BIM (Building Information Modeling)-based large-span post-tensioned bonded prestressed beam construction method

Abstract

The invention discloses a BIM (Building Information Modeling)-based large-span post-tensioned bonded prestressed beam construction method. The BIM-based large-span post-tensioned bonded prestressed beam construction method comprises the following steps: collecting a vibration signal of a working rib and a matt rib reference signal of a matt rib; inputting the dummy rib reference signal into an adaptive filtering algorithm to which phase preserving constraint is applied, and performing noise cancellation processing on the working rib vibration signal to generate a purified pressure wave signal; performing multi-scale decomposition on the purified signal, extracting propagation time delay of the direct wave based on coherence analysis among components, and generating a phase-time delay characteristic matrix; and based on pre-configured geometric parameters of the BIM three-dimensional model, a real-time stress distribution field is calculated through inversion according to the phase-time delay characteristic matrix, and the real-time stress distribution field is synchronously updated to the BIM three-dimensional model. According to the method, the phase distortion problem under noise interference is solved through a physical reference channel and phase preserving filtering, the accuracy of time delay extraction is improved through a multi-path separation technology, and high-precision and high-robustness real-time monitoring of the prestress is achieved.

Description

Technical Field

[0001] The present invention belongs to the field of civil engineering construction monitoring, and in particular to a BIM-based large-span post-tensioned bonded prestressed beam construction method. Background Art

[0002] Post-tensioning bonded prestressing technology is the core process for realizing large-span prestressed concrete structures, and its construction quality directly determines the long-term safety and service life of the structure. During the post-tensioning process, the prestressed tendons will experience significant prestress loss from the tensioning end to the anchoring end due to factors such as friction between the prestressed tendons and the duct wall and the deviation of the prestressed tendon path, resulting in a deviation between the actual effective prestress distribution along the line and the design value. If this deviation exceeds the allowable range, it will at least affect the normal use function of the structure, causing excessive deflection or cracks, and at worst may induce structural damage. Therefore, real-time and accurate monitoring and feedback control of the actual stress distribution of prestressed tendons during the construction phase is a key link in ensuring the construction quality and fundamental safety of large-span structures, and has engineering significance and technical value.

[0003] Currently, monitoring the post-tensioning construction process primarily relies on a combination of theoretical calculations and on-site fixed-point measurements. In theoretical calculations, engineering practice generally uses preset friction coefficients and pipe deflection coefficients based on relevant specifications to estimate prestress losses along the prestressing path. In on-site measurements, mainstream techniques include installing pressure sensors at the tensioning and anchoring ends to monitor the tensioning force of the jacks and the retraction of the anchors. Strain gauges are also attached to the concrete surface of key beam sections to indirectly infer the stress state of the prestressed tendons by measuring the strain in the concrete. Some research has begun exploring dynamic nondestructive testing techniques based on wave propagation principles. These techniques apply an excitation at one end of the prestressed tendon and measure the response signal at the other end to analyze the stress state. In terms of data processing, collected data is often post-processed offline, and stress conversion relies on simplified linear elastic material assumptions. Construction control primarily relies on open-loop control based on a pre-defined tensioning plan, using dual control indicators of tension force and elongation to determine whether tensioning is in place.

[0004] However, existing technologies still face several specific and profound technical bottlenecks when addressing the dynamic and nonlinear challenges of complex construction sites. These bottlenecks primarily focus on the precise perception of signal-level information and the accurate inversion of physical states. In summary, the core challenges facing existing technologies are the inability to preserve signal phase information, which is crucial for accurate measurement, in the presence of strong noise interference, and the difficulty in effectively distinguishing the direct wave path that carries key time delay information in complex wave fields. Summary of the Invention

[0005] The purpose of the invention is to provide a BIM-based large-span post-tensioned bonded prestressed beam construction method to solve the above-mentioned problems existing in the prior art.

[0006] Technical solution: A BIM-based construction method for large-span post-tensioned bonded prestressed beams, including:

[0007] Collecting the working reinforcement vibration signal of the working reinforcement and the dumb reinforcement reference signal of the dumb reinforcement;

[0008] The noise of the working rib vibration signal is canceled by using the dumb rib reference signal to generate a purified pressure wave signal.

[0009] Analyze and purify the pressure wave signal, extract the propagation delay and generate the phase-delay characteristic matrix;

[0010] Based on the phase-time delay characteristic matrix and the geometric parameters of the preconfigured BIM 3D model, the real-time stress distribution field is inversely calculated and updated to the BIM 3D model.

[0011] Beneficial effect: The present invention solves the phase distortion problem under noise interference through physical reference channels and phase-preserving filtering, and improves the accuracy of time delay extraction through multi-path separation technology, realizing high-precision and high-robustness real-time monitoring of prestress. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 A flowchart of the steps of a BIM-based construction method for large-span post-tensioned bonded prestressed beams provided in an embodiment of the present application.

[0013] Figure 2 A flowchart of the steps for applying phase-hold constraints provided in an embodiment of the present application.

[0014] Figure 3 A flowchart of the step size determination process provided in an embodiment of the present application.

[0015] Figure 4 A flowchart of the steps for generating an adaptive frequency sweep excitation signal provided in an embodiment of the present application. DETAILED DESCRIPTION

[0016] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0017] It should be noted that the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or apparatus that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units that are not explicitly listed or are inherent to these processes, methods, products or apparatus.

[0018] Research has discovered that phase information is distorted during noise filtering. The accuracy of dynamic measurement methods based on wave propagation relies heavily on precise measurement of the wave packet propagation delay, which in turn depends on accurate analysis of the signal's instantaneous phase. Construction sites are rife with strong common-mode noise generated by mechanical vibrations, electromagnetic equipment, and other sources, necessitating filtering of the acquired signals. However, traditional filtering algorithms (such as standard LMS adaptive filtering or bandpass filtering) are designed to minimize signal amplitude errors. While filtering out noise, they inevitably introduce nonlinear perturbations to the signal's phase spectrum, resulting in phase distortion. This distortion undermines the physical basis for subsequent delay calculations, creating a fundamental technical contradiction: filtering operations designed to improve the signal-to-noise ratio compromise the most critical information required for measurement, leading to persistent bottlenecks in measurement accuracy. Furthermore, there's the issue of interference from multipath propagation within complex structures. When the excitation wave propagates within the complex acoustic waveguide composed of prestressed tendons, grouting, and corrugated pipes, in addition to the direct wave propagating along the shortest path, a large number of reflected waves are also generated due to interface reflection and scattering. These multipath signals are superimposed at the receiving end, causing the received waveform to exhibit multiple complex peaks. Simple time-domain peak detection methods can easily misjudge the reflected wave, which may have stronger energy, as the first wave (i.e., the direct wave), leading to serious errors in the calculation of propagation delay. These errors are directly transmitted to the final stress inversion results, causing misjudgment of the structural state.

[0019] like Figure 1 As shown in the figure, a BIM-based construction method for large-span post-tensioned bonded prestressed beams is proposed, which includes the following steps:

[0020] The working reinforcement vibration signal of the working reinforcement and the dumb reinforcement reference signal of the dumb reinforcement are collected.

[0021] In this embodiment, the working reinforcement refers to the prestressed reinforcement that actually bears the tensile force in the prestressed structure and serves as a wave propagation medium. The dumb reinforcement refers to a prestressed reinforcement that is identical or substantially identical to the working reinforcement in terms of path, material, diameter, etc., but deliberately does not apply tensile force. Its function is to serve as a physical reference channel for measuring the environmental noise and vibration interference of the construction site that is shared with the working reinforcement, that is, common mode noise. In other words, the working reinforcement refers to the steel bars that participate in the structural stress, such as the main reinforcement, distribution reinforcement, stirrups, etc. in structural components such as beams, slabs, columns, and walls, all of which are working reinforcement. The dumb reinforcement refers to the steel bars that do not participate in direct stress, such as the negative bending moment reinforcement in the slab, the frame reinforcement in the beam web, the structural reinforcement in the column, etc.

