Abstract
Current SupplierManufacturer (SM) networks are highly complex and susceptible to local and global disruptions, due to connectivity and interdependency among suppliers and manufacturers. Resiliency of supply chains is critical for organizations to remain operational in the face of disruptive events. Existing quantitative analyses oversimplify the mutualistic nature of SM networks, in which failure of individual entities affects not only the directly connected entities but also those connected indirectly. In this work we investigate resiliency of SM networks using the quantitative methods employed to study mutualistic ecological systems. Much like in ecological systems, catastrophic failures of SM networks are difficult to predict due to high dimensionality of their interactive space. To address this, first we create a bipartite representation and generate a multidimensional nonlinear model that captures the dynamics of a SM network. We transform the multidimensional model into a twodimensional model without sacrificing the model’s ability to predict the point of collapse. We extensively validate the model using realworld global automotive SM networks. We observe that the resiliency of a SM network depends on both the network structure and parameters. The current work offers a means for designing resilient supply chains that can remain robust to local and global perturbations.
Similar content being viewed by others
Introduction
Large manufacturing organizations rely on many suppliers spread across globally to acquire parts for their products. At factory level, internal perturbations in the form of labor unrest, machine reliability issues, or material shortage affect the optimum performance of a factory and the associated SM network. At meso level, geographical conditions surrounding the factory, local perturbations in the form of natural disasters, material shortage or restrictive economic policies tend to affect the performance of suppliers and manufacturers. Finally, at enterprise level, global perturbations in the form of reduction in overall demand or weak economic markets reverberate across the entire supply chain. Disruptions in supply chain can be detrimental to a firm’s performance and can negatively impact the overall financial gains^{1,2,3,4,5}. Organizations are proactively paying attention to improving their Supply Chain Resiliency (SCR)^{5,6,7,8}. SCR enables SM networks to withstand disruptions at multiple scales to mitigate performance and economic losses^{5}. For example, it was SCR that enabled Japanese firms to rapidly respond and restore production post a massive earthquake in 2011 through resilient resource mobilization^{5,9}. The common notion of SCR is the ability of a supply chain to cope with the unavoidable and unpredictable performance disruptions and bounce back to the original operating state or to a new desirable operating state^{7,10,11,12,13,14}.
The existing studies on resiliency of supply chains focus on both qualitative and quantitative approaches, though qualitative ones are more common. These strategies involve heuristics and rule of thumb strategies built on past experience. At factory level, redundancy in the form of parallel machines and storage buffers improves resiliency^{15}. At system level, the risks of supplier disruptions are mitigated by building redundancy through multisource strategies. The common approach in all the qualitative studies is identifying vulnerabilities and building capabilities to counter the vulnerabilities^{16}. Quantitative approaches on the other hand focus on operational resiliency. They monitor postdisruption performance of supply chains using key performance metrics (KPIs)^{17,18}. The reactive and preventive strategies for improving resiliency minimize the related costs including the cost associated with recovery and performance loss^{17}. SM networks are multidimensional, and the existing lowdimensional analyses of SCR do not effectively incorporate structural information of interconnected components and do not sufficiently consider the dependencies between the suppliers and manufacturers. Several studies have investigated the robustness of highly interconnected complex networks in the event of disruption^{19,20,21,22}. Novel strategies are proposed to deal with cascading failures in engineered connected systems such as power grid, and communication networks^{21,22}. Supply chains have been investigated as a complex adaptive system^{23}. This allows researchers to model supply chains as a network and analyze them using techniques from the field of nonlinear dynamics, statistical physics and information theory. However, current supply chain network models do not consider the mutualistic interaction effects of supplier and manufacturer on each other. The phenomenon of mutualism creates an environment in which two different groups benefit mutually due to interactions among members across the groups. The existing analysis only focuses on manufacturer’s resiliency as a function of supplier disruptions. However, because of mutualism, the survival (and growth) of one manufacturer positively influences other manufacturers, and similarly the survival (and growth) of one supplier positively influences other suppliers. Intuitively speaking a much better configuration to study SCR would be to treat manufacturers and suppliers as different classes having betweenclass interactions. Such configurations are commonly investigated in mutualistic ecological networks. The dynamical properties of such mutualistic systems are represented by a mathematical model that combines the network structure and the parameters to assess the resiliency of the system against perturbations^{24,25,26}. The mathematical models presented in these studies involve growth, decay, competition, and mutualistic effects. Previous studies have reduced the multidimensional models into one or twodimensional models to effectively represent the resiliency function^{24,25}. In such dimension–reduced model the resiliency of the system depends on the network structure and parameters alone. We believe that the same analysis can be extended to SM networks where one can treat the manufacturers and suppliers as different species who benefit from mutualistic interactions. The dynamics that govern the abundance of a species in a mutualistic ecological network share similarity with the abundance of throughput (production quantity) in a SM network. Similar to ecological systems, SM networks include a growth component which is bounded by the maximum capacity of the production center, decays in form of internal reliability loss that reduces the production quantity, competition in the form of pricing and technology, and finally the mutualistic effects can be captured using the network structure. Apart from the similarity in the dynamics, interestingly, the structural properties of ecological networks and SM networks are strikingly similar^{27,28,29}. These provide an impetus to emulate the approaches used for ecological networks.
