2. Literature Review
For the rectification and reinforcement of existing buildings, Wang and Li [
3] conducted a risk assessment based on Bayesian networks for rectification and reinforcement of an 11-story high-rise building. Zhou et al. [
4] used the integrated bundling technique for the first time to correct the deviation of the pagoda body of Dinglin Temple in China. Peng et al. [
5] introduced a special technique to rectify the deviation of the building near the foundation pit, and investigated the performance of the technique through field tests. Xiao et al. [
6] proposed two under-excavation calculation formulas for hole spacing and hole diameter applicable to general deviation rectification projects and verified them by examples. Xiao et al. [
7] introduced plane strain numerical simulation to systematically probe the under-excavation mechanism for building rectification, and derived the two key parameters of optimal hole spacing and hole diameter. Chai et al. [
8] estimated the jacking force required for the rectification of rigid inclined piles in cohesive soil using the
numerical method and carried out a detailed parametric analysis, which provided a basis for safer and more economical rectification design. The above literature focuses on the research at the technical level: one is to propose an innovative rectification and reinforcement scheme based on the analysis of the causes of uneven settlement of the foundation under complex conditions [
4,
5]; the other is to pay more attention to the mechanism behind the deviation rectification and provide scientific theoretical formulas for the rectification design [
6,
7,
8]. However, there have been very few studies on the identification of potential risk factors for building rectification and reinforcement and the impact of these risk factors on construction quality.
The quality state of the rectification and reinforcement system will be affected by a variety of factors. For risk modeling of complex engineering systems, probabilistic risk analysis (PRA) can not only be used to identify possible failure scenarios, but can also be used to estimate the probability of failure occurrence [
9]. The PRA provides access to critical risk factors and related causality, which can help to provide decision support for quality or safety assurance in advance. PRA has been widely used in complex systems with various risk factors under uncertainty, such as construction projects, operating subways and other civil infrastructure systems [
10,
11]. Fault tree analysis (FTA), event tree analysis (ETA) and Bayesian networks (BNs) are common methods in PRA [
12]. However, FTA faces difficulties in establishing dependencies for events, updating probabilities, and handling uncertainty, and ETA cannot handle complex dependency models well to obtain success or failure probabilities [
13]. As the fault tree or event tree grows, its structure becomes not intuitive and the computational complexity increases. The limitations of FTA and ETA necessitate the utilization of BNs, a probabilistic graphical model, to effectively capture causal relationships between risk factors through conditional probability tables (CPT). Due to its ability to update and correct probabilities, BNs can perform predictive and diagnostic reasoning to support decision making on construction risk management measures [
14]. Wu et al. [
15] conducted a dynamic safety analysis of tunnel-induced pavement damage based on dynamic Bayesian networks. Zhang et al. [
16] analyzed the safety performance of buried pipelines adjacent to the Wuhan Yangtze River Tunnel Project using the fuzzy Bayesian network (FBN). Wang and Chen. [
17] evaluated the safety risk of a metro construction project based on fuzzy comprehensive Bayesian networks (FCBN). Pan et al. [
18] conducted a risk analysis on the structural health status of operating subway tunnels based on the combination of interval-valued fuzzy sets, the D-S evidence theory and FBN. Guan et al. [
19] adopted the fuzzy Bayesian belief network (FBBN) approach to systematically assess the risk of an international engineering project. Xiang et al. [
20] established a Bayesian network based on real case data from a Canadian pipeline operator and existing fault trees in the literature to assess the risk probability of third-party excavation activities damaging existing pipelines.
