Assessing Construction Safety Performance in Urban Underground Space Development Projects from a Resilience Enhancement Perspective

Abstract

Urban underground space construction frequently encounters issues of inadequate prevention and ineffective resistance to various disturbances, resulting in safety accidents that are difficult to recover from. Resilience pertains to a system’s capacity to absorb, resist, recover, and adapt when faced with disruptions. Enhancing the construction safety resilience of underground spaces can effectively tackle the issue of frequent accidents and the challenge of pre-controlling risks at construction sites. Utilizing systems engineering theory, this paper investigates the factors that affect the construction safety resilience of underground spaces and establishes a general framework for evaluating the safety performance of the construction process. Utilizing a large-scale underground construction project as a case study, the Bayesian network inference technique is applied to ascertain the project’s safety resilience value. Through reverse reasoning, the method identifies the most likely sequence of causes leading to construction safety incidents, and subsequently, the resilience assessment framework’s efficacy is tested. The research findings suggest that the core of construction safety management is the prevention of unsafe human behaviors and that the key to enhancing resilience lies in the optimization of response capabilities. The proposed “PFR-EFR-LFR” whole-process resilience analysis method can be applied to safety assessments for various types of underground space construction projects.

1. Introduction

Urban underground spaces incorporate a multitude of functions, including rail transit, commercial zones, civil defense, and more, providing essential services that sustain the daily operations of cities. Recently, the scarcity of available land on the surface has accelerated the development of underground spaces. Despite extensive construction efforts, numerous issues persist in construction management, but they are often hidden from view: natural disasters such as floods [1] and typhoons [2,3], unsafe human behavior [4], insufficient construction technology and experience, and the interplay of internal and external disturbances pose significant challenges to the management of underground space development projects, frequently resulting in safety incidents during construction. Many projects encounter risks of needing repair, being suspended, or even causing casualties: for instance, the collapse of a deep foundation pit during the construction of the Hangzhou Metro Line 1 Xianghu Station resulted in 21 fatalities. In the construction of underground spaces, there is a prevalent issue of inadequate prevention, weak resilience, and slow recovery in the face of various disturbances. The question of how to effectively enhance the construction system’s ability to prevent, respond, and recover throughout the entire lifecycle of disasters has emerged as a new research topic.

Presently, research on infrastructure construction safety management has transitioned from a traditional paradigm centered on post-event analysis and retrospective causality to a comprehensive paradigm that emphasizes proactive prevention and recovery systems. The concept of resilience underscores that systems should have the capability to sustain or rapidly regain their functions when faced with disruptions, and to adeptly manage the uncertainty of risks and disturbance events through adaptation and self-learning [5]. Current research outcomes on infrastructure resilience still heavily focus on conceptual differentiation and qualitative analysis, mostly targeting large-scale infrastructure with strong interconnectivity, such as ports [6,7], power grids [8,9], and rail transit systems [10]. The goal is to improve the operational service quality of infrastructure by considering resilience, with a particular emphasis on the operational phase as the main area of study. Currently, there is a dearth of research that centers on project management components and zeroes in on the construction phase to examine the factors that affect the safety resilience of the construction process. Research into the resilience of underground space construction safety is, essentially, an uncharted territory.

The construction process of underground space projects is fraught with significant uncertainties. Failures in any process, such as technical, management, or mechanical equipment issues, can lead to project failure. Despite numerous cases of engineering accidents, the historical data available for study are limited, and many of these data are fragmented and incomplete. Bayesian networks have been proposed to address issues of uncertainty and incompleteness and are ideal tools for reasoning based on incomplete, imprecise, or uncertain knowledge and information [11]. They are now widely used in infrastructure safety risk assessment [12,13,14,15], engineering reliability analysis [16,17,18], seismic assessment [19], and major project decision-making [20,21,22]. This paper employs Bayesian network as a modeling tool to integrate the five major elements involved in the construction process of underground spaces, namely, “manpower–material–machine–environment–management” (4M1E), with the “robustness–resourcefulness–rapidity–redundancy” (4Rs) attributes in resilience analysis. It identifies the factors influencing safety resilience during the construction phase of underground spaces and proposes a resilience measurement method for the entire project construction process. Bayesian networks are employed for forward and backward reasoning based on the identification and dynamic evolution of resilience capabilities. This analytical framework can ascertain the longest chain of causes for safety risks and pinpoint the most hazardous stages of project construction. The outcomes of the resilience assessment offer a foundation for enhancing project safety performance.

The logical structure of this article is as follows: Section 2 provides a literature review on the latest quantitative methods for infrastructure safety management and resilience analysis; Section 3 focuses on assessment methods and research framework for the resilience of underground space construction safety; Section 4 introduces a case study to verify the effectiveness of the proposed research framework; Section 5 discusses and summarizes the research findings of this article.

2. Literature Review

The safety management for urban underground spaces projects encompasses multiple dimensions, such as technology, management, resources, and the environment [23,24]. The resilience of engineering systems highlights the significance of prevention before disasters, efficient responses during disasters, and thorough recovery after disasters [25,26]. This section reviews the logical framework, application scenarios, and research progress of resilience theory, systematically expounding the identification, measurement, and improvement methods of resilience.

