A system dynamics simulation model of chemical supply chain transportation risk management systems

Accepted Manuscript Title: A system dynamics simulation model of chemical supply chain transportation risk management systems Author: Chaoyu Li Jun Ren Haiyan Wang PII: DOI: Reference:

S0098-1354(16)30056-4 http://dx.doi.org/doi:10.1016/j.compchemeng.2016.02.019 CACE 5387

To appear in:

Computers and Chemical Engineering

Received date: Revised date: Accepted date:

11-2-2015 11-2-2016 29-2-2016

Please cite this article as: Li, C., Ren, J., and Wang, H.,A system dynamics simulation model of chemical supply chain transportation risk management systems, Computers and Chemical Engineering (2016), http://dx.doi.org/10.1016/j.compchemeng.2016.02.019 This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

A system dynamics simulation model of chemical supply chain transportation risk management systems

1

Liverpool Logistics, Offshore and Marine (LOOM) Research Institute, Liverpool John Moores

cr

University, Liverpool, L3 3AF, UK

School of Transport, Wuhan University of Technology, 1040 Heping Ave., Wuhan, 430063, China a

[email protected], [email protected], [email protected]

us

2

ip t

Chaoyu Li 1, a, Jun Ren 1, b and Haiyan Wang 2, c

an

Abstract

Unforeseen events can interrupt the operational process and have a negative impact on the chemical

M

supply chain transportation (CSCT) system. A number of research studies have addressed the risk management issues in chemical supply chain (CSC) or CSCT. However, most of the existing

d

methodologies lack inbuilt and practical techniques that take into consideration the complex

Ac ce pt e

interactions and dynamic feedback effects, which can significantly affect the reliability of risk management outcomes. This paper suggests a novel modelling and simulation method to address the dynamic risks effects in the CSCT, especially the consideration of time-dependent system behaviour in different operational conditions. Furthermore, the flexibility of the model modification is adapted to enhance the practice in risk mitigation. A transparent decision support tool is provided to compare the outcomes of different risk mitigation processes, which offers decision makers an alternative CSCRM mitigation package.

Key words: Chemical supply chain transportation, Risk management, System dynamics, Modelling and simulation Chemical supply chain transportation: CSCT Chemical supply chain: CSC Chemical supply chain risk management: CSCRM System dynamics: SD

Page 1 of 41

1. Introduction The chemical supply chain (CSC) involves the sourcing, conversion, transportation and warehousing of raw materials into final chemical products and their delivery to customers across national boundaries (Tsiakis and Papageorgiou, 2008). Due to the geographic dispersion of the members of the

ip t

supply chain, CSC requires highly coordinated material, information and finance flows, with the conveyance of hazardous substances between the members of the CSC – chemical supply chain

cr

transportation (CSCT) - regarded as one of the most significant operational processes in terms of risk (Reiskin et al., 1999). Hazardous characteristics of chemical substances, uncertainties and disruptions

us

pose significant challenges to CSCT operations as well as the surrounding environment, which potentially threatens ecological balance and endangers human health (Bonvicini et al., 1998;

an

Papageorgiou, 2009). In response, the supply chain members have to implement a large variety of methods to manage their supply chains in order to maintain the effectiveness and efficiency of supply

M

chain operations (Thun and Hoenig, 2009). In addition, governments and authorities have introduced a substantial body of legislation, regulatory guidance and recommendations to ensure the safety of

d

CSCT operations (Furuhama et al. 2011; Fisk, 2014; Scruggs et al., 2014). Both academia and CSC

Ac ce pt e

operators appreciate the need to improve safety and reliability of CSCT chains, to prepare for, respond to and recover from risk scenarios, especially after the 9/11 attacks in 2001 (Mullai, 2009) and the frequent natural disasters (Rao and Goldsby, 2009; Ehlen et al., 2014). In spite of the challenges arising from the internal system and external environment, the CSC is required to deliver a competitive business performance (Manuj and Mentzer, 2008). To manage undesired events, a large number of studies have been devoted to extending current knowledge and enhancing the application of CSCRM. It is important to credit the previous publications that have developed various conceptual or analytical models to investigate different kinds of risks in the CSC, as these have provided the dynamic analysis that underpin this proposed modelling and simulation research. In particular, García-Flores and Wang (2002) and Gao et al. (2009) developed agent-based models to simulate the dynamic behaviour of the CSCs and estimate the compromised risk management decisions. Taking into consideration the characteristics of chemical products, a dual 2

