Monitoring Method and Performance Analysis of Climbing Scaffolds in Super High-Rise Buildings Based on BeiDou/GNSS Technology

Abstract

Monitoring the stability and safety of climbing scaffolds in super-high-rise construction is critical to ensuring construction quality and worker safety. This study proposes a Global Navigation Satellite System (GNSS)-based real-time monitoring method to track scaffold displacement and assess structural performance. A multi-level data optimization framework integrating gross error elimination, data interpolation, robust Kalman filtering, and a Cumulative Sum Control Chart (CUSUM)-based early warning system is developed to enhance monitoring accuracy. The key objectives of this research are to improve real-time displacement tracking, suppress measurement noise, and establish an automated anomaly detection mechanism for climbing scaffolds under complex construction conditions. The proposed method was validated in a super-high-rise construction project in Tianjin, China. Experimental results demonstrated that the system effectively reduced high-frequency noise and gross errors, achieving root mean square error (RMSE) reductions of 51.4% in the E direction, 45.5% in the N direction, and 49.6% in the U direction. The system successfully tracked vertical climbing displacements of 4.4 m per ascent and horizontal deviations of 4 cm (E direction) and 2 cm (N direction). Additionally, the multi-level warning mechanism identified displacement anomalies based on predefined thresholds, providing an early warning function to enhance scaffold safety management. Compared to conventional monitoring methods, the proposed BeiDou/GNSS-based system provides higher precision, real-time adaptability, and enhanced automation, offering a scalable solution for intelligent construction safety management. The findings contribute to structural health monitoring (SHM) applications and can serve as a reference for future high-rise construction safety assessments.

1. Introduction

With the acceleration of urbanization and breakthroughs in modern construction technology, the scale of super-high-rise buildings (height ≥ 300 m) has been continuously expanding, particularly in China, where the number of such structures ranks among the highest globally. While these buildings showcase remarkable engineering achievements, their construction presents unique challenges, including complex high-altitude working conditions, fluctuating dynamic loads, and heightened structural safety risks (Liasidis et al., 2023 [1]; Mishra et al., 2022 [2]). Among them, climbing scaffolds serve as essential structural supports in super-high-rise construction, requiring frequent vertical ascents, with their stability directly affecting both safety and efficiency (Li et al., 2023 [3]). However, as super-high-rise buildings are often located in dense urban areas, the “urban canyon” effect exacerbates the impact of strong wind loads, construction vibrations, and temperature fluctuations on climbing scaffold stability. Traditional monitoring methods, such as manual inspections and static sensors, struggle to meet the requirements for real-time tracking and risk assessment in such dynamic environments (Szymański, et al., 2017 [4]; Mneymneh et al., 2019 [5]). Consequently, developing a high-precision, automated monitoring system suitable for complex super-high-rise environments has become an urgent research challenge in engineering safety.

The current research on climbing scaffold monitoring primarily focuses on two aspects. The first is structural condition monitoring, specifically stress distribution and deformation analysis (Resende et al., 2023 [6]). Studies have investigated key scaffold components such as main frames, attachment nodes, and crossbeams, aiming to identify stress concentration zones and overload risks (Sakhakarmi, 2022 [7]). Additionally, displacement and inclination sensors have been used to track scaffold deformations and displacements during operation, assessing structural stability (Zhou et al., 2023 [8]). Błazik-Borowa et al. [9] proposed a novel axial force measurement method, validating its effectiveness for detecting scaffold anomalies through 115 experimental cases. Similarly, Xi et al. [10] deployed fiber Bragg grating static level meters to monitor scaffold ascent displacement and evaluate the influence of differential support point movements on the climbing process. However, these methods rely on discrete sensor deployment, making it difficult to achieve full-scale dynamic monitoring, and their effectiveness is often limited by the signal transmission stability.

The second major area of research concerns the impact of external loads on scaffold stability. Studies have shown that wind speed, wind direction, and turbulence effects significantly influence scaffold dynamics (Peng et al., 1996 [11]). Typically, environmental factors are assessed using meteorological sensors and vibration monitoring devices to evaluate how external forces affect scaffold stability in real time, ensuring safe operation in complex environments (Ramezantitkanloo et al., 2024 [12]; Chuang et al., 2024 [13]). S. Huang et al. [14] and Agarwal et al. [15] employed computational fluid dynamics (CFD) simulations and wind tunnel experiments to derive wind load effects on scaffolds and proposed wind pressure calculation models for safety-net-covered scaffolding. Additionally, temperature and humidity sensors have been used to analyze material degradation under extreme weather conditions, preventing issues such as corrosion or brittleness. Wei et al. [16] deployed meteorological sensors to assess wind thrust and vibration risks, contributing to extreme weather disaster prevention. However, these studies primarily focus on environmental parameters without directly correlating them to scaffold displacement responses.

Although emerging technologies have been leveraged to enhance scaffold safety monitoring (Dzeng et al., 2024 [17]; Park et al., 2025 [18]; Hu et al., 2024 [19]), fully intelligent monitoring has yet to be realized. Recently, artificial intelligence (AI) and multi-source data fusion have provided new perspectives on scaffold monitoring. Kim et al. [20] and Hui et al. [21] integrated BIM models with multi-sensor data to achieve the real-time visualization of scaffold planning. Zhao et al. [22] introduced a ConvLSTM-based multi-source information fusion model, which enables the dynamic analysis of bolt failure and steel pipe deformation in scaffolding. In addition, deep-learning-based image recognition methods have been developed to detect scaffold safety hazards through video surveillance (Zhang et al., 2020 [23]).

However, existing approaches still present notable limitations: (1) a heavy reliance on dense sensor networks, leading to high costs and complex deployment requirements; (2) inadequate real-time tracking capabilities for scaffold ascent and short-term dynamic displacements; and (3) a lack of an effective early warning mechanism, making it difficult to respond to sudden structural anomalies. Additionally, wind-induced vibrations and environmental disturbances significantly affect monitoring accuracy, but existing methods have not fully accounted for these error sources. In recent years, Global Navigation Satellite Systems (GNSS) have emerged as a powerful tool for structural health monitoring due to their high-precision positioning and real-time tracking capabilities (Egea-Roca et al., 2022 [24]; Chen et al., 2024 [25]). However, GNSS applications in dynamic climbing scaffold monitoring remain largely unexplored. The existing research has primarily focused on the long-term deformation monitoring of entire structures, with little emphasis on the high-precision tracking of short-period dynamic movements during scaffold ascent (Wang et al., 2024 [26]; Wang et al., 2025 [27]).

