Digital Twin Empowers Electric Vehicle Supply Chain Resilience

Abstract

To reveal how digital twin empowers electric vehicle supply chain resilience, this study first proposes a novel “Human–Machine–Material–Environment” system architecture. Then, it employs dynamic fsQCA on data from 27 electric vehicle companies to explore the underlying configurational mechanisms. The results reveal that digital twin empowers electric vehicle supply chain resilience not through singular factors, but through multiple, equally effective configurations of its core dimensions. This study identifies six types of high-resilience pathways, such as “dual-driven by twin and safety” and “comprehensive upgrade digital twin”. This demonstrates that no universal best pathway exists. This finding of equifinality is complemented by causal asymmetry, as the paths leading to non-high resilience are not mere opposites of the successful ones. Across time periods, data security management, human–machine collaboration, and digital twin applications consistently emerge as core prerequisites for improving supply chain resilience. By introducing digital twin, this study expands the theoretical boundaries of electric vehicle supply chain resilience research and provides new analytical perspectives and frameworks.

1. Introduction

In recent years, a variety of highly uncertain events have continued to impact the electric vehicle supply chain [1,2,3]. The electric vehicle supply chain is characterized by complexity, length, unpredictability, high technological dependency, and information asymmetry. It is highly susceptible to disruption at any point, potentially triggering a “chain breakage” crisis. The widespread semiconductor shortage of 2020 and the Russia-Ukraine conflict of 2022 have subjected the electric vehicle industry to widespread pressures of reduced production, halted manufacturing, and supply disruptions. The development of electric vehicles is influenced by factors like the evolution of emerging technologies and shifts in the competitive international landscape. Challenges, including low transparency in risk monitoring, cross-level data interoperability, and the security of software-hardware integration, remain largely unresolved. Digital twin, with its capabilities for virtual–physical interaction, real-time mapping, and model integration, is being increasingly applied in supply chain resilience management. However, research on precisely how it enhances the resilience of electric vehicle supply chains remains limited.

The traditional electric vehicle supply chain suffers from information silos and transmission delays, causing demand signals to become distorted during multi-tiered transmission and creating a “bullwhip effect”, which brings risks such as overcapacity or resource waste to enterprises [4]. Digital twin serves as a virtual counterpart where artificial intelligence (AI) learns, understands, and optimizes the physical world. Through the integration of physical models, sensor data, and AI, it facilitates dynamic mapping, diagnostics, prediction, and decision-making for the real world. Moreover, digital twin emphasizes real-time interaction and dynamic synchronization between physical entities and virtual models. Its high-fidelity simulations empower managers to promptly perceive the state of supply chain networks and anticipate risk propagation pathways. Taking the electric vehicle battery supply as an example, under traditional models, production disruptions or resource wastage often occur between battery manufacturing and vehicle assembly due to issues like delayed information transmission and inventory mismatches. However, digital twin integrates real-time data collection devices such as IoT sensors, RFID, and GPS to dynamically monitor battery status across all stages, including production parameters, inventory locations, in-transit trajectories, and assembly nodes, forming a “digital thread” that spans the entire battery lifecycle. The digital twin platform deeply integrates multi-source heterogeneous data with 3D geometric models to construct a holistic visual interface spanning from battery cell production to complete vehicle assembly. It can also simulate material consumption and supply rhythms based on real-time data, issuing early warnings for potential supply disruptions, inventory bottlenecks, or transportation delays. These alerts are highlighted in the 3D view, enabling identification and response to potential issues before they occur. Thus, the digital twin transforms fragmented and lagging supply chain information into an integrated, real-time, interactive, and transparent system, injecting data-driven decision-making power to enhance supply chain resilience.

Hu et al. [5] investigated a digital twin-driven trajectory tracking system for autonomous vehicles, introducing a digital twin function to compute the position of actual autonomous vehicles. Rajesh et al. [6] explored the potential and significance of digital twins in advancing electric vehicle battery technology, promoting sustainability and efficiency within the electric vehicle ecosystem. Waseem et al. [7] proposed a digital twin surrogate model incorporating machine learning to predict machine status during battery assembly. Naseri et al. [8] conducted a comprehensive review of battery DT use cases, development and integration platforms, and software requirements for implementing battery DT. This includes electrical topics related to modeling and algorithmic approaches, software architecture, and DT development and integration. We found that scholars have conducted research on digital twins in areas such as electric vehicle driving and batteries, but studies on digital twins in the field of supply chain resilience remain limited. Existing research has primarily focused on the impact of digital technologies or data elements on supply chain resilience [9,10,11]. The discussion has centered on the mechanisms through which the digital economy influences supply chain resilience [12,13]. The above studies confirm the crucial role of the digital economy and digital technologies in enhancing supply chain resilience from multiple perspectives, yet they predominantly focus on the independent effects of individual factors. Furthermore, existing studies predominantly employ quantitative methods to identify optimal solutions, lacking detailed mechanism analysis, theoretical framework construction, and studies on the synergistic effects of multiple factors interacting with the digital economy and digital technologies to enhance supply chain resilience. Additionally, they overlook digital twin as a key technological approach.

Based on the above arguments, this study aims to answer the following research question:

How does digital twin empower electric vehicle supply chain resilience?

To answer this question, the study seeks to uncover the “black box” of the underlying mechanism through which digital twin enhances electric vehicle supply chain resilience. The main contributions are threefold:

(1)

To our knowledge, this is the first study to explicitly investigate the impact of digital twins on enhancing the resilience of electric vehicle supply chains. It integrates digital twin and supply chain resilience into a unified analytical framework, providing novel perspectives and a methodological approach for supply chain resilience research.

