: ;in, The normalized morphology error from step S4065 constitute, It is composed of Modified Bingham constitutive residuals. Under the dual constraints of the measured morphology and physical equations of the current sample, the network iteratively optimizes the output rheological parameters starting from the initial values ​​provided by CFD pre-estimation.

[0236] It is worth noting that the shape error weighting coefficient The value of should be adjusted according to the dimensions and order of magnitude of the morphological error. When the typical value of the morphological error is 0.01–0.1, It is recommended to use 0.5-1.0; when the morphology error When less than 0.01, It is recommended to use a value of 0.1-0.3 to avoid the additional weighting term dominating the loss function.

[0237] It is understandable that step S414 incorporates the macroscopic appearance prediction accuracy directly into the network optimization objective by using the shape error as an additional weighting term in the loss function, so that the physical information neural network model can optimize rheological parameters while taking into account the shape prediction accuracy.

[0238] S415: Based on the input, independent variables, and additional weighting terms, run the preset physical information neural network model, and obtain the numerical results of the rheological property parameters of steel fiber reinforced concrete by inversion calculation through the constraints of the preset rheological physical equations embedded in the preset physical information neural network model.

[0239] In this embodiment, the pre-defined rheological physical equation constraint refers to the constitutive equation describing the relationship between shear stress and shear rate, established based on the Modified Bingham model, the Herschel-Bulkley model, or a conventional power function model. This constitutive equation is embedded as a soft constraint in the loss function of the physical information neural network model. The inversion calculation refers to the process of solving for unknown rheological parameters by means of optimization algorithms starting from observation data. The numerical results of the rheological property parameters refer to the specific values ​​of yield stress, plastic viscosity, and fiber network contribution coefficient obtained by model calculation.

[0240] Modified Bingham constitutive relations are embedded within the physical information neural network model:

[0241] ;

[0242] And embeds dynamic constitutive constraints of fiber networks: ;Will and the strain rate scalar obtained from CFD forward modeling Substituting into the Modified Bingham constitutive equation, the theoretical shear stress is calculated. Let the shear stress predicted by PINN be... The normalized physical residual is defined by the output of the stress decoding subnetwork:

[0243] ;

[0244] in, The stress reference value can be the maximum shear stress, typical shear stress, or preset unit stress from the training data; Take 10% of the standard deviation or maximum shear stress in the training dataset to ensure normalized residuals. Typical values ​​are distributed in to The interval should be chosen to avoid both excessively weak physical constraints and their over-dominance of the total loss.

[0245] Physical loss can be defined as:

[0246] ;

[0247] in, This represents the number of physical residual sampling points.

[0248] For dynamic constitutive constraints, the fiber network reinforcement residual is defined as follows:

[0249] ;

[0250] in The equivalent shear stress is directly predicted by the network. The theoretical stress is the result of enhanced fiber network connectivity; the difference between the two reflects the network's learning bias towards the fiber reinforcement effect, and this bias is incorporated into... .

[0251] The specific implementation process is as follows: Based on the input of step S412, the independent variables of step S413, and the additional weighting terms of step S414, the physical information neural network model is run, and inversion calculation is performed through the constraints of the preset rheological physical equations embedded in the physical information neural network model. This inversion calculation uses the Adam optimizer for iterative optimization. The Adam optimizer refers to an adaptive moment estimation optimization algorithm, specifically a gradient descent optimization algorithm used in this application to jointly optimize network weights and rheological parameters. It is used to achieve parameter updates based on adaptive estimation of first and second moments, with the loss function gradient as input and the parameter update amount as output. The first moment decay rate of the Adam optimizer... Set to 0.9, second moment decay rate Set to 0.999, numerical stability constant Set as The learning rate is set to 0.001-0.01, dimensionless; the number of iterations is set to 1000-5000, dimensionless; a validation set loss evaluation is performed every 100 iterations. The validation set is randomly divided from the training dataset in an 8:2 ratio, with at least 200 validation set samples; if the validation set loss does not decrease for 50 consecutive iterations, an early stopping mechanism is triggered, and the current optimal model weights are saved; the training batch size is set to 16-64, dimensionless; the training dataset contains morphological features and fiber network connectivity combinations under different rheological parameter groups, with at least 1000 training samples, dimensionless, and covering a wide range of conditions. From 50Pa to 500Pa From 1 Pa·s to 10 Pa·s From 0.1 Pa·s n up to 1 Pa·s n , The parameter space ranges from 0.01 to 0.5. After pre-training, fine-tuning is performed using actual test data, with the fine-tuning learning rate set to 1% to 10% of the pre-training learning rate.