To address the aforementioned challenges, this study proposes a BeiDou/GNSS-based monitoring framework for climbing scaffolds in super-high-rise buildings and conducts field validation in an actual construction project. The main contributions of this study include the following: (1) High-Precision Real-Time Monitoring Method: A BeiDou/GNSS-based 3D dynamic monitoring system is developed to achieve the real-time tracking of scaffold movements, enhancing measurement accuracy and system stability. (2) Multi-Level Data Optimization Strategy: A combination of gross error elimination, adaptive filtering, and Kalman filtering is applied to improve GNSS data reliability, mitigating environmental noise effects. (3) CUSUM-Based Anomaly Detection Mechanism: A Cumulative Sum (CUSUM) control chart model is adopted to accurately identify scaffold operation anomalies, improving the sensitivity and reliability of the early warning system. (4) Field Validation and Performance Assessment: The monitoring system is tested in a real-world super-high-rise construction project to evaluate its applicability and stability in complex construction environments, verifying its effectiveness in climbing scaffold safety monitoring.

The remainder of this paper is structured as follows: Section 2 introduces the proposed BeiDou/GNSS monitoring framework and multi-level warning mechanism. Section 3 details the experimental setup, data processing methods, and system performance validation results. Section 4 discusses the research findings, system limitations, and future optimization directions. Finally, Section 5 summarizes the study and provides an outlook for future research.

2. Climbing Scaffold Status Monitoring Method

2.1. GNSS/BeiDou Positioning Solution Model with Height Constraint for Climbing Scaffolds

2.1.1. Data Preprocessing

(1)

Outlier Removal

GNSS signals used in monitoring external climbing scaffolds for super-high-rise buildings are subject to non-line-of-sight errors and multipath effects, due to obstructions and reflections from buildings in typical urban canyon environments. During real-time processing, occasional issues such as unfixed ambiguities may arise, leading to significant solution drift that deviates greatly from normal observation values. These erroneous solutions cannot be considered valid measurements. Therefore, outlier removal is necessary during the data preprocessing stage. By setting a specific threshold, data outside this threshold are identified as outliers and removed. The current methods for outlier removal in monitoring data include the 3σ criterion, statistical tests, Grubbs’ test, and the t-test [28].

Outlier removal from the monitoring data is performed using the 3σ criterion, which is applicable to sample data that follow a normal or approximately normal distribution. Following this, we address the challenge of data loss in GNSS data processing, which can occur during the acquisition, transmission, and preprocessing stages. Furthermore, we assume that the monitoring data $X_1,X_2,X_3,\cdots ,X_n$ are mutually independent and identically distributed, following the distribution $X\sim (\mu ,{\sigma }^2)$. The standard deviation of the sample is calculated using the following formula:

$$\sigma ={\left({\displaystyle \frac{1}{n-1}}\sum _{i=1}^n{\left(x_i-\overline{x}\right)}^2\right)}^{{\displaystyle \frac{1}{2}}}$$

(1)

where $\sigma$ represents the standard deviation of the sample, n is the number of monitoring data points, $X_i$ denotes the i-th monitoring data point, and $\overline{X}$ is the mean of the monitoring data.

The 3σ criterion works well in the post-processing of monitoring data; however, it is ineffective for real-time monitoring. In practical monitoring scenarios, if deformation occurs, the traditional 3σ criterion may incorrectly classify the deformation data as outliers, leading to erroneous removal. To address this issue, the paper introduces a delayed processing method based on the 3σ criterion. Since the improved 3σ criterion processes data over a defined interval, it provides a confidence interval when the data volume is a multiple of a set value. In the absence of deformation, data points outside the confidence interval are directly treated as outliers and removed. When deformation occurs, the confidence interval is adjusted to prevent deformation data from being misclassified as outliers. The occurrence of deformation is determined by whether the amount of data removed falls within the threshold, enabling the effective real-time monitoring of deformation data.

(2)

Data Interpolation

To address data loss, data interpolation is required. Common interpolation methods include linear interpolation, Lagrange interpolation, cubic spline interpolation, and piecewise cubic Hermite interpolation. Common interpolation methods include linear interpolation, Lagrange interpolation, cubic spline interpolation, and piecewise cubic Hermite interpolation. This study adopts the piecewise cubic Hermite interpolation method for data interpolation, as it not only avoids the divergence of higher-order interpolation functions but also does not require complex computational processes. The piecewise cubic Hermite interpolation polynomial $I_k(x)$ satisfies $I_k(x)\in C^{\prime }[a,b]$; $I_k\left(x_k\right)=f_k,I_k^{\prime }\left(x_k\right)=f_k^{\prime },k=0,1,\cdots ,n$; in the context of piecewise cubic Hermite interpolation, the polynomial $I_k(x)$ defined over each subinterval $\left[\begin{array}{c}{X}_k,X_{k+1}\end{array}\right]$ is a cubic polynomial. When $n=1$:

$$\begin{array}{c}{I}_k\left(x\right)={\left({\displaystyle \frac{x-x_{k+1}}{x_k-x_{k+1}}}\right)}^2\left(1+2\cdot {\displaystyle \frac{x-x_k}{x_{k+1}-x_k}}\right)f_k+{\left({\displaystyle \frac{x-x_k}{x_{k+1}-x_k}}\right)}^2\left(1+2\cdot {\displaystyle \frac{x-x_{k+1}}{x_k-x_{k+1}}}\right)f_{k+1}\\ +{\left({\displaystyle \frac{x-x_{k+1}}{x_k-x_{k+1}}}\right)}^2\left(x-x_k\right)f_k^{\prime }+{\left({\displaystyle \frac{x-x_k}{x_{k+1}-x_k}}\right)}^2\left(x-x_{k+1}\right)f_{k+1}^{\prime }\end{array}.$$