First, we formulate a dynamical model for the multidimensional SM network combining structural information and network parameters. Next, we reduce the multidimensional formulation to an effective twodimensional model. Using this dimensionreduced model, we introduce local and global perturbations and assess their impact on the throughput of the SM network. We observe that the dimensionreduced model, like multidimensional model, is insensitive to external noise and parameter variations, and is able to generate the same point of collapse as the one generated by the multidimensional model. In addition, we observe that the SM network resiliency is highly sensitive to the network structural properties, namely, nestedness and density. The proposed approach is accurate in identifying the point of collapse (tipping point) beyond which a SM network will be unable to deliver the desired throughput and unable to regain its production capacity.
Results
Nonlinear dynamical model of SM mutualistic network
Figure 1a shows a mutulistic SM network, and Figure 1b shows the bipartite representation of the SM network. The manufacturers and the suppliers are represented as nodes. In Figure 1b, M represents a manufacturer and S represents a supplier. A manufacturer is connected to a supplier if the supplier provides parts/products to the manufacturer. The link is unweighted, i.e., even if the supplier supplies multiple components to the manufacturer, the link will still have a weight of one.
Mathematically the bipartite network is represented as adjacency matrix A whose element y_{ij} is 1 if manufacturer i is connected to supplier j; 0 otherwise. Equations (1) and (2) represent the dynamics of a mutualistic SM network. While Eq. (1) models the nonlinear dynamic change of the throughput of manufacturer i, Eq. (2) similarly models the nonlinear dynamic change of the throughput of supplier i.
Eq. (1) follows the proposed logistic growth model for our SM network application. Here M_{i} is the throughput of manufacturer i; \({K}_{i}^{(M)}\) is the production capacity of manufacturer i acting as plant capacity (i.e., capacity limiting term)^{30,31,32,33}; \({\alpha }_{i}^{(M)}\) is an internal perturbation parameter; \({\beta }_{ij}^{(M\mathrm{1)}}\) and \({\beta }_{ij}^{(M\mathrm{2)}}\) are the effects of price competition and technology competition between manufacturers i and j; Q_{M} is the number of manufacturers; \({\gamma }_{ik}^{(M)}\) represents the mutualistic interaction effect between manufacturer i and supplier k; the saturation parameter h is used to characterize the squashing effect; Q_{S} is the number of suppliers; and \({\mu }_{i}^{(M)}\) indicates the production subcontracting/outsourcing intensity. Same notations are applied to Eq. (2) except that the superscript stands for suppliers.
Growth term
The first term on the righthand side of Eq. (1) represents the growth term. In this research we are measuring growth of a manufacturer or supplier in terms of throughput (production volume). Figure 2 shows the commonly adopted growth models, which are either unbounded, bounded, and bell shaped curves. A bell shaped growth model will fail to represent the realty that when manufacturers decommission specific product lines, they compensate for the production loss at the enterprise level by producing other products. A bell shape growth model for the proposed approach will indicate that, over time the throughput of a manufacturer declines to zero (which translates to extinction of manufacturer) after reaching a peak production.
In this context, an unbounded (e.g., exponential) growth model indicates that throughput keeps increasing for ever without a limit. The unbounded growth models, although mathematically appealing, they are not supported by empirical evidence^{34}. For a given plant capacity, the unbounded growth model is feasible only when the throughput of a manufacturing plant scales linearly or superlinearly with the plant utilization, which is not the case in reality. In practice, when the plant utilization increases, the throughput of the plant also increases. However, beyond a certain point, any increase in utilization is followed by a decrease in the throughput. The root cause for this trend is that the increase in utilization comes at the expense of avoiding equipment maintenance which adversely affects machine reliability. This in turn leads to increase in the frequency of failures (F) and longer repair time (T_{d}). An increase in T_{d} and/or F increases downtime, work in progress (WIP), and cycle time (T_{c})^{31,32}. The fact that throughput does not scale linearly with the plant utilization is not represented correctly in the unbounded growth model.