2.1. Framework for Infrastructure Resilience Assessment

Resilience, connoting elasticity and the ability to rebound, characterizes a system’s capacity to respond and recover after being disturbed. The construction process of infrastructure projects is complex, fraught with numerous disturbance factors, and prone to risks. Numerous scholars have proposed corresponding resilience evaluation methods tailored for different types of infrastructure. Hossain et al. [6,9,27] take port systems and associated power infrastructure systems as research subjects, using natural disasters and physical attacks as disturbance entities, and define resilience in terms of absorption, recovery, and adaptation capabilities through capacity characterization methods. They identify key risk factors in the operation of various infrastructure systems using Bayesian network reasoning. Dixit et al. [28] assess supply chain resilience based on networked structural parameters, considering network resilience as the composite effect of network density, centrality, connectivity, and network size. Lu et al. [29] analyzes the resilience of rail transit networks under normal operating conditions, combining network topological features and passenger traffic characteristics, explicitly considering factors that cumulatively affect passenger flow, and quantifying the resilience of rail transit networks over time under different events. Shi et al. [30] builds a framework for complex urban systems based on the Complex Adaptive System (CAS) theory. This operates via three aspects: the system environment, system elements, and system structure.

2.2. Identification and Measurement of Infrastructure Resilience

Resilience, a novel concept in the realm of safety management, underscores the importance of effective risk prevention and prompt recovery. It plays a pivotal role in optimizing design schemes for infrastructure construction, ensuring safety throughout the construction process and enhancing operational performance [31,32,33]. There are already numerous theoretical studies on the definition of resilience, but the identification of factors influencing systemic resilience and methods for numerical measurement still need to be further developed. Much of the existing research is based on the characterization of resilience capabilities and the calculation of performance response functions for resilience identification and measurement. For example, Toroghi et al. [34] established a quantitative framework to assess the resilience of power infrastructure systems. This includes five dimensions of resilience—robustness, resourcefulness, redundancy, rapidity, and adaptability—and implements a visual interface for the computational process. Zeng et al. [35] modeled and analyzed the resilience of nuclear power plants under seismic threats, defining four indicators for describing different aspects of system resilience, namely, resistance, absorption, recovery, and overall resilience. Additionally, they proposed a simulation-based multiphase energy system resilience analysis algorithm. Specking et al. [36] proposed a definition of resilience for engineering systems, dividing system resilience into platform resilience and mission resilience and employing multi-objective decision analysis to assess the resilience of systems with multiple performance indicators.

Ahmadian et al. [37] proposed a method for quantitatively measuring the resilience of systems with networked operational characteristics, defining the resilience of network components as a function of severity, disturbance frequency, disturbance impact, and recovery capability. Chen et al. [38] constructed an agent-based land use model, as a typical example of socio-economic system resilience, to assess the resilience state, and studied how the resilience of the system is affected by external disturbances and internal dynamics. Bao M. et al. [39] proposed a time-varying resilience assessment framework for multi-state energy systems, considering the energy interaction in extreme weather events (such as storms). Using multiphase performance curves to describe the system’s response behavior at different stages under the influence of storms, they established an optimal energy flow model based on service, minimizing the impact of storms by coordinating different energy subsystems. Chen et al. [40] drew on the resilience evolution curve by analyzing the feedback relationship between four subsystems of China’s oil import system, established a system dynamics (SD) simulation model, and studied the resilience of China’s oil import system under external shocks.

2.3. Review of Existing Research

Current research on infrastructure resilience primarily focuses on large and complex engineering systems, such as power systems and rail transit networks, and on the interconnectivity of urban infrastructure. Research stages often target the operational processes of these systems, aiming to enhance their ability to respond to and recover from internal and external disturbances. The identification and measurement of resilience are primarily based on methods that characterize resilience capabilities, and mostly use the 4R attributes as the identification principle to construct an evaluation index system with the specific attributes of the research subject. Alternatively, they are based on resilience evolution curves, employing improved performance response function models to measure resilience. There is a scarcity of the systematic deconstruction of the construction process for a specific type of infrastructure, beginning with safety management during the construction phase, guided by the enhancement of resilience, and utilizing project management methods to address construction safety issues.

Therefore, the innovations and primary contributions of this article are mainly reflected in the following points:

(1)

A three-stage resilience evolution model, encompassing prevention for resilience (PFR), emergency for resilience (EFR), and learning for resilience (LFR), is proposed in order to address the temporal characteristics of infrastructure construction processes. This model integrates the entire disaster process with the response characteristics of the construction organization system to disturbances, representing an effective enhancement and upgrade compared to existing resilience assessment models.

(2)

Combining the theories and practices of existing safety resilience, this study couples technical factors with management factors, identifying the components of safety resilience in underground space construction from multiple perspectives of “4M1E”, and constructs a reasonable evaluation index system. The assessment framework optimizes the system safety management model, representing an important improvement in the current stage of infrastructure safety management, which tends to emphasize technical factors over management factors.

(3)

Utilizing the Bayesian network reasoning method, the numerical value of construction safety resilience during the construction phase of underground space projects is proactively calculated. This approach effectively assesses the safety status at the early stages of the project. Conversely, backward reasoning uncovers the longest chain of causes for accidents that occur during the construction of underground space development projects, identifying weak links in the management process. It addresses the issues of unclear focus and illogical reasoning in the safety management process.