[0022] Specifically, this step involves optimizing sensor placement based on BIM data. For example, by reading a Building Information Model (BIM) .ifc file, the three-dimensional spatial curve equation P(s) = [x(s), y(s), z(s)] of the prestressed tendon and the beam's geometric properties can be parsed. (x, y, z) represents the three-dimensional spatial coordinates, and s represents the arc length parameter along the tendon path. Based on this, a curvature analysis algorithm can be used to identify potential stress concentration areas along the tendon path, such as locations where the curvature κ exceeds a certain threshold (e.g., 0.02 / m). Accelerometers are then prioritized at these key locations, as well as representative locations such as beam endpoints and mid-span. To ensure that the excitation signal avoids strong noise frequencies at the construction site and improve the signal-to-noise ratio, a period of on-site background noise (e.g., 30 seconds) can be collected by sensors before formal acquisition. Fast Fourier transform (FFT) analysis is performed on this noise signal to obtain the noise power spectrum N(f). Based on this power spectrum, an adaptive frequency sweep excitation signal can be generated. The generation logic is: in the frequency band with large noise power spectrum amplitude, the sweep rate is accelerated; in the clean frequency band with small noise, the sweep rate is slowed down, so that more signal energy is injected into the frequency band with high signal-to-noise ratio. The adaptive sweep excitation signal generated above is applied to the jack or anchor at the tensioning end through the piezoelectric exciter; at the same time, the vibration signals x of multiple measuring points on the working reinforcement are synchronously collected at a high sampling rate (for example, 10kHz). i (t) and the reference signal r at the corresponding position on the dummy rib i (t). These collected geometric data, time domain signals, noise spectra, etc. together constitute a structured multi-source data set, providing input for subsequent processing.

[0023] The dumb rib reference signal is used to perform noise cancellation processing on the working rib vibration signal to generate a purified pressure wave signal.

[0024] In this embodiment, the purified pressure wave signal refers to a signal that mainly reflects the propagation characteristics of the excitation signal along the prestressed tendon after the common-mode noise component is removed from the original working tendon vibration signal. Specifically, an adaptive filtering algorithm is used, and the reference signal r(t) from the dummy tendon is used as the input of the adaptive filter. The filter continuously adjusts its own weights to generate the best estimate of the common-mode noise, and subtracts this estimate from the working tendon signal x(t) to obtain the purified signal y(t). By combining differential measurement and adaptive processing, the traditional noise processing paradigm is changed, and the noise is transformed from estimation to actual measurement, thereby improving the accuracy of noise cancellation. Noise interference can be eliminated to the greatest extent possible without damaging the phase information of the useful signal.

[0025] The pressure wave signal is analyzed and purified, the propagation delay is extracted, and a phase-delay characteristic matrix is ​​generated.

[0026] In this embodiment, the propagation delay refers to the time required for the purified pressure wave signal to propagate between two different measuring points on the working rib. The phase-delay characteristic matrix is ​​a matrix T=[τ ij ], where τ ij represents the time delay of the signal propagating from measurement point i to measurement point j. Specifically, phase-locked wave packet tracking analysis technology is used. Since the signal-to-noise ratio of the purified signal has been improved, its instantaneous phase can be accurately extracted through methods such as the Hilbert transform. By analyzing the gradient of the phase variation with frequency (i.e., group delay), the precise propagation time of the wave packet between different measurement points can be calculated. To address the reflection and multipath problems that may arise when waves propagate in the structure, multipath separation algorithms can also be included. For example, based on the coherence differences of the signals along different paths, direct waves and reflected waves can be distinguished, and the multipath information can be fused to obtain a robust propagation delay estimate.

[0027] Based on the phase-time delay characteristic matrix and the geometric parameters of the preconfigured BIM three-dimensional model, the real-time stress distribution field is inversely calculated and updated to the BIM three-dimensional model.

[0028] In this embodiment, the real-time stress distribution field refers to the continuous function or discrete vector σ(s, t) of the stress along the entire working rib path. Specifically, an inversion model is established based on the wave velocity-stress relationship. The propagation speed (wave velocity) of sound waves in solid media is related to the stress on the medium. Generally, the greater the stress, the faster the wave velocity. Using this physical principle, a set of equations can be established, where each equation represents the relationship between a pair of measuring points: τ ij =∫ pathds / v(σ, T), that is, the measured time delay is equal to the integral of the path length divided by the wave velocity. The wave velocity model v(σ, T) here preferably takes into account the coupled effects of stress σ and temperature T, and exhibits a nonlinear relationship. By solving this set of inverse equations, the stress values ​​at each point on the prestressed tendon can be obtained. The calculated real-time stress distribution field data is mapped back to the BIM model, and the corresponding properties of the prestressed tendon unit are updated (for example, creating a property set named Pset_TendonStress), realizing digital twin visualization and recording of the construction status.

[0029] According to one aspect of the present application, the collection of the dumb rib reference signal originates from a physical dumb rib that has the same path and material as the working rib but is not subject to tension. The physical dumb rib is specifically used as a noise reference channel to measure common-mode noise. The noise cancellation processing is specifically as follows: the dumb rib reference signal is input into an adaptive filtering algorithm to generate a dynamic estimate of the common-mode noise, and the dynamic estimate is subtracted from the working rib vibration signal to generate a purified pressure wave signal.

[0030] In this embodiment, the motivation for setting up dummy bars is that the impact of common-mode noise such as environmental vibration and electromagnetic interference at the construction site on working bars and dummy bars is theoretically the same. Since the dummy bars do not bear tension, the main component of the signal r(t) collected by them is the common-mode noise n common (t) and a portion of random noise n random_ref (t). The signal x(t) of the working rib is the propagated pressure wave signal s prop (t), common mode noise nc ommon (t) and random noise n random Therefore, the dumb signal r(t) provides the adaptive filter with a high-quality reference input that is highly correlated with the noise to be eliminated.

[0031] According to one aspect of the present application, the adaptive filtering algorithm controls the generation of dynamic estimation by imposing a phase preservation constraint; the phase preservation constraint is specifically to minimize the instantaneous phase difference between the working tendon vibration signal and the final generated purification pressure wave signal.

[0032] The objective function of traditional adaptive filtering algorithms (such as LMS algorithm) is usually to minimize the mean square value of the filtering error, that is, minE[|e(n)| 2], where the error e(n) is the difference between the desired signal and the filtered output. This optimization only focuses on the amplitude error, which often causes unpredictable phase offsets while eliminating noise, and phase information is crucial for subsequent delay calculations. In this embodiment, in order to solve this problem, the objective function of the traditional algorithm is improved, and a phase preservation constraint is introduced. That is, when the algorithm iteratively updates its filtering weights, it must not only minimize the amplitude error, but also meet a constraint condition at the same time, that is, the instantaneous phase of the processed purified signal y(t) should be as close as possible to the instantaneous phase of the original working rib signal x(t). This constitutes a dual-constrained optimization problem, which allows the phase characteristics of the signal to be maintained while reducing noise.

[0033] like Figure 2 As shown, according to one aspect of the present application, applying the phase preservation constraint specifically includes the following data processing steps:

[0034] By Hilbert transform, the instantaneous phases of the working rib vibration signal and the purified pressure wave signal are calculated respectively to obtain the instantaneous phase difference;

[0035] Based on the instantaneous phase difference, a phase preservation factor is constructed in the form of a complex exponential function of the phase difference;

[0036] The phase preservation factor is used as a multiplication term and introduced into the updating rule of the time-varying filter weights of the adaptive filtering algorithm.

[0037] Specifically, the working tendon signal x(t) and the current step purification signal (filter output) y(t) are Hilbert transformed to obtain their analytical signals x a (t) and y a (t). The argument of the analytical signal is the instantaneous phase. For example, Φ x (t)=angle(x a (t)) and Φ y (t)=angle(y a (t)), where Φ x (t) is the instantaneous phase (argument) corresponding to the working rib signal, Φ y (t) is the instantaneous phase corresponding to the current purified signal. From this, we can get the instantaneous phase difference ΔΦ(t)=Φ x (t)-Φ y(t). In order to suppress the endpoint effect, a window function (such as a Tukey window) can be applied to the signal before the transformation. In some embodiments, in order to suppress the endpoint effect generated when a finite-length signal is transformed, it is preferred to apply a Tukey window to the signal before the transformation. The roll-off coefficient r of the Tukey window can be selected according to the transient characteristics of the signal. For example, for a signal with strong impact, a smaller r value (such as 0.1) can be selected, and for a relatively stable signal, a larger r value (such as 0.25) can be selected to achieve a balance between suppressing the endpoint effect and maintaining the signal waveform. After the transformation, the argument of the analytical signal is the instantaneous phase. Based on the calculated instantaneous phase difference, a complex phase preservation factor P(t)=exp(j·ΔΦ(t)) can be constructed, where j is an imaginary unit. The modulus of this factor is 1, and its phase directly reflects the phase difference between the two signals.

