A method, system and apparatus for predicting the performance of foamed fibre concrete

Abstract

The present application belongs to the technical field of material performance prediction, and particularly relates to a foam fiber concrete performance prediction method, system and device, the method comprising: obtaining original experimental samples, calculating the theoretical apparent density of each original experimental sample according to phase composition principle, identifying the sparse target value interval in the original experimental samples, performing random linear interpolation on the sample points according to the characteristic vectors of the original experimental samples, generating virtual characteristic vectors, performing physical feasible domain verification, eliminating abnormal virtual characteristic vectors, and obtaining expanded samples; building a random forest regression model, predicting in parallel by multiple decision trees and taking the arithmetic mean to obtain an initial discrete stress sequence; performing two-stage correction; fitting the correction result to obtain a whole-process prediction curve, identifying the compaction starting point and yield point, and calculating the platform width index as the pressure release performance. The foam fiber concrete under complex stress environment can be evaluated and predicted.

Description

Technical Field

[0001] This invention belongs to the field of material performance prediction technology, specifically relating to a method, system, and apparatus for predicting the performance of foamed fiber concrete. Background Technology

[0002] As my country's transportation, water conservancy, and mining construction advances into deeper strata, the number of tunnel projects with high ground stress and large deformation is increasing. In soft rock with large deformation or high stress environments, traditional rigid supports, such as shotcrete and steel arches, often cannot adapt to the release of the massive deformation energy of the surrounding rock, leading to twisting, breakage, or even overall instability of the support structure.

[0003] In the actual research and development and engineering application of existing buffer and pressure relief materials, the performance of foam fiber concrete is still affected by the interaction of multiple factors such as foam content, fiber parameters, water-cement ratio and admixture ratio. Traditional empirical linear fitting or physical semi-empirical formulas are difficult to accurately describe its mechanical characteristics throughout the process, especially when predicting key engineering indicators, the error is large. The cost of obtaining experimental samples is high and the data scale is limited. Due to the limitations of experimental cycle, material preparation process and testing cost, the number of effective experimental samples obtained in the early stage of new material research and development is usually small. Traditional machine learning models are prone to overfitting traps under small sample datasets, lack the physical reality of material mechanics, and cause non-physical jumps in prediction results under boundary ratios. Summary of the Invention

[0004] The purpose of this invention is to provide a method, system, and apparatus for predicting the performance of foamed fiber concrete.

[0005] A method for predicting the performance of foamed fiber reinforced concrete includes the following steps: S1. Obtain the original experimental samples of foam fiber concrete formulations with different components. Calculate the theoretical apparent density of each original experimental sample according to the principle of phase composition. Combine the data of foam volume ratio, fiber volume fraction, water-cement ratio, fly ash replacement rate, coal gangue ratio, and fiber aspect ratio in the original experimental sample to obtain the feature vector of the original experimental sample. S2. Based on the theoretical apparent density, identify the sparse target value interval in the original experimental sample. Perform random linear interpolation on the sample points of the original experimental sample within the sparse target value interval according to the feature vector of the original experimental sample to generate virtual feature vectors. Based on the apparent density of the virtual feature vectors, perform physical feasible region verification on the virtual feature vectors, remove abnormal virtual feature vectors, and obtain expanded samples. S3. Replace the original experimental samples in S2 with the expanded samples and the original experimental samples together. Repeat S2 until the number of new expanded samples is greater than the first threshold. Use the new expanded samples as global samples. Based on global samples, a random forest regression model is built, and multiple decision trees make parallel predictions and take the arithmetic mean to obtain the initial discrete stress sequence; S4. Set the strength threshold through empirical expression. If the value in the initial discrete stress sequence is greater than the strength threshold, replace the value with the strength threshold to obtain the corrected initial discrete stress sequence and the initial discrete stress sequence less than or equal to the strength threshold as the first-level correction sequence. When the fiber volume fraction in the first-level correction sequence is greater than or equal to the preset volume ratio, the first-level correction sequence is subjected to multiplicative attenuation correction to obtain the second-level correction sequence. S5. Fit the second-level correction sequence to obtain the whole process prediction curve, identify the compaction start point and yield point, and calculate the platform width index as the yield performance.