Therefore, a bounded growth model, especially a logistic shape, can better represent throughput growth pattern of a manufacturer or a supplier. In a bounded growth model, the growth becomes bounded as a consequence of the growth itself. Constraints such as space limit, finite talent pool, policies, regulations, resources (water and energy), and the risk of unmanageable complexity, make unending growth untenable. Logistic growth models have extensively been investigated for ecological systems where the environment carrying capacity, and competition act as growth limiting factors^{35,36}. In addition, logistic growth behavior has been observed in new car registrations in Japan^{37}, car population in Italy^{37}, and annual production of natural resources^{34,38}.
Internal perturbation term
The second term in Eq. (1), α_{i}, is an internal perturbation parameter. It represents the production system reliability which is adversely affected by internal perturbations such as machine failure and labor unrest^{39,40,41,42}. Throughput nonlinearly decreases as the production system reliability decreases^{39,40,41,42}. When the throughput is high, a small reduction in reliability due to internal perturbation could result in large throughput reduction; when the throughput is small, a small reduction in reliability can only cause a minor change in throughput^{39,40,41,42}.
Price and technology competition terms
The third and the fourth terms in Eq. (1) reflect the effect of price competition and technology competition between manufacturers (suppliers) i and j on their throughput^{43,44,45}. The throughput of a manufacturer (supplier) declines as the competition becomes fierce. When \({\beta }_{ij}^{(1)} > {\beta }_{ij}^{(2)}\), the price competition dominates the technological advantage. The dominance is reversed when \({\beta }_{ij}^{(2)} > {\beta }_{ij}^{(1)}\).
Mutualistic interaction term
The fifth term in Eq. (1), \({\gamma }_{ik}\), captures the mutualistic interaction effect between manufacturer i and supplier k. When both manufacturers and suppliers are generating high throughput, the interaction effect between them saturates the throughput growth^{24,25,46}. The saturation constant h is used to characterize the squashing effect. The term \({\gamma }_{ik}\) is defined as^{26,47}
where \({y}_{ik}=1\) if manufacturer i and supplier k interact; \({y}_{ik}=0\) otherwise. The term γ_{0} represents the level of mutualistic strength, which in this work is set to 1. The term δ denotes the mutualistic tradeoff modulating the relation between the interaction strength and the number of interactions N_{i}. The interaction strength γ_{ik} is determined by the node degree of manufacturer i. The more connections manufacturer i maintains, the weaker is the mutualistic interaction between manufacturer i and supplier k. This also indicates that when manufacturer i partners with a large number of suppliers, the contribution of each of these suppliers will be smaller.
Outsourcing intensity term
In Eq. (1), μ indicates the production subcontracting/outsourcing intensity. In the long run, μ will have a negligible effect on a manufacturer’s dynamics, but in the short term, μ exerts a considerable influence. However in reality, manufacturers and suppliers rarely outsource their proprietary components/products for the fear of creating unintended opportunities for future competition. So even for short term, one can assume a negligible effect of μ on throughput of a SM network.
As mentioned earlier, the nonlinear dynamics of suppliers are modeled by Eq. (2). All the terms in Eq. 2 have the same interpretation for suppliers as they have for manufacturers in Eq. (1).
Dimension reduction
Equations (1) and (2) represent a multidimensional SM network, where each manufacturer or each supplier is associated with its own dynamics. Multidimensionality increases the computational load and is lack of predictability, and those issues can be resolved by reducing the dimensionality of the model described by Eqs 1 and 2. The dimensionreduced models are given in Eqs (4) and (5). They can be used to capture the SM network’s dynamics and topological information that was described by Eqs (1) and (2). The dimension reduction procedures are given in the Supplementary Information Sections 1–3. In Eqs (4) and (5), M_{eff} and S_{eff} represent the effective average throughput of the SM network aggregated over all manufacturers and suppliers respectively; K is the production capacity; α is the effective internal perturbation parameter; β is the effective competition parameter that combines both the price and technology competitions; and \(\langle {\gamma }_{M}\rangle \) and \(\langle {\gamma }_{S}\rangle \) are the effective mutulistic strengths for manufacturers and suppliers. The terms \(\langle {\gamma }_{M}\rangle \) and \(\langle {\gamma }_{S}\rangle \) can be computed using averaging methods such as unweighted average, degreeweighted average, and eigenvectorbased average^{25}. In this work we consider the degreeweighted average method for computing \(\langle {\gamma }_{M}\rangle \) and \(\langle {\gamma }_{S}\rangle \), since it is robust against noise and parameter variations^{25}.
In the following sections we investigate the effect of network structure on the resiliency function, demonstrate the robustness of the dimensionreduced model against ransom noise and random parameter variations, and discuss the ability of the dimensionreduced model to predict the point of collapse in the structureparameter space.