[0038] Furthermore, in order to increase robustness, a credibility weight α(t) can be introduced so that when the phase difference fluctuates violently, the effect of this factor is weakened. For example, α(t)=exp(-σ ΔΦ (t) / σ0), where σ ΔΦ (t) is the standard deviation of the phase difference within the local window, and σ0 is the normalization coefficient. In this case, the phase preservation factor is P(t) = α(t) exp(j ΔΦ(t)). The constructed phase preservation factor P(n) (discrete form) is multiplied into the weight update formula of the adaptive filter. For example, for the LMS algorithm, its traditional update rule is w(n+1) = w(n) + μ e(n) r(n), where w(n) is the filter weight vector at step n, w(n+1) is the updated filter weight vector at step n+1, μ is the learning rate or step size coefficient, e(n) is the error signal at step n, and r(n) is the reference signal sequence from the dummy rib. After the phase preservation constraint is introduced, the update rule can be improved to w(n+1)=w(n)+μ(n)·e(n)·r(n)·P(n), where μ(n) is the time-varying step size used in the nth iteration; P(n) is the phase preservation factor matrix used in the nth iteration. Alternatively, in a more complex dual-constraint model, the update rule can be decomposed into the amplitude error e amp and phase error e phase The common response is w(n+1)=w(n)+μ(n)·[e amp (n)·r(n)+λ·e phase (n)·r(n)ΘP(n)], where Θ represents the element-wise multiplication and λ is the phase constraint strength coefficient, which is used to balance the proportion of amplitude and phase errors in weight update.

[0039] like Figure 3As shown, according to one aspect of the present application, the adaptive filtering algorithm uses an adaptive step size to update its (adaptive filtering algorithm) time-varying filtering weights, and the process of determining the step size specifically includes:

[0040] Calculate the variance of the filter error sequence generated by the adaptive filtering algorithm in real time to obtain the error variance value;

[0041] According to a preset inverse function relationship, the adaptive step size is determined by the error variance value, so that the step size is in inverse proportion to the error variance.

[0042] In this embodiment, the motivation for using an adaptive step size (or variable step size) is that the noise intensity at a construction site is not constant. When the noise suddenly increases, the filtering error increases, and a smaller step size is required to ensure stable convergence of the algorithm. When the environment is relatively quiet and the filtering error is very small, a larger step size can be used to accelerate convergence and tracking ability. The specific implementation method is as follows: In each iteration, the variance var(e(nL:n)) of the filtering error e(n) of the most recent L samples is calculated. The step size μ(n) is updated according to the inverse function relationship. An exemplary relationship is μ(n) = μ0 / [1+γ·var(e(nL:n))], where μ0 is the initial or maximum step size and γ is a positive constant used to control the sensitivity of the step size to changes in variance. Optionally, to further improve environmental adaptability, the step size calculation can also take into account the power spectrum of the current noise. For example, an environmental adjustment factor β(n) = 1+γ′·mean(N(f current ±Δf)), where N(f current ) is the noise power near the current signal's primary frequency, γ′ is the noise sensitivity factor, and Δf is the frequency offset window width. This factor is then multiplied into the step size calculation formula, as shown in the equation μ(n) = μ0·β(n) / [1+var(n) / var0], where var0 is the reference variance. The step size adjustment considers both the filter's output error and the input noise characteristics of the external environment, making the adaptive process more intelligent and efficient.

[0043] In the preferred scheme for calculating error variance using a multi-scale sliding window, the lengths of the multi-scale sliding windows (e.g., L1 = 50, L2 = 200, L3 = 1000 samples) are selected based on the fact that different time scales can capture the different dynamic characteristics of the error sequence. The short-scale window (L1) is sensitive to error fluctuations caused by sudden strong interference or transient signal changes; the medium-scale window (L2) reflects the algorithm's convergence performance under general conditions; and the long-scale window (L3) characterizes the algorithm's performance under quasi-steady-state conditions. The weighting coefficients (e.g., 0.5, 0.3, 0.2) are determined based on engineering experience or offline simulation experiments, with the basic principle being to give higher weight to the short-scale error variance. This is because in real-time adaptive processes, a rapid response to near-term errors is generally more important than considerations of long-term steady-state conditions, helping the algorithm recover more quickly from unstable conditions.

[0044] This embodiment solves the problem of phase information distortion during noise filtering by introducing a physical dummy rib parallel to the working rib as a noise reference channel, and by applying a phase-preserving constraint in the adaptive filtering algorithm. On the one hand, the dummy rib converts the uncertain common-mode noise that needs to be estimated by the algorithm into a physical quantity that can be measured with high fidelity, providing an extremely accurate reference input for adaptive subtraction. On the other hand, the phase-preserving constraint introduces a penalty term based on the instantaneous phase difference into the weight update rule of the filter, forcing the algorithm to maintain the consistency of the phase characteristics of the purified signal and the original signal while minimizing the amplitude error. This enables the system to effectively filter out common-mode noise in strong interference environments such as vibration and electromagnetic interference at the construction site, while keeping the phase of the signal carrying the core propagation delay information undistorted, providing a high-quality, high-fidelity input signal for subsequent high-precision stress inversion.

[0045] In a preferred embodiment, in order to more rigorously solve the dual-constraint problem of amplitude minimization and phase preservation, an optimization framework based on the alternating direction method of multipliers (ADMM) can be used. The specific process is: construct a Lagrangian objective function L that includes all optimization objectives: L = |||xw⊕r||| 2 +λ1•(angle(y)-angle(x)) 2 +λ2•∣∣▽w∣∣ 2; The first term is the traditional amplitude error term; the second term is the phase-preserving constraint term, which aims to minimize the phase difference between the purified signal y and the original signal x; and the third term is the weight smoothing constraint, which is used to prevent violent oscillations in the filter weight w and improve the stability of the algorithm. r is the reference signal, ⊕ is the convolution operator, and ▽ is the gradient; λ1 and λ2 are two Lagrange multipliers, or regularization parameters, used to balance the relative importance of the three objectives. Their initial values ​​can be set empirically, for example, to 0.1, and are updated during the iteration process based on the dual residual. The ADMM algorithm is used to decompose the above complex optimization problem into a series of more easily solvable subproblems, which are solved through alternating iterations. Each iterative cycle involves: fixing the phase term and other variables to solve the optimal filter output that minimizes the amplitude error; fixing the amplitude term and other variables to solve the optimal phase that minimizes the phase error; and updating the Lagrange multipliers. The ADMM iteration process continues until the algorithm meets the convergence criteria. A feasible convergence criterion is that when the norms of the original residual and the dual residual are both less than a preset small threshold (for example, 10 -4 ), or when the number of iterations reaches the preset maximum value (for example, 10 inner loop iterations), the algorithm is considered to have converged, and the current optimal filtering result is output as the purified pressure wave signal.