[0252] It is worth noting that when the physical information neural network model is first applied to steel fiber reinforced concrete with a new mix proportion, it needs to be pre-trained using alternative calibration materials or training datasets generated by numerical simulation. The alternative calibration materials refer to reference mortar materials, high-concentration corn starch paste, or thick starch paste that have similar rheological properties to the steel fiber reinforced concrete under test but do not contain steel fibers. In this application, they specifically refer to alternative materials used to provide calibration data without fiber interference, which are used to establish the basic rheological parameter-morphological mapping relationship. Their input is the mix proportion design parameters and their output is the calibration rheological parameters. The training dataset generated by numerical simulation refers to the synthetic dataset generated through computational fluid dynamics simulation. In this application, it specifically refers to the simulation data used to expand the training samples and cover the parameter space that is difficult to exhaust through experiments. Its input is the rheological parameter set and boundary conditions and its output is the simulated morphological data.

[0253] Understandably, step S415 performs inversion calculations by running a physical information neural network model with embedded physical equation constraints, which deeply integrates macroscopic morphological features, microscopic fiber network features and underlying rheological physical laws, solving the problem that traditional rheometers cannot directly calibrate steel fiber reinforced concrete due to shear damage to fiber distribution, and realizing intelligent rheological parameter inversion based on physical guidance.

[0254] S416: Obtain rheological property parameters from the output of the preset physical information neural network model. The rheological property parameters include at least one of yield stress, plastic viscosity, and fiber network contribution coefficient.

[0255] In this embodiment, rheological property parameters refer to constitutive parameters that can describe the rheological behavior of steel fiber reinforced concrete, including yield stress. First plastic viscosity Second viscosity coefficient Flow Index and fiber network contribution coefficient At least one of them; the output refers to the last fully connected layer in the physical information neural network model.

[0256] Specific implementation process: Obtain rheological property parameters from the output of the physical information neural network model. These rheological property parameters include yield stress. First plastic viscosity Second viscosity coefficient Flow Index and fiber network contribution coefficient At least one of the following. The output should be denormalized to restore the normalized output to its physical dimension value. The denormalization formula is expressed mathematically as follows:

[0257] .

[0258] in, This represents the restored physical value. Indicates the network output value; This represents the maximum value in the training dataset; This represents the minimum value in the training dataset; To prevent small amounts from being divided by zero.

[0259] The obtained rheological parameters should be physically validated, and the validation conditions are as follows: And less than 1000 Pa, And less than 50 Pa·s, And less than 5 Pa·s n , And less than 5, And less than 1; if any parameter exceeds the verification range, the inversion result should be marked as abnormal, and a prompt should be made to retest or check the model status.

[0260] It is worth noting that the accuracy of the inversion of rheological parameters is affected by the quality of the training data and the state of model solidification. When using the model solidified in step S4074, the inversion error can be controlled within 10%; if the inversion error is greater than 20%, online fine-tuning or complete final-state verification should be performed again.

[0261] It is understandable that step S416 obtains rheological property parameters from the output of the physical information neural network model, transforming the high-dimensional feature representation of the network into rheological constitutive parameters with clear physical meaning and dimensions. The output yield stress, plastic viscosity, and fiber network contribution coefficient can truly characterize the overall rheological constitutive relationship of steel fiber reinforced concrete, providing a reliable rheological basis for mix design and construction performance evaluation in engineering applications.

[0262] In summary, step S4 employs a Poisson surface reconstruction algorithm, a morphology feature extraction algorithm, a fiber identification algorithm based on the Hessian matrix, and a connectivity calculation formula to perform spatiotemporal correlation fusion on the dual-source data to generate a spatiotemporal correlation fusion dataset. Based on a preset observation time step, dynamic boundary updates and morphology prediction corrections are iteratively performed to obtain updated dynamic boundary conditions. Combined with full-field 3D scanning verification after solidification and backpropagation correction of the adjoint equation, the physical information neural network model is corrected. Finally, based on the physical information neural network model constrained by the Modified Bingham constitutive equation, the yield stress, plastic viscosity, and fiber network contribution coefficient are inverted and calculated, realizing intelligent inversion and accuracy improvement of the overall rheological properties parameters of steel fiber reinforced concrete.

[0263] S4067: After the test is completed, the steel fiber reinforced concrete sample shall be left to cure under the preset curing conditions until the preset degree of curing is reached.

[0264] In this embodiment, the curing conditions refer to the standard environmental conditions used to induce hydration and hardening of steel fiber reinforced concrete. These curing conditions are set as a constant temperature and humidity environment with a temperature of 20℃±2℃ and a relative humidity of not less than 95%. The degree of curing refers to the state in which the steel fiber reinforced concrete has reached a predetermined age or strength after hydration and hardening. The predetermined degree of curing is set as a curing age of 28 days or a concrete compressive strength of more than 70% of the design strength.