Remarkably, traditional BNs can only be constructed based on crisp sets and probabilities. Due to the lack of data and incomplete knowledge, it is difficult to obtain accurate information from complex systems [
21]. As a supplement to the measurement data, domain experts can roughly estimate the probability with a series of linguistic terms, and then convert the linguistic terms into quantitative values to improve the traditional BNs. There are three main ways to achieve this transformation: Cantor set, fuzzy set and extension set. The range of Cantor sets is
, where 0 and 1 indicate whether an element belongs to a set, so Cantor sets are often used to solve binary problems. By introducing the concept of membership degree, fuzzy sets can not only describe whether an element belongs to a certain set, but can also describe its degree by the number between 0 and 1. Therefore, fuzzy sets are able to deal with the problem of unclear boundaries of linguistic concepts. Extension sets extend the range of values to
, making it possible to use raw data directly without normalization, which effectively avoids information loss [
22]. The extension theory (ET) can solve the contradiction problem by adding the description of matter properties. However, it cannot address the uncertainty caused by the change in risk factors and the fuzziness of prior knowledge. The cloud model (CM) can realize the conversion between qualitative concepts and quantitative representations through a cloud generator, and has been demonstrated to be an effective approach for dealing with the uncertainty of stochasticity and ambiguity [
23]. Therefore, the integration of the ET and the CM can not only resolve the paradoxical problem of standardizing raw data, but can also double the uncertainty during the conversion of qualitative concepts and quantitative values.
Another issue that needs attention is how to integrate the judgments from different experts in a more effective way. The Dempster–Shafer (D-S) evidence theory, one of the most exemplary information fusion techniques, has been widely used in expert systems to integrate a wide range of knowledge and data into a generalized framework due to its excellent performance in dealing with cognitive uncertainty, conflict, and bias [
24]. The prevalence of evidence conflicts is the biggest obstacle to the greater application of the evidence theory, which results in counterintuitive outcomes. In recent years, some studies have improved the evidence theory from the evidence itself [
25] and the fusion rules [
18,
26]. However, another problem is how to allocate conflicts when the contribution of experts to the decision results is different. Most of the existing studies use an averaging of the evidence set to eliminate conflicts [
18,
27], while few studies focus on the importance of different evidence sources.
With the development of construction informationization, a large amount of data has been collected during construction, such as digging data and monitoring data. How to effectively use these data to control the occurrence of construction quality accidents is the key issue of current research [
28]. Most scholars utilize the patterns of monitoring data to validate the reasonableness of the assessment results [
29,
30], without directly combining monitoring data with risk analysis [
31]. Wu et al. [
32] determined the risk factors through the data learning method based on the existing monitoring data to parameterize the risk assessment model. However, the model could not update the risk assessment results in real time according to the newly acquired monitoring data, and the relationship between the acceptance criteria for risk assessments and the early warning of monitoring indicators was not clear. Practice has shown that the risk early warning results calculated by risk assessment models are often inconsistent with the warning results when monitoring indicators reach thresholds. The single use of monitoring indicators for early warning cannot reflect the overrun of monitoring fluctuations generated by accidental factors during construction, which is easy to miss the report and affect the construction process. Therefore, it is necessary to incorporate the monitoring indicator thresholds into the risk assessment model to establish a unified early warning model to facilitate dynamic risk management of the construction process by project managers.
Existing research focuses on exploring the innovative implementation and scientific design of rectification and reinforcement techniques, and a gap still exists in the body of knowledge related to risk analysis under uncertainty. Therefore, this study attempts to construct a novel hybrid analysis approach for the quality risk perception of building rectification and reinforcement. This approach focuses on (a) constructing basic probability assignments (BPA) for multisource evidence using the ET and the CM, (b) improving the D-S evidence theory by taking into account expert contributions, and (c) incorporating monitoring indicators and their thresholds into the DBN model.
The structure of the rest of this paper is as follows:
Section 3 provides the theoretical basis for risk assessment;
Section 4 describes the proposed risk assessment model;
Section 5 conducts a case empirical study;
Section 6 analyzes and discusses the results of the risk assessment; and
Section 7 presents the conclusions and future perspectives.