3. Resilience Assessment Methods and Framework

Infrastructure projects, such as the construction of underground spaces, are not a single module but require the coordination of various factors, including personnel management, materials and equipment, and the construction environment [41]. Moreover, it is not a disjointed process; rather, it requires seamless connection throughout the entire lifecycle, from project planning, investment, decision-making, design, and construction to operation and maintenance. According to the principles of system engineering, the construction process of underground space projects can be viewed as a construction safety system composed of five modules: personnel, materials, machinery, environment, and management. This section elaborates on the assessment and measurement methods for construction safety resilience. The research framework is a generic architecture that can be utilized for resilience analysis during the construction phase of various types of infrastructure.

3.1. Bayesian Network Theory

Bayesian networks are risk analysis modeling and inference tools that fully consider the uncertainty and incompleteness of system information. By updating the prior probabilities of basic events, statistical inference is performed to provide a scientific basis for decision-making regarding significant system events [11,42]. This paper employs Bayesian network inference methods to identify and calculate the resilience of infrastructure construction safety.

A Bayesian network consists of a network structure G and a set of conditional probability tables (CPTs). The network structure G = <V, E> represents a directed acyclic graph composed of nodes and directed arcs between nodes, where V represents the nodes and E represents the directed arcs, indicating the existence of causal relationships between variables. The conditional probability tables (CPTs) represent the probability distribution of a node given the set of states of its parent nodes. The theoretical foundation of Bayesian networks is Bayes’ theorem, as shown in Equation (1):

$$P(AB)=P(A)\times P(B/A)$$

(1)

P(AB) represents the probability of both events A and B occurring, P(B/A) represents the probability of event B occurring given that A has occurred, and P(A) represents the prior probability of A, which is the probability of event A occurring on its own. Assuming there are N mutually independent variables in a Bayesian network, X₁, X₂, X₃, …, X_N, the joint probability distribution of these is represented by Equation (2):

$$P(X_1,X_2,X_3,\dots X_N)=P(X_1/X_2,X_3,\dots X_N)P(X_2/X_3,\dots X_N)\dots P(X_{N-1}/X_N)P(X_N).$$

(2)

This equation can also be transformed into (3):

$$P(X_1,X_2,X_3,\dots X_N)={{\displaystyle \prod }}_{i=1}^{N}P(X_i/X_{i+1},\dots X_N)={{\displaystyle \prod }}_{i=1}^{N}P(X_i/parents(X_i))$$

(3)

Parents(X_i) represents the set of parent nodes of variable X_i. Taking Figure 1 as an example, the Bayesian network contains a set of nodes S = {X₁, X₂, X₃, X₄, X₅}, with the directed edges that connect the nodes representing the logical relationships between them. X₁ and X₃ are root nodes, X₅ is a leaf node, and X₂ and X₄ are intermediate nodes. According to Bayes’ theorem, the joint distribution probability can be represented as Equation (4):

$$P(X_1,X_2,X_3,X_4,X_5)=P(X_1)P(X_3)P(X_2/X_1)P(X_4/X_2,X_3)P(X_5/X_4)$$

(4)

3.2. Assessment Model and Framework

Since its inception, the concept of resilience has undergone a cognitive transformation from engineering resilience [36] to ecological resilience [43], and then to evolutionary resilience [44], shifting from single-system equilibriums to multimodal and dynamic equilibriums. This paper investigates the ability of the construction organization management system to maintain or regain its safe operational status in the face of individual or combined risk factors, such as “manpower, materials, machinery, environment, and management”, during the construction of underground space engineering. This ability is referred to as the construction safety resilience of the underground space construction process.

Based on the stage division of the construction process and the characteristics of resilience evolution, construction safety resilience is divided into three sub-blocks: prevention for resilience, emergency for resilience, and learning for resilience [45]. These factors correspond to the preventive process before disturbance, the responsive process during disturbance, and the retrospective and improvement process after disturbance. The relationship between system resilience and various indicators is shown in Equation (5).

$$R=\{PFR[C_{abs}]+EFR[C_{res}+C_{rec}]+LFR[C_{ada}]\}/3$$

(5)

Among these factors, R represents the construction safety resilience, PFR, EFR, and LFR are the preventive resilience, emergency resilience, and learning resilience functions, while Cabs, Cres, Crec, and Cada represent influencing factors under dimensions of absorption, resistance, recovery, and adaptation.

PFR (prevention for resilience) refers to the initial stage of resilience in a system where disturbances are generated but have not yet caused damage. The preventive resilience of a system is reflected in its ability to absorb disturbances, focusing on the timely elimination of potential disturbances, reducing the probability of failure in safety construction systems, and demonstrating the effect of prevention.

EFR (emergency for resilience) refers to the resilience of a system following disturbances, encompassing its ability to withstand impacts and recover swiftly. This corresponds to the system’s resistance and recovery capacities. Resistance capacity focuses on timely responses, controlling the damage caused by disturbances to the system, as well as minimizing efficiency loss and other destructive consequences. Recovery capacity emphasizes the rapid mobilization of resources, organizational optimization, and the restoration of normal performances.

LFR (learning for resilience) refers to the resilience of a system that learns to improve after disturbances, corresponding to the system’s adaptive capacity. Adaptive capacity enhances the system’s ability to respond to disturbances by changing its structure and optimizing the component configuration. It reflects the system’s essential goal of continuously adapting to disaster disturbances and improving disaster resilience through learning and innovation.