[0265] The specific implementation process is as follows: After testing, the steel fiber reinforced concrete samples are removed from the loading container and placed in a standard curing room with a temperature of 20℃±2℃ and a relative humidity of not less than 95% for static curing. Curing continues for 28 days or until the concrete compressive strength reaches more than 70% of the design strength, achieving the preset degree of curing. During curing, the samples should be laid flat on the curing rack with a spacing of not less than 5cm to ensure uniform moisture circulation. The temperature and humidity in the curing room should be recorded every 4 hours. If the temperature deviation exceeds ±2℃ or the humidity is lower than 90%, the temperature control and humidification equipment should be adjusted promptly. To reduce additional fiber migration between the end of the test and the start of curing, the samples should be placed in a static curing state as soon as possible after the test. If it is necessary to remove the samples, the sample coordinate marks should be retained, and vibration, flipping, and impact should be avoided. In a preferred embodiment, the samples are statically cured in a loading container or a transparent curing mold to maintain consistency between the final scan coordinates and the coordinates during the testing process.

[0266] It is worth noting that if the curing conditions do not meet the standard temperature and humidity requirements, the internal structure of the cured sample may differ from the dynamic state during the test, leading to distorted verification results. Violent vibrations should be avoided when removing the sample from the loading container to prevent damage to the internal fiber distribution.

[0267] Understandably, step S4067 uses standard curing to allow the steel fiber reinforced concrete to complete hydration and hardening, fixing the dynamic rheological state during the test into a static solid structure, providing a solidified sample for subsequent full-field three-dimensional scanning to obtain the measured final state of fiber distribution.

[0268] S4068: A lateral non-destructive radiographic scanning device is used to perform a full-field three-dimensional scan of the cured sample to obtain the spatial distribution state of the cured fibers, which is taken as the measured final state of the fiber distribution. In this embodiment of the application, full-field three-dimensional scanning refers to using an industrial computed tomography (CT) scanner to perform overall three-dimensional reconstruction of the cured sample and obtain complete three-dimensional volume data; the measured final state refers to the spatial distribution state of the cured steel fibers actually measured by the full-field three-dimensional scanning; the fiber spatial distribution state refers to the set of three-dimensional coordinate positions and direction vectors of each steel fiber inside the cured sample.

[0269] The specific implementation process is as follows: A full-field 3D scan is performed on the sample that has reached the preset curing degree in step S4067. This full-field 3D scan uses a microfocus industrial computed tomography (CT) scanner with a scanning resolution set to 0.05mm–0.1mm / voxel. During scanning, the sample is placed on a rotating stage, the X-ray source and detector are fixed, and the sample rotates 360 degrees, acquiring at least 720 two-dimensional projection images at different projection angles. The 3D volume data is reconstructed using a filtered back-projection algorithm or an iterative reconstruction algorithm to obtain the spatial distribution state of the cured fibers, which serves as the measured final state of the fiber distribution. In another embodiment, the cured sample can also be cut, and the cut section can be photographed or scanned. Image analysis can then be used to extract the fiber distribution on the cut section as a destructive alternative verification method for the full-field 3D scan.

[0270] It is worth noting that the scanning resolution of the full-field 3D scan must be no less than 0.1 mm / voxel to ensure the identifiability of individual fibers; if the fiber diameter is less than 0.3 mm, the scanning resolution should be increased to 0.05 mm / voxel; sample vibration should be avoided during the scanning process, and the rotation speed of the rotary stage should be set to 0.5 to 2 revolutions per minute.

[0271] Understandably, step S4068 obtains the measured final state of fiber distribution through full-field three-dimensional scanning, providing real and reliable benchmark data for subsequent comparison with the model-predicted final state, and realizing cross-state data acquisition from dynamic testing to static verification.

[0272] S4069: Extract the dynamic distribution evolution trajectory of steel fibers during the overall deformation process from the CT tomographic image sequence, and recursively derive the model-predicted final state of fiber distribution based on the dynamic distribution evolution trajectory. In this embodiment, the dynamic distribution evolution trajectory refers to the spatiotemporal path extracted from the industrial computed tomographic image sequence acquired during the measurement process, describing the spatial position of the steel fibers at each moment during the overall deformation process; the model-predicted final state refers to the theoretical final state of fiber distribution obtained based on the dynamic distribution evolution trajectory using the exponential decay model of hydration hardening rate; recursion refers to the process of calculating the final state based on the known dynamic distribution evolution trajectory using the exponential decay model of hydration hardening rate.

[0273] Specific implementation process: From the industrial computed tomography (CT) image sequence acquired during the measurement process, the dynamic distribution evolution trajectory of steel fibers during the overall deformation process is extracted. This extraction process employs a fiber tracking algorithm based on particle filtering or optical flow, recording the spatial coordinates and orientation vector of each fiber in each frame of the industrial CT image. In one implementation, the Lucas-Kanade sparse optical flow method is used: taking feature points on the centerline of each fiber as the tracking object, the optical flow displacement of the feature points is solved using the gray-level gradient information of adjacent frames, and the dynamic distribution evolution trajectory is formed by concatenating frames one by one. (Test end time) Corresponding to the Frame, number Root fibers in Speed ​​of time Calculated using the central difference method:

[0274] ;

[0275] in The time sampling interval of the CT image sequence is defined. If endpoint frames are missing, forward or backward differencing is used as a substitute. Before calculation, the coordinate sequence is smoothed using Savitzky-Golay filtering with a window width of 5-7 frames to suppress velocity oscillations caused by tracking noise.

