A Method for Reverse Casting and Reconstruction of Tunnel Arch Wall Concrete Based on Multi-Source Monitoring

By reconstructing the front edge of the reverse pouring of tunnel arch concrete using multi-source monitoring data, and combining the pumping volume and design section mapping, a quality risk prediction model was established. This solved the problem of difficulty in observing the filling status inside the closed template, and enabled dynamic control and quality risk identification during the tunnel construction process.

CN122899496BActive Publication Date: 2026-11-17HUNAN INSTITUTE OF ENGINEERING
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202611386599.6
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2026-09-08
Publication Date
2026-11-17
Estimated Expiration
2046-09-08

AI Technical Summary

Technical Problem

In the reverse pouring construction of secondary lining in tunnels, it is difficult to directly observe the concrete filling status inside the closed formwork. A single pressure monitoring index cannot effectively characterize the propagation of the filling front edge and the unevenness between left and right sides, and there is a lack of effective quality risk early warning means.

Method used

By acquiring information such as pressure timing and pumping flow rate at each pressure measuring point on the template interface through multi-source monitoring data, the equivalent filling front is reconstructed. Consistency verification is carried out by combining the pumping volume and the geometric volume mapping of the design section, a quality risk prediction model is constructed, and a comprehensive dynamic control index is established to divide the construction status.

Benefits of technology

It enables dynamic perception of the concrete filling status inside the closed formwork and real-time monitoring of quality risks, improving the quality control capability of the construction process and reducing the difficulty and cost of rectification after the fact.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122899496B_ABST
    Figure CN122899496B_ABST
Patent Text Reader

Abstract

This invention discloses a method for reconstructing the reverse pouring and filling of concrete in tunnel arch walls based on multi-source monitoring, belonging to the field of tunnel construction monitoring and control technology. The method includes: acquiring multi-source monitoring data; correcting background noise for the pressure time series at each measuring point, identifying the start-up time based on an adaptive threshold, and calculating the propagation delay and jacking speed; using the start-up time and elevation as control points, reconstructing the equivalent filling leading edges on the left and right sides using piecewise cubic Hermite interpolation, and introducing volume-geometric mapping for consistency verification; extracting multivariate features of pressure response, propagation time series, leading edge state, volume verification, construction process, and material properties, constructing a risk prediction model to obtain risk probabilities; calculating anomaly scores based on normal cyclic statistical calculations, constructing a weighted comprehensive control index, and jointly establishing a three-level construction state zoning with the risk probability. This invention achieves visualized reconstruction and process perception of the concrete filling state within a closed formwork.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of tunnel construction monitoring and control technology, and in particular to a method for reverse pouring and reconstructing of tunnel arch wall concrete based on multi-source monitoring. Background Technology

[0002] The construction quality of secondary lining in tunnels directly affects structural safety and durability. With the development of tunnel construction towards mechanization, digitalization, and intelligence, construction quality control is gradually shifting from traditional post-construction inspection to process perception, condition evaluation, and dynamic regulation. Existing research has explored various aspects, including tunnel lifecycle monitoring, multi-source information fusion, construction condition prediction, and intelligent optimization, providing a foundation for data-driven analysis of the construction process.

[0003] Existing research on lining quality mainly focuses on defect detection and intelligent identification after lining formation. Researchers have utilized methods such as ground-penetrating radar, machine vision, and deep learning to detect defects such as cracks, voids, and insufficient compaction in the lining. Simultaneously, machine learning and multi-source information fusion methods are increasingly being applied to surrounding rock deformation prediction, construction risk assessment, and grouting status identification. In recent years, technologies such as drones, Transformers, and lightweight target detection have further enhanced the automatic identification capabilities of tunnel structures and construction status. Overall, current research has made good progress in lining formation quality detection and intelligent construction perception, but the dynamic identification of the filling status inside the closed formwork during secondary lining concrete pouring remains relatively insufficient.

[0004] Reverse continuous pouring is one of the main methods for constructing linings in large-section tunnels. In this method, concrete enters the closed formwork from the lower part of the sidewall and is continuously pumped upwards. However, this process faces the following technical challenges in actual construction:

[0005] First, the concrete filling interface inside the closed formwork cannot be directly observed. During the reverse pouring process, the concrete rises from bottom to top within the completely sealed formwork cavity. Construction workers cannot directly determine the true position and shape of the filling front edge through visual or conventional means, and can only rely on indirect monitoring signals for estimation. This poses significant challenges to quality control during the construction process.

[0006] Second, existing single pressure monitoring indicators are insufficient to comprehensively characterize the filling status. Currently, construction sites typically only focus on the amplitude changes of pumping pressure, but a single pressure peak or average pressure cannot reflect the propagation differences of the filling front on the left and right sides, the jacking efficiency at different elevation intervals, and the continuity of the filling process. In actual pouring, concrete is continuously jacked from the sidewalls to the arch crown, and its filling front is affected by factors such as pumping, pumping stoppage, left and right material supply, vibration, and venting, exhibiting complex spatiotemporal evolution characteristics. There is an urgent need to extract equivalent arrival time and propagation information from multi-point pressure time series.

[0007] Third, there is a lack of effective early warning mechanisms for quality risks. Current quality assessment mainly relies on non-destructive testing after demolding (such as ground-penetrating radar scanning), which falls under the category of post-construction inspection. Once defects are discovered, rectification is difficult and costly, and intervention and adjustments cannot be made during construction. Although some research has carried out intelligent sensing of the construction process, a technical solution that systematically integrates filling status sensing with quality risk prediction and dynamic construction control is still lacking. Summary of the Invention

[0008] In view of the technical problems in the prior art that it is difficult to directly observe the filling state of the concrete inside the closed formwork during the reverse casting construction of the secondary lining of the tunnel, and that a single pressure monitoring index cannot effectively characterize the propagation of the filling front edge and the left-right imbalance, the present invention provides a method for reconstructing the filling of the tunnel arch wall concrete in reverse casting based on multi-source monitoring.

[0009] To solve the above-mentioned technical problems, the technical solution proposed by this invention is as follows:

[0010] A method for reverse pouring and reconstructing of tunnel arch wall concrete based on multi-source monitoring includes the following steps:

[0011] Step S1: Obtain multi-source monitoring data. The multi-source monitoring data includes at least the pressure timing, pumping flow rate, cumulative pumping volume, pumping start / stop status, left and right displacement of the trolley, material placement station switching signal, vibration status, and exhaust alarm signal of each pressure measuring point on the template interface.

[0012] Step S2: Estimate background noise and correct baseline for the pressure timing of each pressure measuring point, and identify the pressure start-up time of each measuring point; calculate the pressure propagation delay, equivalent jacking speed, and left and right start-up time difference based on the pressure start-up time; reconstruct the left and right equivalent filling front edges using the pressure start-up time and measuring point elevation as control points; introduce volume-geometric volume mapping to construct the volume front edge, and use the volume front edge to perform consistency verification on the reconstructed pressure front edge;

[0013] The consistency check includes:

[0014] Establish a mapping relationship between the leading edge height and the theoretical filling volume based on the design cross-section: ,in For the leading edge height to reach Theoretical filling volume at that time This represents the effective lining length for the current cycle. For elevation The equivalent concrete section width at that location. To the equivalent filling leading edge height, The elevation of the measuring point from the pouring reference surface;

[0015] On-site effective cumulative pumping volume : ,in This is the converted instantaneous pumping flow rate. In pumping mode;

[0016] Front edge of the square: ,in To measure the height of the leading edge, It is the inverse function of the mapping relationship;

[0017] Based on the volumetric leading edge and combined with construction events, a field verification leading edge is formed. The average absolute deviation between the pressure front and the field verification front, the average deviation between the pressure front and the volume front, and the relative error of the final volume of the cycle are calculated to evaluate the consistency of the reconstructed front.

[0018] Step S3: Extract multivariate process features based on a complete casting cycle. These features include at least pressure response features, pressure propagation features, leading edge and volume features, construction process features, structural response features, and material performance features. Construct a binary classification quality risk prediction model and obtain the external risk probability of each cycle through grouped cross-validation.

[0019] Step S4: Calculate the individual abnormal score based on the statistical distribution of the normal cycle, construct a weighted comprehensive dynamic control index, and combine the comprehensive dynamic control index with the external risk probability to establish a three-level construction status zoning.