(2)

To construct a set of piecewise cubic interpolation basis functions on the interval [a, b],

$$I_k\left(x\right)=\sum _{j=0}^n\left[{\alpha }_j\left(x\right)f_j+{\beta }_j\left(x\right)f_j^{\prime }\right]$$

(3)

2.1.2. Filtering Algorithm

(1)

Discretized Model of the Observation System

The external climbing scaffold monitoring is a short-baseline positioning system, and, for most of the time, it remains stationary, only transitioning into a moving state during the ascent. The instantaneous acceleration $\ddot{\mathit{x}}(t)$ can be considered as random interference, i.e., treated as system noise w(t). The state vector of the observation point is taken as

$$\mathit{X}(t)={[\mathit{x}(t) \dot{\mathit{x}}(t)]}^{\mathrm{T}},$$

(4)

the following continuity equation can be used to represent

$$\stackrel{.}{\mathit{X}}(t)=\left[\begin{array}{cc}0& 1\\ 0& 0\end{array}\right]\mathit{X}(t)+\left[\begin{array}{c}0\\ 1\end{array}\right]\mathit{w}(t),$$

(5)

its solution is

$$\mathit{X}(t)=\left[\begin{array}{cc}1& t-t_0\\ 0& 1\end{array}\right]\mathit{X}(t_0)+{\displaystyle {\int }_{t_0}^t\left[\begin{array}{cc}1& t-\tau \\ 0& 1\end{array}\right]}\left[\begin{array}{c}0\\ 1\end{array}\right]\mathit{w}(\tau )\mathrm{d}\tau$$

(6)

Discretize the above equation:

$$\left.\mathit{X}(t_{k+1})=\left[\begin{array}{cc}1& t_{k+1}-t_k\\ 0& 1\end{array}\right.\right]\mathit{X}(t_k)+{\displaystyle {\int }_{i,}^{t_{k+1}}\left[\begin{array}{cc}1& t_{k+1}-\tau \\ 0& 1\end{array}\right]}\left[\begin{array}{c}0\\ 1\end{array}\right]\mathit{w}(\tau )\mathrm{d}\tau ,$$

(7)

use the subscripts $k+1$ and $k$ to represent $t_{k+1}$ and $t_k$, respectively, to solve the above equation, and define:

$${\mathit{w}}_k={\displaystyle \frac{1}{\Delta t}}{\displaystyle {\int }_{t_k}^{t_{k+1}}\mathit{w}}(\begin{array}{c}\tau )\end{array}\mathrm{d}\tau ,$$

(8)

This gives the state equation of the observation system:

$${\mathit{X}}_{k+1}=\left[\begin{array}{cc}1& \Delta t\\ 0& 1\end{array}\right]{\mathit{X}}_k+\left[\begin{array}{c}{\displaystyle \frac{1}{2}}\Delta t^2\\ \Delta t\end{array}\right]{\mathit{w}}_k,$$

(9)

where $\Delta t$ is the observation time interval, and, since the time intervals between adjacent observations are the same, it is set to 1. The observation equation is

$${\mathit{y}}_k=[1 0]\left[\begin{array}{c}{x}_k\\ ⋮\\ x_k\end{array}\right]+{\mathit{\nu }}_k.$$

(10)

The discretized model of the above observation system can be represented as

$$x_{k+1}=\mathit{A}{x}_k+\Gamma w_k,$$

(11)

$$y_k=\mathit{B}{x}_k+v_k.$$

(12)

(2)

Robust Kalman Filtering

In classical Kalman filtering, the observation noise is typically assumed to be system errors, white noise with a uniform power spectral density distribution, and a random Gaussian–Markov process, with errors exhibiting randomness and independence. However, in practical observation processes, the noise is not randomly distributed, and continuous large outliers may exist in the data. In such cases, classical Kalman filtering is not effective at eliminating or adjusting for the continuous data fluctuations caused by persistent outliers, leading to significant biases in the filtered results as time progresses.

When large, continuous deviations appear in the data, robust Kalman filtering can be employed. Building upon the classical filtering approach, robust Kalman filtering adjusts the observation noise covariance matrix by replacing the original observation noise covariance matrix with the observation covariance matrix, thus mitigating the impact of outliers on the filtering process. The Kalman filtering process can be summarized into two stages: prediction and update.

Prediction: In the prediction step of the Kalman filter, the system’s state and its uncertainty (covariance matrix) are predicted based on the previous state estimate and the system dynamics. The state prediction equation is as follows:

$${\widehat{\mathit{x}}}_{k+1|k}={\mathit{A}}_d{\widehat{\mathit{x}}}_{k|k}+{\mathit{B}}_d{u}_k,$$

(13)

$${\mathit{P}}_{k+1|k}={\mathit{A}}_d{\mathit{P}}_{k|k}{\mathit{A}}_d^{\mathrm{T}}+{\mathit{Q}}_d,$$

(14)

where ${\widehat{\mathit{x}}}_{k+1|k}$ is the predicted state at time $k+1$ based on the state at time $k$, ${\mathit{A}}_d$ is the state transition matrix that models the system’s dynamics, ${\widehat{\mathit{x}}}_{k|k}$ is the state estimate at time k, B is the control input matrix, $u_k$ is the control input. ${\mathit{P}}_{k+1|k}$ is the predicted state covariance matrix at time $k+1$, ${\mathit{P}}_{k|k}$ is the state covariance matrix at time $k$, and ${\mathit{Q}}_d$ is the process noise covariance matrix.