(2)

This study empirically explores the complex mechanisms through which digital twin empowers the resilience of electric vehicle supply chains, addressing gaps in existing research regarding the underlying mechanisms of this empowerment. It also proposes corresponding recommendations, offering valuable insights for enterprises seeking to enhance supply chain resilience.

(3)

This study employs the dynamic fsQCA method in the field of supply chain resilience, enriching the research methodology in this domain.

The structure of this paper is as follows. Section 2 reviews the relevant theoretical foundations. Section 3 analyzes the construction and operational mechanisms of the intelligent digital twin system architecture for electric vehicle supply chains. Section 4 introduces the research methodology and design. Section 5 presents empirical analysis. Section 6 discusses findings and conclusions. Section 7 outlines management implications. Finally, Section 8 outlines research limitations and future directions.

2. Theoretical Foundation

2.1. Intelligent Supply Chain Theory

Digital-intelligent supply chain theory outlines a blueprint for supply chains to transition from data-driven to intelligence-driven and autonomous decision-making through deep digital and intelligent transformation [14]. Its three core characteristics establish advanced objectives for digital twin-enabled electric vehicle supply chains. Dynamically interconnected networks demand real-time visibility across the entire chain, forming the perceptual foundation of resilience. Closed-loop human–machine integration enables systems to adaptively manage uncertainty, forming the core capability of resilience. The integration of economic, social, and environmental values ensures resilience strategies are inherently sustainable. This theory elucidates that the essence of digital twin-enabled supply chain resilience lies in driving systemic transformation towards intelligent, adaptive, and sustainable operations.

2.2. Complex Adaptive Systems Theory

Complex adaptive systems theory posits that systems comprise proactive agents that, through interaction, learning, and co-evolution, generate emergent behavior at the “edge of chaos” [15]. The electric vehicle supply chain exemplifies such a system, where resilience is manifested as the capacity for adaptive adjustment and recovery in the face of disturbances. By constructing an end-to-end virtual mirror of the supply chain, a digital twin enables managers to understand the system’s emergent properties, predict disturbance propagation pathways, and test as well as optimize strategies within the virtual space. This capability facilitates the steering and regulation of complex adaptive systems, thereby enhancing their inherent resilience.

2.3. Dynamic Capability Theory

Dynamic capability theory emphasizes that enterprises must continuously perceive, capture opportunities, and reconfigure resources to maintain competitive advantage [16]. Supply chain resilience is fundamentally a dynamic capability, encompassing multidimensional concepts such as absorption, adaptation, and recovery, with a focus on addressing supply chain disruptions within dynamic environments. Digital twin significantly enhances an organization’s perception capabilities by providing comprehensive real-time data and early warnings. Through simulation and modeling, they enable rapid evaluation of multiple response strategies, supporting scientific decision-making and resource reconfiguration, thereby transforming resilience from passive response to proactive planning.

3. Intelligent Digital Twin System for Electric Vehicle Supply Chains

3.1. Hierarchical Architecture of an Intelligent Digital Twin System for the Electric Vehicle Supply Chain

The electric vehicle supply chain intelligent digital twin system (EVSCIDTS), which integrates the “human–machine–material–environment” framework, comprises components such as personnel, machinery, materials, environmental factors, facilities and equipment, intelligent systems, and twin simulation systems. The system emphasizes bidirectional real-time interaction and dynamic mapping between physical and virtual spaces, constructing a virtual mirror image spanning raw materials, components, and complete electric vehicle integration. It forms a self-organizing evolutionary closed-loop system centered on “mapping-iteration-decision” [17]. Through real-time monitoring and predictive fault warning, EVSCIDTS enables intelligent supply chain management for electric vehicle enterprises, enhancing their resilience against sudden risks. As shown in Figure 1, the system features a four-tier architecture:

(1)

The physical layer represents the collection of physical entities within the supply chain of electric vehicle enterprises, encompassing physical elements such as personnel, facilities, and equipment;

(2)

The perception layer collects, processes, stores, and transmits data through sensor devices, GPS positioning, and other equipment, digitally mapping logistics and abnormal conditions in the physical space;

(3)

The digital twin layer is built upon a digital twin platform, representing virtual counterparts of physical scenarios, including twin models;

(4)

The application layer refers to services provided throughout the entire supply chain life cycle from raw materials and components to vehicle manufacturing, maintenance, and recycling—encompassing monitoring and early warning systems, simulation optimization, and so on.

3.2. Operational Mechanism of the Intelligent Digital Twin System for the Electric Vehicle Supply Chain

The operational mechanism of the EVSCIDTS demonstrates an intelligent, closed-loop logic for the integration, optimization, and control of physical and virtual spaces. It begins by leveraging multi-sensor systems, GPS, and other technologies to capture critical node information, such as logistics and positioning data, from the physical environment of the electric vehicle supply chain. This data is then collected and fused using perception technologies, enabling the system to dynamically profile operational states and accurately detect anomalies [18,19]. By utilizing a digital twin platform, a virtual twin space can be constructed. The virtual models are continuously updated to optimize decision-making processes, conduct simulation analyses, and implement monitoring and early-warning systems, thereby controlling and optimizing the physical space. Specific details are illustrated in Figure 2. EVSCIDTS elevates the electric vehicle enterprise supply chain from “passive response” to “self-organizing evolution”. Through the integration of virtual and physical systems and AI-driven capabilities, it achieves comprehensive enhancements in resource efficiency, resilience, and sustainability.