[0046] In a specific embodiment, assuming that at a discrete moment n=1, there are the following inputs: the working rib signal x(1)=1.2, which can be understood as the true pressure wave signal s(1)=0.5 and the common mode noise n c (1) = 0.7; the dumb rib reference signal r (1) = 0.72, this value is the common mode noise n c The measured value of (1) contains a small measurement error; the weight of the adaptive filter at the previous moment is w(1)=0.9; the parameter settings used in the algorithm are: initial step size μ0=0.1, step size adjustment sensitivity γ=0.5, phase constraint strength λ=0.2. The calculation process is as follows: Calculate the filter output: y(1)=w(1)·r(1)=0.9·0.72=0.648, which is the filter's estimate of the common-mode noise. Calculate the error signal: filter error e(1)=x(1)-y(1)=1.2-0.648=0.552, which is the preliminary purified signal. Calculate the instantaneous phase and phase error (conceptual demonstration): To simplify the calculation, the real Hilbert transform is not performed, but it is assumed that the following instantaneous phase value is obtained through the transformation: The instantaneous phase of the working rib signal: Φ x (1) = 0.8π (radians); the instantaneous phase of the filter output: Φ y (1) = 0.75π (radians) Phase error e phase (1) = Φ x (1)-Φ y(1) = 0.05π. Calculate the amplitude error: Amplitude error e amp (1)=|x(1)||-|y(1)||=1.2-0.648=0.552 (in this simplified example, the amplitude error is equal to the filtering error); Calculate the adaptive step size μ(2): First, the error variance is required. Assume that the error variance of the last L samples is calculated to be var(e)=0.04. According to the formula: μ(2)=μ0 / [1+γ·var(e)]=0.1 / [1+0.5·0.04]=0.1 / 1.02≈0.098, it can be seen that due to the existence of the error, the step size is slightly reduced from the initial 0.1 to 0.098. Calculate the phase holding factor P(1): Assume that through the dynamic analysis of the phase difference, the credibility weight α(1)=0.95 is obtained. The phase difference ΔΦ(1) can be approximately understood here as e phase (1). P(1)=α(1)·exp(j·e phase (1))=0.95·exp(j·0.05π)=0.95·(cos(0.05π)+j·sin(0.05π))≈0.95·(0.9877+j·0.1564)≈0.9383+j·0.1486. ​​Update the filter weights w(2): Adopt the update rule including the amplitude and phase double error (for simplicity, element-wise multiplication is ignored here): w(2)=w(1)+μ(2)·[e amp (1)+λ·e phase (1)]·r(1)·P(1)* (the conjugate of the reference signal and the phase factor P(1)* is usually used in the actual algorithm); assuming that the role of P(1)* in weight update is mainly reflected in its phase correction, here is the process of demonstrating the numerical calculation. The physical meaning of this step is simplified and only the calculation is performed: weight increment Δw = 0.098·[0.552+0.2·0.05π]·0.72≈0.098·[0.552+0.0314]·0.72≈0.098·0.5834·0.72≈0.0411; w(2) = w(1)+Δw = 0.9+0.0411=0.9411. It can be seen that in a single iteration of n=1, the filter weight is updated from 0.9 to 0.9411. This new weight is closer to 1.0, which means that in the next iteration, the filter's estimate of the noise y(2) will be closer to the true noise value n c (1) = 0.7. This process continues until the weights converge, minimizing the energy of the filtering error e(n) (i.e., the purified signal) and minimizing the phase difference between it and the original signal.

[0047] like Figure 4As shown, according to one aspect of the present application, the vibration signal of the working rib and the reference signal of the dumb rib are collected by applying an adaptive sweep frequency excitation signal. The generation process of the adaptive sweep frequency excitation signal specifically includes:

[0048] Before applying the adaptive sweep frequency excitation signal, the background noise signal of the construction site is collected in advance;

[0049] Perform spectrum analysis on the background noise signal to generate a noise power spectrum, which is used to mark the peak frequency band of the environmental noise;

[0050] An adaptive frequency sweep excitation signal is generated based on the noise power spectrum. The instantaneous frequency of the adaptive frequency sweep excitation signal changes the rate at which it passes through the noise peak frequency band during the frequency sweep process.

[0051] In this embodiment, the adaptive frequency sweep excitation signal is a non-uniform frequency sweep signal, and the speed of its frequency sweep changes dynamically according to the spectrum distribution of the ambient noise. Due to the traditional fixed frequency or linear frequency sweep signal, its energy may unfortunately fall into the strong noise frequency band generated by the operation of large-scale equipment on site, resulting in serious signal pollution. This embodiment actively avoids this and strategically puts the signal energy into a relatively clean frequency band. Specifically, a short period of time (for example, 30 seconds) before the start of the tensioning operation, the acceleration sensor to be used for measurement is used to collect a section of the background noise signal n(t) on site. In order to obtain a stable noise spectrum estimate, the collected signal can be segmented, for example, with 2048 sampling points per segment, and a 50% overlap between segments. Each segment is windowed (such as a Hanning window) and FFT calculation is performed to obtain the power spectral density PSD of each segment. i (f). By averaging the power spectral density of all segments, a smooth and representative noise power spectral density N(f) is obtained: N(f)=(1 / M)∑ i=1 M |FFT(n i (t))∣ 2 / N; where n i (t) is the background noise after weighting the i-th segment window function; N is the normalization factor, and M is the total number of segments. The Hanning window function is preferably used because it strikes a good balance between good frequency resolution and low sidelobe leakage. The overlap rate of 50% is selected to ensure the statistical stability of the spectrum estimation while reducing the loss of data edge information caused by windowing. Read the noise power spectral density N(f) obtained in the previous step and analyze it to identify the noise peak. An exemplary method is to calculate the noise mean μ of the entire spectrum N and standard deviation σ N , and all the N +2σ NThe frequency point f is identified as the noise peak frequency f peak , these frequencies constitute the range that needs to be avoided.

[0052] According to one aspect of the present application, a process of generating an adaptive frequency sweep excitation signal based on a noise power spectrum specifically includes:

[0053] Based on the noise power spectrum, a frequency avoidance weight vector is constructed, where the weight value corresponding to the noise peak frequency band is lower;

[0054] Multiplying the reference sweep rate by the frequency avoidance weight vector to obtain a time-varying instantaneous sweep rate;

[0055] The instantaneous sweep frequency rate is integrated to generate a sweep frequency function; and an adaptive sweep frequency excitation signal is synthesized based on the sweep frequency function.

[0056] Specifically, based on the noise power spectrum density N(f), a frequency avoidance weight function W(f) is constructed. The goal of this function is to assign low weights to frequencies with large noise and high weights to frequencies with small noise. An exemplary function form is a S-shaped function, for example, W(f)=1 / [1+exp((N(f)-N threshold ) / k′)], where N threshold is the preset noise threshold, and k' is a coefficient that controls the steepness of the transition band. Thus, when N(f) is much larger than the threshold, W(f) approaches 0; when N(f) is much smaller than the threshold, W(f) approaches 1. The values ​​of this function at discrete frequency points are combined to form the frequency avoidance weight vector W. A baseline constant sweep rate k0 is set. During the sweep process, the sweep rate is dynamically adjusted by querying the weight value W(f(t)) corresponding to the current instantaneous frequency f(t). The time-varying instantaneous sweep rate can be calculated as k(t) = k0 × W(f(t)), where k0 is the baseline sweep rate, a constant. This means that when the instantaneous frequency of the signal is about to enter the noise peak region, the corresponding W(f(t)) decreases, resulting in a smaller instantaneous rate k(t). This should be understood as a greater frequency change per unit time, meaning that the noise region is swept through more quickly. When in a clean frequency band with less noise, W(f(t)) is larger, k(t) is also larger, the signal stays here for a relatively longer time, and more energy is injected. The instantaneous frequency f(t) is obtained by integrating the instantaneous sweep rate k(t): f(t)=f0+∫0 t k(τ)dτ, where f0 is the starting frequency of the frequency sweep, τ is the time variable, and d is the differential element of the integral operator. The f(t) function is also called the frequency sweep function. Substituting the frequency sweep function into the sine function, the adaptive frequency sweep excitation signal s(t) = A·sin(2π∫0 t f(τ)dτ), where A is the amplitude of the excitation signal. This signal has the ability to intelligently avoid noise.

[0057] This embodiment achieves a paradigm shift from passive noise reduction to active anti-interference by pre-sensing the background noise spectrum of the construction site before applying the excitation, and based on this, generates an adaptive sweeping excitation signal that can actively avoid the noise peak frequency band, thereby improving the source quality of signal acquisition. The spectrum sensing concept in the field of cognitive radio is innovatively applied to structural excitation. Specifically, the on-site noise map is obtained through FFT analysis, and the sweeping rate is dynamically adjusted to quickly pass through the frequency band with high noise to reduce energy injection, and appropriately stay in the clean frequency band with low noise to increase energy injection. In the complex electromagnetic and mechanical vibration environment of prestressed construction, the excitation energy can always be effectively coupled to the frequency band with the highest signal-to-noise ratio, thereby improving the signal-to-noise ratio of the initial acquisition signal. This not only directly improves the stability and accuracy of all subsequent processing steps, but also enhances the anti-interference robustness of the entire system.