[0020] A further improvement to the above technical solution is as follows:

[0021] Preferably, identifying the pressure start-up time of each measuring point in step S2 specifically includes:

[0022] The stable period before the start of pouring is taken as the background period. The median background pressure and background noise level are calculated for each pressure measuring point. The median background pressure is used to perform baseline correction on the original pressure time series to obtain the corrected pressure time series.

[0023] The pressure start-up threshold is calculated based on the background noise level of each measuring point. The moment when the correction pressure first exceeds the start-up threshold and continues to increase for a preset duration is determined as the pressure start-up moment of that measuring point.

[0024] Preferably, in step S2, reconstructing the left equivalent filling front edge and the right equivalent filling front edge includes:

[0025] The left control point is constructed as The control points on the right are constructed as follows ,in, This represents the start time of the current effective pumping cycle. and These represent the pressure start-up times at each measuring point on the left and right sides, respectively. and These represent the elevations of each measuring point from the pouring reference surface; For measurement point index, ; This represents the total number of pressure measurement points on a single-sided template interface.

[0026] The left and right pressure leading edges are constructed using piecewise cubic Hermite interpolation that maintains monotonicity.

[0027]

[0028]

[0029] in, and These are the equivalent filling leading edge heights for the left and right sides, respectively. and These represent the time and high-order program sequence for the left and right control points, respectively.

[0030] Mean pressure front is defined as: ,in for The average pressure leading edge height at any given time.

[0031] Preferably, the consistency check in step S2 includes:

[0032] Establish a mapping relationship between the leading edge height and the theoretical filling volume based on the design cross-section: ,in For the leading edge height to reach Theoretical filling volume at that time This represents the effective lining length for the current cycle. For elevation The equivalent concrete section width at that location. To the equivalent filling leading edge height, The elevation of the measuring point from the pouring reference surface;

[0033] On-site effective cumulative pumping volume : ,in This is the converted instantaneous pumping flow rate. In pumping mode;

[0034] Front edge of the square: ,in To measure the height of the leading edge, It is the inverse function of the mapping relationship;

[0035] Based on the volumetric leading edge and combined with construction events, a field verification leading edge is formed. The average absolute deviation between the pressure leading edge and the field verification leading edge, the average deviation between the pressure leading edge and the volume leading edge, and the relative error of the final volume of the cycle are calculated to evaluate the consistency of the reconstructed leading edge.

[0036] Preferably, the multivariate process features extracted in step S3 include:

[0037] Pressure response characteristics, including the average pressure, peak pressure, pressure standard deviation, and growth rate within the preset time window after pressure start-up at each measuring point;

[0038] Pressure propagation characteristics include left and right start-up time difference of sidewalls, left and right start-up time difference of arch shoulder to arch waist, propagation time delay from sidewall to arch shoulder to arch waist, propagation time delay from arch shoulder to arch waist to arch crown, propagation time delay throughout the process, local equivalent jacking speed, and equivalent jacking speed throughout the process.

[0039] Leading edge and volume characteristics, including the asynchronous area index of the leading edge, the maximum height difference between the left and right leading edges, the average deviation between the pressure leading edge and the volume leading edge, the maximum deviation between the pressure leading edge and the volume leading edge, the final volume relative error, and the time difference between the exhaust alarm and the crown pressure activation.

[0040] Construction process characteristics include standard deviation of pumping flow rate, coefficient of variation of pumping flow rate, cumulative pump stop time, maximum continuous pump stop time, vibration duration, number of vibrations, average pumping pressure, peak pumping pressure, and standard deviation of pumping pressure.

[0041] Structural response characteristics, including the maximum displacement of the trolley and the maximum difference between left and right displacements;

[0042] Material performance characteristics, including slump, spread, pour temperature, and initial setting time.

[0043] Preferably, the construction of the binary classification quality risk prediction model in step S3 specifically includes:

[0044] Taking a complete casting cycle as the sample unit, the quality status label is defined as follows: This indicates a normal cycle. This represents a risk cycle, in which The cyclic sample number;

[0045] No. Input feature vectors of each cycle , respectively representing pressure response characteristics, propagation timing characteristics, leading edge state characteristics, volumetric verification characteristics, construction and structural response characteristics, and material performance characteristics;

[0046] Output the first The probability of each cycle belonging to the risk class ,in For the first Each cycle represents a probability prediction belonging to the risk category. It is a probability function;

[0047] Using the construction batch number as the grouping variable, K-fold repeated stratified grouping cross-validation was employed.

[0048] Preferably, step S3 further includes ablation analysis and displacement importance evaluation:

[0049] Four feature combinations were set up for ablation analysis:

[0050] M1: Characteristics of the construction process and material properties;

[0051] M2: Characteristics of M1's mean, peak, standard deviation, and growth rate under applied pressure;

[0052] M3: M2 pressure propagation time lag, left and right start time difference, pressure start asynchronous indicators and equivalent jacking speed;

[0053] M4: M3 plus leading edge asynchrony, volume consistency, exhaust alarm and trolley displacement characteristics;

[0054] We used a random forest model and the same construction batch grouping method to perform repeated cross-validation and compared the F1 values ​​of different feature combinations.

[0055] The importance of permutation is defined as ,in For the first The importance of permutations of each feature For the ROC-AUC test before replacement, For the first ROC-AUC after feature permutation.

[0056] Preferably, step S4, which involves calculating individual anomaly scores and constructing a weighted comprehensive dynamic control index, specifically includes:

[0057] Using the normal cycle as the statistical benchmark, the first Individual abnormal scores for each indicator:

[0058]

[0059] in, For the first The individual abnormal score of each indicator, This indicates that the calculated value will be truncated to an interval. Inside, and These are the median and 90th percentile of the normal baseline sample, respectively; For the current loop number The actual value of the indicator; For the first The smallest scale parameter of the indicator;

[0060] The indicators include the asynchronous area of ​​the leading edge, the average deviation between the pressure leading edge and the volume leading edge, the maximum left and right displacement difference, and the maximum continuous pump stop time.

[0061] Comprehensive dynamic control index for:

[0062]

[0063] in, Scoring is given for asynchronous behavior at the leading edge; The score is the deviation between the pressure leading edge and the volume leading edge. The score is based on the maximum difference in displacement between the left and right sides; The score is based on the longest continuous pump stop time.

[0064] The threshold for the comprehensive dynamic control index is taken as the 90th percentile of the comprehensive index for normal cycles.

[0065]

[0066] in, The threshold for the comprehensive dynamic control index; The 90th percentile operation takes the value that is at the 90th percentile after the data is sorted from smallest to largest. To form a comprehensive dynamic control index, For quality status labels.

[0067] Preferably, the establishment of a three-level construction status zoning in step S4 specifically includes:

[0068] The comprehensive dynamic control index and the probability of additional risk will be combined to construct the normal zone, the attention zone, and the adjustment zone:

[0069]

[0070] in, The probability of the out-of-bounds risk in the current cycle. This is the risk probability threshold.

[0071] The method for reverse pouring and reconstructing of tunnel arch wall concrete based on multi-source monitoring provided by this invention has the following advantages compared with the prior art:

[0072] (1) The method for reconstructing the reverse pouring of concrete for tunnel arch walls based on multi-source monitoring of the present invention takes the multi-point pressure response of the template interface as the core information source. Through background noise estimation and adaptive threshold determination, the pressure start time of each measuring point is accurately captured from the pressure time sequence affected by mechanical disturbance and noise pollution. Then, piecewise cubic Hermite interpolation that maintains monotonicity is adopted to transform the sparse and discrete left and right pressure start points into a continuous, monotonic and physically reasonable equivalent filling front curve.

[0073] (2) The method for reconstructing the reverse pouring of tunnel arch concrete based on multi-source monitoring of the present invention transforms the original pressure time sequence into propagation characteristics with clear physical meaning, such as propagation time delay, left and right start-up time difference, equivalent jacking speed, and asynchronous area of ​​the leading edge, rather than simply using pressure amplitude statistics. It has greater identification value than simple pressure amplitude characteristics and effectively solves the technical problem that a single pressure monitoring index is difficult to comprehensively characterize the propagation of the filling leading edge and the left and right unbalanced state.