Analysis of real world SM networks
Studies have revealed a strong similarity between structural properties of mutualistic ecological networks and SM networks^{27,28,29}. Network properties, mainly, nestedness, degree distribution, and modularity are significantly similar for ecological networks and SM networks^{27}. Nestedness is a property of a bipartite network where specialists (e.g., pollinators that visit a few specific type of plants) interact with species of other class with whom generalists (e.g., pollinators that visit many types of plants) interact^{48}. Similarly SM networks are characterized by the presence of “generalist” and “specialist” suppliers^{29}. The generalists have the technical capabilities to manufacture a wide variety of products at lower volumes (job shop production) while specialists have the capabilities to manufacture products with low design variation at higher volumes (mass production). The nestedness in SM networks aligns with that of ecological networks where, the specialist suppliers interact with manufacturers with whom the generalist suppliers interact^{29}. This redundancy (nestedness) ensures the resiliency of SM networks, which otherwise will fail with the demise of a few suppliers. Apart from the nested property, network density plays a major role in ensuring the resiliency of a mutualistic system^{24}.
We validate the models proposed in Eqs (4) and (5) using the global automotive supply chain network data. There are 376 manufacturers and 5229 suppliers who supply approximately 300 components including critical and noncritical components. The critical components like piston and connecting rod are essential for the function of an automobile. The noncritical components like lamp wiper and exterior are required although their failures do not affect the primary function of an automobile. The data spans from 1999 to 2020 (future orders are also considered). We divide the automotive SM networks into 21 subnetworks according to AA classification scheme as given in Supplementary Table S1. Subnetworks are considered for two reasons: (a) automotive assemblies are highly modular and independent of each other (e.g., an engine subassembly has no dependence on a door panel for its production completion), and (b) to allow simulation runs to complete in a reasonable amount of time. We create two versions of automotive SM networks for analysis. In the first one, we consider all the manufacturers and suppliers across multiple years (longitudinal data). In such networks, manufacturers do not always get parts/products from the same suppliers every year. We form 21 longitudinal networks corresponding to each of the 21 subnetworks. The network properties (densisty and nestedness) for longitudinal networks are given in Supplementary Table S2. In the second version of networks, we consider two automotive SM subnetworks just for the year 2017 (cross sectional data). The network properties (densisty and nestedness) for cross sectional networks are given in Supplementary Table S3. Resiliency patterns of networks using cross sectional data show the pointintime (or current) resiliency of the automotive SM networks. The networks formed using longitudinal data are not useful to assess the true resiliency of the networks compared to the networks formed using the cross sectional data. However, the networks formed using longitudinal data serve as an ideal reference for evaluating resiliency of the network that can be obtained using networks built with the cross sectional data.
In tune with the previous studies^{27,28,29}, we observe that automotive SM networks generated using both longitudinal and cross sectional data exhibit high nestedness and low network density (see Supplementary Information Tables S2 and S3), striking similarity to the network properties exhibited by ecological networks. It is not unusual for automotive SM networks to exhibit such network properties. It is in the best interest of manufacturers to have many suppliers, while this is constrained by the supplier capabilities and production economics. A supplier may take many years to upgrade technology and skills to align with the manufacturer’s requirements. Even though that there are many capable suppliers, manufacturers will work only with optimum number of suppliers to allow suppliers to derive the benefit of economies of scale. In the event of low capacity utilization at the supplier end, the costs cannot be maintained competitive which hampers the survivability of the suppliers. Therefore a network with the structure of low density and high nestedness is created.
Effect of network structure
In order to reduce the time complexity of rewiring real world SM networks to understand the effects of nestedness, we use simulated data (having same nestedness and density properties of real world networks) to observe the structural effects on the resiliency of the SM network. We simulate synthetic SM networks with 10 manufacturers and 26 suppliers. The network density, \(\bar{a}\), of a bipartite network is computed as \(\bar{a}=\frac{m}{I\times J}\), where m is the total number of interactions for all the nodes in the SM network, I is the number of manufacturers and J is the number of suppliers^{49}. The value of density varies from 0 to 1, 0 being low density and 1 being high density. The nestedness is computed using the NTC method^{50}. The normalized value of nestedness varies form 0 to 1, 0 being low nestedness and 1 being high nestedness.
We remove f_{n} fraction of suppliers and observe the effects of the perturbation on the throughput of suppliers. Similarly, we remove f_{n} fraction of manufacturers and observe the effects of the perturbation on the throughput of manufacturers. We remove f_{l} fraction of suppliers and see the effect of perturbation on manufacturer throughput, and finally we remove f_{l} fraction of manufacturers and see the effect of perturbation on supplier throughput. In addition to local disturbances, we also introduce global perturbations by reducing f_{w} which is the fraction of interaction strength loss in Eqs (4) and (5). For all the perturbations, the effective throughput of the manufacturers and suppliers are computed using Eqs (4) and (5). We assume that in the event of a node or link loss the manufacturers do not redistribute their workload among the remaining suppliers and suppliers do not form a new link with remaining manufacturers. It is a realistic assumption considering that the last minute load distribution by manufacturers is impeded by the capacity constraints of the suppliers and, demand and product constraints prevent suppliers to form links with new manufacturers in a very short period of time.