[0006] In S4, a multiplicative decay correction is performed on the first-level corrected sequence to obtain the second-level corrected sequence, specifically as follows: Obtain the fiber volume fraction in the first-level correction sequence. If it is greater than a preset ratio, calculate the degradation operator. , in, To degrade the operator, For degradation sensitivity coefficient, This represents the fiber volume fraction in the first-order corrected sequence. This is the critical fiber content threshold value; Perform multiplicative decay correction: , in, This is the final output value for the strength of the secondary corrected sequence. This is the intensity output value of the first-level corrected sequence.

[0007] In S2, based on the apparent density of virtual feature vectors, a physical feasible region check is performed on the virtual feature vectors to eliminate abnormal virtual feature vectors. Specifically: Calculate the intensity of the virtual eigenvector and compare it with the physical boundary: , in, For virtual feature vector strength, are geometric constants. The nominal strength of the matrix cement paste. The apparent density of the virtual feature vector. The density of the material in its fully compacted state is its solid density. It is an experience index; If the ratio of the virtual feature vector strength to the apparent density of the virtual feature vector deviates from the physical boundary by more than a preset deviation threshold, then the virtual feature vector is determined to be abnormal, and the abnormal virtual feature vector is removed.

[0008] In S2, random linear interpolation is performed on the sample points of the original experimental samples within the sparse target value interval based on the feature vector of the original experimental samples to generate virtual feature vectors, specifically: ; in, For virtual feature vectors, These are the sample points of the original experimental sample. As the nearest neighbor, A random weighting factor between 0 and 1.

[0009] In S3, a random forest regression model is built, which uses multiple decision trees to make parallel predictions and take the arithmetic mean to obtain the initial discrete stress sequence, specifically: Build Each of the four independent decision trees obtains an initial predicted value during the prediction phase, based on the global sample. The initial discrete stress sequence is obtained by taking the arithmetic mean of the initial predictions from all decision trees: ; Where i represents the i-th tree, For the initial discrete stress sequence, These are the initial predicted values.

[0010] In S5, the secondary correction sequence is fitted into a full-process prediction curve, specifically as follows: The second-order correction sequence is fitted into the whole process prediction curve by using cubic spline interpolation or fifth-order polynomial fitting algorithm.

[0011] In S5, the compaction initiation point and yield point are identified as follows: The second derivative algorithm is used to perform a full scan of the entire process prediction curve. The shear point where the second derivative value changes from positive to negative and the absolute value reaches its peak is the yield point. The integral area under the predicted curve throughout the entire process is calculated in real time, and the derivative of the strain is used as the inflection point where the slope changes the most.

[0012] In S5, the platform width metric is used as a yielding performance indicator, specifically: , in, For platform width metrics, As the starting point for compaction, This is the yield point.

[0013] A foamed fiber reinforced concrete performance prediction system, used to implement a method for predicting the performance of foamed fiber reinforced concrete, includes: The data acquisition module acquires the original experimental samples of foam fiber concrete formulations with different components, calculates the theoretical apparent density of each original experimental sample based on the principle of phase composition, and splices the data of foam volume ratio, fiber volume fraction, water-cement ratio, fly ash replacement rate, coal gangue ratio, and fiber aspect ratio in the original experimental sample to obtain the feature vector of the original experimental sample. The sample expansion module identifies sparse target value intervals in the original experimental samples based on theoretical apparent density. It performs random linear interpolation on the sample points of the original experimental samples within the sparse target value intervals according to the feature vectors of the original experimental samples to generate virtual feature vectors. Based on the apparent density of the virtual feature vectors, it performs physical feasible region verification on the virtual feature vectors, removes abnormal virtual feature vectors, and obtains expanded samples. The prediction module replaces the original experimental samples in the sample expansion module with expanded samples and original experimental samples. The sample expansion module is executed repeatedly until the number of new expanded samples is greater than the first threshold. The new expanded samples are then used as global samples. Based on global samples, a random forest regression model is built, and multiple decision trees make parallel predictions and take the arithmetic mean to obtain the initial discrete stress sequence; The correction module sets a strength threshold using an empirical expression. If a value in the initial discrete stress sequence is greater than the strength threshold, the strength threshold is used to replace the value, resulting in a corrected initial discrete stress sequence and initial discrete stress sequences less than or equal to the strength threshold, which are used as the first-level correction sequence. When the fiber volume fraction in the first-level correction sequence is greater than or equal to the preset volume ratio, the first-level correction sequence is subjected to multiplicative attenuation correction to obtain the second-level correction sequence. The performance acquisition module fits the secondary correction sequence to obtain the full-process prediction curve, identifies the compaction start point and yield point, and calculates the platform width index as the yield performance.