Update: In the update step of the Kalman filter, the predicted state is corrected using new measurements to obtain a more accurate state estimate. The process involves the following steps:

$${\mathit{K}}_{k+1}={\mathit{P}}_{k+1|k}{\mathit{H}}^{\mathrm{T}}{[\mathit{H}{\mathit{P}}_{k+1|k}{\mathit{H}}^{\mathrm{T}}+{\mathit{R}}_d]}^{-1},$$

(15)

$${\widehat{\mathbf{x}}}_{k+1|k+1}={\widehat{\mathbf{x}}}_{k+1|k}+{\mathit{K}}_{k+1}[z_{k+1}-\mathit{H}{\widehat{\mathit{x}}}_{k+1|k}],$$

(16)

$${\mathit{P}}_{k+1|k+1}=[\mathit{I}-{\mathit{K}}_{k+1}\mathit{H}]{\mathit{P}}_{k+1|k},$$

(17)

where ${\mathit{K}}_{k+1}$ is the Kalman gain, $\mathit{H}$ is the measurement matrix, ${\mathit{R}}_d$ is the measurement noise covariance matrix. ${\widehat{\mathbf{x}}}_{k+1|k+1}$ is the updated state estimate, $z_{k+1}$ is the actual measurement, $\mathit{H}{\widehat{\mathit{x}}}_{k+1|k}$ is the predicted measurement, and $\mathit{I}$ is the identity matrix.

(3)

Equivalent Weight Function

In robust Kalman filtering, the key aspect of data processing lies in the selection of an appropriate equivalent weight function. Different equivalent weight functions result in different equivalent observation covariance matrices, ultimately leading to varying filtering outcomes. Commonly used equivalent weight functions in practical applications include the Huber, IGG1, and IGG3 functions. In this study, the IGG3 equivalent weight function is adopted for robust estimation. The IGG3 weight function is an improved approach based on the bounded nature of measurement errors. It extends the two-segment division of IGG1 by introducing an additional segment, classifying the data into three distinct regions: the normal region, the suspicious region, and the elimination region. In the normal region, observations are retained with their original weights, without any modification. In the suspicious region, the weights of the observations are reduced to mitigate the impact of potential gross errors. In the elimination region, observations with excessively large residuals are completely discarded, with their weights set to zero. This weighting strategy effectively suppresses the influence of gross errors, making it particularly effective in scenarios where a significant number of gross errors exist in the observation data. It is expressed as follows:

$${\overline{P}}_i=\left\{\begin{array}{cc}{P}_i& \left|v\right|⩽k_0\\ P_i{\displaystyle \frac{k_0}{\left|v\right|}}({\displaystyle \frac{k_1-\left|v\right|}{k_1-k_0}})& k_0<\left|v\right|<k_1\\ 0& \left|v\right|⩾k_1\end{array}\right.,$$

(18)

where $P_i$ represents the weight matrix corresponding to the observation value. $v$ denotes the standardized residual. $k_0$ is typically set within the range of 1.0 to 1.5. $k_1$ is generally chosen between 3.0 and 4.5. After each filtering update, iterative calculations are performed until the difference between the state estimate at iteration t(${\widehat{\mathit{X}}}_k$) and the estimate at iteration t − 1 (${\widehat{\mathit{X}}}_{k,t-1}$) is smaller than a predefined threshold, at which point the iteration process terminates.

2.1.3. Climbing Scaffold Height Constrained Model

Climbing frame monitoring falls under short-baseline positioning. Short-term multipath effects typically refer to those occurring within a very short time span (e.g., milliseconds to seconds) and are particularly pronounced in urban canyon environments or areas with dense high-rise buildings. When monitoring a station located in the middle of a city, short-term multipath effects can cause rapid fluctuations in the signal strength and phase received by the receiver, thereby affecting positioning accuracy. Introducing constraints on the vertical direction during positioning can effectively suppress the impact of short-term multipath effects. Therefore, a height constraint estimation is incorporated into each iteration of the Kalman filter. Due to the structural design of high-rise buildings, the initial stage of external climbing scaffolding ascent does not produce deviations, maintaining a strictly vertical trajectory. However, in the later stages of construction, some high-rise buildings feature a torsional design, causing the ascent path of the climbing scaffolding to deviate by an angle $\theta$ relative to the horizontal direction. To address this, the following inequality constraint is introduced into each iteration of the Kalman filter to refine height estimation:

$$\begin{array}{r}|U_k-U_{k-1}|=\sqrt[2]{{\left(N_k-N_{k-1}\right)}^2+{\left(E_k-E_{k-1}\right)}^2}\cdot \tan \theta \\ \le \sqrt[2]{{\left(N_k-N_{k-1}\right)}^2+{\left(E_k-E_{k-1}\right)}^2}\cdot \tan {\theta }_{th}\end{array}.$$

(19)

In the early stages of monitoring, as the climbing scaffolding ascends almost vertically, a direct constraint on the height threshold can be set. Therefore, the overall filtering height constraint for the climbing scaffolding is formulated as follows:

$$f(U_k,U_{k-1},N_k,N_{k-1},E_k,E_{k-1})=\left\{\begin{array}{cc}\Delta U\le \tan ({\theta }_{th})\cdot \sqrt{\Delta N^2+\Delta E^2}& \mathrm{if} later stages\\ \Delta U\le h_{th}& \mathrm{if} early stages \end{array}\right..$$

(20)

2.2. Climbing Scaffold Status Detection Algorithm

When monitoring the status of the climbing scaffolding, the Cumulative Sum Control Chart (CUSUM) principle can be utilized. The sliding window-based CUSUM algorithm is an effective event detection method that, with appropriately set parameters, can automatically detect state changes in the input time-series signal. This method is characterized by its simple logic and strong anti-interference capability.

As illustrated in Figure 1, the U-direction coordinate time series monitored by BeiDou defines a reference window $w_{\mathrm{m}}$ and a climbing detection window $w_{\mathrm{n}}$. The climbing detection window follows immediately after the reference window. To ensure a sufficient coupling degree between height variations within the windows, a certain overlapping region exists between the two windows.

The window lengths are set to m and n, respectively, with an overlapping length set to x. Considering that the climbing scaffolding moves only vertically, with an approximate climbing height of 4.5 m per ascent and a sampling interval of 1 s, the reference window and climbing detection window lengths are set to 60, while the overlapping length is set to 30. The mean values of the two windows are calculated using the following formula:

$$M_{\mathrm{m}}={\displaystyle \frac{1}{m}}\sum _{j=k-m-n+x+1}^{k-n+x}p(j),$$

(21)

$$M_{\mathrm{n}}={\displaystyle \frac{1}{n}}\sum _{j=k-n+1}^{k}p(j),$$

(22)

where K represents the starting sampling point for calculation, located at the end of the ascent detection window.