Most studies on the influencing factors of resilience identify them from four dimensions—technology, organization, society, and economy—known as TOSE [46]. This paper improves upon the traditional TOSE four-dimensional perspective by adopting a “manpower–material–machine–environment–management” (4M1E) five-dimensional identification framework suitable for the construction process of infrastructure projects, encompassing most risk factors that affect safety production at the construction site.

The evolution of resilience is a dynamic process. Figure 2 illustrates the evolution curve of construction safety resilience for infrastructure projects, depicting resilience as the sequential ability to absorb, resist, recover, and adapt to various disturbances [45]. Bruneau et al. [46] introduced the 4R attributes of resilience, encompassing robustness, resourcefulness, rapidity, and redundancy. These 4Rs have become widely accepted criteria in resilience research. This paper adopts a method that integrates resilience capacity characterization with the 4R attributes to identify factors influencing construction safety resilience. Sources such as industry standards, corporate standards, accident cases, and the academic literature are utilized to pinpoint factors affecting construction safety resilience. A detailed analysis of common causes of construction safety accidents is conducted from five perspectives: “manpower, materials, machinery, environment, and management”. In conjunction with expert opinions, the factors influencing construction safety resilience are further refined.

The research framework [45] for the construction safety resilience of infrastructure construction systems is depicted in Figure 3.

4. Case Study

4.1. Project Overview

To verify the rationality of the research framework, the Nanjing JiangBei underground space development project was adopted as a case study. This project is the largest single underground space project in China at present, and is divided into two phases of construction, with this case study focusing on the second phase. The construction mainly includes underground parking lots, sunken squares, metro-supporting facilities, and underground loop ramps. The total underground construction area is about 300,000 m², including tunnel works of approximately 41,000 m², metro-supporting facilities of about 52,000 m², and loop ramp areas of about 36,000 m². The total investment for the project is estimated at about CNY 8.85 billion. The layout of the construction site is shown in Figure 4, with H1–H8 representing the ramp works and numbers 1–23 marking the plot works. Among these, the Binjiang Metro Station is a key controlled project in this development project.

During the construction of underground spaces, the primary focus is on deep foundation pit engineering. In the construction process of deep foundation pit groups within underground spaces, a series of significant and challenging problems is inevitably encountered due to the combined effects of various factors, such as geological conditions, construction methods, and project management. These problems encompass technological, managerial, and environmental areas, as well as other factors. A detailed analysis of the significant and challenging issues encountered during the development of underground spaces in Nanjing is presented in

Table 1. Difficulty points in underground space development project.

Classification of Influencing Factors	Restricted Content	Key Points of the Content
Environmental constraints	The construction of deep foundation pits by the river involves significant dewatering challenges.	The project is near the Yangtze River, 0.7 to 1 km away, with a shallow water table. It features a 20 m thick silt layer prone to water and mud gushing. The design uses a fully sealed barrier with 10 m deep diaphragm walls and relies on drainage wells extending into the gravel layer.
	Poor geological conditions make it difficult to control deformation and water leakage in the foundation pit.	BinJiang Station Line 13 is part of an ultra-deep foundation pit project with a maximum excavation depth of 51 m. The stratum at a 10 m depth consists mainly of silty clay, which is highly compressible and presents poor engineering geological conditions, posing challenges and risks for deep foundation pit construction.
	Pipeline protection.	The project involves municipal pipelines and underground space design with multiple facilities. Construction requires high protection standards for pipelines, posing challenges.
	The construction site is narrow, with limited work space.	BinJiang Station Line 13 lacks an auxiliary construction site, with various pieces of construction equipment present during the construction phase. Adjacent plots 22# and 23# are used to construct retaining structures, involving significant work with large amounts of special equipment. Simultaneous multiple processes make on-site construction organization challenging.
Organizational Management	Completely enclosed construction.	To protect the surrounding environment, the construction phase of the foundation pit earthwork excavation and structural construction adopts fully enclosed construction.
	The excavation of earthwork from the foundation pit is difficult, requiring secondary transfer.	At BinJiang Station, daily soil excavation is significant. Continuous excavation is not possible due to site conditions, and there is limited space for soil stockpiling, which hinders comprehensive construction. The proper management of soil secondary transportation is necessary.
	Many process overlaps.	The project involves large-scale construction, including subway stations, road ramps, rail lines, tunnel structures, and ancillary works. Multi-profession and multi-trade collaboration is challenging.
Construction technology	The construction schedule handover is complex, and there is mutual influence between the foundation pits.	BinJiang Station Line 13 has an excavation depth of about 44.79 m, with the large shield construction well at 51 m. The Central Business District Station’s excavation depths are approximately 29 m as a standard, 30 m at the end, and 31 m for local deep pits. The project construction sequence is complex, with excavations having a sequential relationship and mutual influence.
	The coexistence of the open-cut excavation method and top-down construction method.	Binjiang Station’s A area uses open-cut construction for the upper three layers and reverse construction for the lower three layers. The B area uses only reverse construction. The Central Business District Station employs full top-down construction. Combining open-cut and top-down methods is complex and requires high technical and management standards.
	The demolition of diaphragm walls causes significant disturbance to adjacent excavation pits.	In the project’s first and second phases, shared diaphragm walls exist. Demolishing these walls causes disturbances between adjacent excavation pits. The removal of steel and concrete supports leads to stress release in the pit, making soil displacement and deformation control challenging.
	Removal of wall surfaces and straightening of wall ties.	Binjiang Station on Line 13 has a six-level underground structure with 1500 mm diaphragm walls. During excavation, walls are chiseled for texture and reinforcement steel is exposed and straightened, which extends the construction period.