[0058] According to one aspect of the present application, a process of analyzing and cleaning a pressure wave signal to extract a propagation delay includes:

[0059] Perform multi-scale decomposition on the purified pressure wave signal to obtain a set of narrowband component signals;

[0060] Analyze the coherence between each signal in the narrowband component signal set;

[0061] Based on coherence, signal components are classified into direct wave paths and reflected wave paths; propagation delays are extracted from the direct wave paths.

[0062] In this embodiment, multi-scale decomposition is a signal processing technique that can decompose a broadband signal into a series of narrowband components with different center frequencies and bandwidths. For example, wavelet packet decomposition technology can be used. i (t) Performing a five-layer wavelet packet decomposition, for example, can obtain signal components in multiple frequency bands. Based on the energy or signal-to-noise ratio of each frequency band, several (for example, three) frequency bands with the most concentrated energy can be automatically selected and reconstructed into a narrowband component signal set {y i (b) (t)}, b = 1, 2, 3. The direct wave path is the path that minimizes energy loss and propagation time for the excitation signal from the excitation point to the measurement point. The reflected wave path is the path that the signal takes after one or more reflections from structural boundaries (such as the interface between concrete and steel bars, the edge of a beam, etc.) before reaching the measurement point. The propagation delay of the direct wave most directly reflects the stress state between two points. However, in the received signal, the direct wave is often superimposed with various reflected waves, making it difficult to directly identify them. By decomposing the signal into different frequency bands and exploiting the differences in frequency domain coherence between signals along different paths, it is possible to distinguish them.

[0063] Furthermore, the process of analyzing coherence and classification paths specifically includes:

[0064] Calculate the cross power spectrum and auto power spectrum between different signals in the narrowband component signal set;

[0065] Based on the cross power spectrum and auto power spectrum, the coherence function used to characterize the coherence is calculated;

[0066] A clustering algorithm is applied to the values ​​of the coherence function to automatically classify the signal components into direct wave paths and reflected wave paths based on a preset coherence threshold.

[0067] Specifically, select the same narrowband component signal at different measurement points (such as measurement point i and measurement point j), or different narrowband component signals at the same measurement point, and calculate the cross-power spectrum S between them. ij (f)=E[X i *(f)X j (f)] and their respective autopower spectra S ii (f), S jj (f); where X i *(f) is the conjugate frequency domain expression of the narrowband signal at the i-th measurement point at frequency f, X j (f) is the frequency domain expression of the narrowband signal at the jth measurement point at frequency f. In order to improve the accuracy of the estimation, the Welch method can be used, that is, the calculation is performed by averaging the segmented periodogram. On this basis, the coherence function is calculated: γ 2 (f)=|S ij (f)∣ 2 / [S ii (f)S jj (f)]. The value range of the coherence function is between [0, 1]. The closer the value is to 1, the stronger the linear correlation between the two signals at that frequency. Since the direct wave signal has a single propagation path, its different frequency components usually maintain a high degree of coherence. However, since the reflected wave has experienced a complex path, the phase relationship between its different frequency components will become chaotic, resulting in a decrease in coherence. Using this feature, the calculated coherence function values ​​can be clustered. For example, the K-means clustering algorithm can be used to automatically group the coherence function values. The coherence threshold can be preset to define different clusters. An exemplary classification rule is: direct wave cluster: average coherence γ 2 >0.8 signal component; primary reflection wave cluster: average coherence is 0.4<γ 2 <0.8 signal components; multiple reflections / noise clusters: average coherence γ 2<0.4. In this way, each narrowband signal component is labeled as a direct wave or reflected wave path classification. It should be noted that before calculating the cross-power spectrum and auto-power spectrum, accurate analytical signal construction is required for each narrowband component signal. This process should strictly follow the method based on Bedrosian's theorem test and frequency domain processing to ensure the accuracy of subsequent coherence analysis.

[0068] Furthermore, the process of extracting the propagation delay specifically includes:

[0069] Extract the initial path delay for the direct wave path and the reflected wave path respectively;

[0070] Based on the path classification results, a path credibility weight is assigned to each path delay, where the weight attributed to the direct wave path is higher than the weight attributed to the reflected wave path;

[0071] For the path delays of all paths, a weighted fusion calculation is performed based on the path credibility weights corresponding to the path delays to obtain the final propagation delay.

[0072] In this embodiment, for each signal component classified in the previous step, the initial path delay τ is calculated by the phase gradient method or the cross-correlation method. i According to its path classification label, it is assigned a path credibility weight w path，i For example, the weight of the direct wave path can be set to 1.0, the weight of the single reflection wave path can be set to 0.5, and the weight of the multiple reflection wave path can be set to 0.1. In order to obtain the final propagation delay that is both accurate and robust, it is necessary to fuse the delay information of all paths. A preferred fusion method is maximum likelihood estimation or weighted least squares. The final propagation delay τ final It can be calculated by the following formula: final =∑(w i 2 ·τ i ) / ∑w i 2 , where w i is the credibility weight of the i-th path. Furthermore, the weight w i It can be a comprehensive weight that takes into account not only the path type but also the certainty of the measurement. i =w path，i ·exp(-σ τ，i / σ0), where σ τ，iis the uncertainty of the i-th time delay measurement (e.g., the variance during its calculation), and σ0 is a normalization constant. This fully utilizes all acquired signal information while highlighting the most reliable direct wave information through a weighted approach. This effectively suppresses the interference of reflected waves on the time delay measurement, resulting in a highly accurate final propagation delay and laying a solid foundation for subsequent stress inversion.

[0073] In some embodiments, the allocation of path credibility weights is specifically as follows: after the path classification is completed, a fixed, empirical path credibility weight w can be allocated to different categories of paths. path，i A feasible allocation scheme is: for the paths classified as direct wave clusters, their weight w path，i The weight of the path classified as a single reflection cluster is set to 1.0; the weight of the path classified as a multi-reflection cluster is set to 0.5; the weight of the path classified as a multi-reflection cluster is set to 0.1. This reflects the decreasing reliability of signals from different paths in representing the true propagation delay.

[0074] This embodiment addresses the interference problem of multipath propagation in complex structures by introducing a multipath separation and fusion method based on coherence analysis. Specifically, a broadband received signal is decomposed into multiple narrowband components through multi-scale decomposition. By calculating the coherence function based on the differences in the stability of the phase relationships between the frequency components of signals propagating along different paths, signals with highly stable phase relationships (high coherence) are identified as direct waves, while those with chaotic relationships (low coherence) are identified as reflected waves. When fusing delay information, different credibility weights are assigned to different paths to highlight the contribution of direct waves. This allows this embodiment to effectively distinguish the often weaker but crucial first-wave signal, which is obscured by strong reflected waves, in a complex acoustic waveguide composed of prestressed tendons, grouting, and corrugated pipes. This ensures that the extracted propagation delay truly corresponds to the shortest physical path between the two points, providing a key guarantee for the accuracy of the final stress inversion.

[0075] According to one aspect of the present application, inverse calculation of a real-time stress distribution field includes:

[0076] Additional temperature data along the prestressed tendon path is collected;

[0077] Establish a wave velocity-stress-temperature model and express the wave velocity as a nonlinear function of stress and temperature;

[0078] Based on the model, an inversion equation system with stress as the unknown variable is established, where one side of the equation is the propagation delay from the phase-delay characteristic matrix, and the other side is the path integral of the model using geometric parameters and temperature data;

[0079] Solve the inverse equations to obtain the real-time stress distribution field.