[0014] A foamed fiber concrete performance prediction device includes a processor and a memory, wherein the processor executes a computer program stored in the memory to implement a foamed fiber concrete performance prediction method.

[0015] Compared to existing technologies: This invention effectively overcomes the bottleneck of small sample data in the research and development of foamed fiber concrete. For the limited original experimental data, a regression-type oversampling algorithm coupled with the mechanical laws of porous materials is used to scientifically interpolate and filter the original sparse data for physical consistency, eliminating non-physical virtual samples. This allows the limited original data to exert the predictive stability and generalization ability of hundreds of samples without the need to add a large number of additional experimental samples. This significantly reduces the experimental preparation, testing and time costs in the research and development of new materials, and improves the reliability of material performance prediction in small sample scenarios.

[0016] This invention ensures the mechanical authenticity of the performance prediction results of foamed fiber concrete. By embedding a two-level correction mechanism in the integrated learning regression model, it effectively corrects the non-physical fluctuation problem that is prone to occur in pure data-driven models under boundary ratio conditions. It forces the prediction results to conform to the basic laws of material mechanics, ensuring that the output stress-strain prediction curve strictly follows the three-stage deformation characteristics of porous materials, thus avoiding the problem of the prediction results being out of sync with the actual mechanical behavior of engineering.

[0017] This invention realizes a closed-loop design for foam fiber concrete from research and development to engineering application. Its built-in reverse design function based on adaptive genetic algorithm addresses the problem of designing material proportions on demand in tunnel engineering. It can reverse-engineer the material mix ratio and allowable construction error range based on engineering requirements such as specific surrounding rock pressure and expected pressure relief stroke of the tunnel, which greatly improves the response speed of dynamic support in underground engineering and enhances the pertinence and efficiency of support material design. Detailed Implementation

[0018] To further understand the content of this invention, the invention will be described in detail with reference to the embodiments.

[0019] This invention relates to a method for predicting the performance of foamed fiber concrete, which includes the following steps: S1. Obtain the original experimental samples of foam fiber concrete formulations with different components. Calculate the theoretical apparent density of each original experimental sample according to the principle of phase composition. Combine the data of foam volume ratio, fiber volume fraction, water-cement ratio, fly ash replacement rate, coal gangue ratio, and fiber aspect ratio in the original experimental sample to obtain the feature vector of the original experimental sample.

[0020] Thirty-two sets of original experimental samples were obtained, including data on foam volume ratio, fiber volume fraction, water-cement ratio, fly ash replacement rate, coal gangue ratio, fiber aspect ratio, and theoretical apparent density. The original feature data consists of a set of feature vectors X and a set of corresponding label vectors Y, where the label vectors... Includes: the compressive strength of the specimen (MPa) and the initial strain of the pressure relief plateau. Theoretical apparent density It serves both as input features for the random forest model and as a feature for subsequent physical consistency verification.

[0021] The theoretical apparent density of each sample group was calculated based on the principle of phase composition. The formula is ,in The true density of each component in solid form is given. The phase composition calculation principle is based primarily on the solid framework and porous system, without explicitly considering the transient effects of the free liquid phase volume. The aqueous phase is equivalently converted to the solid structure density calculation using the water-binder ratio parameter. The absolute volume percentage excluding foam is then normalized and mapped.

[0022] Subsequently, all the obtained raw feature data are input into the preprocessing unit for normalization, specifically using a linear scaling method to uniformly map the values ​​to... Within the interval, the normalization formula is: ,in, The normalized feature vectors, It is the smallest eigenvector. It is the largest eigenvector.

[0023] This operation aims to eliminate the negative interference caused by differences in the dimensions of different features, such as the large difference in magnitude between foam content and water-gluoride ratio, on the decision boundary division during the subsequent training of the random forest model.