.

The construction process of this case project is technically challenging, with stringent quality requirements, complex organizational management, and a tight project schedule. These factors impose extremely high demands on the organizational and management capabilities of all parties involved in the project. Resilience is an inherent attribute of a system. For engineering systems, enhancing resilience involves addressing the sources of various influencing factors in project management, improving the engineering system’s inherent ability to withstand disturbances, avoiding unnecessary safety investments in the later stages, and maximizing both the engineering safety system’s capacity to withstand risks before an accident occurs and its recovery power after an accident.

4.2. Identification of Influencing Factors

Building upon the analytical framework outlined in Section 3.2, and employing methods such as literature review and accident case analysis, the factors influencing safety resilience during the construction of deep foundation pits in underground space engineering are identified. Primary indicators include PFR, EFR, and LFR, while secondary indicators encompass absorption, resistance, recovery, and adaptation capacity. Tertiary indicators correspond to human, material, machinery, environment, and management factors, which altogether account for 35 influencing factors that are encoded individually. The coding method for influencing factors, such as AbsR1, represents factor 1 of the personnel under the dimension of absorptive capacity, where “Abs” stands for absorptive capacity, “R” stands for personnel, “C” stands for materials, “J” stands for machinery, “H” stands for environment, and “G” stands for management. Similarly, this coding method can be applied to indicators across various dimensions, including resistance, restoration, and adaptation. In

Table 2. Influencing factors for construction safety resilience.

Resilience Classification	Resilience Characterization	Identify Dimensions	Specific Manifestations	Summary of Influencing Factors	Encoding
PFR	Absorptive capacity	manpower	All personnel entering the construction site should have the qualifications.	Personnel entry Security check system	AbsR1
			Workers operating specialized machinery on-site should possess the corresponding operating qualifications.	Specialized machinery Operator qualification Review	AbsR2
			Construction personnel and management possess adequate safety awareness, adhere to the construction plan correctly, and refrain from operating in violation of regulations.	Personal safety awareness	AbsR3
		material	All materials utilized for deep foundation pit construction must undergo safety inspections for quality upon entry, and strict acceptance procedures should be applied to materials processed on-site to prevent safety hazards.	Material entry and acceptance during the process	AbsC1
			Site managers should conduct routine inspections of material storage on construction sites to reduce safety hazards.	Safety inspection of material stacking conditions	AbsC2
		machinery	The dewatering equipment at the construction site should be maintained and its safety should be checked to reduce equipment hazards and ensure normal equipment conditions.	Rainfall equipment maintenance and safety inspection	AbsJ1
			Large-scale mechanical equipment at the construction site must undergo maintenance and safety inspections to ensure its normal condition. The parking location of the equipment should adhere to regulations.	Maintenance and safety inspection of machinery equipment	AbsJ2
		environment	The surveying company should provide a comprehensive survey report, allowing the construction unit to accurately understand the geological and hydrological conditions at the construction site.	Inspection report improvement	AbsH1
			Managers should anticipate extreme weather conditions that may arise at the construction site in the future and have corresponding countermeasures.	Extreme weather forecasting and emergency plans	AbsH2
		management	The construction department should conduct effective technical handover for deep foundation pit engineering.	Technical disclosure improvement	AbsG1
			Construction department regularly conduct safety training and assessments for construction workers, and regularly carry out emergency drills and accident simulations.	Emergency safety training	AbsG2
			The management personnel inspect the construction site, find the dangerous points in time, and correct the illegal operations.	Inspections ensure safe and standardized construction operations	AbsG3

, only influencing factors involved in absorption capacity is detailed described.

In the Bayesian network structure with “construction safety resilience” as the leaf node, the 35 influencing factors identified in the results are used as the influencing factor nodes in the Bayesian network. Further, 3 resilience sub-blocks, “PFR”, “EFR”, and “LFR”, are used as the phase resilience nodes, 4 capability characterizations, “Abs absorption capacity”, “Res resistance capacity”, “Rec recovery capacity”, and “Ada adaptation capacity”, are used as network measurement nodes, and 10 auxiliary nodes, such as “SV1 personnel unsafe behavior prevention”, are added to represent the states present in each scenario in order to meet the needs when establishing a causal network. By synthesizing expert opinions, the causal relationships between factors are analyzed to establish the Bayesian network structure. The network is depicted in Figure 5.

In this Bayesian network, all nodes are defined as two-dimensional variables, with node states categorized into failure and non-failure. Node failure signifies that the point has not effectively prevented risks and has not utilized the inherent resilience of the system, potentially resulting in safety accidents during construction. The consequences of node failure are detailed in

Table 3. Consequences of node failure.