[0080] In this embodiment, the wave velocity-stress-temperature model refers to a mathematical expression that describes the quantitative relationship between the velocity v of elastic waves propagating in the prestressed tendons, the stress σ inside the prestressed tendons, and the temperature T of the environment in which they are located. The inverse equation group refers to a set of equations in which the stress values ​​of discrete nodes on the prestressed tendons are unknowns and the measured propagation delays between each node are knowns. Specifically, while performing the time delay measurement, the real-time temperature data T(s) of multiple measuring points along the tendon path are synchronously collected through temperature sensors (such as fiber grating sensors or thermocouples) pre-buried in the bellows or attached to the surface of the prestressed tendons. Based on the acoustic elasticity theory and experimental data, a nonlinear wave velocity-stress-temperature coupling model of the following form can be established: v(σ, T) = v0(T)[1+α(T)σ+β(T)σ 2 ]; where v0(T), α(T), and β(T) are temperature-related parameters in the model, determined by real-time calibration; and v(σ, T) is a nonlinear function of stress σ and temperature T. This model shows that the wave velocity not only has a nonlinear (quadratic) relationship with stress σ, but also its coefficients v0(T), α(T), and β(T) themselves are functions of temperature T. Physically, the propagation delay between two measuring points is equal to the integral of the path length between the two points over the wave velocity. Based on this, an equation can be established for any pair of measuring points i and j: τ ij =∫ si sj ds / v(σ(s),T(s)); where, τ ij is the measured propagation delay and is a known quantity. The geometric parameters of the integration path (such as the curve length s) can be accurately extracted from the BIM model, and the temperature T(s) is a real-time measurement value. The unknown quantity is the stress distribution function σ(s). This is achieved by discretizing the tendon path into several nodes and using the stress function σ(s) at these nodes as the stress value σ k By expressing it through an interpolation function (such as linear or spline interpolation), the above integral equation is transformed into a nonlinear algebraic equation system of the form F(σ)=0. Since this system of equations is usually ill-posed, direct solution may lead to results that are extremely sensitive to measurement errors. Therefore, it is preferred to use a regularization method for solution. For example, the Levenberg-Marquardt algorithm with Tikhonov regularization is used. Its goal is to solve the minimization problem: min||Aσ-τ measured ∣∣ 2 +λ|||Dσ||| 2 ; Among them, the first term is the data fitting term, which represents the difference between the calculated delay and the measured delay; the second term is the regularization term (or smoothness constraint term), which is used to ensure the stress distribution of the solution; σ has physical smoothness, D is the difference operator, A is the propagation response matrix, τmeasured is the propagation delay obtained by actual measurement through signal processing. The value of the regularization parameter λ can be adaptively determined by the L-curve method. After several (e.g., 20) iterative calculations, the stable node stress vector σ can be obtained. nodes .

[0081] In some embodiments, the L-curve method is implemented by iterating over a series of discrete values ​​of the regularization parameter λ, solving the regularized least squares problem for each value of λ, and calculating the norm (or smoothness norm) of the corresponding solution ∣∣Dσ∣∣ 2 and the norm of the residual || |Aσ-τ measured ∣∣ 2 . Plotting these two sets of values ​​on a logarithmic coordinate graph as the y-axis and x-axis coordinates, an L-shaped curve will be formed. The λ value corresponding to the inflection point (corner) of the curve is the optimal regularization parameter, because it achieves the best balance between the smoothness of the solution and the goodness of fit to the data. The initial damping factor in the Levenberg-Marquardt algorithm is usually set to a small positive number, such as 10 -3 This factor will be adaptively adjusted during the iteration process according to the decrease in each step to ensure the stable convergence of the algorithm.

[0082] Furthermore, the process of establishing the velocity-stress-temperature model further includes real-time calibration of temperature-related parameters in the model, specifically:

[0083] Based on the currently collected temperature data and by fitting the historical measurement data, the current values ​​of the temperature-related parameters in the model are determined and a real-time calibration parameter set is generated;

[0084] The wave velocity-stress-temperature model is constructed using the real-time calibration parameter set.

[0085] In this embodiment, real-time calibration means that before each stress inversion calculation, the parameters in the wave velocity model are updated and corrected according to the latest on-site environmental data (mainly temperature), rather than using fixed parameters measured under laboratory conditions. The acoustic properties of materials (such as sound velocity, acoustic elastic coefficient, etc.) are very sensitive to temperature. During a long construction period, the ambient temperature may change by several degrees or even dozens of degrees, which will cause the model parameters to drift. Real-time calibration is precisely to compensate for this error introduced by temperature changes. Specifically, based on historical measurement data (for example, multiple sets of time delay and stress data measured at different construction stages and different temperatures), the specific functional relationship between the model parameters v0(T), α(T), β(T) and the temperature T can be fitted. The exemplary relationship is as follows: Specifically: v0(T)=v 00 [1-αT (T-T0)], α(T)=α0exp(-β T |T-T0|), β(T)=β0[1+γ T (T-T0) 2 ], where v 00 , α T , T0, α0, β T ,β0,γ T is the material constant obtained by least squares fitting the historical data. Specifically, v 00 is the wave velocity at reference temperature T0, α T is the linear decrease rate of wave velocity with temperature; α0 is the stress sensitivity at the reference temperature, β T is the temperature suppression factor; β0 is the reference second-order stress sensitivity at the reference temperature, γ T is the temperature enhancement factor. When performing a new stress inversion, the current temperature distribution T(s) is measured and substituted into the above equation to calculate the parameter set {v0(T), α(T), β(T)} specific to the current temperature environment. Using this parameter set specific to the current temperature environment to construct the velocity-stress-temperature model used in the current calculation improves the accuracy of the inversion.

[0086] In some embodiments, the following supplementary explanation is provided regarding the engineering feasibility details of the real-time calibration portion: To ensure the reliability of the fitting results, the historical measurement data used for parameter fitting should meet certain minimum sample size requirements, for example, at least 30 sets of valid data points collected under different operating conditions (covering a temperature variation range of at least 10°C and a maximum stress variation range of 30%). When fitting historical data to determine temperature-related parameters, clear goodness-of-fit criteria are required. One feasible criterion is to require the coefficient of determination (R-squared) value of the fitting model to be greater than 0.95, so that the established parameter-temperature relationship model can highly explain the variability of the data.

[0087] This embodiment solves the stress inversion inaccuracy problem caused by traditional methods relying on simplified linear models and static parameters by constructing a coupled model that simultaneously considers stress nonlinear effects and temperature influences, and by performing real-time calibration of the temperature-related parameters in the model. Specifically, this nonlinear model (for example, including quadratic stress terms) is closer to the physical nature of sound wave propagation in stressed metal than traditional linear models. More importantly, through real-time calibration, the system can dynamically update key parameters in the model, such as the zero-point wave velocity and acoustoelastic coefficient, using field-measured temperature data. This allows for real-time compensation for disturbances in the material's physical properties caused by temperature changes in the construction environment (such as the difference between day and night). This constitutes a critical closed loop from precise time-delay measurement to precise stress inversion, enabling the entire technology chain to ultimately output stress distribution results that are highly consistent with the actual physical state, thereby improving accuracy.

[0088] Furthermore, when the nonlinear inversion process calculates the stress vector σ of the discrete nodes along the prestressed tendon path, nodes After that, the following steps will be performed:

[0089] Step 1: Continuity and verification of stress field. In order to facilitate visualization in BIM model and subsequent continuous field analysis, it is first necessary to convert the discrete node stress vector σ nodes is converted into a continuous stress distribution function σ(s, t) along the path s. Specifically, cubic spline interpolation or similar smoothing interpolation algorithms can be used to calculate σ nodes Interpolation is performed to obtain the stress value at any point on the entire prestressed tendon path. After obtaining the continuous stress field σ(s, t), a verification step can be performed to ensure the physical rationality and accuracy of the inversion results. An optional verification method is to check the smoothness of the stress gradient, that is, to calculate ▽σ so that its value is within a reasonable range, for example, less than the ratio of the maximum allowable stress to the beam span, to avoid stress mutations that have no physical meaning. Another verification method is to perform virtual time delay backcalculation, that is, to use the obtained stress field σ(s, t) and wave velocity model to reversely calculate the theoretical propagation delay τ between each measuring point. calc and compare it with the measured delay τ measured If the relative error, such as |τ calc -τ measured | / τ measured, is less than a preset threshold (e.g., 5%), the inversion result is reliable. Specifically, when performing cubic spline interpolation, it is preferred to employ the no-knot boundary condition. This condition provides two additional constraints by enforcing the equality of the third-order derivatives of the first and second polynomial segments near the endpoints, eliminating the need to manually specify derivative values ​​at the endpoints. This generally results in an overall interpolation curve that is both visually and mathematically smoother and more natural.