[0024] This step calculates the theoretical apparent density. The core meaning is: First, it provides a physical verification benchmark for data augmentation in S2: when generating virtual samples, it calls... As an independent variable input, it is used to verify the monotonicity of the "density-intensity" relationship and prevent the generation of non-physical noise. Second, it provides a porosity inversion basis for the two-stage modified physical gating in S3: through theoretical apparent density Density of solids under fully compacted conditions By comparing the results, the true internal porosity of foamed concrete can be accurately determined. The porosity It will be used as the first-level correction unit ( The core input is used to limit the predicted range of yield strength, thereby achieving true physical-data dual-drive.

[0025] S2. Based on the theoretical apparent density, identify the sparse target value interval in the original experimental sample. According to the feature vector of the original experimental sample, perform random linear interpolation on the sample points of the original experimental sample within the sparse target value interval to generate virtual feature vectors. Based on the apparent density of the virtual feature vectors, perform physical feasible region verification on the virtual feature vectors, remove abnormal virtual feature vectors, and obtain expanded samples.

[0026] In the implementation of this invention, given the limited experimental data scale in the early stage of foam fiber concrete research and development, as shown in Table 1, the original number of experimental samples was approximately 32 groups. This resulted in sparse sample distribution and insufficient data in some local areas, which could easily lead to poor prediction stability of the data-driven model in some mix proportion ranges.

[0027] Table 1. Test results of mechanical properties of some materials in the original experimental samples. To improve the above situation, the present invention uses the nearest neighbor interpolation method to generate virtual samples during the sample expansion process, and combines the theoretical apparent density to screen the generated samples for rationality, so as to reduce abnormal samples that do not conform to the basic density characteristics of materials, thereby improving the reliability and applicability of the expanded data.

[0028] Data augmentation first identifies sparse target value intervals in the sample feature space. Then, the SMOTER algorithm is used to perform random linear interpolation on the sample points within the intervals, representing the original experimental samples as feature vectors. .

[0029] in, This refers to the foam volume ratio. This refers to the water-to-glue ratio. Apparent density. Calculate the target variable (e.g., compressive strength). The numerical distribution of the sample points is used to identify the low-frequency target value range where the sample points are sparse using the probability density function.

[0030] Furthermore, within the normalized feature space, sparse intervals are identified where the number of sample points in the frequency band is less than the target value. For any original target sample point within the sparse interval... Calculate the Euclidean distance between it and all its neighboring points, and select the nearest one. Sample points (preferred) Construct a nearest neighbor set.

[0031] The virtual feature vector is generated as follows: ; in, For virtual feature vectors, These are the sample points of the original experimental sample. As the nearest neighbor, A random weighting factor between 0 and 1.

[0032] Corresponding intensity target value It is generated synchronously in a linear proportion.

[0033] During the virtual sample generation process, the density-strength correlation law in porous material mechanics is combined to make a physical consistency judgment on the virtual feature vector, so as to ensure that the generated sample satisfies the basic monotonic relationship between material strength and apparent density, and to remove abnormal samples that do not conform to this relationship.

[0034] Based on the apparent density of virtual feature vectors, a physical feasible region check is performed on the virtual feature vectors to eliminate abnormal virtual feature vectors. Specifically: Calculate the intensity of the virtual eigenvector and compare it with the physical boundary: , in, For virtual feature vector strength, are geometric constants. The nominal strength of the matrix cement paste. The apparent density of the virtual feature vector. The density of the material in its fully compacted state is its solid density. It is an experience index; If the ratio of the virtual feature vector strength to the apparent density of the virtual feature vector deviates from the physical boundary by more than a preset deviation threshold, then the virtual feature vector is determined to be abnormal, and the abnormal virtual feature vector is removed.

[0035] The generated virtual samples are subjected to intensity prediction and verification to determine whether they satisfy the mechanical law that intensity increases monotonically with increasing apparent density. If the ratio of intensity to density of the virtual sample deviates from the physical boundary determined by the discrimination criterion by more than 15%, the virtual sample is determined to have a non-physical logical anomaly and is removed.

[0036] During the data synthesis process, a screening mechanism based on the density-strength relationship of porous materials is introduced to make a consistency judgment on the generated virtual samples, so as to remove sample data that does not conform to the basic mechanical laws of materials, thereby improving the stability and reliability of training data.