Page 2 of 41

model was built by Liu et al. (2011) to analyse both internal and external transportation risks in a CSC. Kleindorfer and Saad (2005) have demonstrated a conceptual model to assess and mitigate CSC risks based on the accident data of the U.S. chemical industry from 1995 to 2000. Adhitya and Srinivasan (2010) developed a dynamic model to simulate CSC operations under the changes of customer behaviour, business policies, and environmental issues. Laínez and Puigjaner (2012) discuss the state-

ip t

of-the-art of risk management strategies that attempt to broadly outline the risk perspectives in the

cr

CSC.

However, it is interesting to observe that many studies in CSCRM are carried out to analyse a specific

us

kind of risk, instead of offering a holistic risk management framework. For instance, Ferrio and Wassick (2007), You et al. (2009), Tong et al. (2011), Oliveira et al. (2013) and Cai (2014) provide

an

stochastic programming approaches to investigate CSC planning problems with risk consideration. Laínez et al. (2009), Carneiro et al. (2010), Oliveira and Hamacher (2012), Oliveira et al. (2013), and

M

Ruiz-Femenia et al. (2013) investigate the financial problems in the CSC and provide retrofit actions to control and optimise investment decisions. However, further analysis is required to provide

d

advantageous risk management techniques under a broader context to assess a more exhaustive

Ac ce pt e

variety of extreme and disruptive events in CSCT.

In CSCRM research, it is difficult to understand clearly the complex CSC structure, operating procedures and related aspects from the available quantitative data. Experts use their experiences to judge the risk consequences and to illustrate CSC operations (Tse, 2013). However, the majority of existing risk analysis methods are restricted by using a combination of qualitative and quantitative data. A novel method is required to conduct risk analysis and risk mitigation based on multiple data sources, such as numerical data, expert judgement, and interviews (Kaggwa, 2008). Besides, the developed CSC and CSCT systems are presented with the nature of static models and established upon the sequence event chains (Leveson, 2004). It ignores that the information feedback among the logical loops emerging from the interactive relationships governs the changes of system behaviour, which should be taken into consideration during risk modelling and simulation. Meanwhile, simple algebraic equations are frequently adapted to predict the cause and effect relationships between 3

Page 3 of 41

the supply chain components, but nonlinear relationships exist as the norm rather than the exception. It is imperative to develop a methodology that can capture and represent both linear and nonlinear relationships among the system in order to address the dynamic risk impacts in the complex CSCT system. This study, therefore, provides a novel risk management method employing limited qualitative and

ip t

quantitative data/information to manage a more exhaustive variety of risks. It offers a methodological

cr

approach to deal with the existing causal relations and feedback effects between the CSCT system and its associated hazardous events. Instead of assessing the risks based on expert knowledge or historical

us

data, the risk effects are addressed through combining the modelling approaches for the quantification of the system performance with interactive risk analysis procedures. Taking the advantages of the

an

flexibility of model modification, the outcomes of risk mitigation methods are estimated to ensure that the particular risk mitigation approaches indeed support CSCRM. Furthermore, the proposed method

M

can be generalised to provide a flexible and rigorous RM tool for other industries. The proposed research addresses the risks existing in the transportation stage of the CSC, and predicts

d

the outcomes of alternative risk mitigation decisions using a novel SD-based CSCRM method. It does

Ac ce pt e

this by: (1) developing the CSCT risk models based on the cause and effect relationships within the system boundaries; (2) running created models under different scenarios to explore the system performance; (3) benchmarking the series of system behaviour in the initial situation and in the risk scenarios to screen out the critical hazards; (4) offering a flexible model modification function to measure the outcomes of alternative risk mitigation solutions; and (5) providing a novel approach to evaluate risk management decisions that might improve CSCT system performance. This paper is organised as follows: Section 2 describes the sequential development of the SD model, i.e. by constructing a causal loop diagram, by establishing a stock and flow diagram using Vensim© software, and by model validation. A number of risk experts and analysts were surveyed to generate different sets of input values to estimate risk consequence. In this way, the risks are assessed by benchmarking the comparisons of the system performance in different risk scenarios. Section 3 outlines the simulation results obtained from each of the risk scenario and corresponding risk 4

Page 4 of 41

mitigation scenario, and Section 4 summarises the key findings and their relevance to academia and the chemical industry, together with suggestions for future research.