6. Results and Discussion
From
Figure 7, it can be concluded regarding the overall risk perception that there is nearly a 40% probability to rate the quality risk level for the rectification and reinforcement of this high-rise building at VH (Very High). That is to say, the prediction technology can indicate a profile and related evolution of the risk level in the process of building rectification and reinforcement before the construction quality failures occur, which works well even in the absence of a clear picture of the actual status of the risk factors. The construction of this high-rise building is in a very dangerous state overall, with
. Therefore, relevant risk control measures must be taken in advance to minimize the risk level and even prevent quality failures during rectification and reinforcement. Once the risk factor (
) is adjusted and optimized, the overall risk state (
) of the construction will also be updated. Eventually, the desired
can be obtained with the continuous updating of the relevant risk factors. Set
,
,
, and
as the target nodes.
is considered to be the most unfavorable risk group leading to construction quality failures.
,
and
are identified as the most sensitive risk factors. Therefore, more attention needs to be paid to these sensitivity indicators during the rectification and reinforcement process of this high-rise building, aiming to manage the risks in a more effective way.
Figure 9 illustrates the dynamic changes in the probability of construction risk for the period 27 February–2 September. With the advancement of the rectification and reinforcement process, the possibility of the overall quality risk being in the VH (Very High) and EH (Extremely High) grades is increasing and the possibility of being in the H (High) and below grades is decreasing. When the rectification and reinforcement are essentially complete, the probability of the overall quality risk occurring gradually tends to be stable. The construction dynamic risk probability based on observational evidence is depicted in
Figure 10.
Prior to 1 April, the construction risk level was at SL (Slightly Low), that is, there was a slight risk. Risks at this stage were pervasive, and some damage was acceptable and needed to be managed. In the early stage of the project, construction preparations inevitably caused some disturbance to the inclined building. To prevent the building from continuing to tilt northward and reduce the risk of subsequent construction, two rows of steel pipe piles were added to widen the north- and west-facing raft foundations. As can be seen in
Figure 5 and
Figure 6, the fluctuation of settlement
and settlement rate
from 27 February to 1 April is relatively stable, with
at approximately 5 mm and
less than 0.2 mm/d.
From 1 April to 2 May, the construction risk rose to H level. It is necessary to strengthen management and take preventive measures. From 2 May to 14 July, the risk of rectification and reinforcement of the building reached VH level. Moreover, the probability of being at VH and EH levels was increasing, indicating an elevated likelihood of construction quality failures. The construction risk at this stage was high and damage was unacceptable, requiring increased monitoring and close attention to the dynamics of the risk probability. The middle of the project was in a critical period for the rectification and reinforcement of the building, and there were large-scale construction activities. Once a certain process was not operated properly, it would inevitably pose a great threat to the tilted building. This phase mainly involved the construction of high-pressure rotary spray piles and the tensioning of prestressing anchor cables. High-pressure jet grouting piles were added around the raft foundation to stabilize the geotechnical strata around the building, which inevitably induced significant disturbance to the underlying foundation soil. And since the construction needed to be carried out on the original foundation cap, it was impossible to avoid damage to its original bottom and upper steel bars during the drilling process. The deviation rectification of the building was achieved by first anchoring the prestressed anchor cable into the raft foundation, and then using the jack to apply the downward tension to the side of the raft foundation with small settlement to force the two sides of the raft foundation to reach balance. The tensioning process of prestressed anchor cable was technically complex, which required synchronous control of multiple hydraulic jacks and graded tensioning. Moreover, the settlement change in the main structure of the building under each level of load was uncertain. After each level of load was applied, the settlement must be observed until it is stable to carry out the next level of loading. It can be seen from
Figure 5 and
Figure 6 that the fluctuation of settlement
and settlement velocity
gradually increases. From 7 April to 6 May, the fluctuation is relatively slow, with
less than 10 mm and
maximum 0.24 mm/d. From 6 May to 14 July, the fluctuation increase significantly, with
exceeding 15 mm and
maximum 0.55 mm/d.