During the early stage of climbing, the scaffold moves strictly in a vertical direction with a relatively consistent ascent speed and distance. In this phase, the threshold-based ascent variation can be directly utilized for detection. The difference is used as the characteristic parameter for identifying ascent events, and its calculation formula is as follows:

$$D_{\mathrm{k}}=M_{\mathrm{m}}-M_{\mathrm{n}}.$$

(23)

When the characteristic parameter is positive and remains above the preset threshold $D_m$ for most of a given period, it is considered that the ascent behavior begins at the leading edge of the first monitoring window that exceeds the threshold. Conversely, when the characteristic parameter gradually decreases from continuously exceeding the threshold to stabilizing below it, the endpoint of the last window exceeding the threshold is identified as the end of the ascent.

For detecting the duration of the ascent, since the building currently only undergoes vertical ascent with minimal variation in ascent time, a cumulative summation method is adopted to track the ascent duration. The ascent event cumulative sum $r_k^{+}$ is defined, where the accumulator starts counting upon detecting an ascent event. It continues to accumulate during the ascent phase until the ascent ends, at which point the accumulator resets to zero. This marks a complete ascent event, recording both the total ascent duration and the total change in ascent distance. The specific calculation formula is as follows:

$$r_k^{+}=\left\{\begin{array}{ll}1,& D_k>D_{\mathrm{th}}\\ r_k^{+}+1,& \mid D_{k+1}\mid ⩾\mid D_k\mid \\ 0,& \mid D_{k+1}\mid <\mid D_k\mid \end{array}\right..$$

(24)

By using this method, the monitoring system can effectively track and record the number of ascent events and the corresponding ascent durations of the climbing frame.

2.3. Climbing Scaffold Status Warning Model

To ensure the operational safety of climbing scaffolds in super-high-rise construction, this study introduces a real-time warning model based on the Cumulative Sum (CUSUM) anomaly detection algorithm. The model identifies abnormal displacement trends by continuously monitoring GNSS-based displacement data and issuing warnings when cumulative deviations exceed predefined thresholds. Compared with traditional threshold-based monitoring, CUSUM is more sensitive to small but sustained deviations, making it particularly suitable for early warning applications in dynamic scaffold monitoring.

2.3.1. CUSUM-Based Anomaly Detection Model

The CUSUM algorithm follows a recursive formulation to track cumulative deviations in displacement data. For the coordinate sequence of GNSS monitoring $X(t),t=1,2,\dots ,n$, it is assumed to be mutually independent and follows a normal distribution $N({\mu }_0,\sigma )$, where ${\mu }_0$ represents the mean of the sample sequence and $\sigma$ denotes the standard deviation. If the monitored object undergoes deformation, the coordinate sequence transitions to an approximate distribution of $X\left(t\right)\sim N\left({\mu }_0+\delta \sigma ,\sigma \right)$, with a mean shift of $\delta \sigma$. The null hypothesis H₀ states that the coordinate sequence remains unchanged, meaning its mean does not vary. The alternative hypothesis H₁ posits that an abrupt change occurs at time $s(s<n)$, indicating a deformation at time $s$. It is assumed that, from time $s$ onward, the mean of the monitoring data undergoes a significant change. The probability density functions before and after time $s$ can be expressed as follows:

$$\left\{\begin{array}{c}{f}_0\left(x\right)={\displaystyle \frac{1}{\sigma \sqrt{2\pi }}}\exp \left[-{\displaystyle \frac{{\left(x-{\mu }_0\right)}^2}{2{\sigma }^2}}\right],H_0\mathrm{established}\\ f_1\left(x\right)={\displaystyle \frac{1}{\sigma \sqrt{2\pi }}}\exp \left[-{\displaystyle \frac{{\left(x-{\mu }_1\right)}^2}{2{\sigma }^2}}\right],H_1\mathrm{established}\end{array}\right..$$

(25)

Construct the log-likelihood ratio statistic:

$${\lambda }_n=\sum _{i=1}^n\ln {\displaystyle \frac{f_1\left(x_i\right)}{f_0\left(x_i\right)}}=\sum _{i=1}^n{\displaystyle \frac{1}{2{\sigma }^2}}\left[{\left(x_i-{\mu }_0\right)}^2-{\left(x_i-{\mu }_1\right)}^2\right]$$

(26)

Let the mean shift be ${\mu }_1-{\mu }_0=\delta \sigma =\Delta , k=\Delta /2$. Then, we have ${\lambda }_n={\displaystyle \sum _{i=1}^n}{\displaystyle \frac{\Delta }{{\sigma }^2}}\left(x_i-{\mu }_0-k\right)$.

If $\Delta >0$, the above equation is equivalent to $C_n^{+}=\max {\displaystyle \sum _{i=1}^n}\left(x_n-{\mu }_0-k\right)$, where $C_n^{+}$ represents the likelihood ratio statistic, which can be further expressed as:

$$C_n^{+}=C_{n-1}^{+}+\left(x_n-{\mu }_0-k\right)=\max \left(0,C_{n-1}^{+}+x_n-{\mu }_0-k\right).$$

(27)

This equation is referred to as the CUSUM statistic when the mean shifts upwards. Similarly, for a downward shift in the mean, the corresponding CUSUM statistic is given by

$$C_n^{-}=\min \left(0,C_{n-1}^{-}+x_n-{\mu }_0+k\right),$$

(28)

where k is the reference value, and h is the decision threshold.

Based on the aforementioned upper and lower CUSUM statistics, the corresponding deformation disaster early warning model is constructed as follows:

$$\left\{\begin{array}{l}{C}_0^{+}=0\\ C_N^{+}=\max \left(0,{\mathrm{C}}_{n-1}^{+}+{\mathrm{x}}_n-{\mu }_0-k\right)\\ when C_n^{+}>h,warning\end{array}\right.,$$

(29)

$$\left\{\begin{array}{l}{C}_0^{-}=0\\ C_N^{-}=\min \left(0,C_{n-1}^{-}+{\mathrm{x}}_n-{\mu }_0+k\right)\\ when C_n^{-}<-h,warning\end{array}\right.,$$

(30)

where $C_0^{+}$ and $C_0^{-}$ represent the initial values of the upper and lower CUSUM statistics, typically set to zero. The threshold h serves as the decision criterion for detecting deformation disasters. Specifically, when the upper statistic exceeds h or the lower statistic falls below −h, deformation is considered to have occurred, triggering an early warning.