[0037] S3. Replace the original experimental samples in S2 with the expanded samples and the original experimental samples together. Repeat S2 until the number of new expanded samples is greater than the first threshold. Then, use the new expanded samples as global samples.

[0038] Specifically, S2 is executed repeatedly. Through the above iterative loop, only samples that conform to the monotonicity of mechanics are retained and added to the expanded database. In the end, the training sample size is expanded from the original 32 groups to more than 150 groups. The expanded samples serve as global samples, ensuring that the augmented dataset has both data diversity and conforms to physical reality.

[0039] Preferably, the first threshold is 3-10 times the number of original samples, and in this invention, it is set to 150.

[0040] Based on global samples, a random forest regression model is built. Multiple decision trees make parallel predictions and take the arithmetic mean to obtain the initial discrete stress sequence, specifically: Based on Bootstrap resampling, construct N decision trees. Each of the four independent decision trees obtains an initial predicted value during the prediction phase, based on the global sample. The initial discrete stress sequence is obtained by taking the arithmetic mean of the initial predictions from all decision trees: ; Where i represents the i-th tree, For the initial discrete stress sequence, These are the initial predicted values.

[0041] During the training phase, set the number of decision trees. The maximum path depth is 15 to prevent overfitting; the minimum number of split samples is 2 to ensure the model's sensitivity to fluctuations in trace components.

[0042] S4. Set the strength threshold through empirical expression. If the value in the initial discrete stress sequence is greater than the strength threshold, replace the value with the strength threshold to obtain the corrected initial discrete stress sequence and the initial discrete stress sequence less than or equal to the strength threshold as the first-level correction sequence.

[0043] When the fiber volume fraction in the first-level correction sequence is greater than or equal to the preset volume ratio, the first-level correction sequence is subjected to multiplicative attenuation correction to obtain the second-level correction sequence.

[0044] To ensure the initial discrete stress sequence To avoid deviating from physical facts under extreme boundary conditions, two levels of correction units are connected to the output of the random forest. Perform real-time verification and correction.

[0045] In the first-level correction step, an empirical relationship between porosity and material strength is introduced: Used to impose an upper limit constraint on the prediction results, in order to limit the output value from exceeding the reasonable strength range of the material.

[0046] in, This represents the theoretical upper limit of strength. The nominal strength of the cement stone matrix under fully compacted state (value range: 40-60 MPa); These are empirical constants related to pore structure (with values ​​ranging from 3.5 to 4.5). This refers to the real-time porosity obtained from the theoretical apparent density inversion in step S1.

[0047] Initial predicted value and Perform real-time comparison, if This triggers gating correction, forcibly limiting the predicted output to... This eliminates non-physical numerical divergence caused by data extrapolation.

[0048] The corrected initial discrete stress sequence and the initial discrete stress sequence less than or equal to the strength threshold are used as the first-level correction sequence.

[0049] The second-level correction unit is a multiplicative decay correction used to compensate for the underfitting of machine learning for excessive fiber damage. It obtains the fiber volume fraction in the first-level correction sequence. If it is greater than a preset ratio, for example, when the fiber volume fraction of the input first-level correction sequence is greater than a preset ratio, it will be corrected. At that time, calculate the degradation operator: , in, To degrade the operator, For degradation sensitivity coefficient, This represents the fiber volume fraction in the first-order corrected sequence. This is the critical fiber content threshold value; Perform multiplicative decay correction: , in, This is the final output value for the strength of the secondary corrected sequence. This is the intensity output value of the first-level correction sequence.

[0050] This mechanism simulates the clumping phenomenon, increased internal microcrack sources, and destructive effect on the slurry encapsulation caused by excessive fiber content during actual casting through multiplicative attenuation, ensuring that the prediction results conform to the physical law that excessive fiber leads to strength reduction.

[0051] The two-stage correction unit ensures the mechanical authenticity of the prediction results. This invention changes the black-box prediction mode of traditional machine learning models. By introducing porosity strength constraints in the first-stage correction unit and fiber agglomeration deterioration compensation coefficients in the second-stage correction unit, it successfully integrates classical materials science theory with modern nonlinear algorithms.