3.3. The Mechanism of Digital Twin Empowering Resilience Enhancement in the Electric Vehicle Supply Chain

In the first phase, when supply chain risks impact electric vehicle enterprises, they will struggle to withstand the shock and face a crisis. Companies can utilize RFID and other technologies to establish monitoring networks, gaining real-time insights into the status of automotive components, vehicle production, and distribution processes. This enables the detection of fault signals and anomalies, facilitating early warning capabilities. Digital twin systems enable enterprises to implement targeted response strategies through simulation-based optimization, serving as a “sensory nerve” for supply chain resilience management. This enables enterprises to shift from reactive responses to proactive foresight.

In the second phase, leveraging the closed-loop interaction mechanism of the digital twin, enterprises transform supply chain operational processes and relationships while optimizing decoupling points. This drives cross-scenario collaboration and cross-domain coordination within the supply chain ecosystem, promotes resource sharing, such as production and inventory, and facilitates a shift toward modular complementarity. It enhances critical capabilities like resilience, defense, and adaptability, propelling supply chain resilience from static defense to dynamic transformation and upgrading. The closed-loop interaction and adaptive learning mechanisms of digital twin enhance their effectiveness, enabling supply chains to enhance risk defense capabilities through continuous iteration.

3.4. Analysis and Integration Between Architecture, Operational Mechanisms, and Variable Dimensions

The four-layer architecture depicted in Figure 1 provides a static “skeletal system” for digital twins to enhance the resilience of electric vehicle supply chains. The physical layer comprises the collection of physical entities within the supply chain of electric vehicle enterprises; the perception layer constitutes the sensory nerves of the supply chain, capturing risk signals in real time; the twin layer conducts simulation and scenario analysis; while the application layer serves as the execution terminal for resilience enhancement. The effective operation of this architecture relies upon the dynamic closed-loop revealed in Figure 2: “perception-mapping-simulation-optimization-feedback”. When supply chains encounter disruptions, the digital twin system rapidly identifies anomalies through the perception layer, constructs virtual risk scenarios in the twin layer, and rehearses response strategies via the simulation optimization capabilities of the application layer. This facilitates a shift from reactive response to proactive prediction.

The second phase depicted in Figure 3 enables the system, through continuous virtual–physical interaction and closed-loop learning, not only to manage known and unknown risks but also to drive profound structural transformation within the supply chain. This manifests as the decoupling and restructuring of supply chains alongside ecosystem-wide collaboration. This explains why isolated digital twin conditions alone cannot constitute high supply chain resilience, as enhancing EV supply chain resilience relies on the digital twin’s capacity to trigger the reshaping of organizational processes, business relationships, and even industrial boundaries. The subsequent multi-period fsQCA configuration analysis empirically identifies which key internal configurations of the digital twin can effectively initiate and sustain this resilience enhancement mechanism that transitions from “monitoring and early warning” to “ecological restructuring”.

This study derives empirical metrics from the EVSCIDTS theoretical framework, translating its core elements into observable prerequisites. The operational logic is as follows: the effectiveness of digital twins in empowering supply chain resilience depends both on the performance of each architectural layer and on the resource provision supporting the overall system.

Application layer performance is measured across two dimensions: the depth of system integration and the efficiency of interfaces that translate system insights into management actions, which correspond to the later-mentioned Digital Twin Application (DTA) and Human–Machine Collaboration (HMLCL), respectively.

The technical quality of the perception and twin layers is determined by the maturity and advancement of underlying digital technologies, corresponding to Digital Technology Development (DLDT).

Data security management (DSM) forms a cross-domain foundational dimension, ensuring the trustworthiness and availability of information across all layers.

Sustained resources for system development, maintenance, and upgrades depend on research resources investment (RRA) and digital technology professionals (DTP).

Thus, the six prerequisites are not arbitrarily combined but represent a systematic mapping and measurement of key components within the EVSCIDTS theoretical framework.

4. Research Methods and Design

4.1. Research Methods

The dynamic fsQCA method is particularly suited for handling both interval and ratio-scaled variables by treating all variables as continuous and defining membership degrees for specific sets. Unlike traditional statistical methods, which require large samples to ensure statistical power, fsQCA emphasizes comparative case analysis to identify configurations of sufficient and necessary conditions that lead to a given outcome. Dynamic fsQCA is therefore generally appropriate for small-to-medium-N studies. Given the continuous nature of most variables in this study and its small-to-medium sample size, the multi-period fsQCA method is adopted. Its application serves a dual purpose: to explore the complex causal relationships between the internal elements of digital twin and supply chain resilience, and to investigate the temporal evolution of these causal configurations. The specific theoretical model for this analysis is presented in Figure 4.

4.2. Detailed Procedures for Dynamic fsQCA

4.2.1. Data Calibration and Dynamic Setting

Based on the theoretical framework and case data, fuzzy set calibration is applied to the outcome variables and antecedent conditions across the three time periods, converting them into membership scores between 0 and 1. Calibration anchors were comprehensively determined according to theoretical percentiles and methodological references. The crossover, complete non-affiliation, and complete affiliation anchors were set at the 0.5, 0.05, and 0.95 percentiles, respectively.

4.2.2. Necessity Analysis and Truth Table Construction

Necessity analysis is performed for each time period to identify any necessary conditions for high supply chain resilience. Following this, a separate truth table is constructed for each period, presenting all possible configurations of antecedent conditions alongside their corresponding outcome membership.

4.2.3. Condition Configuration Identification and Organized Analysis

For each period, standardized Boolean minimization of the truth table generated the specific configurations leading to high supply chain resilience. These configurations were then systematically analyzed.

4.2.4. Robustness Test Process

To ensure the reliability of our findings, we conduct multiple tests, including adjusting calibration anchor points and altering consistency thresholds. The core conclusions remain stable under various parameter settings, confirming that the findings of this study possess sufficient robustness.