Node Type	Encoding	Node Name	Consequences of Node Failure
Leaf node	R	Construction safety Resilience of underground spaces	It failed to effectively prevent construction safety accidents; when such accidents occurred, it failed to effectively minimize production and construction efficiency losses and mitigate the destructive consequences; it failed to effectively reduce recovery time; and it failed to effectively improve accident adaptability.
Phase resilience nodes	PFR	Preventive resilience	It failed to effectively prevent the occurrence of construction safety accidents.
	EFR	Response resilience	At the time of the accident, there was an inability to effectively mitigate production and construction losses, to reduce the destructive consequences of the accident, and to effectively shorten the recovery time.
	LFR	Learning resilience	Following the accident, there was an inability to effectively enhance adaptability to the incident.
Resilience measurement node	Abs	Absorptive capacity	It failed to effectively prevent unsafe personnel behaviors and unsafe conditions in terms of materials and equipment, leading to the occurrence of construction safety accidents.
	Res	Resistance capability	The construction safety accident was not controlled in time, resulting in severe damage to the construction site and serious consequences such as casualties.
	Rec	Recovery capacity	The on-site recovery time is too long; the restoration of normal production and construction processes is slow, affecting the normal schedule.
	Ada	Adaptive capacity	Failure to learn from historical events and implement corrective and optimization measures may lead to similar incidents happening again.
Status node	SV1	Prevention of unsafe personnel behavior	Construction accidents in deep foundation pits caused by the unsafe behavior of construction personnel or management staff.
	SV2	Prevention of unsafe conditions in materials and equipment	Accidents caused by unsafe conditions such as material stacking and mechanical equipment in deep foundation pits.
	SV3	Prevention of unsafe environmental factors	Occurs due to hydrogeological factors or extreme weather conditions causing safety accidents during deep foundation pit construction.
	…
Root node	AbsR1	Personnel entry security check system	Failed to detect non-project personnel entering the site, increasing safety hazards.
	AbsR2	Specialized machinery personnel qualification review	Unqualified personnel operating specialized machinery, with illegal operation causing foundation pit construction safety accidents.
	AbsR3	Personal safety awareness	Construction site personnel lack a sense of crisis about potential danger points and cannot respond in time when an actual accident occurs.
	…

, where only some nodes are displayed.

4.3. Construction Safety Resilience Analysis

Bayesian network reasoning encompasses both forward and backward processes. Forward reasoning relies on the prior probabilities of the root nodes and the conditional probability tables of each intermediate node to deduce the failure probability of the leaf nodes. In contrast, backward reasoning utilizes the outcomes of forward reasoning, sets target values for the leaf nodes, and traces backwards to ascertain the occurrence probabilities of the cause nodes, thus evaluating changes in the failure probabilities of these cause nodes. This paper utilizes the underground space development project detailed in Section 4.1 as a case study. It employs a questionnaire survey method to obtain the failure probabilities of various influencing factors, including “manpower, material, machine, environment, and management”, throughout the entire process of deep foundation pit group construction. Causal reasoning is conducted using Bayesian networks, and the safety resilience value of the construction project is calculated. The prior probabilities of the root nodes are obtained according to the trapezoidal fuzzy comprehensive evaluation method [47,48], and the conditional probability tables of non-root nodes are detailed based on the leaky noisy-or model [49,50]. The questionnaire was distributed among the project company and the construction department of the case study group, with 21 management personnel from the project company and 42 from the construction department. The project company investigated has established departments for construction management, safety supervision, planning and contracts, finance and capital, and comprehensive management. The functions of these five departments are sufficient to cover all aspects of “4M1E”. All company personnel hold intermediate or higher professional titles, and some are professors and senior engineers. A total of 63 questionnaires were distributed and 58 valid responses were collected, ensuring there were sufficient sample data for accurate analysis results.

4.3.1. Forward Reasoning Analysis

As indicated by

Table 4. Forward reasoning process of Bayesian network.

Primary Intermediate Node	Failure Probability	Secondary Intermediate Node	Failure Probability	Influencing Factors (Third-Level Intermediate Nodes)	Failure Probability	Influencing Factors (Root Node)	Failure Probability
PFR	21.1	Abs	51.3	SV1	76.1	AbsG3	35.8
						AbsG2	34.1
						AbsG1	32.8
						AbsR1	32.3
						AbsR2	36.0
						AbsR3	42.9
				SV2	50.7	AbsC2	34.0
						AbsJ1	38.6
						AbsJ2	38.6
				SV3	39.5	AbsH1	35.5
						AbsH2	40.1
				AbsC1	38.3
EFR	42.9	Res	57.3	ResG5	59.4	SV4	38.3
						ResJ2	37.0
						ResG1	40.5
						ResG2	36.2
						ResG3	34.6
						ResG4	35.6
				SV5	42.0	ResC1	34.0
						ResJ1	38.4
				SV6	51.8	ResR1	42.5
						ResR2	43.9
						ResH1	40.5
				ResH2	35.3
		Rec	37.3	SV7	55.9	RecC1	36.3
						RecJ1	36.8
						RecH1	37.0
				SV8	25.6	RecG1	34.5
LFR	21.1	Ada	52.7	AdaR1	34.4
				SV9	40.6	AdaC1	37.7
						AdaJ1	40.0
				SV10	27.4	AdaG3	20.1
						AdaG4	26.0
						AdaG2	24.5
						AdaG1	34.0

, the construction safety resilience value of this project is 0.716, which translates to a moderate level on a percentage scale. This suggests that the construction safety risks associated with the deep foundation pit group project are generally manageable; however, safety management efforts should still be intensified. Among them, the failure probability of the absorption capacity during the construction process of the case foundation pit construction system is 51.3%, the failure probability of the resistance capacity is 57.3%, the failure probability of the recovery capacity is 37.3%, and the failure probability of the adaptation capacity is 52.7%; the failure probability of PFR is 21.1%, EFR is 42.9%, and LFR is 21.1%.