[0052] This mechanism can automatically correct for possible overestimation of strength or non-physical fluctuations in the extrapolation range of the model, accurately simulate the microscopic damage law of the matrix caused by the fiber under high content conditions, and make the output performance data fully consistent with the real stress characteristics of underground engineering support materials.

[0053] S5. Fit the second-level correction sequence to obtain the whole process prediction curve, identify the compaction start point and yield point, and calculate the platform width index as the yield performance.

[0054] The corrected second-order correction sequence is fitted into a continuous full-process prediction curve using a cubic spline interpolation algorithm. The curve is then fully scanned using a second-order derivative algorithm, and the shear point where the second derivative changes from positive to negative is defined as the yield point. The energy absorption integral enclosed below the curve is calculated, and the stress surge starting point after the end of the stable energy absorption rate interval is defined as the compaction starting point. Based on this, the platform width index representing the yielding performance is automatically calculated and output.

[0055] Specifically, in this embodiment, step S4 does not simply output a single intensity value, but rather predicts a set of values ​​related to strain in parallel. The corresponding secondary correction sequence divides the strain axis into multiple intervals with a step size of 0.5%, and trains a prediction model for each strain node to obtain a set of stress-strain prediction data points for the entire process under the target ratio.

[0056] After obtaining the discrete second-order correction sequence, cubic spline interpolation or fifth-order polynomial fitting algorithms are used to connect the discrete points into a continuous full-process prediction curve. This curve fully covers the mechanical response characteristics of foamed concrete from loading initiation, elastic deformation, yield failure to the yield plateau stage and the final compaction stage.

[0057] The second derivative algorithm is used to perform a full scan of the fitted whole-process prediction curve to accurately identify the following performance indicators of the material: The shear point where the second derivative changes from positive to negative and the absolute value reaches its peak is defined as the yield point. The integral area under the predicted curve throughout the entire process is calculated in real time, and the derivative of the strain is used as the inflection point where the slope changes the most.

[0058] Preferably, the yield point is defined as the turning point where the curve transitions from the linear elastic stage to the plateau stage. It is preferably identified using the second derivative extremum method, where the strain point corresponding to the continuous and rapid increase of the first derivative of stress with respect to strain, exceeding a preset threshold, is taken as the compaction starting point.

[0059] Furthermore, the integral area represents the energy absorbed. .

[0060] Based on the identification of the above key points, the platform width index is calculated as a pressure relief performance indicator, specifically: , in, For platform width metrics, As the starting point for compaction, This is the yield point.

[0061] Simultaneously, a digital performance profile including elastic modulus, peak strength, and specific energy absorption value is generated, enabling a full-dimensional quantitative description of the material's mechanical behavior.

[0062] This invention achieves automated identification of key indicators such as yield strength, plateau width, and energy absorption efficiency by performing second-order derivative feature scanning on the predicted stress-strain curve. Compared to existing technologies that can only predict a single strength, this invention can provide a complete mechanical profile of the material under extreme compression conditions, providing comprehensive data support for evaluating the "yield-then-resist" mechanism of support structures during large deformation of surrounding rock, and significantly improving the safety factor of underground and tunnel operations.

[0063] A foamed fiber reinforced concrete performance prediction system, used to implement a method for predicting the performance of foamed fiber reinforced concrete, includes: The data acquisition module acquires the original experimental samples of foam fiber concrete formulations with different components, calculates the theoretical apparent density of each original experimental sample based on the principle of phase composition, and splices the data of foam volume ratio, fiber volume fraction, water-cement ratio, fly ash replacement rate, coal gangue ratio, and fiber aspect ratio in the original experimental sample to obtain the feature vector of the original experimental sample. The sample expansion module identifies sparse target value intervals in the original experimental samples based on theoretical apparent density. It performs random linear interpolation on the sample points of the original experimental samples within the sparse target value intervals according to the feature vectors of the original experimental samples to generate virtual feature vectors. Based on the apparent density of the virtual feature vectors, it performs physical feasible region verification on the virtual feature vectors, removes abnormal virtual feature vectors, and obtains expanded samples. The prediction module replaces the original experimental samples in the sample expansion module with expanded samples and original experimental samples. The sample expansion module is executed repeatedly until the number of new expanded samples is greater than the first threshold. The new expanded samples are then used as global samples. Based on global samples, a random forest regression model is built, and multiple decision trees make parallel predictions and take the arithmetic mean to obtain the initial discrete stress sequence; The correction module sets a strength threshold using an empirical expression. If a value in the initial discrete stress sequence is greater than the strength threshold, the strength threshold is used to replace the value, resulting in a corrected initial discrete stress sequence and initial discrete stress sequences less than or equal to the strength threshold, which are used as the first-level correction sequence. When the fiber volume fraction in the first-level correction sequence is greater than or equal to the preset volume ratio, the first-level correction sequence is subjected to multiplicative attenuation correction to obtain the second-level correction sequence. The performance acquisition module fits the secondary correction sequence to obtain the full-process prediction curve, identifies the compaction start point and yield point, and calculates the platform width index as the yield performance.