[0090] Step 2: BIM Property Update and 3D Visualization. The verified real-time stress distribution field σ(s, t) is synchronously updated to the BIM model, enabling real-time refresh of the digital twin. Specifically, this process is implemented through a programmatic interface (API). The stress field data σ(s, t) is mapped to the corresponding prestressed tendon element in the BIM model and updated as a new physical property in the element's IFC (Industry Foundation Classes) property set. For example, a custom property set named Pset_TendonStress can be created and the stress distribution data written to it. To provide intuitive feedback to on-site engineers and managers, the system can perform 3D visualization based on the updated BIM model. Using algorithms such as trilinear interpolation, a continuous stress contour map can be generated on the BIM 3D prestressed tendon model, with different colors representing different stress levels (e.g., red indicates high stress areas, blue indicates low stress areas). This provides a clear overview of the prestress distribution within the structure.

[0091] Step 3: Model prediction and construction control parameter optimization. The ultimate goal of this embodiment is not only to monitor, but also to guide and optimize the construction process through accurate monitoring data to form a closed-loop feedback. Specifically, the system compares the measured real-time stress distribution field σ(s, t) with the design stress distribution field σ stored in the BIM model. d (s) and calculate the deviation field between the two Δσ(s)=σ(s, t)-σ d (s). This deviation field intuitively reflects the gap between the current construction status and the design requirements. Based on this deviation field Δσ(s), the system uses the model predictive control (MPC) algorithm or the classic PID (proportional-integral-differential) control algorithm to calculate the adjustment required for the next tensioning operation. For example, when using the PID algorithm, the correction amount ΔF of the tensioning force can be calculated by the following formula: ΔF=K p •Δσ+K i •∫Δσdt+K d •dΔσ / dt; where K p , K i , K dThey are proportional, integral and differential coefficients respectively. The purpose of this calculation is to eliminate the current stress deviation to the greatest extent possible after the next tensioning or holding operation. The result of the calculation is a series of optimized construction control parameters, such as the recommended jack tensioning force F in the next stage. i , tensioning speed v i , load holding time t i etc., forming an optimized control parameter set {F i , v i , t i Similar to stress data, the optimized control parameter set is also written back to the BIM model, for example, in a property set named Pset_ConstructionControl. On-site intelligent tensioning equipment can directly read these optimized instructions from the BIM model to perform operations, or on-site engineers can manually adjust them based on these suggestions. This completes a closed loop from real-time monitoring to data analysis, intelligent decision-making, and precise control, improving the quality, safety, and automation of prestressed construction.

[0092] In some embodiments, the control parameter optimization calculation is specifically: if the PID control algorithm is used, its parameter K p , K i , K d The key is the tuning. A commonly used method in engineering is the Ziegler-Nichols tuning method based on the step response of the system. By applying a step tension, measuring the response curve of the system, and calculating a set of initial PID parameters according to the characteristics of the curve (such as delay time, rise time, etc.), and then fine-tuning is performed on this basis. If a more advanced model predictive control (MPC) algorithm is used, it is necessary to set its prediction time domain and control time domain. In the application scenario of the present invention, the prediction time domain can be set to 10 to 20 calculation steps in the future to predict the evolution trend of stress in the short term; the control time domain can be set to 2 to 5 calculation steps to optimize the control input of a few key steps in the future. Such a setting can take into account the computational efficiency while ensuring the accuracy of the prediction and meet the requirements of real-time control.

[0093] According to one aspect of the present application, a system for implementing a BIM-based large-span post-tensioned bonded prestressed beam construction method may include:

[0094] BIM data server: stores the BIM model (.ifc format) of the long-span prestressed beam and provides a data interface for the processing module to read the geometry and attribute information.

[0095] Central processing unit: can be a high-performance industrial computer or server, used to run all the data processing and inversion algorithms described in the present invention.

[0096] Excitation signal generation and application module: includes an arbitrary waveform generator for generating an adaptive frequency sweep signal; and a piezoelectric exciter mechanically coupled to the tensioning jack for converting the electrical signal into micro-amplitude mechanical vibration.

[0097] Sensor and data acquisition module: Acceleration sensor: preferably a piezoelectric IEPE accelerometer, such as the 352C33 model from PCBPiezotronics, which has high sensitivity, wide-band response, and good signal-to-noise ratio. The sensor is fixed to the prestressed tendon tensioning end or a specially opened observation hole by epoxy adhesive or magnetic base. Temperature sensor: preferably a fiber Bragg grating (FBG) sensor, which can be connected in series on an optical fiber to achieve distributed temperature measurement along the tendon path. Multi-channel synchronous data acquisition card: such as the PXIe-4499 from NI (National Instruments), which can support up to 24 channels of synchronous sampling with a maximum sampling rate of 204.8 kS / s.

[0098] In this embodiment, the recommended ranges of the key operating parameters of the system are as follows: the sampling rate is preferably 10kHz to 20kHz. A lower sampling rate may not be able to capture the high-frequency details of the signal, and an excessively high sampling rate will produce unnecessary data redundancy. The sweep frequency range is preferably 0.5kHz to 5kHz, which covers the commonly used frequency bands of excitation signals in most civil structure inspections. For the arrangement along the length of the beam, the sensor spacing is preferably 1 meter to 5 meters; a spacing that is too small is not economical, and a spacing that is too large may miss local stress anomalies. The temperature measurement range is -10°C to 60°C, covering the temperatures that may be encountered in most construction environments. The stress measurement range is 0 to 1500MPa, covering the entire stress range of prestressed high-strength steel bars from initial tensioning to final locking.

[0099] The system's workflow and data interface are defined as follows: The central processing unit (CPU) reads model data from the BIM server and generates a sensor layout and initial excitation parameters. These instructions are sent via an Ethernet interface to the excitation signal generator and data acquisition card. The acquisition card transmits the acquired acceleration and temperature data (for example, in TDMS format) back to the CPU. The various algorithm modules within the CPU (such as noise cancellation, phase analysis, and stress inversion) sequentially process the data and write the resulting stress field and optimized control parameters back to the BIM data server via an API, updating the model properties. The CPU orchestrates the entire process, enabling millisecond-level real-time response.

[0100] According to one aspect of the present application, before extracting the propagation delay, an improved path integral method is used to perform phase unwrapping, and the unwrapping integer k(τ) is determined by minimizing the phase derivative jump.

[0101] In this embodiment, when the wrapped phase Φ of the narrowband component signal is obtained by Hilbert transform, w (t), since its value range is limited to the interval [-π, π], phase jumps will occur. The phase unwrapping algorithm must be used to restore its true and continuous phase Φ(t), which is a prerequisite for accurately calculating the group delay. Traditional phase unwrapping algorithms (for example, simply detecting whether the phase difference between adjacent points exceeds π and compensating for it) are very sensitive and prone to errors when the signal-to-noise ratio is low, resulting in an erroneous slope of the phase after unwrapping, which causes a huge deviation in the group delay calculation. This embodiment adopts an improved path integral method to improve the robustness of the algorithm. Specifically, before phase extraction, the narrowband component signal y i (b) (t) Apply the Bedrosian's Theorem to verify that its spectrum does not overlap with the spectrum of the Hilbert transform kernel, which is the prerequisite for constructing an effective analytical signal. After the test is passed, the analytical signal z is accurately constructed by padding the signal spectrum with zeros in the frequency domain and applying Hermitian symmetry. i (b) (t)=y i (b) (t)+j·H[y i (b) (t)], where H[·] represents the Hilbert transform. The continuous phase Φ(t) and the wrapped phase Φ w (t) is related to Φ(t)=Φ w (t)+2πk(t), where k(t) is an integer that changes with time. Taking the derivative of this equation, we get dΦ / dt = dΦ w / dt +2π(dk / dt). Since the real instantaneous frequency (i.e., phase derivative) should change slowly, the jump in the phase derivative mainly comes from dΦ w / dt is a pulse at the wrapping point. The improved path integral method has the following expansion formula: Φ(t)=Φ(0)+∫0 t [dΦ w / dτ+2πk′(τ)]dτ Here k′(τ) is not a direct integer, but represents an integer multiple of 2π that needs to be compensated at time τ. The principle of determining k′(τ) is to minimize the derivative jump of the final unfolded phase Φ(t). The algorithm is used to find a set of integers k i (corresponding to discrete time point i), so that the objective function J=∑ i ∣(Φ i+1 -Φ i )-(Φ i -Φi-1 )∣ 2 Minimization. Essentially, it is an optimization problem of minimizing the second-order difference (i.e., the approximation of the derivative). This problem can be solved by efficient dynamic programming or least squares algorithms, thereby determining an integer compensation value k that best maintains the continuity of the phase slope for each moment when phase wrapping occurs. i Even if there is local noise in the signal causing phase jitter, the algorithm can find a global optimal expansion path, avoiding the error accumulation of traditional algorithms, and thus obtaining an accurate and smooth continuous phase sequence Φ i (b) (t), which provides high-quality input for the subsequent calculation of group delay by derivation.