[0276] Considering that hydration hardening causes the fiber motion velocity to gradually decrease, the velocity model is as follows:

[0277] ;

[0278] in, The end time of the test. This is the hydration hardening attenuation coefficient, in seconds. -1 The velocity model assumes that the fiber's direction of motion remains unchanged after the test, and only the velocity amplitude decays with the hydration and hardening process. Therefore, the displacement can be obtained by integrating the scalar velocity decay relationship along the velocity direction at the end of the test.

[0279] From this velocity model Integrating until the target is solidified ,get:

[0280] ;

[0281] in, For the first Predicted location of root fibers at the target curing time. This is the position at the end of the test. The speed at the end of the test.

[0282] when When it varies with time, it can be expressed as a piecewise constant:

[0283] ;

[0284] The model predicts the final state fiber set as follows:

[0285] ;

[0286] in, To predict the number of fibers in the final state. To predict the location, To predict direction.

[0287] It is worth noting that the bonding effect between the fiber and the matrix should be considered during the recursive process. As the matrix viscosity increases with time, the fiber velocity should decrease accordingly, and the attenuation coefficient should be determined based on the hydration kinetics model. In one embodiment, The value can be determined by performing an exponential least squares fit on the fiber velocity decay curve obtained from tracking the last 5-10 frames in the later stage of the test, with a typical value range of 10. -4 s -1 -10 -2 s -1 If viscosity growth is ignored, the recursive results may overestimate fiber displacement.

[0288] Understandably, step S4069 extends the dynamic fiber tracking data during the test process to the time of curing completion by extracting the dynamic distribution evolution trajectory and predicting the final state based on the exponential decay model of the hydration hardening rate recursive model, thus providing a theoretical prediction benchmark for comparison with the measured final state.

[0289] S4070: Spatial registration of the measured final state and the model predicted final state.

[0290] In the embodiments of this application, spatial registration refers to the rigid transformation process that unifies the model-predicted final state and the measured final state into the same coordinate system.

[0291] Spatial registration can be achieved using a rigid transformation model:

[0292] .

[0293] in, Let be a rotation matrix. It is a translation vector. This is the predicted location after registration.

[0294] The rotation matrix and translation vector can be determined by minimizing the distance between corresponding points between the predicted final state and the measured final state:

[0295] .

[0296] in, To determine the number of corresponding points for registration, in one implementation, this minimization problem is solved using the Kabsch algorithm: first, the centroids of the sets of corresponding points for the measured and predicted final states are calculated and centered; then, singular value decomposition (SVD) is performed on the covariance matrix formed by the centered coordinates; and the optimal rotation matrix is ​​directly calculated from the decomposition results. Finally, the optimal translation vector is determined by the centroid difference. .

[0297] After spatial registration, the two sets of 3D data can be resampled to the same voxel grid for point-by-point comparison. The registration residual can be controlled within the range of 0.1 mm to 0.5 mm, with the specific threshold determined based on the CT resolution. When the sample curing shrinkage exceeds 0.05%, a similarity transformation, i.e., a rigid transformation plus uniform scaling, can be used to compensate for the shrinkage effect. The scaling factor is estimated by the ratio of the centroid radius of the measured final state to the predicted final state.

[0298] S4071: The measured final state after spatial registration is compared point-by-point with the model-predicted final state to construct the fiber distribution error field for the entire field. In this embodiment, the fiber distribution error field refers to the three-dimensional error distribution obtained after point-by-point comparison between the measured final state and the model-predicted final state; point-by-point comparison refers to finding the corresponding fiber with the closest spatial position in the model-predicted final state for each fiber in the measured final state and calculating the spatial coordinate deviation between the two; the entire field refers to the comparison results of all fibers covering the entire three-dimensional spatial domain of the sample.

[0299] The specific implementation process is as follows: the measured final state after spatial registration is compared point by point with the model predicted final state to construct the fiber distribution error field for the entire field.

[0300] In one implementation, the fiber distribution is converted into a fiber occupancy voxel map. Let... To measure the final state at the voxel position The fiber occupancy value or fiber density value at that location. To predict the corresponding value of the final state, the voxel error is: .

[0301] In another implementation, the distance from any voxel location to the nearest fiber centerline is defined as follows:

[0302] .

[0303] .

[0304] The fiber distribution error field can then be defined as: The aforementioned error field can reflect the local deviation of the predicted fiber position relative to the measured fiber position.