4.3. Sample Selection and Data Sources

This study focuses on listed electric vehicle enterprises, as they provide relatively comprehensive and representative industry data. Based on the completeness of data disclosed in annual reports, 27 listed companies were ultimately selected as the research sample.

The years 2020, 2022, and 2024 were chosen as longitudinal comparison points for the following reasons. Research on supply chain resilience gained prominence following the COVID-19 outbreak. Using 2020 as a baseline with two-year intervals effectively captures the “shock-adaptation-evolution” cycle of the electric vehicle supply chain. The pandemic caused severe disruptions in 2020. By 2022, amid ongoing geopolitical and supply chain volatility, firms had entered a phase of adaptive restructuring. By 2024, guided by relevant strategies, the industry was undergoing intelligent and digital transformation. These three points capture the dynamic trajectory of digital twin-enabled resilience as the supply chain adapts to environmental shifts. The two-year span allows observation of meaningful strategic evolution while minimizing external noise from longer periods, thereby clarifying the evolutionary logic of configuration pathways.

Most data is sourced from publicly released annual reports, corporate disclosure documents, news reports, and relevant databases of the sample companies during the specified three-year period. The keyword data required for text analysis was obtained through Python 3.9.5-based full-text extraction and lexical analysis of annual reports.

4.4. Variable Measurement and Calibration

4.4.1. Antecedent Variables Measurement

Research on digital twin remains limited; thus, academia has not yet reached a consensus regarding its metrics. As shown in Figure 4, the variable DLDT was obtained by using Python’s jieba word segmentation package to analyze corporate annual reports [20]. The frequency of 32 keywords related to digital technology was precisely counted. Each value was then incremented by 1, and the results were then transformed using the natural logarithm. The selected keywords are detailed in Appendix A. To ensure content validity, two experts specializing in supply chain were invited to review the keyword list, confirming that it accurately reflects the current digital technology landscape in the electric vehicle manufacturing sector. While the text mining methodology offers a novel and direct measurement perspective, this study acknowledges its intrinsic limitations. The derived metric essentially captures “disclosure intensity” rather than “actual technological capability”. Companies might strategically emphasize popular technical terms, whereas truly advanced firms may underreport mature technologies. To address these limitations, future research could employ more sophisticated natural language processing techniques, such as contextual semantic analysis based on models like BERT, to identify the sentiment and application context of technology mentions. Alternatively, multi-source cross-validation using patent data and case studies could help build more robust measurement metrics. RRA was measured by the proportion of R&D expenditure. DTP was expressed by the ratio of the number of R&D personnel to the total number of employees in the company. The remaining indicators were assigned values using four-valued fuzzy sets based on their varying degrees of intensity, assigning values of 0, 0.33, 0.67, and 1 to represent complete non-affiliation, partial non-affiliation, partial affiliation, and complete affiliation, respectively, as shown in

Table 1. Assignment criteria for four-valued fuzzy sets of relevant variables.

Variables	Assignment Standard	Assignment
DTA	Conducting in-depth digital twin simulation and closed-loop optimization within the supply chain	1
	There are clear technical applications and projects, but their depth is limited	0.67
	Preliminary digital twin applications	0.33
	Not involved at all	0
HMLCL	Deployed human–machine intelligent interaction within the supply chain to realize data-driven collaborative decision-making processes	1
	Provide visual data dashboards, mobile management systems, or decision support tools for the electric vehicle supply chain	0.67
	Preliminary human–machine interaction collaboration	0.33
	Not involved at all	0
DSM	Possesses a comprehensive data security management system	1
	A relatively clear data security management system and technical measures have been established	0.67
	Foundational data security management system	0.33
	Not involved at all	0

Source(s): Authors’ own work.

.

4.4.2. Outcome Variables Measurement

Electric vehicle supply chain resilience (SCR) is a comprehensive metric that encompasses multiple dimensions of a company’s heightened vigilance, resistance, and recovery capabilities in response to external environments, making it difficult to assess from a single perspective. We adopt the widely recognized resilience capability stage theoretical framework within the international academic community [21]. This framework deconstructs the complete journey of a supply chain’s response to disruption into three sequential stages: pre-event, during-event, and post-event, corresponding to three core capabilities: defense, response, and recovery. This three-dimensional framework has been adopted by numerous studies due to its systematic nature and operational feasibility [22]. The present research aligns profoundly with this theoretical framework. Based on this, drawing on the essence of supply chain resilience and referencing research by Qi et al. [23], this study measured supply chain resilience across three dimensions: defense, response, and recovery capability. The CRITIC method was employed to assign weights to each indicator.

(1)

Defense Capability. Drawing upon the research of Fan et al. [24] and Shareef et al. [25], this study employed supply chain visibility and risk prevention mechanisms as measurement indicators. Based on the fsQCA variable assignment criteria, a four-value fuzzy set assignment was applied, assigning values of 0, 0.33, 0.67, and 1 to represent complete non-affiliation, partial non-affiliation, partial affiliation, and complete affiliation, respectively.

(2)

Responsiveness. Inventory turnover rate (times) was used as the metric.

(3)

Recovery. Financial strength influences the speed of recovery and adjustment capabilities within an enterprise’s supply chain. Drawing upon the research of Zhang et al. [26], two metrics, return on assets and debt-paying capacity, were employed to measure recovery. Return on assets was measured by the ratio of net profit to average total assets. Debt-paying capacity is expressed as the ratio of net cash flow from operating activities to current liabilities.