Supplementary Figure S3 shows the result of the SM network resiliency under varying degree of nestedness with constant density. Two networks consisting of 10 manufacturers and 26 suppliers are generated with low (0.19) and high (0.95) nestedness properties. The density is maintained at a constant value of 0.30. Supplementary Figure S4 shows the result of the SM network resiliency under varying degree of density and constant nestedness. The simulation process remains the same as described above except that the two networks are generated with high density (0.68), and low density (0.30). The nestedness value is kept constant at 0.95. First, we observe that irrespective of the network structure, the system is more sensitive to global perturbations than the local perturbations. The network resiliency is sensitive to both nestedness and the network density. High values of nestedness and network density bestow a supply chain with better resiliency. This result is in agreement with the observations for mutualistic ecological networks, where the nested structure allows a resilient system that can accommodate node or link deletion with minimal performance loss^{51}. To better understand the relation between network density and nestdeness, a sensitivity analysis is conducted on a synthetic network (10 manufacturers, 26 suppliers) by varying degree of nestedness and network density. The results (see Supplementary Figures S5 to S7) confirm that both nestedness and density contribute to the network resiliency. When the network density is relatively high, the network resiliency is less sensitive to degree of nestedness. Similarly when the degree of nestedness is high, the network resiliency becomes robust to degree of density. Hence either high degree of nestedness or high network density can guarantee the resiliency of the system.
Robustness against noise and parameter variation
We perform numerical analyses using the multidimensional model (Eqs (1) and (2)) and dimensionreduced model (Eqs (4) and (5)) to examine the sensitivity of the dimensionreduced model to noise and parameter variations, i.e., we examine if the dimensionreduced model is able to predict the point of collapse (point where the SM network collapses unrecoverably) without large deviation. We use a real world SM network with 80 suppliers who supplied drive train parts to 75 manufacturers for the year 2017. The nestedness of the network is 0.9703 and the density is 0.0323. Figure 3 shows the throughput of manufacturers and the suppliers when subjected to local and global perturbations. The parameters for both the multidimensional model and the dimensionreduced model remain the same. The perturbations are introduced randomly and one hundred realizations of each perturbation scenario are performed. We observe that the dimensionreduced model effectively captures the resiliency pattern of the multidimensional model without a significant deviation. The average throughput of both the models converge almost at the same point of collapse.
We then add Gaussian noise of strength 0.1 to the multidimensional model and the dimensionreduced model. We observe that both the models are able to converge at the point of collapse (see Supplementary Figure S8). The observation also extends to scenarios with parameter variation, where a random variation (U\([0.8,1.2]\)) in value of K does not significantly alter the point of collapse for both models (see Supplementary Figure S9). We also introduce random variations in competition parameter β using uniform distribution U\([0.001,0.002]\) and internal reliability α using uniform distribution U\([1,1.2]\). We observe that the variations in α and β do not affect the ability of the dimensionreduced model to predict the point of collapse (see Supplementary Figures S10 and S11). Overall, the effective dimensionreduced model does not deviate from the multidimensional model for predicting the point of collapse and can be used to represent the dynamics of the SM network. For generalizability, we also test the effectiveness of dimensionreduced model using synthetic network. We generate a network of 10 manufacturers and 26 suppliers with a nestedness of 0.95 and a density of 0.19. The observations do not deviate from the earlier investigation done on real world networks and the effective dimensionreduced model does not deviate from the multidimensional model for predicting the point of collapse and can be used to represent the dynamics of the SM network. The results are shown in Supplementary Figures S12 to S15.