[0305] To serve as the terminal condition for the adjoint equation, the error field can be converted into a terminal loss:

[0306] ;

[0307] in, The number of voxels.

[0308] It is worth noting that point-by-point comparison requires that the coordinate systems of the measured final state and the model-predicted final state be consistent. After spatial registration, the coordinate systems of the two are unified. For broken or agglomerated fibers, nearest neighbor matching may fail, and in this case, they should be marked as outliers and analyzed separately. Understandably, step S4071 constructs a fiber distribution error field through point-by-point comparison, quantifying the difference between the measured final state and the model-predicted final state into a spatial distribution map, providing the spatial distribution input of the error for subsequent backpropagation of the adjoint equation.

[0309] S4072: Based on the fiber distribution error field, the adjoint equation backpropagation algorithm is used to trace the inversion error contribution of each time step along the time axis and generate the error propagation weight sequence.

[0310] In this embodiment, the adjoint equation backpropagation algorithm refers to a sensitivity analysis method used in inverse problems of fluid dynamics to trace the source of error backward along the time axis. This algorithm is a commonly used algorithm in the field of solving inverse problems of fluid mechanics. The error propagation weight sequence refers to the sequence of contributions of each time step to the final inversion error. Each element in the sequence represents the sensitivity weight of the inversion error of the corresponding time step to the final fiber distribution error. The contribution of the inversion error refers to the influence of the rheological parameter inversion error or boundary condition update error of a certain time step on the final fiber distribution state.

[0311] Specific implementation process: Based on the fiber distribution error field constructed in step S4071, the adjoint equation backpropagation algorithm is used to trace the inversion error contribution of each time step back along the time axis, generating an error propagation weight sequence. Let the... The state variables at each time step are The state recursion relationship is as follows:

[0312] .in, This represents the forward recursive operator.

[0313] Terminal loss is Accompanying variables Recursively calculating in reverse chronological order, there are two cases:

[0314] Terminal conditions, i.e. And corresponding to the final cured state: .

[0315] Intermediate recursion, i.e. : .

[0316] Due to terminal loss Depends only on the final state It does not directly depend on intermediate states. , Therefore, the intermediate recursion does not contain This backward recursive process propagates the terminal fiber distribution error forward step by step over time, quantifying the sensitivity of the state variables at each historical moment to the final state error.

[0317] No. The error contribution at each time step can be represented by the magnitude of the accompanying variable:

[0318] ;

[0319] Normalizing the error contribution yields the error propagation weights:

[0320] ;

[0321] in, To prevent small amounts from being divided by zero, This represents the total number of time steps.

[0322] The error propagation weight sequence is: .

[0323] It is worth noting that the computational cost of the backpropagation algorithm for the adjoint equation is about 2-3 times that of the forward modeling. When the number of time steps exceeds 1000, it is recommended to use parallel computing or dimensionality reduction approximation to save computation time. During the backpropagation process, the forward modeling state variables of each time step should be saved to avoid repeated calculations.

[0324] Understandably, step S4072 generates an error propagation weight sequence through the adjoint equation backpropagation algorithm, which accurately quantifies the contribution ratio of the inversion error at each time step to the final fiber distribution error, providing a weight basis for subsequent targeted correction of the inversion calculation path.

[0325] S4073: Based on the error propagation weight sequence, the inversion calculation loss corresponding to each time step in the physical information neural network model is weighted and corrected, and the network parameters of the physical information neural network model are updated based on the weighted and corrected inversion calculation loss to suppress the accumulation and amplification of errors in the time progression process.

[0326] In the embodiments of this application, the inversion calculation loss refers to the weighted sum of data loss, physical loss and morphological loss generated by the physical information neural network model in predicting rheological parameters at each time step; the weighted correction refers to the process of re-weighting the loss of each time step according to the error propagation weight sequence; error accumulation and amplification refers to the phenomenon that in the iterative calculation process, the error of the earlier time step is transmitted to the later time step through forward prediction, resulting in the later error being greater than the earlier error.

[0327] Specific implementation process: Based on the error propagation weight sequence generated in step S4072, the inversion calculation loss corresponding to each time step in the physical information neural network model is weighted and corrected. Let the physical information neural network model be at the... The original inversion calculation loss at each time step is When there is no truth label during the online correction phase Consistent with the loss definition in the online inversion stage of step S414, the total loss in the pre-training stage is... The accompanying weighted adjusted loss is: .

[0328] When updating the model, the total corrected loss is defined as: .

[0329] in and These are dimensionless weight coefficients. Original pre-training loss. During the online correction phase, they will no longer participate directly to avoid conflict. The time step loss is repeatedly calculated. Based on The Adam optimizer is used to update the PINN network parameters. The update process can be performed in multiple iterations, and the terminal error and validation loss are recalculated after each iteration. If the validation loss decreases, the update is retained; if the validation loss increases, the learning rate is reduced or the model is reverted to the previous state.