2.3.2. External Climbing Frame Status Warning

The climbing scaffold status warning model is an integral part of the external climbing formwork state monitoring system (Figure 2). It processes real-time BeiDou/GNSS displacement data, enabling automated anomaly detection and risk assessment. The model workflow consists of the following steps:

• Data Input: This step receives filtered displacement data (E, N, and U components) from the GNSS monitoring system.
• CUSUM Calculation: This step continuously computes positive and negative cumulative deviations to detect scaffold movement trends.
• Threshold-Based Alerting: If the CUSUM statistic exceeds a predefined threshold, an alert is triggered.
• Risk Notification: Warning messages are sent to the construction monitoring platform, enabling timely safety intervention.

To ensure adaptability in varying construction conditions, the system implements multi-level warnings based on the deviation magnitude: Level 1 Warning: minor displacement anomaly detected, requiring observation; Level 2 Warning: significant deviation observed, requiring further monitoring; and Level 3 Warning: large abnormal displacement, requiring immediate safety intervention.

By dynamically adjusting thresholds based on historical data and real-time trends, the system effectively minimizes false alarms while providing a timely risk assessment. This approach ensures the safe and stable operation of climbing scaffolds in super-high-rise buildings.

2.4. Climbing Scaffold Monitoring and Processing Workflow

The overall process of monitoring the status of the external climbing framework in super-high-rise buildings based on BeiDou/GNSS technology consists of four key stages: data acquisition, preprocessing, filtering optimization, and state recognition with warning mechanisms.

(1)

Data Acquisition and Transmission: BeiDou/GNSS-integrated receivers are deployed at critical locations on the climbing framework (e.g., four corners), while a reference station is established around the building. Dual-frequency observations from the four satellite constellations are collected to ensure comprehensive coverage. The monitoring data are transmitted to the data processing center in real time at a 1 s epoch interval, ensuring data continuity and timeliness.

(2)

Data Preprocessing: Given that GNSS signals are susceptible to multipath effects and non-line-of-sight errors in urban canyon environments, an improved Rainer criterion is first applied to remove gross errors. The confidence interval is dynamically adjusted to prevent deformation data from being misidentified as gross errors. Additionally, missing data are interpolated using a piecewise cubic Hermite interpolation method to ensure the completeness of the data sequence.

(3)

Filtering Optimization and State Modeling: The preprocessed data are fed into a robust Kalman filter model. By adjusting the observation noise covariance matrix (using the IGG3 equivalent weight function), the impact of continuous gross errors and short-term multipath effects is mitigated. Furthermore, height constraint conditions are introduced—setting threshold limitations during the vertical climbing phase—to further optimize dynamic positioning accuracy. The filtered data effectively separate high-frequency noise from low-frequency displacement trends, providing smooth and reliable time-series data for further analysis.

(4)

State Recognition and Warning Mechanism: Based on the filtered data, a sliding-window CUSUM algorithm is employed to detect the climbing status in real time. The feature quantity is calculated by analyzing the mean difference between a reference window and a detection window, while predefined thresholds are used to determine the start and end times of the climbing motion. The climbing height and duration are also recorded. In both horizontal and vertical directions, a multi-level CUSUM warning model is established to accumulate minor displacement deviations and dynamically trigger different warning levels. If monitoring values exceed the predefined safety thresholds, the system automatically generates warning notifications and transmits them to the construction management platform, facilitating timely intervention by decision-makers.

The overall monitoring framework based on BeiDou/GNSS technology for the external climbing framework in super-high-rise buildings is illustrated in the following diagram:

3. Experimental Verification

3.1. Data Acquisition and Preprocessing

3.1.1. Data Acquisition

To validate the feasibility of the proposed climbing formwork state monitoring method, Building 3 of Phase B, China Overseas City Plaza, Tianjin (2024) was selected as the monitoring object. The main office tower is designed to reach a height of approximately 339.9 m and is located in the city center, where the construction environment is highly complex. During construction, an attached climbing scaffolding system is utilized for enclosure and safety. The monitoring items are shown in Figure 3.

There are a total of four monitoring stations, each installed at one of the four corners of the climbing scaffold, with two reference monitoring stations set up on the building’s exterior. The monitoring instruments are mounted on pre-welded brackets, positioned approximately half a meter away from the building’s exterior surface. The distribution of monitoring stations is illustrated in Figure 4.

3.1.2. Experimental Data Preprocessing

The monitoring system utilizes an integrated receiver capable of observing dual-frequency signals from three satellite navigation systems, with an epoch interval of 1 s. Once installed, the equipment continuously performs the real-time monitoring of the scaffolding structure.

During the satellite-based monitoring, interference from surrounding high-rise buildings and other complex environmental factors can degrade the quality of satellite signals during certain periods. This degradation results in the inability to obtain precise fixed solutions, leading to significant deviations in the monitoring data. To illustrate the effectiveness of gross error elimination, 12 h of monitoring data from a selected day were analyzed. As shown in Figure 5a, the raw monitoring data exhibit significant outliers at the fourth and eighth hours, along with additional deviations of approximately 1 m at certain other time points. To address these anomalies, gross error elimination was applied to the original monitoring data. As shown in Figure 5b, after outlier removal, the fluctuation range of the monitoring data is stabilized within ±0.05 m. This improvement indicates a significant enhancement in data stability and reliability, allowing for a more accurate representation of the actual monitoring conditions. The refined dataset provides a robust foundation for subsequent data analysis and research.

3.2. Monitoring Performance Analysis

3.2.1. Monitoring Performance During the Climbing Scaffold Stationary Phase

Monitoring data for 20 h on 29 June, when the climbing frame was station-ary, was selected for analysis. As shown in Figure 6b, at any given time, at least 15 BeiDou satellites were continuously tracked. Nevertheless, the number of visible satellites varied significantly over time, indicating fluctuations in satellite availability during the monitoring process.