4.4.3. Variables Calibration

For variables lacking external calibration standards, we applied the direct calibration method to reduce subjective bias, aligning with methodological precedents [27,28,29]. The crossover, complete non-affiliation, and complete affiliation anchors were set at the 0.5, 0.05, and 0.95 percentiles, respectively. This commonly used scheme provides a sound trade-off; it ensures sufficient discrimination between clearly “in” and “out” cases while avoiding the excessive information loss that more extreme thresholds might cause. Detailed calibration values are listed in

Table 2. Variables calibration anchors for each time period.

Variables	2020			2022			2024
	CA	IP	CNA	CA	IP	CNA	CA	IP	CNA
DLDT	4.013	3.091	0.901	4.074	3.135	1.710	4.504	3.526	1.664
RRA	0.258	0.047	0.017	0.181	0.050	0.019	0.186	0.049	0.035
DTP	0.389	0.157	0.059	0.344	0.162	0.058	0.346	0.176	0.098
SCR	0.277	0.141	0.093	0.297	0.168	0.079	0.325	0.210	0.081

Notes: CA—complete affiliation; IP—intersection point; CNA—complete non-affiliation. Source(s): Authors’ own work.

.

5. Empirical Analysis

5.1. Necessity Analysis of a Single Condition

This study employs the fsQCA 3.0 software to analyze the essential conditions for achieving high and non-high supply chain resilience. Existing literature suggests that when the consistency level of an individual variable falls below 0.9, that variable may be regarded as non-essential to the outcome [30]. As shown in

Table 3. Analysis of the necessity for high supply chain resilience across different time periods.

Condition Variable	High Supply Chain Resilience
	2020		2022		2024
	Consistency	Coverage	Consistency	Coverage	Consistency	Coverage
DTA	0.721	0.804	0.922	0.817	0.770	0.843
~DTA	0.559	0.505	0.425	0.430	0.475	0.496
DLDT	0.723	0.730	0.798	0.663	0.779	0.818
~DLDT	0.603	0.594	0.482	0.527	0.511	0.556
HMLCL	0.596	0.775	0.649	0.757	0.675	0.891
~HMLCL	0.672	0.543	0.615	0.488	0.591	0.531
RRA	0.588	0.638	0.610	0.597	0.627	0.740
~RRA	0.676	0.625	0.674	0.616	0.643	0.628
DTP	0.609	0.630	0.624	0.596	0.671	0.736
~DTP	0.699	0.673	0.696	0.650	0.624	0.650
DSM	0.779	0.705	0.688	0.827	0.808	0.880
~DSM	0.468	0.520	0.669	0.520	0.524	0.550

Notes: ~ stands for the logical “negation” or “absence” of a condition. Source(s): Authors’ own work.

and

Table 4. Analysis of the necessity conditions of non-high supply chain resilience in each period.

Condition Variable	Non-High Supply Chain Resilience
	2020		2022		2024
	Consistency	Coverage	Consistency	Coverage	Consistency	Coverage
DTA	0.453	0.508	0.495	0.491	0.446	0.425
~DTA	0.825	0.748	0.816	0.921	0.836	0.760
DLDT	0.590	0.599	0.612	0.569	0.532	0.487
~DLDT	0.734	0.727	0.638	0.779	0.801	0.760
HMLCL	0.438	0.573	0.423	0.551	0.400	0.460
~HMLCL	0.828	0.673	0.814	0.722	0.905	0.708
RRA	0.596	0.649	0.624	0.682	0.563	0.579
~RRA	0.667	0.619	0.631	0.644	0.747	0.636
DTP	0.662	0.689	0.664	0.710	0.614	0.587
~DTP	0.645	0.623	0.622	0.649	0.724	0.657
DSM	0.570	0.518	0.448	0.602	0.507	0.482
~DSM	0.675	0.754	0.871	0.758	0.874	0.798

Notes: ~ stands for the logical “negation” or “absence” of a condition. Source(s): Authors’ own work.

, most conditions exhibit a consistency coefficient below 0.9 for both high and non-high supply chain resilience. Although two conditions show a consistency coefficient above 0.9, they do not serve as absolute prerequisites for influencing either high or non-high supply chain resilience. This implies that the six conditional variables are not the only prerequisites for enhancing or diminishing supply chain resilience.

5.2. Configuration Analysis

5.2.1. Configuration of Conditions for Achieving High Supply Chain Resilience

Referencing existing research [31], consistency thresholds of 0.8 and 0.7 were set for PRI, with the case-frequency threshold set to 1. Given that there is no conclusive literature to support the direction of influence for conditional variables, no direction was predetermined. Results primarily utilize intermediate solutions supplemented by simple solutions, distinguishing between core and peripheral conditions. The configuration outcomes achieving high supply chain resilience are shown in

Table 5. Configuration results for achieving high supply chain resilience.

Condition Variable	2020				2022		2024
	A1	A2	A3	A4	B1	B2	C1	C2	C3	C4
DTA	⬤	⬤	⬤	$\otimes$	⬤	●	●		⬤	⬤
DLDT		$\otimes$	●	$\otimes$	⊗	●	●	●	⬤	⬤
HMLCL	$\otimes$	$\otimes$	⬤	⬤	$\otimes$	⬤	⬤	⬤	⬤
RRA	$\otimes$		●	●	$\otimes$	$\otimes$		●		⬤
DTP	$\otimes$	●		●	$\otimes$	$\otimes$	●	●	⊗	⊗
DSM	⬤	⬤	⬤	⬤	⬤	⬤	$\otimes$	$\otimes$	●	⬤
Consistency	0.862	0.895	0.892	0.959	0.953	0.960	0.982	0.949	0.996	0.953
Raw coverage	0.357	0.247	0.366	0.206	0.304	0.359	0.302	0.294	0.378	0.337
Unique coverage	0.081	0.002	0.118	0.016	0.072	0.127	0.0001	0.016	0.062	0.044
Total consistency	0.829				0.967		0.946
Total coverage	0.545				0.431		0.552

Notes: ⬤—core conditions present; ⊗—core conditions absent; ●—peripheral conditions present; $\otimes$—peripheral absent, same below. Source(s): Authors’ own work.