[0330] It is worth noting that weighted corrections should be performed in batches. After each batch of corrections, forward modeling and error calculations should be re-executed to verify the correction effect. If the overall error decreases, the correction is accepted; if the error increases, the correction is rolled back and the correction step size is reduced. The initial value of the correction step size is set to 0.01 and can be dynamically adjusted according to the convergence situation.

[0331] Understandably, step S4073, by weighting and correcting the inversion calculation loss according to the error propagation weight sequence, specifically suppresses the error output of time steps with high error contribution, thereby achieving a closed-loop improvement in the inversion accuracy of the physical information neural network model.

[0332] S4074: The updated network parameters are embedded into the physical information neural network model for use in the inversion of rheological property parameters in subsequent test batches.

[0333] In this embodiment, "fixing" refers to saving the updated network weight parameters as a model checkpoint file, making the corrected inversion calculation path a fixed component of the physical information neural network model; "subsequent test batches" refers to new steel fiber reinforced concrete rheological property tests conducted under the same or similar mix proportions; and "model checkpoint file" refers to a binary storage file containing all network weight matrices, bias vectors, and optimizer states.

[0334] The specific implementation process is as follows: The network parameters corrected in step S4073 are solidified into the physical information neural network model. This solidification refers to saving the updated weight matrix, bias vector, and necessary normalization parameters as a model checkpoint file. The file format is HDF5 or TensorFlow SavedModel. The saved content includes the weights of the boundary encoding branch, the weights of the macroscopic morphology encoding branch, the weights of the fiber network constraint branch, the weights of the physical constraint fusion layer, the weights of the stress decoding subnetwork, and the residual weight coefficients. The solidified physical information neural network model can be directly loaded for the inversion of rheological property parameters in subsequent test batches without having to re-execute the solidification verification and model correction steps, which greatly improves the efficiency of subsequent tests.

[0335] It is worth noting that the solidified model is only applicable to steel fiber reinforced concrete with the same or similar mix proportions as those used during training and validation. If the mix proportions change significantly, such as a relative change of more than 30% in fiber volume content with an absolute change of more than 0.5 percentage points, or a relative change of more than 20% in water-cement ratio with an absolute change of more than 0.1, online fine-tuning or full final-state validation should be performed again. The model checkpoint file should be version-managed, recording the corresponding mix proportion parameters and validation errors.

[0336] Understandably, step S4074 achieves persistent storage and reuse of model knowledge by solidifying the updated network parameters into the physical information neural network model, enabling subsequent test batches to directly utilize the model verified by actual measurements of the final state, thereby improving the accuracy and reliability of rheological characteristic parameter inversion in subsequent test batches.

[0337] It is understood that the measurement method and system of this application are not only applicable to steel fiber reinforced concrete, but also to other non-Newtonian multiphase fluid materials containing coarse particles or easily agglomerated and oriented fibers, such as 3D printed concrete, polymer fiber mortar, tailings slurry or industrial high-viscosity slurry. During implementation, only the size of the loading container, the initial range of rheological parameters and the training dataset need to be adjusted.

[0338] In summary, this application constructs a complete technical chain from sample preparation, overall deformation, internal and external joint monitoring to intelligent inversion through step S1 (material pretreatment), step S2 (overall deformation triggering), step S3 (synchronous dual-source monitoring), and step S4 (spatiotemporal correlation fusion and physical information neural network model inversion). This solves the measurement distortion problem caused by fiber agglomeration and directional arrangement in traditional shear testing, and achieves true and accurate measurement of the overall rheological properties of high-viscosity steel fiber reinforced concrete without destroying the original random distribution of steel fibers. The embodiments of this application also provide a steel fiber reinforced concrete rheological property measurement system, including a sample loading module, a deformation driving module, a synchronous acquisition module, and a fusion inversion module.

[0339] The sample loading module is used to load pre-mixed steel fiber reinforced concrete into a pre-set loading container;

[0340] The deformation driving module is used to apply an overall force to the steel fiber concrete through a preset overall deformation triggering mechanism, so that the steel fiber concrete will undergo overall deformation; during the overall deformation process, natural shear is generated inside the steel fiber concrete, accompanied by the evolution of appearance and internal flow field.

[0341] The synchronous acquisition module is used to simultaneously acquire external morphology evolution data using a preset non-contact scanning device during the overall deformation process, and simultaneously acquire internal flow field evolution data using a preset lateral non-destructive perspective scanning device.

[0342] The fusion inversion module is used to perform spatiotemporal correlation fusion of external morphology evolution data and internal flow field evolution data, and to invert and calculate the rheological characteristic parameters of steel fiber reinforced concrete based on a preset physical information neural network model with embedded preset rheological physical equation constraints.