The corresponding outliers are removed, and the preprocessed data after correction can be used for filtering and smoothing. As shown in Figure 7, a comparison is made between the preprocessed data and the corresponding filtered data.

As shown in Figure 7, the raw monitoring data exhibit significant high-frequency noise across all three directions, with noticeable gross errors at certain time points. These fluctuations lead to considerable data instability, affecting the real-time monitoring accuracy of the climbing scaffold system. Among them, the U direction (Figure 7c) demonstrates the lowest monitoring precision, with a fluctuation range of approximately [−0.08 m, 0.08 m], where high-frequency noise is particularly prominent. In contrast, the N direction (Figure 7b) shows the smallest fluctuation range, approximately [−0.02 m, 0.02 m], indicating the highest precision with relatively less noise. After applying filtering and data optimization techniques, high-frequency noise in all three directions was significantly suppressed, leading to a noticeable improvement in data smoothness. In the U direction, the fluctuation range effectively converged to [−0.04 m, 0.04 m], successfully eliminating abnormal spike noise. The optimized data not only preserve the original displacement trend but also mitigate high-frequency noise interference, resulting in a more stable dataset. The removal of spike anomalies through filtering ensures that the processed data better reflect the actual structural displacement patterns.

To quantify the accuracy assessment, the mean value of a long-duration static dataset, during which the scaffold did not ascend, was taken as the ground truth. The root mean square errors (RMSEs) in the E, N, and U directions were then calculated. After applying filtering and other data optimization processes, the RMSEs in the E, N, and U directions were reduced by 51.4%, 45.5%, and 49.6%, respectively, compared to the original displacement sequence. Specific values are shown in

Table 1. Static phase monitoring accuracy analysis table.

RMSE	E	N	U
Preprocessed Data	0.0107	0.0044	0.0137
Filtered Data	0.0052	0.0021	0.0069

:

3.2.2. Monitoring Performance During the Climbing Scaffold Ascending Phase

For scaffolding structures, the majority of the time is spent in a stationary state. However, as the building progresses, monitoring the climbing status of the scaffold becomes crucial. To comprehensively evaluate the monitoring system’s dynamic detection capability under different displacement magnitudes, this study analyzes two typical climbing events: a large displacement scenario on 30 July, characterized by significant variations in both the horizontal and vertical directions, and a small displacement scenario on 30 June, where vertical displacement dominates with minor horizontal perturbations. By comparing the data processing results from these two scenarios, the system’s robustness and sensitivity are separately validated.

Figure 8 presents the monitoring data from 30 July before and after filtering. The raw data exhibited spike noise at approximately 1.5 h and 5.5 h, but this noise was effectively suppressed through filtering, demonstrating the system’s strong data smoothing capability. Around 10 h, the climbing scaffold began its ascent (Figure 8c), resulting in noticeable changes in the horizontal directions. In the E direction (Figure 8a), displacement gradually increased after the climbing process began, reaching a maximum of approximately 0.04 m. In the N direction (Figure 8b), displacement reached approximately 0.02 m. Despite minor horizontal disturbances, the overall trend remained stable. The monitoring results for the U direction (Figure 8c) indicate that, after climbing initiation, the vertical displacement increased rapidly, ultimately reaching approximately 4.4 m, which is consistent with the expected climbing height. Overall, the results from 30 July confirm that the system can accurately capture both horizontal and vertical scaffold movements while effectively suppressing noise interference, thereby improving data reliability. Additionally, all measured parameters meet design specifications, and the BeiDou/GNSS monitoring results align closely with the actual climbing progress. This further validates the system’s reliability and anti-interference capability in super-high-rise construction monitoring.

Figure 9 illustrates the climbing scaffold monitoring process on 30 June. Compared to the results from 30 July, this dataset exhibits more subtle horizontal displacements. Throughout the climbing phase, the system successfully captured uniform short pauses, presenting an overall linear displacement pattern, which aligns well with the actual climbing scaffold movement. Additionally, after the climbing phase ended, the system stabilized in a shorter period, indicating that the improved algorithm demonstrates better adaptability in both dynamic and static phases. During the pre-climbing static phase, noticeable noise fluctuations were observed around 1 h, 2 h, and 5 h (Figure 9a,b). After applying filtering techniques, these fluctuations were effectively suppressed. Directly observing the E and N direction data in Figure 9a,b shows minimal displacement deviations, making it difficult to obtain accurate analytical results. Therefore, a more detailed analysis was conducted using the dataset presented in

Table 2. Hourly mean displacement values in the E and N directions.

	1	2	3	4	5	6	7	8	9	10
E	−0.0041	−0.0023	−0.0068	−0.0006	0.0005	0.0071	0.0101	0.0109	0.0144	0.0127
N	0.0044	0.0010	0.0047	0.0017	0.0029	−0.0020	−0.0075	−0.0042	0.0003	0.0030

, which provides hourly mean displacement values for the E and N directions. From the processed data, it is evident that the E-direction displacement exhibits a more significant variation after preprocessing, increasing from −0.0041 m at 1 h to 0.0127 m at 10 h, with an overall increase of 0.0168 m. This trend demonstrates a gradual upward displacement with a decreasing fluctuation magnitude. In contrast, the N-direction displacement remains relatively stable, changing only slightly from 0.0044 m at 1 h to 0.0030 m at 10 h, with a total variation of just 0.0014 m. These results indicate that the applied preprocessing method is not only effective for detecting large-scale displacement variations but also highly sensitive to subtle displacement changes, ensuring accurate motion detection in complex construction environments.

The results indicate that the BeiDou/GNSS-based monitoring system can reliably capture large-scale scaffold displacements (as observed in the 30 July dataset) while also precisely detecting subtle horizontal perturbations and vibrations (as demonstrated by the 30 June dataset). The consistency of data across these two scenarios validates the system’s high sensitivity and robustness in dynamic monitoring, providing a crucial technical safeguard for the safety of super-high-rise construction.