[0074] (3) The method for reverse pouring and reconstructing of tunnel arch wall concrete based on multi-source monitoring of the present invention combines the comprehensive dynamic control index with the probability of external risk to establish a three-level construction state zoning of normal zone, attention zone and adjustment zone. The comprehensive dynamic control index integrates four indicators with complementary physical meanings: leading edge asynchrony, leading edge-volume deviation, left and right displacement difference and maximum continuous pump stop time, which can comprehensively reflect the filling balance, filling consistency, structural stress balance and construction continuity. The phased control strategy further clarifies the monitoring and control focus of the sidewall stage, arch shoulder-arch waist stage and arch crown stage, realizing the transformation from post-event detection to process perception-risk identification-dynamic regulation. Attached Figure Description

[0075] Figure 1 This is a diagram showing the reconstruction and on-site verification of the equivalent filling front of a representative cycle in the experimental verification.

[0076] Figure 2 This is to verify the deviation distribution between the pressure leading edge in the experiment and the field verification leading edge.

[0077] Figure 3 The diagram shows the representative cyclic pressure response and start-up time in the experimental verification.

[0078] Figure 4 To verify the peak pressure distribution of measuring points at different pouring elevations throughout the entire process in the experiment.

[0079] Figure 5 The results show the comparison of the main monitoring indicators of the normal and risk cycles in the experiment. Among them, (a) is the comparison of the asynchronous area index of the leading edge, (b) is the comparison of the deviation between the pressure leading edge and the volume leading edge, (c) is the comparison of the left and right start time difference between the arch shoulder and the arch waist, and (d) is the comparison of the maximum continuous pump stop time.

[0080] Figure 6 ROC curves were used to identify the on-site quality risks of different models in the experiment.

[0081] Figure 7 This is used to verify the confusion matrix for identifying quality risks in the out-of-bounds prediction process.

[0082] Figure 8 This study aims to verify the ablation analysis of different combinations of on-site monitoring information during the experiment.

[0083] Figure 9 This is a partitioned area for verifying the dynamic control state of reverse casting in the experiment. Detailed Implementation

[0084] The following provides a detailed description of specific embodiments of the present invention. It should be understood that the specific embodiments described herein are for illustrative and explanatory purposes only and are not intended to limit the scope of the invention.

[0085] In this invention, the orientations of "left" and "right" are based on the conventional orientation definitions used in tunnel construction. Specifically, when facing the tunnel's forward direction (the direction of major mileage), the left-hand side is considered the left side, and the right-hand side is considered the right side. This is a commonly understood and followed orientation convention among technicians in the tunnel construction field. The "left" and "right" in this invention are primarily used to distinguish between symmetrically arranged monitoring information on both sides (such as the left and right pressure front edges, or the left and right wall measuring points). The technical essence lies in the comparative relationship and differences between the two sides, rather than the absolute position of the left or right side itself. The key is the relative difference between the left and right sides, not their absolute orientation.

[0086] The present invention provides a method for reconstructing the reverse pouring of concrete for tunnel arch walls based on multi-source monitoring. This method uses the multi-point pressure response at the template interface as the core information source, reconstructing the left and right equivalent filling leading edges by identifying the pressure initiation time at each measuring point. It uses the mapping of pumping volume and design cross-sectional geometric volume as auxiliary verification methods to validate the consistency of the pressure leading edges. Based on this, it extracts process features with physical meaning, such as propagation delay and asynchronous indicators, to construct a quality risk prediction model. Finally, it combines the risk probability with a comprehensive dynamic control index to establish a three-level construction state zoning, achieving a complete closed loop of filling state perception, quality risk identification, and dynamic construction control.

[0087] Specifically, the following steps are included:

[0088] Step S1: Obtain multi-source monitoring data.

[0089] This establishes a data foundation under a unified time reference for subsequent filling front reconstruction and quality risk analysis, ensuring that information on multiple physical quantities such as pressure, flow rate, displacement, and construction events is strictly aligned in time sequence.

[0090] The tunnel arch wall concrete reverse pouring trolley mainly consists of formwork support, hydraulic travel, rotating concrete placing, attached vibratory compaction, exhaust and monitoring systems, and is equipped with several concrete placing positions and exhaust devices. The concrete enters the closed formwork from the lower part of the sidewall and is continuously lifted upwards by pumping.

[0091] The on-site monitoring system consists of pressure sensors, laser displacement sensors, a PLC acquisition module, and a communication module. Several pressure measuring points are arranged on the formwork interface on the trolley, distributed along the left and right sidewalls, the transition area between the arch shoulder and arch waist, and the center of the arch crown. Simultaneously, it monitors construction information such as the trolley's left and right displacement, pumping flow rate, cumulative volume, pumping start / stop, material placement switching, vibration, and exhaust alarms. All equipment uses a unified PLC time reference. The main monitoring variables for a single pouring cycle are as follows:

[0092] (1)

[0093] in, For time, This refers to the instantaneous pumping flow rate; This represents the cumulative pumped volume. for Pressure on the template interface of each measuring point; and These represent the left and right displacements of the trolley, respectively. This represents the total number of pressure measurement points on the template interface.

[0094] Based on the cross-sectional elevation and the location of pressure measuring points, the pouring area is divided into several sections along the elevation direction, including the sidewall area, arch shoulder area, arch waist area, and arch crown area. During construction, the principles of bottom-up, lateral balance, continuous jacking, and zoned vibration are maintained.

[0095] Step S2, filling the leading edge reconstruction.

[0096] Based on acquiring multi-source monitoring data and establishing a unified time benchmark, the macroscopic jacking process of concrete inside the closed template, which cannot be directly observed, is transformed into a quantifiable and traceable equivalent filling front curve, realizing the structured extraction from the original pressure signal to the filling state characteristics.

[0097] S2-1, Pressure start point identification.

[0098] By accurately capturing the instantaneous time of the concrete leading edge reaching the elevation of each sensor from the pressure time series contaminated by mechanical disturbance and background noise, the original pressure signal is transformed into a leading edge arrival event with clear physical meaning, providing key control points for subsequent leading edge reconstruction.

[0099] Concrete is pushed upwards from the bottom of the sidewall. When the front edge of the filling reaches the vicinity of the measuring point, the pressure changes from background fluctuation to continuous response. Therefore, the starting time of the continuous pressure response is defined as the equivalent arrival time of the corresponding elevation.

[0100] S2-1-1, Background noise estimation and baseline correction.

[0101] The stable period before the start of pouring is taken as the background period. The stable period refers to a time window without construction disturbance before the formal pumping begins in each pouring cycle, when the pressure sensors at the formwork interface are installed and in place, but before the concrete enters the closed formwork. The median background pressure is used for baseline correction.

[0102] (2)

[0103] in, For the first Background pressure sequence at each measuring point; The pressure after baseline correction. The physical meaning is that by deducting background components such as initial sensor drift and environmental noise, the zero reference of the pressure signal is unified to the stable state before the concrete arrives. For the first Each measuring point is at The original template interface pressure timing values ​​at each moment (without baseline correction).

[0104] Background noise is characterized by the median absolute deviation:

[0105] (3)

[0106] in, For the first The median absolute deviation of the background pressure at each measuring point This is for median calculation; 1.4826 is the reciprocal of the ratio of the absolute deviation of the median to the standard deviation of the standard normal distribution, used to convert the absolute deviation of the median into an unbiased estimate of the standard deviation. The median is used instead of the mean to reduce the impact of sensor installation disturbances or abnormal spike noise before pouring on the background estimation.

[0107] S2-1-2, Startup threshold and judgment logic.

[0108] No. The pressure trigger threshold for each measuring point is defined as follows: Where 2.8 is the pressure threshold in kPa, an empirical benchmark threshold determined based on the signal-to-noise ratio of the on-site pressure signal and the sensor resolution, used to avoid false triggering due to excessively low start-up thresholds. The moment when the following conditions are first simultaneously met is determined as the pressure start-up moment. :

[0109] (1) Corrected pressure Exceeded the startup threshold;

[0110] (2) Within the next 24 seconds, no less than 90% of the sampling points remain above the threshold, where the 24-second window length corresponds to the shortest duration of the field pressure response from startup to stabilization, used to eliminate instantaneous noise disturbances;

[0111] (3) The maximum pressure increase within the next 90 seconds shall not be less than 4.5 kPa, wherein the 90-second window length is determined based on the minimum time required for the concrete to be lifted through the elevation interval of the adjacent measuring point; 4.5 kPa is the minimum effective pressure increase after the start of the pressure based on the statistical data of the actual field measurement, which is used to distinguish between the actual filling response and the background fluctuation.