[0343] In summary, this application constructs a complete technical chain from sample preparation, overall deformation, internal and external joint monitoring to intelligent inversion through the sample loading module's shear-damage-free loading pretreatment, the deformation driving module's multi-mode overall deformation driving, the synchronous acquisition module's macro-micro spatiotemporal joint acquisition, and the fusion inversion module's physical information neural network model-constrained inversion. This solves the technical problem of fiber agglomeration and directional alignment caused by local forced shearing in traditional rotational shear rheological testing, which leads to distortion in rheological property measurement. It achieves true and accurate measurement of the overall rheological properties of high-viscosity steel fiber reinforced concrete without destroying the original random distribution of steel fibers.

[0344] The embodiments of this application have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of this application. The description of the above embodiments is only for the purpose of helping to understand the technical solutions and core ideas of this application. Those skilled in the art should understand that they can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. These 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 this application.

Claims

1. A method for measuring the rheological properties of steel fiber reinforced concrete, characterized in that, Includes the following steps: The pre-mixed steel fiber reinforced concrete is loaded into a pre-designated loading container; An integral force is applied to the steel fiber reinforced concrete by a pre-set integral deformation triggering mechanism, causing the steel fiber reinforced concrete to undergo integral deformation. During the integral deformation process, natural shearing occurs inside the steel fiber reinforced concrete, accompanied by evolution of appearance and internal flow field. The integral deformation triggering mechanism is a pre-set perforated baffle set at the bottom of the pre-set loading container, a pre-set vertical piston set at the bottom of the pre-set loading container, or a pre-set horizontal piston set inside the pre-set loading container. During the overall deformation process, the starting moment of the overall deformation is used as the synchronization reference, and a preset non-contact scanning device and a preset lateral non-destructive fluoroscopic scanning device are activated simultaneously. Throughout the entire evolution stage of the overall deformation, the preset non-contact scanning device continuously acquires a three-dimensional contour point cloud sequence of the steel fiber reinforced concrete surface as appearance morphology evolution data. At the same time, the preset lateral non-destructive fluoroscopic scanning device continuously acquires a CT tomographic image sequence of the steel fiber reinforced concrete as data characterizing the evolution of the internal flow field. Each frame in the three-dimensional contour point cloud sequence corresponds to the frame at the corresponding moment in the CT tomographic image sequence according to a preset time synchronization relationship. The three-dimensional contour point cloud sequence and the CT tomographic image sequence are matched one by one according to a preset time synchronization relationship to form a data pair at the same time. The surface of the three-dimensional contour point cloud in the data pair is reconstructed to generate a free surface geometric model of steel fiber concrete at the current time, and dynamic boundary conditions and macroscopic morphological feature variables are extracted. Fiber identification is performed based on the CT tomographic image sequence, and the fiber network connectivity characterizing the integrity of the internal skeleton is calculated. The dynamic boundary conditions, fiber network connectivity, and macroscopic morphological feature variables are associated and stored to generate a spatiotemporal associated fusion dataset. Based on a pre-defined physical information neural network model with embedded pre-defined rheological physical equations as constraints, the rheological property parameters of steel fiber reinforced concrete are inverted and calculated.

2. The method according to claim 1, characterized in that, The preset overall deformation triggering mechanism is a preset perforated baffle plate set at the bottom of the preset loading container. The preset overall deformation triggering mechanism applies an overall force to the steel fiber reinforced concrete, including the following steps: Open the hole baffle to open the bottom opening of the preset loading container; The open bottom opening causes the steel fiber reinforced concrete to lose its lower surface support at the bottom opening. The steel fiber reinforced concrete at the bottom opening flows downward through the bottom opening under the action of gravity; During the downward flow, the steel fiber reinforced concrete undergoes overall deformation.

3. The method according to claim 1, characterized in that, The preset overall deformation triggering mechanism is a preset vertical piston located at the bottom of the preset loading container. The preset overall deformation triggering mechanism applies an overall force to the steel fiber reinforced concrete, including the following steps: Control the vertical piston to move downwards at a preset speed; The downward movement of the vertical piston creates a suction force at the bottom of the preset loading container, corresponding to the preset speed. The suction force induces the steel fiber reinforced concrete to flow downwards; During the flow process, steel fiber reinforced concrete undergoes overall deformation.

4. The method according to claim 1, characterized in that, The preset overall deformation triggering mechanism is a preset horizontal piston set inside the preset loading container; the preset loading container is a horizontally placed material cylinder with a preset opening at one end and a preset test platform on the outside of the preset opening. Applying an overall force to steel fiber reinforced concrete through a pre-set overall deformation triggering mechanism includes the following steps: Control the horizontal piston to move in the preset opening direction; A horizontal piston pushes steel fiber reinforced concrete from a pre-set opening onto the test platform; Once the steel fiber reinforced concrete is pushed out of the pre-set opening, it loses the side wall constraint of the cylinder. Steel fiber reinforced concrete that has lost its sidewall restraints will collapse laterally under the action of gravity; During the collapse, the steel fiber reinforced concrete underwent overall deformation.