3.2.3. Performance Analysis of Climbing Scaffold Status Warning

During the monitoring of high-rise building climbing scaffolds, in addition to analyzing the current movement status, it is crucial that we utilize monitoring data for early the warning of potential risks, such as scaffold detachment or falling hazards. Timely warning detection is essential for ensuring construction safety, as it helps mitigate potential losses by providing predictive alerts. Since the monitoring accuracy and risk standards vary across the three directions (E, N, and U), three different levels of early warning thresholds are set for each direction. The following figure presents the three-level warning results in a static state for each direction.

From Figure 10, it is clear that warnings are triggered at peak fluctuation moments (i.e., significant displacement events) and during continuous descent periods (i.e., potential fall risks). This method effectively accumulates minor fluctuations in the displacement sequence, enabling the identification of both abrupt and small-scale anomalous changes. The figure also indicates that monitoring accuracy in the elevation (U) direction is relatively low, while the N direction exhibits the highest accuracy. Consequently, the precision thresholds for warnings should be adjusted accordingly. By establishing different warning standards within the same direction, the system categorizes warning levels based on displacement severity. Furthermore, differentiated thresholds are set across directions to align with the monitoring accuracy and directional requirements, ensuring the precise surveillance of the climbing scaffold’s status. In our implementation, the warning thresholds for the Level 1, Level 2, and Level 3 alerts are set to 0.03, 0.05, and 0.07, respectively. It can be observed that, as the warning threshold increases, the frequency of warnings decreases; certain cumulative displacements that previously triggered a warning are no longer considered significant enough to do so under the defined probability criteria.

4. Discussion

The results confirm that the proposed BeiDou/GNSS-based monitoring method significantly enhances the accuracy and reliability of climbing scaffold monitoring in super-high-rise buildings. Through experimental validation on a real-world construction site, the system demonstrated stable performance in tracking scaffold movements and effectively captured displacement changes under varying environmental conditions. Compared to traditional manual or single-sensor monitoring approaches, the BeiDou/GNSS-based system provides continuous, high-precision positioning, reducing potential human errors and improving the real-time safety assessment. In contrast to the approach reported in [10], which employed a 16-level sensor system solely for analyzing the vertical displacement of the main load-bearing layer and, thus, only provided point-to-point measurements without a comprehensive assessment of the scaffold’s climbing posture, the present study offers a more holistic monitoring solution. Specifically, while [10] used fixed threshold values for the graded warning that directly triggered corrective actions when exceeded, our method leverages a CUSUM-based anomaly detection mechanism to precisely capture and accumulate even minor displacement variations. This allows for the more accurate identification and early warning of anomalies. Furthermore, by setting graded warning thresholds tailored to the monitoring precision in different directions, our approach is better aligned with practical requirements for scaffold safety management.

While the proposed BeiDou/GNSS-based monitoring system demonstrates high accuracy in tracking climbing scaffold movements, several limitations remain. First, the validation was conducted on a single construction project, which may not fully capture the variations in structural designs, construction methods, and environmental conditions. This limited scope raises concerns about the system’s generalizability across different projects. Second, relying solely on BeiDou/GNSS for displacement monitoring presents challenges in capturing short-term dynamic movements due to the inherent sensor limitations. Additionally, external factors such as signal obstructions and multipath effects may impact measurement accuracy, necessitating the integration of complementary monitoring approaches.

To address these limitations, several improvements can be implemented. Expanding validation across multiple high-rise construction projects with varying structural designs and environmental conditions will enhance the system’s generalizability and reliability. Additionally, integrating MEMS-based IMUs with GNSS data can mitigate short-term positioning errors and improve real-time displacement tracking. Furthermore, incorporating cloud computing and IoT technologies will enable real-time data processing, automated anomaly detection, and remote monitoring, enhancing overall safety and operational efficiency.

By addressing these areas, future research can further improve the accuracy, reliability, and applicability of BeiDou/GNSS-based climbing scaffold monitoring systems, contributing to safer and smarter super-high-rise construction.

5. Conclusions

This paper proposes a BeiDou/GNSS-based real-time monitoring system for climbing scaffolds in super-high-rise construction, integrating multi-level data optimization and an anomaly detection mechanism to enhance monitoring accuracy and safety. The system was experimentally validated on an actual super-high-rise construction project, and the findings confirm its effectiveness in tracking scaffold displacement and detecting potential safety risks. Compared to conventional monitoring approaches, the proposed method significantly improves the real-time accuracy, robustness, and early warning capabilities, providing a reliable solution for ensuring scaffold stability in complex construction environments.

Experimental validation demonstrated a high static positioning accuracy and robust dynamic tracking performance. During static monitoring on 29 June, the RMSE values were significantly reduced through adaptive filtering and gross error correction, achieving reductions of 51.4% in the East (E) direction, 45.5% in the North (N) direction, and 49.6% in the Vertical (U) direction, ensuring sub-centimeter stability.

For dynamic monitoring, two climbing events were analyzed. The first event on June 30 confirmed the system’s ability to track the scaffold ascent with a measured displacement of 4.42 m, capturing millimeter-level horizontal perturbations (1.68 cm in E and 0.14 cm in N direction), demonstrating a high sensitivity to minor oscillations. The second event on 30 July recorded a vertical displacement of 4.38 m, with horizontal deviations of 4.0 cm (E) and 2.0 cm (N) due to wind loads and environmental factors. These findings validate the system’s capability to detect both large-scale scaffold movements and fine horizontal oscillations with high precision.

Beyond displacement tracking, the system integrated a CUSUM-based anomaly detection algorithm, which successfully identified deviations in scaffold movement and triggered early warnings for potential structural risks. Compared to conventional manual inspection and single-sensor monitoring approaches, the proposed BeiDou/GNSS-based system significantly enhances real-time adaptability, measurement accuracy, and automated risk assessment, making it a practical solution for scaffold monitoring in complex construction environments.

Future research should focus on expanding validation across multiple construction projects with varying structural designs and environmental conditions to improve system generalizability. Additionally, integrating MEMS-based IMUs with GNSS data could enhance short-term displacement tracking. Machine-learning-enhanced anomaly detection could further refine the warning system, improving predictive capabilities for scaffold instability. Furthermore, coupling the BeiDou/GNSS monitoring system with structural health monitoring (SHM) models, such as finite element modeling (FEM), could provide deeper insights into scaffold stability and long-term performance.