[0112] (4) The candidate time is located in the effective pumping stage or near the re-pumping stage, and the proportion of pumping state in the relevant time window is not less than 30%.

[0113] At the same time, the pressure response is required to be continuous and to meet the basic propagation law that the lower measuring point starts before the upper measuring point.

[0114] S2-2, calculate the pressure propagation time delay and jacking speed.

[0115] Pressure propagation time lag and jacking speed.

[0116] For adjacent measuring points on the same side and Pressure propagation delay is defined as:

[0117] (4)

[0118] in, Adjacent measuring points on the same side and Pressure propagation lag between them For the first Pressure start-up time at each measuring point For the first Pressure start-up time at each measuring point.

[0119] Under normal continuous jacking conditions, the upper measuring point should be started later than the lower measuring point, and the effective starting interval between adjacent elevation measuring points should generally not be less than 20 seconds; otherwise, the propagation path should be marked as abnormal. The local equivalent jacking velocity is calculated based on the elevation difference of the measuring points and the propagation time delay.

[0120] (5)

[0121] in, For the local equivalent lifting speed, and These represent the elevations of the two measuring points from the pouring reference surface. This index indicates the macroscopic filling and propulsion speed between two finite elevation measuring points, and is not equivalent to the actual local flow velocity of the concrete.

[0122] The time difference between the left and right pressure start-up of the sidewall is defined as follows:

[0123] (6)

[0124] in, The time difference between the start-up pressure on the left and right sides of the sidewall. The pressure start-up time at the measuring point on the left wall. This is the pressure start-up time at the measuring point on the right wall.

[0125] The time difference between left and right pressure initiation for arched shoulders and arched back is defined as follows:

[0126] (7)

[0127] in, The difference in lateral pressure during shoulder and waist arching is the starting time difference. The pressure start-up moment at the left shoulder-waist measuring point. The pressure start-up moment is the measuring point at the right shoulder-waist.

[0128] The propagation time delay from the sidewall to the transition point between the arch shoulder and the arch waist is:

[0129] (8)

[0130] in, The propagation time lag from the sidewall to the arch shoulder and arch waist.

[0131] The propagation time lag from the shoulder-waist transition to the crown is:

[0132] (9)

[0133] in, The propagation time lag from the shoulder to the waist to the crown is considered. This is the pressure start-up time at the measuring point on the arch.

[0134] The propagation time delay throughout the entire process is:

[0135] (10)

[0136] in, This refers to the time lag in the entire transmission process.

[0137] The corresponding equivalent jacking speed for the entire process is:

[0138] (11)

[0139] in, The equivalent lifting speed throughout the entire process, Elevation of the measuring point at the top of the arch, The elevation of the measuring point on the left wall.

[0140] and Describe the differences in left and right filling in the sidewall area and the abutment-waist area respectively; , and The propagation time delay and local lifting efficiency are described separately for different elevation intervals; and Describe the entire process of concrete propagation from the sidewalls to the arch and the overall lifting speed.

[0141] S2-3, Reconstruct the equivalent filling front edge.

[0142] The obtained sparse and discrete left and right pressure initiation points are transformed into continuous, monotonous and physically reasonable equivalent filling interface curves, realizing a visual representation of the macroscopic lifting state of concrete inside the closed template.

[0143] S2-3-1, Control point construction.

[0144] Based on the pressure initiation time and the installation elevation of the measuring points, left and right pressure leading edge control points were constructed respectively. The left control point is... The control point on the right is .in, This represents the start time of the current effective pumping cycle. and These represent the pressure start-up times at each measuring point on the left and right sides, respectively. and These represent the elevations of each measuring point from the pouring reference surface; For measurement point index, ; This represents the total number of pressure measurement points on a single side (left or right) of the template interface.

[0145] S2-3-2, piecewise cubic Hermite interpolation reconstruction.

[0146] The left and right pressure leading edges are constructed using piecewise cubic Hermite interpolation (PCHIP) that maintains monotonicity:

[0147] (12)

[0148] (13)

[0149] in, and These are the equivalent filling leading edge heights for the left and right sides, respectively. and These represent the time and high-order program sequence for the left and right control points, respectively. PCHIP (Piecewise Cubic Hermite Interpolating Polynomial) interpolation can avoid non-physical fallback at the leading edge under sparse measurement point conditions.

[0150] Mean pressure front is defined as:

[0151] (14)

[0152] in, for The average pressure leading edge height at any given time.

[0153] The pressure front is mainly interpreted up to the start time of the highest measuring point, while the area above the highest measuring point is used for auxiliary judgment in combination with volume and construction events.

[0154] S2-3-3, exception handling mechanism.

[0155] Two reconstruction methods are set up: pure pressure and volume-assisted reconstruction. When key measurement points are missing or the propagation sequence is abnormal, pure pressure leading-edge reconstruction is not performed in this cycle. For complete cycle characteristic analysis, when the local pressure control point is invalid, auxiliary backoff can be performed using the moment when the volume leading-edge reaches the corresponding elevation.

[0156] (15)

[0157] in, For the leading edge of the volume reaches the first At the time of measuring the elevation of each point, For the infimum operation, For time variables, for The height of the leading edge of the volume at that moment. For the first The elevation of each measuring point from the pouring reference surface.

[0158] The control points after rollback still need to satisfy the vertical time-incrementing constraint. Pressure information is mainly used to identify differences in filling between the left and right sides, while volume information is used to assist in missing control points and for consistency verification. The two are not simply weighted continuously.

[0159] S2-4, Verification of Geometric Volume.

[0160] By introducing the cumulative pumping volume as an independent mass conservation constraint, and by inverting the geometric volume of the design section, a volume front and an on-site verification front are constructed. The reconstruction accuracy and consistency of the pressure front are cross-validated to avoid misjudgments caused by pressure monitoring anomalies.

[0161] S2-4-1, Constructing the front edge of the volume.

[0162] Based on the secondary lining design cross-section and effective lining length corresponding to each pouring cycle, a mapping relationship between the leading edge height and the theoretical filling volume is established:

[0163] (16)

[0164] in, For the leading edge height to reach Theoretical filling volume at that time This represents the effective lining length for the current cycle. For elevation The equivalent concrete cross-sectional width after considering over-excavation and formwork installation errors; The equivalent filling front height is the macroscopic height by which the concrete rises from the pouring reference surface within the formwork. The elevation of the measuring point from the pouring reference surface.

[0165] The effective cumulative pumping volume on site is:

[0166] (17)

[0167] in, To calculate the effective cumulative pumped volume on site, i.e., from the start of the circulation to... At any given moment, the cumulative volume of concrete injected into the formwork during pumping. This is the instantaneous pumping flow rate converted to m³ / s; In pumping mode, the value is 1 when pumping is in progress and 0 when pumping is stopped. This is the integration time variable, used for time integration in the calculation of cumulative pumped volume.

[0168] Obtain the front edge of the volume through the inverse function of the geometric volume function:

[0169] (18)

[0170] in, To measure the height of the leading edge; It is the inverse function of the theoretical volume-leading height mapping function; This is the effective cumulative pumping volume on site.

[0171] The theoretical volumes and mapping relationships for different cycles are calculated based on the actual design cross-sections of the corresponding mileages, without using a uniform empirical volume ratio as a substitute. For continuous cycles with the same cross-sectional parameters, piecewise monotonic interpolation relationships can be pre-established to improve computational efficiency.

[0172] S2-4-2, On-site verification of the leading edge construction.

[0173] Based on the volumetric leading edge as a continuous foundation, and combined with the switching of material placement positions, exhaust alarm, and construction completion records, key time nodes are checked to form an on-site verification leading edge. The on-site verification leading edge is based on the volumetric leading edge and obtained by verifying key nodes in conjunction with concrete placement station switching, venting alarms, and construction completion records. Specifically, when the concrete placement station is switched, the corresponding time node is marked on the volumetric leading edge; when the venting alarm is triggered, it is confirmed that the filling near the crown is nearly complete at that moment; the final filling height is constrained in conjunction with the construction completion records, forming a segmented correction on-site verification leading edge. The on-site verification leading edge is only used to evaluate the consistency of the pressure leading edge and does not participate in pressure start-up identification or quality risk model training.