5. The method according to claim 1, characterized in that, After generating the spatiotemporal correlation fusion dataset, dynamic boundary updates and shape prediction corrections are performed based on time progression, including the following steps: With a preset time step as the period, the free surface geometric model corresponding to the current moment is obtained sequentially in time order, and the corresponding dynamic boundary conditions are extracted from the spatiotemporal correlation fusion dataset. Obtain the rheological parameter set corresponding to the current time step; wherein, in the first time step, the rheological parameter set is the preset initial rheological parameter set; in subsequent time steps, the rheological parameter set is the rheological parameter set obtained after optimization in the previous time step; Based on the dynamic boundary conditions and rheological parameter set at the current moment, the free surface morphology at the next moment is predicted by forward modeling the preset fluid dynamics equations, and this morphology is defined as the predicted free surface morphology at the next moment. Obtain the free surface geometric model corresponding to the next moment, and define the geometric model as the measured free surface morphology at the next moment; Calculate the morphology error between the predicted morphology of the free surface and the measured morphology of the free surface; The rheological parameter set is optimized based on the morphological error to obtain the optimized rheological parameter set; The dynamic boundary conditions are forcibly updated based on the measured free surface topography at the next moment, and the updated dynamic boundary conditions at the next moment are used as the calculation input for the next moment after that.

6. The method according to claim 5, characterized in that, Based on a pre-defined physical information neural network model constrained by pre-defined rheological physical equations, the rheological property parameters of steel fiber reinforced concrete are inverted and calculated, including: Obtain the fiber network connectivity and macroscopic topography feature variables corresponding to the next time step, and obtain the updated dynamic boundary conditions and topography error for the next time step; The updated dynamic boundary conditions and macroscopic morphological feature variables are used as inputs to the preset physical information neural network model; The fiber network connectivity is used as the independent variable of the dynamic constitutive constraint in the pre-defined physical information neural network model; The shape error is used as an additional weighting term in the loss function of the pre-defined physical information neural network model; Based on the input, independent variables, and additional weighting terms, a preset physical information neural network model is run. Through the constraints of the preset rheological physical equations embedded in the preset physical information neural network model, the numerical results of the rheological property parameters of steel fiber reinforced concrete are obtained by inversion calculation. Rheological property parameters are obtained from the output of a pre-defined physical information neural network model. These parameters include at least one of yield stress, plastic viscosity, and fiber network contribution coefficient.

7. The method according to claim 6, characterized in that, This also includes steps following the inversion calculation of the rheological property parameters of steel fiber reinforced concrete: After the test is completed, the steel fiber reinforced concrete samples are left to cure under the preset curing conditions until the preset degree of curing is reached. The cured sample was scanned in three dimensions using a lateral non-destructive scanning device to obtain the spatial distribution of the fibers after curing, which was taken as the measured final state of the fiber distribution. The dynamic distribution evolution trajectory of steel fibers during the overall deformation process is extracted from CT tomographic image sequences, and the final state of fiber distribution is predicted by model recursion based on the dynamic distribution evolution trajectory. Spatial registration is performed between the measured final state and the model predicted final state. The spatially registered measured final state is then compared point by point with the model predicted final state to construct the fiber distribution error field for the entire field. Based on the fiber distribution error field, the adjoint equation backpropagation algorithm is used to trace the inversion error contribution of each time step along the time axis and generate an error propagation weight sequence. Based on the error propagation weight sequence, the inversion calculation loss corresponding to each time step in the physical information neural network model is weighted and corrected, and the network parameters of the physical information neural network model are updated based on the weighted and corrected inversion calculation loss to suppress the accumulation and amplification of error in the time progression process. The updated network parameters are embedded into the physical information neural network model for use in the inversion of rheological property parameters in subsequent test batches.

8. A system for measuring the rheological properties of steel fiber reinforced concrete, characterized in that, The method described in any one of claims 1 to 7 includes: The sample loading module is used to load pre-mixed steel fiber reinforced concrete into a pre-set loading container; The deformation driving module is used to apply an overall force to the steel fiber concrete through a preset overall deformation triggering mechanism, so that the steel fiber concrete will undergo overall deformation; during the overall deformation process, natural shear is generated inside the steel fiber concrete, accompanied by the evolution of appearance and internal flow field. The synchronous acquisition module is used to simultaneously acquire external morphology evolution data using a preset non-contact scanning device during the overall deformation process, and simultaneously acquire internal flow field evolution data using a preset lateral non-destructive perspective scanning device. The fusion inversion module is used to perform spatiotemporal correlation fusion of external morphology evolution data and internal flow field evolution data, and to invert and calculate the rheological characteristic parameters of steel fiber reinforced concrete based on a preset physical information neural network model with embedded preset rheological physical equation constraints.

Images