Predicting point of collapse in real world SM networks
We subject both longitudinal networks and cross sectional networks to local and global perturbations. The resiliency patterns for all the networks using longitudinal data are shown in Supplementary Figures S16 to S24. The resiliency patterns for all the networks using cross sectional data are shown in Supplementary Figures S25 and S26. As expected, we observe that the networks formed using longitudinal data exhibit higher resiliency than the networks formed using cross sectional data. One can see that each realization generates a different point of collapse when subjected to perturbations in f_{n}, f_{l}, or f_{w} direction. However the emergence of the point of collapse (the first occurrence of the point of collapse does not change with respect to random perturbations. The exact point of collapse can be captured in the structureparameter space. In \([{f}_{n}\alpha ]\) space and \([{f}_{w}\alpha ]\) space we are able to accurately predict the point of collapse (the stability analysis, derivation of the stable state, and computation of the point of collapse is explained in detail in Supplementary Information). So using the structureparameter space one is able to predict the point of collapse in a space that involves local (f_{n}) and global (f_{w}) perturbations that occur outside the manufacturing facility and internal perturbations that is reflected in the parameter α. This is important since the aforementioned perturbations are measurable in real life. Figures 4 and 5 show the traceability of point of collapse in the structureparameter space for networks created using longitudinal data and cross sectional data respectively. We observe that both internal perturbation and external perturbation deteriorate SM network resilience. Large internal perturbation shifts the point of collapse, where a small fraction of local perturbation (supplier or manufacturer removal) causes the system to collapse faster. Hence using the proposed approach we are able to reduce the multidimensions to manageable two dimensions that compute the effective throughput of manufacturers and suppliers in the SM network. In addition, we are able to accurately predict the point of collapse in a space that needs only the information about the underlying network structure and the internal reliability of the manufacturer and supplier facilities.
Discussion
In this work we present an approach to measure the resiliency of a mutualistic suppliermanufacturer (SM) network. The proposed approach helps create a nonlinear dynamical model by integrating the network structure and the network parameters. The proposed model incorporates growth, internal reliability, price competition, technology competition and mutualistic interaction between the suppliers and manufacturers. The simplification approach helps reduce the multidimensional model to a manageable twodimensional model, which is robust against noise and random parameter variations. Our results show that the resiliency of a SM network is a function of network structure and parameters. The resiliency is sensitive to structural properties of the network, namely, nestedness and density. Our model enables an accurate detection of emergence of the point of collapse as a measure of network resiliency which is highly predictable in the structureparameter space. We observe that the SM network collapses in an accelerated fashion with the proportion of node or link removal and the reliability degrades resulting in decreased resiliency. For realworld applications, we observe that the global automotive industry is more resilient to local perturbations than to global perturbations, mainly due to high nestedness resulting from network design principles of redundancy adopted by the automotive industry. Global supply chains are becoming more integrated via cyber and physical interconnections. The emergence of Internet of Things (IOT) enables manufacturers and suppliers to measure system parameters in real time, given that current machines are equipped with advanced sensors for monitoring machine health. In addition, advanced statistical and machine learning models enable accurate prediction of machine failures. As time progresses, manufacturers and suppliers are able to obtain an accurate estimate of their reliability. The same cyberphysical principles help manufacturers and suppliers to obtain real time information regarding the network structure throughout the entire supply chain. Hence the current work is suitable for real world applications for predicting the collapse of SM networks. The proposed model can be adapted to any mutualistic SM networks. The proposed model fills an important existing theoretical gap in understanding and analyzing the resiliency of supply chains. We provide a comprehensive analytical framework from which further supply chain resiliency metrics can be developed.
Recovery patterns are generally time related behavior indicating the time to regain the original performance post a disruptive event^{18}. The behavior of recovery patterns is influenced by factors such as speed of disruption onset, levels of disruptions and delays between intervention and performance response. As disruptions vary in severity and duration, so do the recovery patterns, which can be linear, concave, convex and nonspecific. Hence combining all factors into a single metric for investigating SCR is challenging. Although recovery patterns were not covered within the scope of the current work, we believe this work will motivate and form a foundation for future research on recovery patterns of supply chains.
Methods
Real world SM networks
We use proprietary global automotive supply chain data obtained from Marklines database. The dataset has 376 manufacturers and 5229 suppliers who supply approximately 300 components. The data spans from 1999 to 2020. In the first version the dataset is divided into 21 subnetworks according to AA classification scheme as given in Supplementary Table S1. The classification scheme describes 21 categories where each category has multiple automotive components associated with it. These 21 subnetworks considers all the manufacturers and suppliers in each category across multiple years (longitudinal data). In such networks, manufacturers do not always get parts/products from the same suppliers every year. The network properties (densisty and nestedness) of these 21 subnetworks are given in Supplementary Table S2. In the second version of networks, we consider two automotive SM subnetworks just for the year 2017 (cross sectional data). The network properties (densisty and nestedness) are given in Supplementary Table S3.
Data source
https://www.marklines.com/portal_top_en.html.
Computations
For testing the effectiveness of the dimension reduced model in presence of noise, Gaussian noise was added to the Eqs (1), (2), (3) and (4). The noise was added using the function awgn() in MATLAB computational software. For testing the effectiveness of the dimension reduced model in presence of random parameter variations, variations were introduced in form of uniform distribution using rand() function in MATLAB computational software. For all the synthetic networks and the real world SM networks, the network density, \(\bar{a}\), of a bipartite network is computed as \(\bar{a}=\frac{m}{I{\rm{\times }}J}\)^{49}, where m is the total number of interactions for all the nodes in the SM network; I is the number of manufacturers and J is the number of suppliers. The value of density varies from 0 to 1, 0 being low density and 1 being high density. The nestedness is computed using the NTC method^{50}. The normalized value of nestedness varies form 0 to 1, 0 being low nestedness and 1 being high nestedness.