[0064] A foamed fiber concrete performance prediction device includes a processor and a memory, wherein the processor executes a computer program stored in the memory to implement a foamed fiber concrete performance prediction method.

Claims

1. A method for predicting the performance of foamed fiber reinforced concrete, characterized in that, Includes the following steps: S1. Obtain the original experimental samples of foam fiber concrete formulations with different components. Calculate the theoretical apparent density of each original experimental sample according to the principle of phase composition. Combine the data of foam volume ratio, fiber volume fraction, water-cement ratio, fly ash replacement rate, coal gangue ratio, and fiber aspect ratio in the original experimental sample to obtain the feature vector of the original experimental sample. S2. Based on the theoretical apparent density, identify the sparse target value interval in the original experimental sample. Perform random linear interpolation on the sample points of the original experimental sample within the sparse target value interval according to the feature vector of the original experimental sample to generate virtual feature vectors. Based on the apparent density of the virtual feature vectors, perform physical feasible region verification on the virtual feature vectors, remove abnormal virtual feature vectors, and obtain expanded samples. S3. Replace the original experimental samples in S2 with the expanded samples and the original experimental samples together. Repeat S2 until the number of new expanded samples is greater than the first threshold. Use the new expanded samples as global samples. Based on global samples, a random forest regression model is built, and multiple decision trees make parallel predictions and take the arithmetic mean to obtain the initial discrete stress sequence; S4. Set the strength threshold through empirical expression. If the value in the initial discrete stress sequence is greater than the strength threshold, replace the value with the strength threshold to obtain the corrected initial discrete stress sequence and the initial discrete stress sequence less than or equal to the strength threshold as the first-level correction sequence. When the fiber volume fraction in the first-level correction sequence is greater than or equal to the preset volume ratio, the first-level correction sequence is subjected to multiplicative attenuation correction to obtain the second-level correction sequence. S5. Fit the second-level correction sequence to obtain the whole process prediction curve, identify the compaction start point and yield point, and calculate the platform width index as the yield performance.

2. The method for predicting the performance of foamed fiber reinforced concrete according to claim 1, characterized in that, In S4, a multiplicative decay correction is performed on the first-level corrected sequence to obtain the second-level corrected sequence, specifically as follows: Obtain the fiber volume fraction in the first-level correction sequence. If it is greater than a preset ratio, calculate the degradation operator. ， in, To degrade the operator, For degradation sensitivity coefficient, This represents the fiber volume fraction in the first-order corrected sequence. This is the critical fiber content threshold value; Perform multiplicative decay correction: ， in, This is the final output value for the strength of the secondary corrected sequence. This is the intensity output value of the first-level corrected sequence.

3. The method for predicting the performance of foamed fiber reinforced concrete according to claim 1, characterized in that, In S2, based on the apparent density of virtual feature vectors, a physical feasible region check is performed on the virtual feature vectors to eliminate abnormal virtual feature vectors. Specifically: Calculate the intensity of the virtual eigenvector and compare it with the physical boundary: ， in, For virtual feature vector strength, are geometric constants. The nominal strength of the matrix cement paste. The apparent density of the virtual feature vector. The density of the material in its fully compacted state is the solid density. It is an experience index; If the ratio of the virtual feature vector strength to the apparent density of the virtual feature vector deviates from the physical boundary by more than a preset deviation threshold, then the virtual feature vector is determined to be abnormal, and the abnormal virtual feature vector is removed.