2. Modelling approach 2.1 Characteristics of the SD method

ip t

Forrester (1961) first proposed SD theory to predict the behaviour of dynamic systems and analyse the

cr

efficacy of decision-making by modelling and simulation. In the literature, SD method has been widely applied to analyse the industrial risks (Garbolino et al., 2009; Oehmen et al., 2010; Garbolino

us

et al., 2010; Bouloiz et al., 2013). In the supply chain management discipline, SD modelling technology has been introduced to deal with inventory management, bullwhip effect, strategy

an

assessment and information delays (Ge et al. 2004; Janamanchi and Burns, 2007; Campuzano et al., 2010; Peng et al., 2014). Although there are alternative risk management tools that can be employed

M

to investigate supply chain risks, SD method is implemented to deal with the CSC risks in this research because of its potential as an analytical method to address dynamic system behaviour

d

governed by the information feedback (Kumar and Yamaoka, 2007; Tako and Robinson, 2012).

Ac ce pt e

In the theory, SD is a broad concept that can be divided into two aspects: ‘system’ represents the structure of the system and the concept of feedback effect, while ‘dynamics’ reflects the changes in the behaviour of the system components over time. In a developed causal loop diagram, the assumed interactions between the variables are formalised to demonstrate the interdependence within the system boundaries. A closed chain of causal relations is defined as the feedback loop, which could be positive or negative (Lertpattarapong, 2002). The positive loop is unstable and oscillated that triggers systems to grow, evolve and collapse, while the change of negative loop towards a stable situation. SD is concerned with the qualitative and quantitative analysis of the dynamic performance in largescale systems, both retrospectively and prospectively. The application of SD method to CSCRM not only simulates the CSCT operations, but also predicts dynamic behaviour as the system changes under different risk circumstances (Angerhofer and Angelides, 2000). It takes account of the logical interaction of the components in the supply chain system and observes the causal effects between the 5

Page 5 of 41

risk factors and the system behaviour when time is factored into the sequence. Changes and new situations can be adapted in developed SD models to explore the dynamic outcomes in different scenarios. SD methodology comprises a set of rigorous procedures to describe the supply chain structure and its behaviour in terms of process, information, decision-making and organizational limits. Yeo et al.

ip t

(2013) suggest that the SD modelling process can be characterised by three phases: logical modelling,

cr

model quantification and model application. Fig. 1 describes a novel SD-based CSCRM framework, which is developed from the SD modelling process described in Yeo et al.

us



an

The first step in SD modelling process is problem definition, which indicates the purpose of the study and specifies the system boundaries. Secondly, the key variables are identified and their behaviours

M

are address to illustrate the interactions within the system boundaries. Due to the potential for disruptions and uncertainties in the supply chain, the materials, information and monetary flows can

d

be affected by the occurrence of hazardous events. It is necessary to address the cause and effect

Ac ce pt e

relationships between the developed system and unexpected events. Step 3 formalises these interactions to develop a causal loop diagram (Lertpattarapong, 2002). In step 4, the causal loop diagram is translated into a stock and flow diagram using computer language. It follows the procedures of characterising elements, writing equations, assigning values to parameters, and building the model (Campzano and Mula, 2011). The developed SD model is tested before application in Step 5. By running the verified SD model, the system performance can be obtained with a set of risk input values in Step 6. Benchmarking the system performance in different scenarios allows the risks to be quantitatively assessed and identify the unacceptable risks in Step 7. Incorporating the ability to modify the formalised model in both the design and operational phases provides a method of measuring the effects of potential risk mitigation methods and suggests beneficial risk mitigation decisions in Step 8. 2.2. Defining the causal relations between factors in risk affected CSCT systems 6