[0102] In a specific embodiment, experiments were conducted at an actual long-span continuous beam bridge construction site, with a 50-meter-long prestressed tendon being the test subject. The BIM-based long-span post-tensioned bonded prestressed beam construction method (denoted as this method) was compared with a traditional monitoring method based on fixed-frequency excitation and strain gauge measurement (denoted as the traditional method). After tensioning, a high-precision pressure sensor measured the true stress of the prestressed tendon at mid-span as 1155 MPa. The traditional method, using strain gauges attached to the concrete surface, calculated the stress value to be 1020 MPa, with a relative error of (1155-1020) / 1155≈11.7%. The stress value obtained by inversion using this method was 1148 MPa, with a relative error of (1155-1148) / 1155≈0.6%. This method demonstrates significantly higher measurement accuracy than the traditional method. A high-power electric welder was used near the tensioning pedestal to simulate a strong electromagnetic noise environment. The signal-to-noise ratio (SNR) of the cleaned / processed signal was used as an evaluation metric. In the absence of interference, the SNR of the signal acquired using the conventional method was approximately 25dB, while the SNR of the signal acquired using this method (after noise cancellation) was approximately 45dB. In the presence of welding machine interference, the SNR of the signal acquired using the conventional method dropped sharply to approximately 8dB, with the signal waveform almost completely drowned out by the noise. However, due to the use of a physical dummy reference channel and phase-preserving adaptive noise cancellation, the SNR of the processed signal was maintained at approximately 35dB, ensuring the reliability of subsequent analysis. This demonstrates the robustness of this method against common-mode noise interference. 10 seconds of acquired data (sampling rate 10kHz) was processed on an industrial computer with the same hardware configuration (Intel i7-10700 CPU, 32GB RAM). The conventional method (primarily filtering and peak detection) took approximately 0.8 seconds in total. The proposed method (including all steps of adaptive excitation generation, noise cancellation, phase analysis, and stress inversion) took approximately 2.5 seconds in total. While this method is computationally more complex, its processing time is significantly shorter than the data acquisition time, meeting the real-time requirements of construction monitoring (typically requiring a response time of minutes). Furthermore, the increased computational cost is well worth it given the improvements in accuracy and robustness it brings.

[0103] The preferred embodiments of the present invention are described in detail above. However, the present invention is not limited to the specific details in the above embodiments. Within the technical concept of the present invention, various equivalent transformations can be made to the technical solutions of the present invention, and these equivalent transformations all fall within the scope of protection of the present invention.

Claims

1. A BIM-based construction method for large-span post-tensioned bonded prestressed beams, characterized in that: include: Collect the working reinforcement vibration signal of the working reinforcement and the dumb reinforcement reference signal of the dumb reinforcement; the working reinforcement refers to the reinforcement that participates in the structural stress; the dumb reinforcement refers to the reinforcement that does not participate in the direct stress; The noise of the working rib vibration signal is canceled by using the dumb rib reference signal to generate a purified pressure wave signal. Analyze and purify the pressure wave signal, extract the propagation delay and generate the phase-delay characteristic matrix; Based on the phase-time delay characteristic matrix and the geometric parameters of the preconfigured BIM 3D model, the real-time stress distribution field is inversely calculated and updated to the BIM 3D model.

2. The method according to claim 1, characterized in that The dummy bar reference signal is collected from a physical dummy bar that has the same path and material as the working bar but is not subject to tension. The physical dummy bar is used as a noise reference channel to measure common mode noise. Generating a purified pressure wave signal includes: inputting a dumb rib reference signal into an adaptive filtering algorithm to generate a dynamic estimate of common mode noise, and subtracting the dynamic estimate from the working rib vibration signal to generate a purified pressure wave signal.

3. The method according to claim 2, characterized in that The adaptive filtering algorithm controls the generation of dynamic estimates by imposing phase preservation constraints; The phase preservation constraint is specifically to minimize the instantaneous phase difference between the working rib vibration signal and the final generated purification pressure wave signal.

4. The method according to claim 3, characterized in that The application of phase preservation constraints includes: By Hilbert transform, the instantaneous phases of the working rib vibration signal and the purified pressure wave signal are calculated respectively to obtain the instantaneous phase difference; Constructing a phase preservation factor in the form of a complex exponential function of the instantaneous phase difference; The phase preservation factor is used as a multiplication term and introduced into the updating rule of the time-varying filter weights of the adaptive filtering algorithm.

5. The method according to claim 2, characterized in that The adaptive filtering algorithm uses an adaptive step size to update its time-varying filtering weights. The step size determination process is: Calculate the variance of the filter error sequence generated by the adaptive filtering algorithm in real time to obtain the error variance value; According to a preset inverse function relationship, the adaptive step length is determined by the error variance value, so that the step length is in inverse proportion to the error variance value.

6. The method according to claim 1, characterized in that The vibration signal of the working rib and the reference signal of the dumb rib are collected by applying an adaptive sweep frequency excitation signal. The generation process includes: Before applying the stimulus, the background noise signal of the construction site is collected in advance; Perform spectrum analysis on the background noise signal to generate noise power spectrum; An adaptive swept frequency excitation signal is generated based on the noise power spectrum, and its instantaneous frequency changes at a rate that passes through the noise peak frequency band during the sweep process.

7. The method according to claim 6, characterized in that Generates an adaptive swept frequency stimulus signal based on the noise power spectrum, including: Based on the noise power spectrum, a frequency avoidance weight vector is constructed, where the weight value corresponding to the noise peak frequency band is lower than the preset threshold; Multiplying a preset reference sweep rate by a frequency avoidance weight vector to obtain a time-varying instantaneous sweep rate; Integrate the instantaneous sweep rate to generate a sweep function; An adaptive swept frequency excitation signal is synthesized based on the swept frequency function.

8. The method according to claim 1, characterized in that Extract propagation delay, including: Perform multi-scale decomposition on the purified pressure wave signal to obtain a set of narrowband component signals; Analyze the coherence between each signal in the narrowband component signal set; Based on coherence, the signal components are classified into direct wave paths and reflected wave paths; Extract the propagation delay from the direct wave path.

9. The method according to claim 8, characterized in that Analyze coherence and classify signal components into direct and reflected paths, including: Calculate the cross power spectrum and auto power spectrum between different signals in the narrowband component signal set; Based on the cross power spectrum and auto power spectrum, the coherence function used to characterize the coherence is calculated; A clustering algorithm is applied to the values ​​of the coherence function to automatically classify the signal components into direct wave paths and reflected wave paths based on a preset coherence threshold.

10. The method according to claim 9, characterized in that Extract the propagation delay from the direct wave path, including: Extract the initial path delay for the direct wave path and the reflected wave path respectively; Based on the path classification results, a path credibility weight is assigned to each path delay; the weight attributed to the direct wave path is higher than the weight attributed to the reflected wave path; The path delays of all paths are weighted and fused according to their corresponding path credibility weights to obtain the final propagation delay.

Images