Using 40% and 50% as the standard values for judging failure probability, according to

Table 5. Critical nodes in forward reasoning.

PFR				EFR				LFR
40–50%		>50%		40–50%		>50%		40–50%		>50%
Node Name	Failure Probability	Node Name	Failure Probability	Node Name	Failure Probability	Node Name	Failure Probability	Node Name	Failure Probability	Node Name	Failure Probability
AbsR3	42.9%	Abs	51.3%	ResR1	42.5%	Res	57.3%	AdaJ1	40.0%	Ada	52.7%
AbsH2	40.1%	SV1	76.1%	ResR2	43.9%	ResG5	59.4%	SV9	40.6%
		SV2	50.7%	ResH1	40.5%	SV6	51.8%
				ResG1	40.5%	SV7	55.9%

, in the PFR sub-block, nodes with a failure probability greater than 50% include the following: Abs—absorption capacity; SV1—the prevention of unsafe behaviors by personnel; SV2—the prevention of unsafe conditions of materials and equipment. Nodes with a failure probability between 40% and 50% include the following: AbsR3—self-safety awareness of personnel; AbsH2—prefabrication and emergency plans for extreme weather. In the EFR sub-block, nodes with a failure probability greater than 50% include the following: Res—resistance capacity; ResG5—emergency management actions; SV6—personnel safety assurance; SV7—recovery of the workplace. Nodes with a failure probability between 40% and 50% include the following: ResR1—worker awareness of danger and escape skills; ResR2—safety protection measures for personnel; ResH1—unobstructed rescue channels, ResG1—management personnel patrol and detect instability alerts. In the LFR sub-block, the nodes with a failure probability greater than 50% include the following: Ada—adaptability. Nodes with a failure probability between 40% and 50% include the following: AdaJ1—the improvement and enhancement of mechanical and equipment resource allocation; SV9—the improvement and enhancement of resource allocation.

From the above nodes with a higher probability of failure, it can be observed that personnel management factors at the construction site remain a key cause of accidents, such as nodes SV1, AbsR3, SV6, ResR1, ResR2, ResG1, etc., which are all related to personnel factors. Therefore, the unsafe behavior of personnel continues to be the primary cause of safety accidents at construction sites. Current research on construction safety mostly focuses on technical research, such as scholars Huang et al. [51,52,53,54,55], who conducted in-depth studies on tunnel construction safety, attempting to improve construction safety benefits through methods such as on-site monitoring and information platform management and control. To enhance the safety attributes of the construction process, the application of technology should be combined with optimized management. Improvements in resilience should still focus on personnel management, correcting workers’ non-compliant behaviors during daily operations, refining the daily safety inspections of project managers, and promptly identifying safety hazards during the construction process. By continuously strengthening the proactive management awareness of people at all stages of the construction process, the safety resilience of the system itself can be enhanced more than is possible when relying solely on advanced construction technology.

4.3.2. Backward Reasoning Analysis

The most significant outcome of backward reasoning in Bayesian networks is deducing the most probable causal chain that leads to construction accidents. As depicted in Figure 6, by setting the resilience value to 100%, the maximum causal chain can be derived based on PFR, EFR, and LFR links. Among these, the maximum causal chain under the PFR sub-block is from AbsR3, the personnel self-safety awareness, to SV1, unsafe behavior prevention to Abs absorption capacity, ultimately leading to a PFR failure probability of 15.7% when the resilience value is 1. The EFR sub-block has two chains, resistance and recovery, and so the maximum causal chains are, respectively, from ResG1, management inspection to discover instability alerts, to SV4, discovering instability to ResG5 emergency management actions, and to Res, resistance capacity, and from RecH1, on-site professional rescue, to SV7, face recovery, and to Rec recovery capacity. Ultimately, with both chains in effect, the EFR failure probability is 35%; the maximum causal chain under the LFR sub-block is from AdaJ1 improvement of mechanical equipment resource allocation to SV9 resource allocation improvement to Ada adaptability, ultimately leading to an LFR failure probability of 15.1%. Among the three resilience sub-nodes, PFR, EFR, and LFR, the one with the highest failure probability is EFR, indicating that response capability is the key link in determining construction safety resilience.

Compared with the results of forward reasoning, the failure probability of EFR decreased by 7.9%, while the failure probabilities of PFR and LFR decreased by 5.4% and 6%, respectively. To achieve maximum construction safety resilience, it is essential to concentrate on improving the system’s resistance and recovery capabilities during accidents. This conclusion aligns closely with the resilience development process depicted by the resilience evolution curve. If the system’s resistance capacity is enhanced, the curve’s lowest point rises. Similarly, if the system’s recovery capacity is improved, the resilience recovery time is reduced, and the curve shifts upwards in its entirety. As illustrated in Figure 2, line 2 shows less severe consequences from disturbances and a higher recovery efficiency compared to line 1, indicating a substantial improvement in resilience. Concurrently, to elevate the construction safety level of underground space projects, it is crucial to bolster personnel safety awareness, conduct stringent safety management inspections, enhance on-site professional rescue capabilities, and ensure the proper allocation of equipment resources.