Page 6 of 41

2.2.1 The impact of hazardous factors In supply chain risk management, risk is a significant factor in the exploration of the dynamic effects caused by undesired events. Ren et al. (2009), Mokhtari et al. (2012), Rausand (2013) investigate the risk in three dimensions: the probability of a risk event occurring, consequence probability and consequence severity. The probability of a risk event occurring describes the frequency of the

ip t

occurrence within a finite time period. Consequence probability indicates the probability of the

cr

consequence given the occurrence of a risk event. Consequence severity refers to the magnitude of possible negative consequences when a risk event does occur, which is commonly measured in a

us

specific aspect based on expert judgement, actuarial calculations or historical data. However, the translation of different kinds of risk consequences into a specific aspect is a challenging and

an

vulnerable element in risk management (Waters, 2011). As shown in Fig. 2, four dimensions of a risk are explored in this study: the probability of a hazardous event occurring; the probability of a given

M

variable being affected by the occurrence of a hazardous event; consequence probability for a given variable; and consequence severity for a given variable. In particular, the severity of consequence is

d

analysed by variable. This not only avoids translating different consequences into one attribute but

Ac ce pt e

also improves the accuracy of the risk evaluation results.

In a developed causal loop diagram, the arrow indicates the direction of a cause and effect relationship where the causative variable is represented by the origin of the arrow and the effected variable is on the point of the arrow. Furthermore, “+” and “-” signs on the arrows are used to describe the effects between the variables, which can be positive and negative. In positive mode, the changes for both variables share the same tendency. For instance, a reinforcing effect between the consequence severity and variable damage rate, so that an increase of causative variable will contribute to the improvement of the effected variable. In contrast, the negative effect presents a reduction effect. For instance, an increase in the variable recovery rate could lead to a reduction in the variable damage stack.

7

Page 7 of 41

2.2.2. The dynamic inventory system In the CSC, large volumes of chemical substances are transported across the regional boundaries in response to periodic ordering (Reiskin et al., 1999). Special containers are used to store the chemical materials, but the features of immiscibility and incompatibility dictate that the containers cannot be mixed during transportation and storage (Erera et al., 2005). Unexpected events could interrupt the

ip t

transportation process and result a decrease in service level. In particular, the inventory system could

cr

be significantly affected due to the feedback effects among the logical loops emerging from the interactive relations (Manuj and Mentzer, 2008). To manage the inventory and improve the utilisation

us

of the storage capacity, the SD technique that incorporates time as a variable can assist with this coordination task. Fig. 3 represents the causal loop diagram of a CSCT inventory system.

an



M

There is a feedback loop found in the diagram, which governs the changes in the inventory system. The developed model appears to be stable, and dominated by a negative loop (containing three of the

d

negative relationships). Any actions that attempt to change the variables result in a self-correction of

Ac ce pt e

the system. The transporter inventory level is calculated by the arrival flow of products from the suppliers and the outflow of shipped products to the customers within a given period of time. If the shipped orders fail to keep pace with the customer demand, the backlog of orders will appear and a reduction in the order fulfilment rate can be observed during the simulation. The arrow with the symbol “||” is used to represent the delay in order processing, and a time lag between the interactive variables.

2.2.3. The dynamic transportation capacity

The capacity of transportation systems is particularly vulnerable in the event of a natural disaster, terrorism or other disturbances (Peng et al., 2014). The factors that determine CSCT capacity are described graphically in a causal loop diagram, shown in Fig. 4.