[0174] S2-4-3, Consistency Evaluation Indicators.

[0175] Define three quantitative indicators:

[0176] (1) Mean absolute deviation between the pressure front and the field verification front:

[0177] (19)

[0178] in, The mean absolute deviation between the pressure front and the field verification front is used to evaluate the degree of consistency between the pressure front and the field verification front. This represents the number of sampling points; for The average pressure leading edge height at any given time, for On-site verification of leading edge height at all times. For the first Each sampling time, This is the sampling point number.

[0179] (2) Average deviation between pressure leading edge and volume leading edge:

[0180] (20)

[0181] in, The average deviation between the pressure front and the volume front is used to evaluate the degree of coordination between pressure propagation information and cumulative volume information. for The height of the leading edge at any given moment.

[0182] (3) Relative error of the final cyclic quantity:

[0183] (twenty one)

[0184] in, The relative error of the final volume of the loop is used to determine whether the final volume of the loop is significantly insufficient or excessive; This represents the effective cumulative pumped volume at the end of the current pouring cycle. This represents the theoretical volume of concrete calculated based on the design cross-section for the current cycle. This represents the end time of the current loop.

[0185] Evaluate the consistency between the pressure front and the on-site information. Evaluate the coordination between pressure transmission and volume accumulation. Determine whether the final volume is significantly insufficient or excessive.

[0186] Step S3: Construct a quality risk prediction model.

[0187] By fusing features from multiple sources and training an ensemble learning model with grouped cross-validation, the probability of the current loop belonging to the risk class is quantified, thus achieving a leap from state perception to risk identification.

[0188] S3-1, Partition temporal feature extraction.

[0189] After completing the start-up identification and leading edge reconstruction, six categories of features are extracted on a complete casting cycle basis:

[0190] (1) Pressure Response Characteristics: For each pressure measuring point, the average pressure, peak pressure, pressure standard deviation, and growth rate within 150 seconds after pressure initiation were calculated. The 150-second period was an empirical window length determined based on the time required for the measured pressure response to stabilize from initiation, used to uniformly calculate the pressure growth rate at each measuring point. Based on the single-point characteristics, peak pressure at the sidewall measuring point, peak pressure at the arch shoulder-arch waist transition measuring point, peak pressure at the arch crown measuring point, the maximum pressure growth rate throughout the process, and the average, peak, and standard deviation of the pumping pressure were further constructed. These characteristics respectively characterize the interface pressure level, fluctuation degree, and pressure growth state in different elevation areas.

[0191] (2) Pressure propagation characteristics: including the time difference between the left and right pressure initiation on the sidewalls. 1. Shoulder arching-waist arching left and right pressure start-up time difference Propagation time lag from sidewall to abutment-waist Propagation time lag from shoulder to waist to crown Propagation time lag throughout the entire process Local and full-process equivalent jacking speed and Indicators of asynchronous pressure startup .

[0192] To evaluate the balance of concrete jacking on both sides, the height difference between the left and right leading edges at any given time is defined as:

[0193] (twenty two)

[0194] in, for The height difference between the equivalent filling leading edges on the left and right sides at any given moment. for The equivalent filling leading edge height on the left side at any given moment. for The height of the equivalent filling leading edge on the right side at any given moment.

[0195] To comprehensively reflect the cumulative differences throughout the entire process, the leading edge asynchronous area index is defined as:

[0196] (twenty three)

[0197] in, This refers to the area index of asynchronous leading edges. To monitor the total height of the cross-section; and These are the start and end times when the left and right pressure leading edges are both effective.

[0198] Define the pressure start-up asynchronous index as:

[0199] (twenty four)

[0200] in, Indicators of asynchronous pressure activation; The effective duration of the current pouring cycle is the total duration from the start time of effective pumping to the end time of the current pouring cycle. The time difference between the start-up pressure on the left and right sides of the sidewall. The difference in pressure between the left and right sides of the arched shoulders and arched waist during startup.

[0201] The model does not directly use the absolute start time of each measuring point in order to reduce the impact of cycle preparation time, total duration and overall construction rhythm on the model.

[0202] (3) Leading edge and volumetric characteristics: including asynchronous area indices of the leading edge Maximum elevation difference between the left and right leading edges, and average deviation between the pressure leading edge and the volume leading edge. Maximum deviation between pressure leading edge and volume leading edge; relative error of final volume. The exhaust alarm is activated relative to the crown pressure. Instead of using the absolute alarm time, the time difference between the exhaust alarm activation and the crown pressure activation is used to improve the cross-cycle stability of the characteristic.

[0203] (4) Construction process characteristics: including standard deviation of pumping flow rate, coefficient of variation of pumping flow rate, cumulative pump stop time, maximum continuous pump stop time, duration of vibration, number of vibrations, average value of pumping pressure, peak value and standard deviation.

[0204] (5) Structural response characteristics: including the maximum displacement of the trolley and the maximum difference between the left and right displacements.

[0205] (6) Material performance characteristics: including slump, spread, pour temperature, and initial setting time. Material performance parameters that can be accurately matched with specific cycles through on-site records should be used.

[0206] S3-2, Construction of a binary classification prediction model.

[0207] Based on the on-site quality inspection results, a complete pouring cycle was used as the sample unit. The original quality status included normal, slightly abnormal, and significantly abnormal. When the number of samples in each category was relatively limited, slightly abnormal and significantly abnormal samples were merged into a risk category, constructing a normal-risk binary classification task.

[0208] (25)

[0209] in, For the first Quality status label for each pouring cycle This is the sample number for the pouring cycle.

[0210] No. The input feature vector of each cycle is represented as: superscript These represent pressure response, propagation timing, leading edge state, volumetric verification, construction and structural response, and material performance characteristics, respectively.

[0211] Logistic regression, random forest, and gradient boosting tree were selected for comparison. Comparing multiple models improves the objectivity of model selection by comparing the predictive performance of different algorithms. Logistic regression is used to evaluate the approximately linear relationship between multi-source features and quality risk; random forest is used to characterize nonlinear relationships and feature interactions; and gradient boosting tree is used to test the applicability of boosting ensemble models with the current sample size. The main parameter settings for the three models are shown in Table 1.

[0212] Table 1 Quality Risk Prediction Model and Main Parameters

[0213]

[0214] Model output number The probability that each cycle belongs to the risk class:

[0215] (26)

[0216] in, For the first Each cycle represents a probability prediction belonging to the risk category. Let be a probability function. For the first Quality status label for each pouring cycle For the first The input feature vector of each cycle.

[0217] The final model is comprehensively compared in the following order based on the average F1 score (harmonic mean of precision and recall) in repeated grouped cross-validation, PR-AUC (area under the precision-recall curve), and ROC-AUC (area under the receiver operating characteristic curve), prioritizing the comprehensive identification ability of risk categories, rather than determining the optimal model solely based on ROC-AUC.

[0218] S3-3, Model Training and Cross-Validation.

[0219] Using the construction batch number as the grouping variable, K-fold repeated stratified grouping cross-validation was employed. All pouring cycles within the same construction batch could only be included in either the training set or the test set to avoid similar concrete supplies, equipment conditions, construction personnel, and environmental conditions from being included in both sets simultaneously.

[0220] Cross-validation is repeated multiple times, each time using a different random seed for evaluation. In each iteration, multiple test-fold predictions are obtained through repeated cross-validation, and the out-of-fold risk probability is the average of the corresponding test-fold prediction probabilities. The repeated validation results are expressed as mean and standard deviation, used to evaluate model performance and its sensitivity to data partitioning.

[0221] Risk classification thresholds are searched based on the out-of-bounds predicted probabilities of all cycles. Candidate thresholds that simultaneously meet the following conditions are prioritized: ; Among the candidate thresholds that meet the constraints, the one with the highest balancing accuracy is selected; if the balancing accuracy is the same, the scheme with the higher F1 value and the higher threshold are selected in order to balance the risk of false alarms and the false alarms under normal operating conditions.

[0222] S3-4, Ablation analysis of monitoring information.