Data Availability
The data for synthetic networks used in this work will be provided upon request.
Code Availability
All the codes are generated using MATLAB computational software. Code will be made available upon request.
References
Hendricks, K. B. & Singhal, V. R. The effect of supply chain glitches on shareholder wealth. Journal of operations Management 21, 501–522 (2003).
Hendricks, K. B. & Singhal, V. R. Association between supply chain glitches and operating performance. Management science 51, 695–711 (2005).
Wagner, S. M. & Bode, C. An empirical examination of supply chain performance along several dimensions of risk. Journal of business logistics 29, 307–325 (2008).
Narasimhan, R. & Talluri, S. Perspectives on risk management in supply chains (2009).
Ambulkar, S., Blackhurst, J. & Grawe, S. Firm’s resilience to supply chain disruptions: Scale development and empirical examination. Journal of operations management 33, 111–122 (2015).
Melnyk, S. A., Davis, E. W., Spekman, R. E. & Sandor, J. Outcomedriven supply chains. MIT Sloan Management Review 51, 33 (2010).
Jüttner, U. & Maklan, S. Supply chain resilience in the global financial crisis: an empirical study. Supply Chain Management: An International Journal 16, 246–259 (2011).
Wieland, A. & Marcus Wallenburg, C. The influence of relational competencies on supply chain resilience: a relational view. International Journal of Physical Distribution & Logistics Management 43, 300–320 (2013).
Olcott, G. & Oliver, N. Social capital, sensemaking, and recovery: Japanese companies and the 2011 earthquake. California Management Review 56, 5–22 (2014).
Peck, H. Drivers of supply chain vulnerability: an integrated framework. International journal of physical distribution & logistics management 35, 210–232 (2005).
Christopher, M. & Peck, H. Building the resilient supply chain. The international journal of logistics management 15, 1–14 (2004).
Ponomarov, S. Y. & Holcomb, M. C. Understanding the concept of supply chain resilience. The international journal of logistics management 20, 124–143 (2009).
Briano, E., Caballini, C. & Revetria, R. Literature review about supply chain vulnerability and resiliency. In Proceedings of the 8th WSEAS international conference on System science and simulation in engineering, 191–197 (World Scientific and Engineering Academy and Society (WSEAS) Genova, Italy, 2009).
Smith, R. Operational capabilities for the resilient supply chain. Supply Chain Practice 6, 24–35 (2004).
Gu, X., Jin, X., Ni, J. & Koren, Y. Manufacturing system design for resilience. Procedia CIRP 36, 135–140 (2015).
Pettit, T. J. Supply chain resilience: development of a conceptual framework, an assessment tool and an implementation process. Tech. Rep., Ohio State Univ Columbus (2008).
Vugrin, E. D., Warren, D. E. & Ehlen, M. A. A resilience assessment framework for infrastructure and economic systems: Quantitative and qualitative resilience analysis of petrochemical supply chains to a hurricane. Process Safety Progress 30, 280–290 (2011).
Munoz, A. & Dunbar, M. On the quantification of operational supply chain resilience. International journal of production research 53, 6736–6751 (2015).
Albert, R., Jeong, H. & Barabási, A.L. Error and attack tolerance of complex networks. Nature 406, 378 (2000).
Iyer, S., Killingback, T., Sundaram, B. & Wang, Z. Attack robustness and centrality of complex networks. PloS One 8, e59613 (2013).
Chen, Z., Wu, J., Xia, Y. & Zhang, X. Robustness of interdependent power grids and communication networks: A complex network perspective. IEEE Transactions on Circuits and Systems II: Express Briefs 65, 115–119 (2017).
Wu, J., You, W., Wu, T. & Xia, Y. Abnormal phenomenon in robustness of complex networks with heterogeneous node functions. Physica A: Statistical Mechanics and its Applications 506, 451–461 (2018).
Surana, A., Kumara, S., Greaves, M. & Raghavan, U. N. Supplychain networks: a complex adaptive systems perspective. International Journal of Production Research 43, 4235–4265 (2005).
Gao, J., Barzel, B. & Barabási, A.L. Universal resilience patterns in complex networks. Nature 530, 307 (2016).
Jiang, J. et al. Predicting tipping points in mutualistic networks through dimension reduction. Proceedings of the National Academy of Sciences 115, E639–E647 (2018).
Rohr, R. P., Saavedra, S. & Bascompte, J. On the structural stability of mutualistic systems. Science 345, 1253497 (2014).