8

Page 8 of 41

According to the choice of transport features, such as mode of transport, type of container and route of transport, the capacity of a transportation system is based on the combination of infrastructure capacity and transporter capacity (Chen and Kasikitwiwat, 2011). Infrastructure capacity is determined by the selected route and environmental factors, which cannot be changed by transporters (Chen et al., 2002). In contrast, the capacity of a transporter can be controlled and managed by itself,

ip t

which depends on the capacity of the equipment and the size of the labour force within the system. In

cr

the developed diagram, two distinct loops can be observed that represent the feedback effects related to equipment capacity and labour force, respectively. Equipment capacity represents the ability of a

us

transporter to respond to dynamic orders. The labour force comprises the specified number of operators required to handle the existing equipment.

an

2.2.4. Dynamic transportation time

Hazardous events can interrupt the flow of CSCT operations and result in significant disturbance to

M

the CSCT system. It is crucial to understand how the dynamic variables within the model evolve in response to time delays. In transportation science, transport time is influenced by the volume of

d

products in shipment and the capacity of the available infrastructure. Transportation time can be

Ac ce pt e

calculated by (Bureau of Public Road (USA)):

V T = T0 1 + α C

β

(1)

where T0 represents the travel time when there is zero traffic flow on the road, C is set as the capacity of the infrastructure, V is the current volume of the products being transported, and α and β are two variable parameters (Schreckenberg et al., 2005). However, it is suggested that there will be an overexaggeration when the ratio of V/C is larger than 1.2, so that the function utilisation is obstructed in conditions when infrastructure capacity decreases sharply. To fill this gap, a segment function is provided to estimate transportation time in the post-seismic supply chain (Peng et al., 2014). The authors have tailored this segment function to ensure that it can account for the transportation time in a risk affected CSCT system, which is represented as:

9

Page 9 of 41

Tblock , Ct = 0 T=

(1 + ∂)T0 ,

(1 + ∂ )T0

Vt ≤1 Ct

(2)

Vt Vt , >1 Ct Ct

ip t

where T represents an operator-estimated transportation time. T0 is described as the initial transportation time. Vt is the current volume of products in transit and Ct is the current infrastructure

cr

capacity. Ct = 0 represents that the transportation route is blocked, while Tblock is the length of

us

blockage time. ∂ describes the increased transportation time associated with a hazardous event. Transportation time under different conditions can be estimated. As shown in Fig. 4, the dashed line

an

between the transportation capacity and infrastructure capacity represents the indirect interaction between the variables. Incorporating the temporal basis for representing the evolution of the dynamic

M

equations, the transportation time is represented to dynamically simulate the transport operations. For instance, an earthquake could affect the infrastructure capacity of the CSCT system, which would

Ac ce pt e

2.3 SD Model building

d

result in an increase in transportation time in the system.

The developed causal loop diagram is then converted to a stock and flow diagram using Vensim© (commercial software), which demonstrates the causal relations using stocks, flows and control variables with more specific quantitative information. A CSCT system connects the material flows between the CSC members and distributes the chemical substances from upstream suppliers to downstream customers. In traditional CSC, the supplier responds to downstream requirements by providing the requisite materials to the transporter, in the anticipation that the transporter has sufficient available capacity. A stock and flow diagram of the risk affected CSCT system is presented in Fig. 5.

10

Page 10 of 41

The variable in the rectangle is a stock, which is regarded as the structural element in the built model. In the diagram, the variables of inventory level, the quantity of labour and equipment, backlogged orders and variable values reduced are created as stocks to describe the accumulation of a material, information, or financial behaviour over time. A flow only passes the information that governs the change of stock. The developed CSCT model is a system that allows for the occurrence of a major

ip t

incident, which disrupts the supply chain operations and impairs system performance. Control is used

cr

to describe the hazardous events that govern the changes of CSCT model. Using simulation techniques, the system behaviour in different scenarios is obtained, which provides quantitative

us

results to investigate the risk effects in the system level. Table 1 illustrates the definitions of the

an

major variables used to build SD model.