, yielding 10 condition configurations across three time periods.

(1)

Configuration Analysis for 2020

The configuration “dual-driven by twin and safety” consists of configurations A1 and A2, which are centered on advanced DTA and robust DSM. A prime example is Tesla. In digital twin applications, its “virtual factory” model achieves real-time mapping and simulation of the entire production line process, fulfilling the depth requirements of high DTA. Its implementation of end-to-end encryption for all twin data meets the demands of high DSM.

The “comprehensive upgrade digital twin” configuration A3 indicates that high DTA, high HMLCL, and high DSM serve as core conditions, while non-high DLDT and non-high RRA act as peripheral conditions. BMW Group has employed digital twin technology to establish a “virtual factory”. The company prioritizes human–machine collaboration efficiency by deploying lightweight robotic arms in its facilities to perform precision door welding alongside workers. It has established a comprehensive data security management system covering the entire life cycle.

The “human–machine safety-driven” configuration A4 prioritizes high levels of HMLCL alongside robust DSM as core requirements, while non-intensive investment in high-level research resources and non-reliance on specialized digital technology professionals serve as peripheral conditions.

(2)

Configuration Analysis for 2022

Configuration B1 is the same as A1 and A2, classified as “dual-driven by twin and safety”. Configuration B2 belongs to the same category as A4, representing a “human–machine safety-driven” approach. Configuration B2 is supplemented by non-high DTA and non-high DLDT as marginal conditions.

(3)

Configuration Analysis for 2024

“Human–machine interaction dominant” encompasses configurations C1 and C2. It centers on high levels of HMLCL as its core condition, while non-high DLDT and non-high DTP serve as peripheral conditions. Configuration C1 incorporates the peripheral condition of a non-high DTA, while configuration C2 incorporates the peripheral condition of RRA. Seres Automobile addresses technological shortcomings through human–machine collaboration, adopting a model where human–machine collaboration drives the use of scenario-based digital tools to enhance supply chain resilience.

The “advanced interactive digital twin environment” configuration C3 centers on three core conditions: advanced DTA, high DLDT, and superior HMLCL. DSM serves as a peripheral condition. Tesla serves as a representative case study for this configuration and will not be elaborated upon further.

The “four-dimensional driven digital twin” configuration C4 centers on four core conditions: advanced DTA, high DLDT, substantial RRA, and rigorous DSM. BYD’s fully self-developed technology system demonstrates profound internal digital expertise and sustained R&D investment. Furthermore, its deep application of digital twin technology across the R&D and production chain showcases leading implementation capabilities. Finally, its comprehensive data life cycle management system ensures rigorous data security controls. This integrated approach enables BYD to establish a closed-loop system encompassing proprietary technology, secure data management, and deep application capabilities. This facilitates the simulation of complex scenarios and autonomous supply chain optimization, ultimately achieving highly resilient operational objectives.

(4)

Analysis of Multi-period Configuration Evolution

Dynamic configuration theory encompasses three trajectories: the “dominant trajectory” persists throughout all time periods; the “turning trajectory” shifts from one dominant type to another; and the “hybrid trajectory” incorporates elements of both, featuring minor adjustments without major shifts [30]. The configuration evolution trajectories are shown in

Table 6. Configuration evolution trajectory.

Configuration Type	2020	2022	2024	Trajectory
dual-driven by twin and safety	√	√		Turning trajectory
comprehensive upgrade digital twin	√			Turning trajectory
human–machine safety-driven	√	√		Turning trajectory
human–machine interaction dominant			√	Turning trajectory
advanced interactive digital twin environment			√	Turning trajectory
four-dimensional driven digital twin			√	Turning trajectory

Notes: √ stands for existence. Source(s): Authors’ own work.

.

The dominant set of conditions underwent a systemic shift from 2020 to 2024, which can be explained through the theoretical logic of “shock-response-restructuring” and the practical context of industry development.

In 2020, “security and collaboration” became the survival logic in the face of the abrupt, comprehensive supply chain disruptions caused by the COVID-19 pandemic. Under extreme uncertainty, the overriding corporate goals were “supply assurance” and “disruption prevention”. Operational continuity depended on basic visualization and emergency scheduling via human–machine collaboration. This prioritization explains why “technological advancement” was often a peripheral or absent condition: firms simply lacked the bandwidth to pursue long-cycle innovation during an acute crisis.

Path divergence and exploration during the 2022 transition period: while security-centric pathways persisted, their dominance gave way to more complex pathways. The industry entered a “new normal of volatility”, with supply chain risks evolving from acute shocks to chronic pressures, such as geopolitical tensions. Enterprises transitioned from “survival mode” to “adaptation mode”, experimenting with digital twins for more granular management and localized optimization. This expansion in configuration types reflects the diverse explorations undertaken by businesses as they seek new equilibrium points.

By 2024, the dual-core drive of “technology-R&D” had emerged as the new norm, with the “advanced interactive digital twin environment” and “four-dimensional driven digital twin” pathways gaining prominence. This shift aligned with the industry’s wave of digital and intelligent transformation. As the risk of physical supply chain disruptions was partially mitigated through diversified layouts, the competitive focus shifted towards “enhancing quality, boosting efficiency, and driving innovation”. Enterprises must leverage advanced DLDT and sustained RRA to propel digital twins, enabling supply chains to leap from reactive post-event responses to proactive simulations and autonomous optimization. Meanwhile, human–machine interaction evolved into an internalized foundational competency, its previous centrality now supplanted by the more strategic drivers of technology and R&D.