[0223] To evaluate the incremental contribution of different monitoring information, the following four feature combinations are set:

[0224] M1: Characteristics of the construction process, such as pumping flow rate, pump shutdown, and vibration, as well as material performance characteristics;

[0225] M2: Pressure response characteristics of M1, including mean, peak, standard deviation, and growth rate of applied pressure;

[0226] M3: M2 pressure propagation time lag, left and right start time difference, pressure start asynchronous indicators and equivalent jacking speed;

[0227] M4: M3 plus leading edge asynchrony, volume consistency, exhaust alarm and trolley displacement characteristics.

[0228] Ablation analysis consistently employed a random forest model and the same construction batch grouping method, repeated multiple times to minimize the impact of model type and single data partitioning on comparisons of different feature combinations. This design was used to examine whether features with clear process physical significance, such as pressure propagation and equivalent leading edge, could provide more stable risk identification gains compared to the original pressure amplitude.

[0229] S3-5, Importance of Repeated Grouping Permutations.

[0230] For the finally selected model, the importance of permutations is calculated on the cross-validation test fold. The models are then randomly shuffled one by one. The characteristics were analyzed, and the changes in ROC-AUC before and after the replacement were compared:

[0231] (27)

[0232] in, For the first The importance of permutations of each feature ROC-AUC was tested before replacement. For the first ROC-AUC after feature permutation, For feature indexing. The larger the value, the higher the model's dependence on that feature. To reduce the randomness of a single partition and permutation, repeated group cross-validation is used, with multiple random permutations performed within each test fold.

[0233] Step S4: Establish dynamic control state partitions.

[0234] To improve the on-site interpretability of the model results, a comprehensive dynamic control index was constructed by selecting factors such as leading edge asynchrony, leading edge-volume consistency, trolley left-right displacement difference, and maximum continuous pump stop time. This index was then combined with the model risk probability to partition the construction state.

[0235] By constructing a comprehensive control index covering filling balance, volume consistency, structural stress, and construction continuity, and establishing a three-level zoning system of normal, watch, and adjustment, a closed-loop risk management system can be achieved.

[0236] S4-1, Calculation of single-item abnormal score.

[0237] Select the asynchronous area index of the leading edge and the average deviation between the pressure leading edge and the volume leading edge. Four indicators were considered: maximum left-right displacement difference, maximum continuous pump stop time, etc. Using normal circulation as the statistical benchmark, individual anomaly scores were calculated based on the median and 90th percentile of each indicator.

[0238] (28)

[0239] in, For the first The individual abnormal score of each indicator, This means that the score for each item will be limited to 0-3. and These are the median and 90th percentile of the normal baseline sample, respectively; For the current loop number The actual value of the indicator; For the first The minimum scale parameter of the indicator can be set according to the degree of variation of the normal reference sample, and is generally taken as 0.1 to 0.3.

[0240] S4-2, Comprehensive Dynamic Control Index.

[0241] Comprehensive dynamic control index Defined as:

[0242] (29)

[0243] in, Scoring is given for asynchronous behavior at the leading edge; The score is the deviation between the pressure leading edge and the volume leading edge. The score is based on the maximum difference in displacement between the left and right sides; The maximum continuous pump stop time is scored. Weights of 0.4, 0.3, 0.2, and 0.1 are assigned empirically based on the distinguishing ability (between-group effect size) of each indicator in normal and risk cycles, with higher weights for indicators with stronger distinguishing ability. In this embodiment, the leading-edge asynchrony indicator has a weight of 0.40, the leading-edge-volume deviation indicator has a weight of 0.30, the left-right displacement difference indicator has a weight of 0.20, and the maximum continuous pump stop time indicator has a weight of 0.10.

[0244] Asynchrony at the leading edge reflects the balance of filling from left to right; deviation between the leading edge and volume reflects the consistency of filling; difference in displacement from left to right reflects uneven stress on the trolley; and the maximum continuous pump stop time reflects the continuity of construction. The comprehensive index is only used for construction status evaluation and is not used in quality risk model training. The control threshold is taken as the 90th percentile of the comprehensive index for a normal cycle.

[0245] (30)

[0246] in, The threshold for the comprehensive dynamic control index (taken as the comprehensive index of the normal cycle) The 90th percentile value is used to classify control states. The 90th percentile operation takes the value that is at the 90th percentile after the data is sorted from smallest to largest. To form a comprehensive dynamic control index, For quality status labels.

[0247] In practical engineering applications, relevant statistical parameters and control thresholds should be updated based on historical normal cycles.

[0248] S4-3, Dynamic control state partitioning.

[0249] The comprehensive dynamic control index and the probability of additional risk will be combined to construct the normal zone, the attention zone, and the adjustment zone:

[0250] (31)

[0251] in, The probability of the out-of-bounds risk in the current cycle. This is the risk probability threshold.

[0252] During phased control, the focus is on the pressure response of the measuring points on the left and right side walls, the time difference between left and right starts, and the displacement difference of the trolley during the side wall phase; the focus is on the time difference between left and right arch shoulder and arch waist measuring points, the propagation time lag, and the asynchronous movement of the leading edge during the arch shoulder and arch waist phase; and the focus is on the upper filling status by combining the pressure of the arch waist measuring points, the exhaust alarm, and the cumulative volume. Specifically, the side wall phase corresponds to the side wall area (0–2.20m), the arch shoulder and arch waist phase corresponds to the arch shoulder and arch waist areas (2.20–5.90m), and the arch waist phase corresponds to the arch waist area (above 5.90m).

[0253] In the normal zone, the existing construction parameters can be maintained; in the monitoring zone, the pumping, pumping stop, left and right material supply, vibration and venting status should be checked; in the adjustment zone, the pumping rhythm, material placement sequence and vibration measures should be adjusted in a timely manner according to the source of the abnormality.

[0254] The parameters and thresholds mentioned above are established based on the monitoring conditions upon which this method is based. For other engineering applications, they need to be recalibrated in conjunction with the cross-section, trolley, concrete properties, and sensor layout.

[0255] Experimental verification:

[0256] This verification is based on the reverse continuous pouring construction of the arch wall concrete of a certain tunnel. The tunnel is 3518m long and has a maximum burial depth of about 200m. It adopts a composite lining structure and a reverse continuous pouring process for the arch wall concrete.

[0257] 120 complete pouring cycles with construction records and quality inspections were selected as the research object, with an effective lining length of 12m per cycle. Based on the construction date, concrete supply batch, and continuous equipment operation status, the cycles were divided into 12 construction batches, with 10 cycles in each batch. The pouring cycle served as the analysis and quality labeling unit, while the construction batch served as the model training and validation grouping unit.

[0258] The on-site concrete pouring trolley, consisting of a 12m long arch wall concrete reverse pouring system, mainly comprises formwork support, hydraulic travel, rotating concrete placing, attached vibratory compaction, venting, and monitoring systems. It has 10 concrete placing stations and 5 venting devices. Based on the cross-sectional elevation and pressure measuring point locations, the pouring area is divided into: sidewall area (0–2.20m), arch shoulder area (2.20–4.40m), arch waist area (4.40–5.90m), and arch crown area (5.90–7.20m). During construction, the pumping flow rate is mainly controlled between 0.18 and 0.42 m³ / min.

[0259] The on-site monitoring system consists of pressure sensors, laser displacement sensors, a PLC acquisition module, and a communication module. Five pressure measurement points are set up on the template interface on the trolley (as shown in Table 2), simultaneously monitoring construction information such as the trolley's left and right side displacement, pumping flow rate, cumulative volume, pumping start / stop, material placement switching, vibration, and exhaust alarms. Different equipment uses a unified PLC time reference.

[0260] Table 2. Layout of pressure measuring points on the template interface

[0261]

[0262] Based on the on-site quality inspection results, the 120 cycles were divided into two categories: normal and risky. Among them, there were 74 normal cycles and 46 risky cycles. Both model training and validation adopted a construction batch grouping method to avoid data leakage caused by similar construction conditions.

[0263] (a) Front edge reconstruction results and consistency verification.

[0264] Figure 1 The left and right pressure leading edges, average pressure leading edges, volume leading edges, and field verification leading edges of representative cycles are given. Each leading edge rises monotonically during the pouring process, and the overall trend of change is consistent. The left and right leading edges show some separation in the arch shoulder-arch waist stage, and gradually converge as they approach the arch crown, which can reflect the upward lifting process of concrete.