2.4. Initial operating conditions

M



d

A SD simulation begins with running the developed model under a specified scenario, so that the

Ac ce pt e

initial value of each variable (such as: simulation period; downstream order; transporter capacity; and infrastructure capacity) must be defined at the outset. In this research, the simulation period was set as 365 days, and the time-step for simulation was set as 1 day. A number of assumptions were made in the definition of the established scenario. In reality, customer demand is uncertain and difficult to forecast accurately (Barilas and Gunduz, 2011). Therefore, the downstream order was assumed to be placed every day and followed a normal distribution with a minimum of 50 units, and a maximum of 100 units, with a mean of 85 units, and a standard deviation of 20. The volume of products in transit was determined by the capacity of the transporter and infrastructure. In view of the regulations and policies governing the transportation operations, it was assumed that the volume of products in shipment should not exceed infrastructure capacity in the proposed model. The transporter capacity was set at 400 units in total, and the infrastructure capacity was set at 150 units/day. The CSCT system performance under these baseline conditions is shown in Fig. 6. The downstream order was generated as a probabilistic input based on set policy, which fluctuated between 50 units per 11

Page 11 of 41

day and 100 units per day. Following the receipt of customer orders, the upstream supplier provided the required volume of products to the transporter on time. The inventory comprised the balance of the volume received and volume shipped by the transporter, which fluctuated between 50 units and 110 units. The simulation produced some late deliveries due to a lag in transportation capacity. In this circumstance, the order fulfilment rate was estimated to rise and fall between 0.78 and 1.00 during the

ip t

simulation period. This initial system performance was set as the baseline for benchmarking a series

cr

of system performance involving a variety of risk scenarios.

us



an

2.5. Model validation

The developed SD model should be tested before carrying out experiments to simulate system

M

operations (Qudrat-Ullah and Seong, 2010). The validation is focused on the verification of the correspondence of the model structure and the robustness of the model behaviour. Forrester and Senge

d

(1980) suggest three validation tests - of the structure and parameters; under extreme conditions; and

Ac ce pt e

of the dimensional consistency of SD models.

In addition to the tests, the structure of the proposed model was tested by comparing the variables and the equations against existing literature and available expert knowledge. It was claimed that the model was developed based on the causal relations, thus, the model structure and the contained interactions should be examined against the real system (Barlas, 1996). “Statistical significance” testing is another critical part in the SD model validation process. Regardless of the size of the model, all the variables of concern to the system developers could be tested to address whether the model adequately represented the real system at the operational level. The parameter values under extreme conditions were set by the authors in order to assess whether the performance of the model coincided with the anticipated behaviour of the system in reality. The following principle was applied: “If input A affects the system, then behaviour B should result” (Peterson and Eberlein 1994). Based on the assumption of an independent input value of a variable, it elicited a better performance of the system compared with human beings. However, it was significant 12

Page 12 of 41

to note that the emphasis was on the trend, frequencies and fluctuation prediction, rather than the value of system behaviour prediction (Das and Dutta, 2013). In the examination, a logical result was obtained to verify the developed system. To demonstrate the proposed method, an illustration is provided to describe the SD model validation under extreme conditions. In the CSCT system, the transportation capacity is a significant variable,

ip t

which reflects the ability of the transportation system to ship the orders to the customers. The value of

cr

transportation capacity depends on the combination of the current transporter capacity and infrastructure capacity. If the average number of placed orders is larger than the available

us

transportation capacity, the backlogged orders are going to accumulate to a high level following each simulation step; otherwise the backlogs do not appear. In order to verify this phenomenon, an increase

an

of average downstream orders was set to explore how the built system responded to the unexpected changes. The simulation result of a 5% increase in “Downstream order” is shown in Fig. 7.

M



d

The red line indicates the initial system performance, and the blue line presents the testing of a 5%

Ac ce pt e

increase in downstream orders. As shown, the unexpected customer demand increase puts huge pressure on the transportation system to fulfil the increased requirements. The backlogs frequently appear when the shipped orders fail to maintain the pace with customer demand. It is forecasted to fluctuate between 58.13 units and 104.73 units. The order fulfilment rate represents the performance of the CSCT system, which drops from 82.09% to 75.92%. According to the simulation results, the developed model presents a representation that coincides with logical behaviour in the scenario. In accordance with the system design, the developed CSCT model has spare capacity to gradually adapt to the negative effects of demand increase. Though it is regarded as a kind of waste in normal situation, it provides the backup capacity to deal with the unexpected requirements in risk scenarios. During the “statistical significance” testing, it is significant to identify the extreme value at which the developed system could absorb the negative effects and perform as initially expressed. The testing was carried out using Vensim© software to address experimentally this threshold value in the 13