4. The method for predicting the performance of foamed fiber reinforced concrete according to claim 1, characterized in that, In S2, random linear interpolation is performed on the sample points of the original experimental samples within the sparse target value interval based on the feature vector of the original experimental samples to generate virtual feature vectors, specifically: ； in, For virtual feature vectors, These are the sample points of the original experimental sample. As the nearest neighbor, A random weighting factor between 0 and 1.

5. The method for predicting the performance of foamed fiber reinforced concrete according to claim 1, characterized in that, In S3, a random forest regression model is built, which uses multiple decision trees to make parallel predictions and take the arithmetic mean to obtain the initial discrete stress sequence, specifically: Build Each of the four independent decision trees obtains an initial predicted value during the prediction phase, based on the global sample. The initial discrete stress sequence is obtained by taking the arithmetic mean of the initial predictions from all decision trees: ; Where i represents the i-th tree, For the initial discrete stress sequence, These are the initial predicted values.

6. The method for predicting the performance of foamed fiber reinforced concrete according to claim 1, characterized in that, In S5, the secondary correction sequence is fitted into a full-process prediction curve, specifically as follows: The second-order correction sequence is fitted into the whole process prediction curve by using cubic spline interpolation or fifth-order polynomial fitting algorithm.

7. The method for predicting the performance of foamed fiber reinforced concrete according to claim 1, characterized in that, In S5, the compaction initiation point and yield point are identified as follows: The second derivative algorithm is used to perform a full scan of the entire process prediction curve. The shear point where the second derivative value changes from positive to negative and the absolute value reaches its peak is the yield point. The integral area under the predicted curve throughout the entire process is calculated in real time, and the derivative of the strain is used as the inflection point where the slope changes the most.

8. The method for predicting the performance of foamed fiber reinforced concrete according to claim 1, characterized in that, In S5, the platform width metric is used as a yielding performance indicator, specifically: ， in, For platform width metrics, As the starting point for compaction, This is the yield point.

9. A foamed fiber reinforced concrete performance prediction system, used to implement the foamed fiber reinforced concrete performance prediction method according to any one of claims 1-8, characterized in that, include: The data acquisition module acquires the original experimental samples of foam fiber concrete formulations with different components, calculates the theoretical apparent density of each original experimental sample based on the principle of phase composition, and splices the data of foam volume ratio, fiber volume fraction, water-cement ratio, fly ash replacement rate, coal gangue ratio, and fiber aspect ratio in the original experimental sample to obtain the feature vector of the original experimental sample. The sample expansion module identifies sparse target value intervals in the original experimental samples based on theoretical apparent density. It performs random linear interpolation on the sample points of the original experimental samples within the sparse target value intervals according to the feature vectors of the original experimental samples to generate virtual feature vectors. Based on the apparent density of the virtual feature vectors, it performs physical feasible region verification on the virtual feature vectors, removes abnormal virtual feature vectors, and obtains expanded samples. The prediction module replaces the original experimental samples in the sample expansion module with expanded samples and original experimental samples. The sample expansion module is executed repeatedly until the number of new expanded samples is greater than the first threshold. The new expanded samples are then used as global samples. Based on global samples, a random forest regression model is built, and multiple decision trees make parallel predictions and take the arithmetic mean to obtain the initial discrete stress sequence; The correction module sets a strength threshold using an empirical expression. If a value in the initial discrete stress sequence is greater than the strength threshold, the strength threshold is used to replace the value, resulting in a corrected initial discrete stress sequence and initial discrete stress sequences less than or equal to the strength threshold, which are used as the first-level correction sequence. When the fiber volume fraction in the first-level correction sequence is greater than or equal to the preset volume ratio, the first-level correction sequence is subjected to multiplicative attenuation correction to obtain the second-level correction sequence. The performance acquisition module fits the secondary correction sequence to obtain the full-process prediction curve, identifies the compaction start point and yield point, and calculates the platform width index as the yield performance.

10. A device for predicting the performance of foamed fiber concrete, characterized in that, It includes a processor and a memory, wherein the processor executes a computer program stored in the memory to implement a method for predicting the performance of foamed fiber concrete as described in any one of claims 1-8.

Images