Across all time periods, DSM, HMLCL, and DTA serve as core prerequisites, underscoring their critical importance in empowering supply chain resilience through digital twin. In 2024, DTA, DLDT, and RRA will become even more prominent. Secondly, among the interrelationships between condition configurations across different time periods, the synergistic patterns between DTA and DSM, as well as between DTA and DLDT, remain relatively stable throughout the evolution process. Data security management provides a secure environment for empowering digital twins; digital technology development enables technological upgrades and efficiency improvements, jointly driving the enhancement of supply chain resilience.

5.2.2. Configuration of Conditions for Achieving Non-High Supply Chain Resilience

Taking the 2020 configuration NA1 as an example, when core conditions such as DLDT, RRA, and DSM are lacking, along with other peripheral conditions, it becomes difficult to leverage the impact of digital twin, resulting in low supply chain resilience. The configuration results for other time periods are presented in

Table 7. Configuration results for achieving non-high supply chain resilience.

Condition Variable	2020			2022				2024
	NA1	NA2	NA3	NB1	NB2	NB3	NB4	NC1	NC2
DTA	$\otimes$	$\otimes$	⊗	⊗	⊗	⊗	⊗	⊗	⊗
DLDT	⊗	⊗	⬤	$\otimes$			$\otimes$		⊗
HMLCL	$\otimes$	$\otimes$	⬤	$\otimes$	$\otimes$	$\otimes$		$\otimes$	$\otimes$
RRA	$\otimes$	⬤	●		$\otimes$	●	●	⊗
DTP	⊗	●	●		$\otimes$	●	●
DSM	⊗	⊗	⊗	$\otimes$	$\otimes$	$\otimes$	$\otimes$	⊗	$\otimes$
Consistency	0.945	0.928	0.954	0.948	0.960	0.962	0.962	0.934	0.922
Raw coverage	0.414	0.286	0.215	0.546	0.496	0.415	0.375	0.617	0.687
Unique coverage	0.171	0.033	0.064	0.067	0.033	0.027	0.032	0.049	0.119
Total consistency	0.946			0.956				0.921
Total coverage	0.521			0.683				0.736

Notes: ⬤—core conditions present; ⊗—core conditions absent; ●—peripheral conditions present; $\otimes$—peripheral absent, same below. Source(s): Authors’ own work.

.

5.3. Robustness Test

Drawing upon the research of Fiss [32] and Waldkirch et al. [33], this paper examines the robustness of the configuration results through the following approaches. First, after modifying the calibration anchor point values from the 0.5, 0.05, and 0.95 percentiles to the 0.15, 0.5, and 0.85 percentiles, the configuration results are presented in

Table 8. Robustness test results following modification of calibration anchor point values.

Condition Variable	2020				2022		2024
	A1	A2	A3	A4	B1	B2	C1	C2	C3	C4
DTA	⬤	●	$\otimes$	$\otimes$	●	●	●		⬤	⬤
DLDT		●	●	$\otimes$		●	●	●	⬤	⬤
HMLCL	$\otimes$	⬤	$\otimes$	⬤	$\otimes$		⬤	⬤	⬤
RRA	$\otimes$	●	⬤	●	$\otimes$	$\otimes$		●		⬤
DTP	⊗		⊗	●	$\otimes$	$\otimes$	●	●	⊗	⊗
DSM	●	⬤	$\otimes$	⬤	⬤	⬤	$\otimes$	$\otimes$	●	⬤
Consistency	0.817	0.863	0.871	0.876	0.887	0.901	0.963	0.927	0.998	0.942
Raw coverage	0.296	0.344	0.154	0.112	0.242	0.255	0.258	0.266	0.317	0.271
Unique coverage	0.139	0.134	0.041	0.010	0.088	0.100	0.001	0.034	0.058	0.036
Total consistency	0.813				0.917		0.934
Total coverage	0.562				0.342		0.525

Notes: ⬤—core conditions present; ⊗—core conditions absent; ●—peripheral conditions present; $\otimes$—peripheral absent, same below. Source(s): Authors’ own work.

. From the perspective of the set relations among configurations, the results obtained show no substantive change compared to the aforementioned configuration results. Subsequently, we adjusted the consistency threshold from 0.80 to 0.85, with all other conditions remaining unchanged. Secondly, we raised the PRI consistency from its inherent cutoff value of 0.7 to 0.8, while keeping all other parameters constant. We elevated both the consistency threshold and PRI value to 0.85 and 0.8, respectively. This will not be elaborated further here. The configurations generated from these three tests largely encompassed the existing configurations. These tests demonstrate that the configuration results presented in this study are relatively robust.

6. Discussion and Conclusions

This study addresses the question of how digital twin empowers the enhancement of resilience in electric vehicle supply chains. Through examining the system architecture, operational mechanisms, and functional mechanisms, it reveals the application prospects of the digital twin in enhancing electric vehicle supply chain resilience. The introduction of digital twin expands the theoretical boundaries of supply chain resilience research, offering new analytical perspectives and frameworks.

This study establishes that digital twin technology enhances electric vehicle supply chain resilience not through singular factors, but through multiple, equally effective configurations of its core dimensions. This study identifies six types of high-resilience pathways, such as “dual-driven by twin and safety”, “comprehensive upgrade digital twin”, “human–machine safety-driven”, “human–machine interaction dominant”, “advanced interactive digital twin environment”, and “four-dimensional driven digital twin”, demonstrating that there is no universal “best” pathway. This finding of equifinality is complemented by causal asymmetry, as the paths leading to non-high resilience are not mere opposites of the successful ones.