Page 13 of 41

proposed scenario. It was shown that if the test input of downstream orders increased by 2.54%, the performance of the created system would be affected during the simulation period. Finally, the dimensional consistency tests were carried out to examine the dimensions of provided equations. Based on the causal relations among the system, the dimensions of variables on both sides of the equation are presented through logically calculating the defined mathematical equations. In

ip t

particular, the principle of dimensional homogeneity determines that only commensurable variables

cr

may be compared or calculated. For instance, transporter inventory level was calculated from the arrival flow of products from suppliers (Units/Day) and outflow of shipped products to customers

us

(Units/Day) within a given period of time. Hence, the dimensional unit of transporter inventory level should be defined as “Units”. In some contexts, there are dimensionless variables, such as percentages,

an

service level and risk input, which are expressed as “Dmnl”. This kind of variable does not affect the calculation of dimensional units in the equation.

M

In the research, the software used provides a powerful function of dimension calculation that automatically verifies the dimensional consistency of this model. It verifies the relationships between

d

interacted variables by tracking their fundamental dimension as performed calculations. Once

Ac ce pt e

validation and confidence in the behaviour of the developed SD model had been established, it could be used to address the system performance in a series of risk scenarios and risk mitigation scenarios.

3. Simulation and results 3.1. Risk data collection

Operational risks refer to the specialised internal features of CSCT that may cause transportation delay or damage. Although both academics and practitioners have raised the awareness of CSCRM, the insight from risk issues linked to the transportation process is limited, emerging from increasing challenges in today’s already volatile environment. Transportation activities can be disrupted by physical damage, which not only affects service levels, but also results in cost increases within the CSCT system (Wilson, 2007; Liu et al., 2011). Tatano and Tsuchiya (2008) provide a framework to estimate the economic losses accruing from transportation interruption. Leonelli et al. (2000) and 14

Page 14 of 41

Fabiano et al. (2005) have investigated the CSCT risks relating to available infrastructure capacity, available vehicle capacity, amount (quantity) and type (quality) of damage and transportation time. For this study, thirteen major sources of risks inherent in the CSCT operations are identified from existing literature, which are breakdown in core operations; inappropriate choice of service provider; inappropriate choice of transportation route; inadequate transportation capacity; high levels of

ip t

process variation; the complexity of the products to be transported; lack of/inappropriate inventory

cr

management; lack of/inappropriate container management; lack of qualified labour; the challenge of technological innovation; information sharing delays; information sharing inaccuracies; and

us

financial problems.

Expert intervention was applied to generate risk scenario input values and to suggest risk mitigation

an

methods in the methodology. The data from risk experts and analysts was collected by questionnaire and the responses informed a set of corresponding data - occurrence likelihood; probability of the

M

variable affected by the hazardous occurrence; consequence severity on the variable; and consequence probability on the variable. The data obtained comprised the input values in the proposed SD model,

Ac ce pt e

3.2. Risk scenarios simulation

d

to simulate the system performance under different risk scenarios.

It has been indicated that SD is a scenario-based method and can be used to investigate the impact of parameter changes on system behaviour over time (Yeo et al, 2013). The disturbance is amplified or self-corrected along the existing information feedback loops in the developed system, thus the system behaviour appears to be dynamic. Through comparison with a direct expert judgement, it provides a method of quantitatively estimating the problematic performance as the consequence of the system changes in response to different risk scenarios. Thus, it helps analysts to understand how the CSCT system will perform in different risk scenarios, and estimate the possible risk effects associated with these scenarios on system thinking. In the analysis, thirteen risk scenarios were established and experts were asked to assign input values to each risk scenario regarding the probability and consequence severity. Inserting the obtained data into the developed model simulates time-dependent CSCT system performance. Table 2 represents 15

Page 15 of 41

the system performance with the maximum, the minimum and the average values in the risk scenarios during the simulation period.

The software generated the hazardous events and their consequence severity following the specific

ip t

distributions, according to the particular features of the risk. In the developed SD model, a level

cr

variable was built to represent the accumulation of the damaged value to the affected variable. Due to inherent causal relations, the undesired consequence passed into the CSCT system and damaged

us

operational performance. To describe how the system performs under the developed risk scenario, an illustrative example is shown in Fig. 8, which demonstrates the risk generation mechanism in the

an

“Breakdown in core operations” scenario.