[0265] Figure 2The distribution of absolute deviations between the pressure front and the field verification front is further presented. Out of 120 cycles, 118 met the pure pressure front reconstruction requirements, with an effective reconstruction rate of 98.3%. The median absolute deviation was 0.072 m, and the 90th percentile was 0.094 m, indicating that a finite number of pressure measurement points can stably characterize the macroscopic filling and propulsion process. This deviation reflects the consistency between the pressure front and the field verification information, and is not a direct measurement error of the actual three-dimensional flow interface.

[0266] (ii) Pressure propagation characteristics of different pouring zones.

[0267] Figure 3 The data shows that pressure measuring points P1 to P5 were activated in an overall ascending order, consistent with the construction process of concrete being continuously jacked from the sidewalls to the arch crown. There were varying degrees of time lag between the activation of P1 and P2, and between P3 and P4, with the asynchrony being more pronounced in the arch shoulder-arch waist stage. Local pressure drops occurred during pump shutdowns, and pressure increased again after pumping resumed, reflecting the correlation between pressure response and construction status.

[0268] Figure 4 The results show significant differences in peak pressure throughout the entire process at different elevations: 61–64 kPa at the sidewall monitoring points, approximately 41–45 kPa at the abutment-waist transition monitoring points, and approximately 14–28 kPa at the crown monitoring points. The peak pressure generally decreases with increasing monitoring point elevation, but the dispersion is relatively large at the crown monitoring points. It should be noted that these results represent the peak pressure throughout the entire process at fixed monitoring points and cannot be directly used as the control pressure for phased construction.

[0269] (III) Comparison between normal and risky operating conditions.

[0270] Figure 5 (a) to (d) show the comparison results of the main monitoring indicators between the normal and risk cycles. In the risk group, the leading edge asynchrony area is significantly increased. , ,in Corrected by the Benjamini-Hochberg method The value is used to control the false discovery rate in multiple comparisons; The nonparametric effect size represents the probability that a randomly selected pair of data from two groups is greater than the probability that the data from the two groups is less ...

[0271] Similarly, the difference in left-right starting time difference between arched shoulders and arched back is significant. , This indicates that the area is a sensitive stage of uneven filling between the left and right sides. The maximum continuous pump stop time also showed significant differences between groups. , Of these, the proportion of normal cycles without pump shutdowns was 72.97%, while the proportion of risk cycles was only 32.61%, indicating a strong correlation between the decline in construction continuity and the risk status.

[0272] In contrast, the average deviation between the pressure front and volume front was not significantly different. , This indicates that the indicator is more suitable for process consistency verification than as a sole criterion for quality risk assessment. In summary, asynchronous filling on the left and right sides, the time difference between the start-up of the arch shoulder and the arch waist, and the continuity of pump shutdown are the core process characteristics that distinguish normal and risky operating conditions.

[0273] (iv) Results of quality risk prediction.

[0274] Figure 6 ROC curves for three models are presented. The average ROC-AUCs for logistic regression, random forest, and gradient boosting tree are 0.766±0.018, 0.764±0.012, and 0.744±0.015, respectively. Logistic regression and random forest show similar overall performance, while gradient boosting tree performs slightly worse, indicating that increasing model complexity does not provide a significant advantage with the current sample size. Based on a comprehensive evaluation of F1, PR-AUC, and ROC-AUC, random forest will be used for feature importance and ablation analysis in subsequent analyses.

[0275] Figure 7 The outlier prediction results are presented when the risk probability threshold is 0.40. 52 out of 74 normal cycles were correctly identified, with a normal identification rate of 70.3%; 33 out of 46 risky cycles were correctly identified, with a risk recall rate of 71.7%. The overall accuracy is 70.8%, the risk precision is 60.0%, the F1 score is 65.3%, and the balanced accuracy is approximately 71.0%. The model of this invention shows a relatively balanced identification of the two types of cycles, but there are still 22 false positives in normal cycles and 13 missed detections in risky cycles. Therefore, the method of this invention is mainly applied to construction quality risk screening and auxiliary early warning, rather than replacing on-site quality inspection.

[0276] (v) Ablation analysis of monitoring information.

[0277] Figure 8 The ablation results for different combinations of monitoring information are presented. When only construction process and material characteristics are used, the F1 value of M1 is 0.577±0.012; after adding pressure amplitude characteristics, M2 decreases to 0.525±0.021, indicating that simply increasing the statistical quantities such as the mean and peak pressure did not improve the model performance.

[0278] After incorporating pressure propagation time lag, left and right start-up time difference, and equivalent lifting speed, the F1 value of M3 increased to 0.612±0.039; after further integrating the information of leading edge, volume, exhaust, and trolley displacement, M4 reached 0.635±0.015.

[0279] Therefore, the performance improvement of the model in this invention mainly stems from transforming the original pressure time history into features with process physical meaning, such as propagation delay, left-right asynchrony, and equivalent leading edge, rather than simply increasing the pressure amplitude variable. This is also the main advantage of the multi-source monitoring method of this invention compared to single pressure monitoring.

[0280] (vi) Dynamic control state zoning and phased control strategy.

[0281] According to the method described in step S4 of this invention, the comprehensive dynamic control index and the external risk probability are jointly used for state partitioning, and the results are shown in [the table below]. Figure 9 The threshold for the comprehensive dynamic control index of the project is... The risk probability threshold is . Figure 9 This indicates that normal cycles are mainly concentrated in the low control index and low risk probability areas; as both indicators increase, the proportion of risky cycles increases significantly. When both the comprehensive dynamic control index and risk probability increase, and both indicators exceed their respective thresholds simultaneously, the cycle mainly enters the adjustment zone; while when only the control index exceeds the threshold, it enters the attention zone, which is consistent with the three-level partitioning logic of this invention.

[0282] The key control points for different construction stages can be summarized as follows: In the sidewall stage, focus on the pressure response of P1 and P2, the left and right start time difference and the left and right displacement difference of the trolley; in the arch shoulder-arch waist stage, focus on the start time difference of P3 and P4, the propagation time lag and the asynchronous nature of the leading edge; in the arch crown stage, focus on judging the upper filling status by combining the P5 pressure response, exhaust alarm and cumulative volume.

[0283] In the normal zone, the current construction parameters can be maintained; in the monitoring zone, the pumping, pumping stop, left and right material supply, vibration and venting status should be checked; in the adjustment zone, the pumping rhythm, material placement sequence and vibration measures should be adjusted in a timely manner according to the source of the abnormality.

[0284] The effectiveness of the method of the present invention was verified based on multi-source monitoring data from 120 reverse casting cycles:

[0285] (1) The multi-point template interface pressure start-up sequence can effectively characterize the macroscopic jacking process of concrete. 118 out of 120 cycles met the pure pressure leading edge reconstruction requirements, with an effective reconstruction rate of 98.3%; the median absolute deviation between the pressure leading edge and the field verification leading edge was 0.072m, and the 90th percentile was 0.094m.

[0286] (2) The main manifestations of the risk cycle are the asynchrony of the left and right leading edges, the increased start-up time difference between the arch shoulder and the arch waist, and the decrease in the continuity of pump shutdown.

[0287] (3) Ablation analysis shows that pressure propagation with physical significance, left-right asynchrony, and equivalent leading edge features are more valuable for identification than simple pressure amplitude; after integrating relevant features, the F1 value increased from 0.525 to 0.635. By further combining the comprehensive dynamic control index and risk probability, a three-level state zoning of normal, attention, and adjustment can be established to realize the connection between filling status perception, quality risk identification, and construction dynamic control.

[0288] The above embodiments are merely preferred examples of the present invention and are not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Therefore, any simple modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention should fall within the protection scope of the present invention.