Multi-period analysis shows that the synergistic relationship among DTA, DSM, and DLDT remains relatively stable. Furthermore, enhancing the contribution of digital twin to corporate supply chain resilience requires the coordinated development of multiple factors, among which DSM and DLDT play critical roles.

This study incorporates the temporal dimension into the research. It reveals the dynamic evolutionary mechanism of digital twins in enhancing the resilience of electric vehicle supply chains, thereby advancing the application and development of the dynamic configuration approach and dynamic QCA methodology within the context of electric vehicle supply chain resilience. This paper analyses the multiple concurrent causal relationships through which digital twin multi-factor collaboration enhances the resilience of electric vehicle supply chains from a configuration perspective, effectively addressing previous research’s oversight of how concurrent multi-factor interactions may impact supply chain resilience.

7. Managerial Implications

(1)

For Electric Vehicle Enterprise Managers: Selecting Suitable Configuration Paths

Electric vehicle enterprises should undertake self-assessment and path adaptation. Those with robust technological foundations and ample funding should pursue pathways such as the “four-dimensional driven digital twin”, “advanced interactive digital twin environment”, or “comprehensive upgrade digital twin”. Such enterprises should leverage their R&D and integration strengths to establish autonomous, controllable digital twin platforms covering the entire supply chain. Coordinate data security management with technological development to achieve deep integration between artificial intelligence and twin models. This enables AI to undertake standardized and high-frequency data processing tasks, thereby continuously enhancing the adaptive, self-organizing resilience of the electric vehicle supply chain.

Electric vehicle enterprises with sound foundations in human–machine collaboration but limited technical resources should prioritize deploying lightweight, visualized digital twin interaction front-ends when pursuing pathways such as “human–machine interaction dominant”. Conducting “human–machine collaboration sandbox simulations” to periodically model disruption scenarios based on twin data will optimize emergency response procedures, maximize synergies between existing technology and human capital, and unlock the full potential of human–machine interaction in enhancing the resilience of electric vehicle supply chains.

For electric vehicle enterprises at the nascent stage of digitalization, following the “dual-driven by twin and safety” and “human–machine safety-driven” pathways, it is advisable to establish a unified data security governance framework. Rather than pursuing comprehensive platform development, they should identify critical bottlenecks and introduce pilot initiatives for modular digital twin projects and human–machine collaboration schemes.

(2)

For Relevant Government Departments: Provide Support and Formulate Policies

Relevant government departments play a crucial role in bolstering corporate supply chain resilience by providing foundational support and formulating targeted policies. This involves cultivating essential resources, including technology, talent, policy frameworks, and market access, while proactively diagnosing critical bottlenecks that hinder enterprise development.

A differentiated, pathway-specific approach to policy support is recommended. For instance, to assist firms adopting a “human–machine interaction dominant” pathway, the government should foster partnerships with universities and vocational institutions to design specialized curricula that cultivate multidisciplinary talent in digital technology and supply chain management. The government should allocate funding to establish a repository of best-practice cases for human—machine interaction applications in supply chains. Furthermore, the development of clear policy frameworks, such as those governing digital twin technologies, can significantly lower the barriers to their adoption for enhancing supply chain resilience.

Importantly, policy should avoid promoting a single “benchmark electric vehicle enterprise”. It would be more effective to identify and support a diverse set of exemplary firms, each representing a distinct, successful configuration pathway. This would provide a spectrum of relevant benchmarks for electric vehicle enterprises at varying stages of maturity.

(3)

For Electric Vehicle Industry: Address the Risks of Emerging Technologies

When embracing the potential of digital twins, the electric vehicle industry should remain acutely aware of the emerging risks they entail. Model bias and data inaccuracies may lead to erroneous decision-making; over-reliance on systems could result in the erosion of personnel skills and alert fatigue. Consequently, future technological governance must concurrently establish a dual safeguard mechanism consisting of “technology validation” and “human-centered design”.

8. Limitations and Future Research

First, given that digital twins are still in their infancy in supply chain resilience applications, there is a lack of discussion on their specific mechanisms for enhancing supply chain resilience. Consequently, there are few research findings to draw upon, and the proposed measurement indicators require further refinement and deepening. Future research may transcend text frequency analysis and expert-assigned scoring methods by employing natural language processing techniques to conduct more nuanced semantic analysis of annual reports and technical documentation. Alternatively, integrating multi-source data such as patent records and supply chain digital maturity audit reports could facilitate the development of more robust, multi-dimensional measurement frameworks.

Secondly, this study primarily focuses on the internal factors of digital twins, while overlooking the influence of external environmental, organizational, and human factors. Factors such as organizational change management capabilities, geopolitical risks, and market dynamics all have a significant impact on supply chain resilience. Future research should adopt a broader, multi-level analytical perspective to investigate how the external environment, organizational context, and internal technological components dynamically interact and synergistically work together. This approach will facilitate a more comprehensive understanding of the ways to enhance the resilience configuration of electric vehicle supply chains.

Thirdly, although the QCA method is highly suitable for exploring complex causal configurations in small-to-medium sample sizes, the generalizability of this study’s conclusions remains constrained by the number of cases examined. Future research could expand sample sizes and combine the QCA method with quantitative approaches to validate and supplement these findings.

Finally, the deep integration of artificial intelligence and digital twin technologies is accelerating the release of synergistic effects. Future research can further explore AI-enabled digital twin frameworks to enhance the resilience of electric vehicle supply chain networks. How will AI enable digital twins to evolve from “describing the present state” to “generating strategies”? This necessitates developing new metrics for “AI–twin collaborative autonomy”.