From 14 July to 2 September, the probability of being at each risk level increased or decreased at a slower rate, but the building remained in a dangerous condition. In the later stage of the project, the rectification and reinforcement results were further consolidated and improved. The foundation beneath the raft slab was further reinforced through high-pressure inclined hole grouting. The shear walls on the negative one and negative two floors were strengthened with R100 reactive powder concrete (RPC) by increasing the cross-section, while interspersing crack reinforcement and repair work on the original walls, beams, slabs and other components of the main structure. From
Figure 5 and
Figure 6, it can be seen that the settlement
shows a slight increase, and the settlement velocity
gradually decreases to within 0.1 mm/d. It should be noted that the building had not yet fully stabilized after the rectification and reinforcement were completed. Subsequent monitoring should be continued to ensure that the settlement deformation trend of each measurement point has been stabilized without significant sudden changes, and that the settlement rate has reached the judgment standard for the stability of settlement deformation of buildings.
The dynamic changes in construction quality risk presented in
Figure 10 are generally consistent with the actual rectification and reinforcement process and the fluctuating status of monitoring indicators. Therefore, the established improved DBN risk assessment model can timely and accurately reflect the dynamic change characteristics of construction risk, and provide real-time decision support for the quality management of building rectification and reinforcement.
In the pre-construction phase, decision makers can employ the prediction technique to optimize the initial construction scheme by continuously adjusting the corresponding construction parameters to reduce the quality risk level of the system. The reinforcement company should prepare emergency supplies, equipment and personnel in advance to respond to very likely risk events. During the construction phase, abnormal contingencies can be diagnosed in real time based on the dynamic change characteristics of risks. It should be noted that the monitoring frequency should meet the requirements of real-time feedback analysis of measured data. Must be based on the feedback information to decide whether to proceed to the next procedure, adhere to the “dynamic design, information construction”. By taking effective risk prevention and control measures, the rectification and reinforcement of buildings can be carried out smoothly. After the completion of the rectification and reinforcement, the overall verticality of the building should meet the requirements of the current relevant national specifications (inclination rate ≤ 2.5‰), and does not affect the later use of the building.
7. Conclusions
The rectification and reinforcement of high-rise buildings are difficult, technically demanding and comprehensive, thus high-risk engineering. To evaluate the risk level of construction quality, this study proposes a novel hybrid risk analysis approach by integrating the ECM, the improved D-S evidence theory and the DBN. Important risk issues are empirically analyzed in the context of the rectification and reinforcement for a high-rise building. The main conclusions are as follows:
Based on the analysis of the rectification and reinforcement scheme adopted by the project, three key construction processes are screened out and 14 risk factors prone to inducing quality failures are identified, from which a construction quality risk assessment index system is constructed.
The proposed hybrid risk analysis approach combines the advantages of the ET, the CM, the D-S evidence theory and the DBN, which is expected to enrich risk management in the field of building rectification and reinforcement. The ECM can not only address the ambiguity and randomness in risk assessment, but can also avoid the loss of potential information. The improved D-S evidence theory takes into account differences in the contributions of different experts to the fusion results when resolving evidence conflicts, which helps to reduce errors caused by ignoring individual differences. The DBN model integrating the monitoring indicators realizes the unification of risk early warning, which can timely and accurately reflect the dynamic changes in construction risk and the dynamic impact of unexpected events on the probability of risk occurrence.
The construction quality risk perception results validate the applicability and effectiveness of the proposed risk analysis approach. The approach can also be used as a decision-support tool for systematic risk analysis of other complex projects with multiple evaluation indicators and high uncertainty.
It is important to note that the risk assessment index system determined in this study is for a specific project, and there are some differences under different rectification and reinforcement methods. The knowledge and experience of domain experts have made important contributions to the development of the risk assessment model in this study. However, the process is laborious and relies heavily on domain experts. How to establish an information management system in the field of construction engineering to realize the automatic integration of different knowledge resources is a problem to be solved.