5. Discussion and Conclusions

5.1. Discussion

The current methods for identifying infrastructure resilience primarily involve dividing resilience into several key subsystems, based on principles of systems engineering. For each subsystem, the influencing factors of resilience are identified, and an evaluation index system is constructed corresponding to the research subject. Common methods include expert interviews, literature reviews, and case analysis. This article treats the construction process of underground space engineering as an integrated system, dividing resilience into three major categories—prevention for resilience (PFR), emergency for resilience (EFR), and learning for resilience (LFR)—corresponding to the emergence of capabilities at each stage. It focuses on the safety of engineering construction and uses a combination of literature review and case analysis to identify the influencing factors of resilience for underground space engineering construction safety.

Current methods for measuring infrastructure resilience predominantly depend on functional and inferential computational approaches. Functional methods encompass performance response functions, whereas the most prevalent inferential methods are Bayesian networks or system dynamics. This paper concentrates on underground space engineering and classifies the influencing factors of the infrastructure construction process, beginning with five categories: “manpower, material, machinery, environment, and management”. It seeks to establish the logical relationships among these factors to create a Bayesian network structural model. Subsequently, by distributing questionnaires to gather expert knowledge, the failure probabilities of each node are determined, and network inference is conducted.

The enhancement of infrastructure resilience should be grounded in the outcomes of resilience calculations. This paper integrates Bayesian network forward reasoning to compute the numerical value of construction safety resilience during the construction phase of underground space engineering, assessing the overall system performance; it then utilizes the outcomes of backward reasoning to pinpoint the maximum cause chain that could precipitate construction safety incidents. By identifying the most hazardous nodes, the construction safety management of the project becomes traceable, and construction management becomes more focused. This approach can rectify the shortcomings of traditional safety management, which often lacks direction and clear objectives.

To verify the effectiveness of the resilience analysis framework proposed in this paper, a large-scale underground space development project under construction in China was selected as a case study to analyze the impact of the coupling effect of management factors and technical factors. The resilience assessment method proposed aims to enhance the inherent safety attributes of the construction process, focusing on accident prevention, and treats the construction process of deep foundation pits as an integrated system for analysis. This represents an effective attempt to apply systems engineering analysis methods to traditional civil engineering problems.

5.2. Conclusions

This article focuses on the construction safety resilience of underground space projects. Through methods such as literature analysis, the examination of case studies, construction scheme analysis, and field research, it identifies the key factors that influence construction safety resilience across five dimensions: “manpower, materials, machinery, environment, and management”. By synthesizing expert knowledge, a table of the factors influencing infrastructure resilience assessment is created. The study establishes a construction safety resilience analysis framework using preventive resilience, emergency resilience, and learning resilience as capability indicators. In the case study, questionnaires are developed using the table of factors influencing resilience, and the experiential knowledge of underground space construction management experts is utilized to ascertain the conditional probabilities of each influencing factor node. Bayesian network forward and backward reasoning is then conducted to calculate the system resilience value and the primary causal chain of accidents. The primary research conclusions are as follows:

(1)

The research framework for underground space construction safety resilience proposed in this paper, which encompasses the identification of influencing factors and Bayesian network computation, is both reasonable and effective. It can be utilized for safety assessments in actual projects. The numerical value of resilience indicates the system’s ability to withstand risks and recover from accidents. In the case study, the construction safety resilience value of the system was 0.716, signifying a medium level. This suggests that the overall construction safety risk of the project is manageable, yet risk monitoring should be intensified. During the construction process, the failure probabilities were as follows: absorption capacity was at 51.3%, resistance capacity was at 57.3%, recovery capacity was at 37.3%, and adaptability was at 52.7%. This indicates that, throughout the entire process of accident evolution, resistance capacity is the weakest link, representing the system’s inherent ability to respond to disaster events.

(2)

Construction technology is often regarded as a crucial method of enhancing safety performance. Efforts to update and optimize mechanical equipment, along with the implementation of information management platforms, have indeed significantly contributed to the improvement of construction efficiency. However, this article demonstrates, through Bayesian network inference based on questionnaire survey results, that personnel management factors at the construction site continue to be major contributors to accidents. In the outcomes of Bayesian network forward inference, nodes with elevated failure probabilities, such as SV1, AbsR3, etc., are all associated with personnel factors. Unsafe behavior by personnel persists as the primary cause of safety accidents at construction sites. The primary reasons for unsafe behavior are as follows: first, there is a lack of safety awareness, and second, unsafe behaviors are closely linked to the formation of the safety culture and the safety atmosphere at the construction site.

(3)

This article proposes a resilience analysis process of “PFR-EFR-LFR” based on the temporal characteristics and evolution rules of system responses following disturbances. In the case study, the failure probability of PFR is 21.1%, the EFR is 36.5%, and the LFR is 21.1%. Among the three resilience sub-modules, EFR has the highest probability of failure, indicating that response capability is the key factor determining the level of construction safety resilience. The response process mainly relies on resistance and recovery capabilities. To improve the construction safety level of underground space projects, it is essential to strengthen personnel safety awareness, safety management inspections, on-site professional rescue capabilities, and the allocation of mechanical equipment resources.