Claims

1. A multi-source monitoring-based tunnel arch wall concrete reverse pouring filling reconstruction method, characterized in that, Includes the following steps: Step S1: Obtain multi-source monitoring data. The multi-source monitoring data includes at least the pressure timing, pumping flow rate, cumulative pumping volume, pumping start / stop status, left and right displacement of the trolley, material placement station switching signal, vibration status, and exhaust alarm signal of each pressure measuring point on the template interface. Step S2: Estimate background noise and correct baseline for the pressure timing of each pressure measuring point, and identify the pressure start-up time of each measuring point; calculate the pressure propagation delay, equivalent jacking speed, and left and right start-up time difference based on the pressure start-up time; reconstruct the left and right equivalent filling front edges using the pressure start-up time and measuring point elevation as control points; introduce volume-geometric volume mapping to construct the volume front edge, and use the volume front edge to perform consistency verification on the reconstructed pressure front edge; The consistency check includes: Establish a mapping relationship between the leading edge height and the theoretical filling volume based on the design cross-section: ,in For the leading edge height to reach Theoretical filling volume at that time This represents the effective lining length for the current cycle. For elevation The equivalent concrete section width at that location. To the equivalent filling leading edge height, The elevation of the measuring point from the pouring reference surface; On-site effective cumulative pumping volume : ,in This is the converted instantaneous pumping flow rate. In pumping mode; Front edge of the square: ,in To measure the height of the leading edge, It is the inverse function of the mapping relationship; Based on the volumetric leading edge and combined with construction events, a field verification leading edge is formed. The average absolute deviation between the pressure front and the field verification front, the average deviation between the pressure front and the volume front, and the relative error of the final volume of the cycle are calculated to evaluate the consistency of the reconstructed front. Step S3: Extract multivariate process features based on a complete casting cycle. These features include at least pressure response features, pressure propagation features, leading edge and volume features, construction process features, structural response features, and material performance features. Construct a binary classification quality risk prediction model and obtain the external risk probability of each cycle through grouped cross-validation. Step S4: Calculate the individual abnormal score based on the statistical distribution of the normal cycle, construct a weighted comprehensive dynamic control index, and combine the comprehensive dynamic control index with the external risk probability to establish a three-level construction status zoning.

2. The method for reverse casting, filling, and reconstruction of tunnel arch wall concrete based on multi-source monitoring according to claim 1, characterized in that, The step S2, which identifies the pressure start-up time of each measuring point, specifically includes: The stable period before the start of pouring is taken as the background period. The median background pressure and background noise level are calculated for each pressure measuring point. The median background pressure is used to perform baseline correction on the original pressure time series to obtain the corrected pressure time series. The pressure start-up threshold is calculated based on the background noise level of each measuring point. The moment when the correction pressure first exceeds the start-up threshold and continues to increase for a preset duration is determined as the pressure start-up moment of that measuring point.

3. The method for reverse casting, filling, and reconstruction of tunnel arch wall concrete based on multi-source monitoring according to claim 2, characterized in that, In step S2, reconstructing the left equivalent filling front edge and the right equivalent filling front edge includes: The left control point is constructed as The control points on the right are constructed as follows ,in, This represents the start time of the current effective pumping cycle. and These represent the pressure start-up times at each measuring point on the left and right sides, respectively. and These represent the elevations of each measuring point from the pouring reference surface; For measurement point index, ; This represents the total number of pressure measurement points on a single-sided template interface. The left and right pressure leading edges are constructed using piecewise cubic Hermite interpolation that maintains monotonicity. ; ; in, and These are the equivalent filling leading edge heights for the left and right sides, respectively. and These represent the time and high-order program sequence for the left and right control points, respectively. Mean pressure front is defined as: ,in for The average pressure leading edge height at any given time.

4. The method for reverse casting, filling, and reconstruction of tunnel arch wall concrete based on multi-source monitoring according to claim 3, characterized in that, The multivariate process features extracted in step S3 include: Pressure response characteristics, including the average pressure, peak pressure, pressure standard deviation, and growth rate within the preset time window after pressure start-up at each measuring point; Pressure propagation characteristics include left and right start-up time difference of sidewalls, left and right start-up time difference of arch shoulder to arch waist, propagation time delay from sidewall to arch shoulder to arch waist, propagation time delay from arch shoulder to arch waist to arch crown, propagation time delay throughout the process, local equivalent jacking speed, and equivalent jacking speed throughout the process. Leading edge and volume characteristics, including the asynchronous area index of the leading edge, the maximum height difference between the left and right leading edges, the average deviation between the pressure leading edge and the volume leading edge, the maximum deviation between the pressure leading edge and the volume leading edge, the final volume relative error, and the time difference between the exhaust alarm and the crown pressure activation. Construction process characteristics include standard deviation of pumping flow rate, coefficient of variation of pumping flow rate, cumulative pump stop time, maximum continuous pump stop time, vibration duration, number of vibrations, average pumping pressure, peak pumping pressure, and standard deviation of pumping pressure. Structural response characteristics, including the maximum displacement of the trolley and the maximum difference between left and right displacements; Material performance characteristics, including slump, spread, pour temperature, and initial setting time.

5. The method for reverse casting, filling, and reconstruction of tunnel arch wall concrete based on multi-source monitoring according to claim 4, characterized in that, The construction of the binary classification quality risk prediction model in step S3 specifically includes: Taking a complete casting cycle as the sample unit, the quality status label is defined as follows: This indicates a normal cycle. This represents a risk cycle, in which The cyclic sample number; No. Input feature vectors of each cycle , respectively representing pressure response characteristics, propagation timing characteristics, leading edge state characteristics, volumetric verification characteristics, construction and structural response characteristics, and material performance characteristics; Output the first The probability of each cycle belonging to the risk class ,in For the first Each cycle represents a probability prediction belonging to the risk category. It is a probability function; Using the construction batch number as the grouping variable, K-fold repeated stratified grouping cross-validation was employed.

6. The method for reverse casting, filling, and reconstruction of tunnel arch wall concrete based on multi-source monitoring according to claim 5, characterized in that, Step S3 also includes ablation analysis and displacement importance evaluation: Four feature combinations were set up for ablation analysis: M1: Pumping flow rate, pump shutdown, vibration and material properties; M2: Characteristics of M1's mean, peak, standard deviation, and growth rate under applied pressure; M3: M2 pressure propagation time lag, left and right start time difference, pressure start asynchronous indicators and equivalent jacking speed; M4: M3 plus leading edge asynchrony, volume consistency, exhaust alarm and trolley displacement characteristics; We used a random forest model and the same construction batch grouping method to perform repeated cross-validation and compared the F1 values ​​of different feature combinations. The importance of permutation is defined as ,in For the first The importance of permutations of each feature For the ROC-AUC test before replacement, For the first ROC-AUC after feature permutation.

7. The method for reverse casting, filling, and reconstruction of tunnel arch wall concrete based on multi-source monitoring according to claim 6, characterized in that, Step S4, which calculates individual anomaly scores and constructs a weighted comprehensive dynamic control index, specifically includes: Using the normal cycle as the statistical benchmark, the first Individual abnormal scores for each indicator: ; in, For the first The individual abnormal score of each indicator, This indicates that the calculated value will be truncated to an interval. Inside, and These are the median and 90th percentile of the normal baseline sample, respectively; For the current loop number The actual value of the indicator; For the first The smallest scale parameter of the indicator; The indicators include the asynchronous area of ​​the leading edge, the average deviation between the pressure leading edge and the volume leading edge, the maximum left and right displacement difference, and the maximum continuous pump stop time. Comprehensive dynamic control index for: ; in, Scoring is given for asynchronous behavior at the leading edge; The score is the deviation between the pressure leading edge and the volume leading edge. The score is based on the maximum difference in displacement between the left and right sides; The score is based on the longest continuous pump stop time. The threshold for the comprehensive dynamic control index is taken as the 90th percentile of the comprehensive index for normal cycles. ; in, The threshold for the comprehensive dynamic control index; The 90th percentile operation takes the value that is at the 90th percentile after the data is sorted from smallest to largest. To form a comprehensive dynamic control index, For quality status labels.

8. The method for reverse casting, filling, and reconstructing of tunnel arch wall concrete based on multi-source monitoring according to claim 7, characterized in that, The establishment of a three-level construction status zoning in step S4 specifically includes: The comprehensive dynamic control index and the probability of additional risk will be combined to construct the normal zone, the attention zone, and the adjustment zone: ; in, The probability of the out-of-bounds risk in the current cycle. This is the risk probability threshold.

Citation Information

Patent Citations

  • Tunnel arch wall reverse construction parameter optimization method and system based on reinforcement learning

    CN121562432A

  • Construction method of cast-in-place concrete pile

    JP2022165705A