Saavedra, S., ReedTsochas, F. & Uzzi, B. A simple model of bipartite cooperation for ecological and organizational networks. Nature 457, 463 (2009).
Brintrup, A., Kito, T., Alzayed, A. & Meyer, M. Nested patterns in largescale automotive supply networks. Capturing Value Int. Manuf. Supply Networks, Institute for Manufacturing (2012).
Brintrup, A., Barros, J. & Tiwari, A. The nested structure of emergent supply networks. IEEE Systems Journal 1–10 (2015).
Leitch, R. A. Effect of stochasticity, capacity and lead time cost drivers on wip and throughput in a pull production environment. Management Accounting Research 12, 167–196 (2001).
Banker, R. D., Datar, S. M. & Kekre, S. Relevant costs, congestion and stochasticity in production environments. Journal of Accounting and Economics 10, 171–197 (1988).
Hopp, W. J. & Spearman, M. L. Factory physics (Waveland Press, 2011).
Clark, A. J. et al. Production forecasting with logistic growth models. In SPE annual technical conference and exhibition (Society of Petroleum Engineers, 2011).
Höök, M., Li, J., Oba, N. & Snowden, S. Descriptive and predictive growth curves in energy system analysis. Natural Resources Research 20, 103–116 (2011).
McCarthy, M. The allee effect, finding mates and theoretical models. Ecological Modelling 103, 99–102 (1997).
Holland, J. N., DeAngelis, D. L. & Bronstein, J. L. Population dynamics and mutualism: functional responses of benefits and costs. The American Naturalist 159, 231–244 (2002).
Marchetti, C. The automobile in a system context: The past 80 years and the next 20 years. Technological Forecasting and Social Change 23 (1983).
Modis, T. & Debecker, A. Chaoslike states can be expected before and after logistic growth. Technological forecasting and social change 41, 111–120 (1992).
Jacobs, D. & Meerkov, S. M. A systemtheoretic property of serial production lines: improvability. International Journal of Systems Science 26, 755–785 (1995).
Gershwin, S. B. & Gershwin, S. Manufacturing systems engineering. (1994).
Chiang, S.Y., Kuo, C.T. & Meerkov, S. M. cbottlenecks in serial production lines: identification and application. Mathematical problems in engineering 7, 543–578 (2001).
Li, J., Blumenfeld, D. E. & Alden, J. M. Comparisons of twomachine line models in throughput analysis. International journal of production research 44, 1375–1398 (2006).
Uzzi, B. The sources and consequences of embeddedness for the economic performance of organizations: The network effect. American sociological review 674–698 (1996).
Uzzi, B. Social structure and competition in interfirm networks: The paradox of embeddedness. Administrative science quarterly 35–67 (1997).
Walker, G. & Weber, D. Supplier competition, uncertainty, and makeorbuy decisions. Academy of Management journal 30, 589–596 (1987).
Ohtsuki, H., Hauert, C., Lieberman, E. & Nowak, M. A. A simple rule for the evolution of cooperation on graphs and social networks. Nature 441, 502 (2006).
Lever, J. J., van Nes, E. H., Scheffer, M. & Bascompte, J. The sudden collapse of pollinator communities. Ecology letters 17, 350–359 (2014).
Staniczenko, P. P., Kopp, J. C. & Allesina, S. The ghost of nestedness in ecological networks. Nature communications 4, 1391 (2013).
Dormann, C. F., Fründ, J., Blüthgen, N. & Gruber, B. Indices, graphs and null models: analyzing bipartite ecological networks. The Open Ecology Journal 2 (2009).
Atmar, W. & Patterson, B. D. The measure of order and disorder in the distribution of species in fragmented habitat. Oecologia 96, 373–382 (1993).
Bastolla, U. et al. The architecture of mutualistic networks minimizes competition and increases biodiversity. Nature 458, 1018 (2009).
Author information
Authors and Affiliations
Contributions
All authors designed and performed the research. M.X. performed analytical and numerical calculations, S.R. analyzed the empirical data. S.R. wrote the manuscript. M.X., X.J. and S.R. reviewed and revised the manuscript.
Corresponding author
Ethics declarations
Competing Interests
The authors declare no competing interests.
Additional information
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Supplementary information
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Xu, M., Radhakrishnan, S., Kamarthi, S. et al. Resiliency of Mutualistic SupplierManufacturer Networks. Sci Rep 9, 13559 (2019). https://doi.org/10.1038/s41598019499321
Received:
Accepted:
Published:
DOI: https://doi.org/10.1038/s41598019499321
This article is cited by

Equality of access and resilience in urban populationfacility networks
npj Urban Sustainability (2022)
Comments
By submitting a comment you agree to abide by our Terms and Community Guidelines. If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.