M



Hazardous events can interrupt the operational process and have a negative impact on the CSC system

d

in terms of time delays, financial or reputational losses. Using SD modelling method, the risk impacts

Ac ce pt e

were explored taking a number of steps along the time axis. The broken line represented the variable behaviour based on the simulation step. It indicated that the given risk consequence influenced the created system on five occasions, causing 28.9% damage to the affected variable each time. The changes of the variables resulted in the fluctuation of system performance, which were used to explore the risk effects in the system level.

In order to provide meaningful insights, Table 3 extracts the results from Table 2 that depicts the comparison of the system performance within each risk scenario with the initial value. In the risk scenario of “Breakdown in core operations”, it was estimated that there was a 1.90% decrease in the average transportation capacity and a sharp increase of transportation time due to the occurrence of the hazardous events. In previous research, Barlas and Gunduz (2011), Liu and Papageorgiou (2013) suggest that the changes to the capacity and an increase of lead-time will cause oscillations in inventory level along the supply chain. The SD simulation results confirmed these effects, with the 16

Page 16 of 41

average inventory level increasing approximately 1.2 times compared with the initial value, due to the transporter being unable to fully satisfy the customer requirements, and the developed CSCT system presenting a lower order fulfilment rate. The average order fulfilment rate decreased to 82.1% during the simulation period. Table 3 presents the effects of the captured risks, which provides quantitative information to screen out the significant risks. In the developed CSCT system, the major sources of

ip t

risks are breakdown in core operations, inappropriate choice of service provider, choice of

cr

transportation route, lack of/inappropriate inventory management, and lack of/inappropriate



an

3.3. Risk mitigation scenarios simulation

us

container management.

M

Ensuring that a particular risk mitigation approach does indeed support CSCRM, it is worthwhile simulating the system operations to explore the potential risk mitigation outcomes on the changes of

d

system behaviour (Oliveira et al., 2013). An SD model can be modified both in the design and the

Ac ce pt e

operational phases based on the proposed mitigation measures, which is able to help decision makers estimate the system performance in different risk mitigation scenarios. In particular, it does not require a decision-maker to directly assess different risks and provide arbitrary decisions based on past experience or historical data (Yeo et al., 2013). Instead, the experts propose the mitigation measures, and then the whole SD modelling approaches (shown in Fig. 1) are iterated to investigate the possibility of these risk mitigation outcomes. Based on the risk analysis results, the risk of a breakdown in core operations is the most serious risk in the CSCT system. The obtained system behaviour was set as a baseline for benchmarking the outcomes of risk management methods. Two potential risk mitigation methods were suggested by experts in response to the undesired risk effects: increasing transportation equipment capacity and increasing the amount of transportation equipment. The outcomes of the proposed mitigation cases are presented in Table 4.

17

Page 17 of 41

Based on the suggested risk mitigation solutions, the values of both the structure and the variables within the developed SD model were modified to investigate the effects of the implemented approaches. In each case, the effects were simulated in four different scenarios - a 5%, 10%, 15% and 20% increase of the exogenous parameter values were inserted to assess the performance of the built

cr

implemented risk mitigation approach in the real world (Bouloiz et al., 2013).

ip t

CSC system. It should be noted that a range of variations is widely used to test the performance of the

Fig. 9 depicts graphically the simulation results of the inventory level and the average order fulfilment

us

rate under the 5% increase scenario. Three lines are provided to describe the simulation results: Line 1 indicates the effect of increasing transportation equipment capacity by 5%; Line 2 describes the effect

an

of adding 5% to the transportation equipment number; Line 3 represents the system performance in the initial risk scenario.

M



d

The simulation found that both of the implemented methods could significantly improve system

Ac ce pt e

performance in terms of inventory level and order fulfilment ability. However, the method of increasing equipment number performed better in that the negative impacts of the hazardous events that occurred were absorbed and resolved within a shorter time period. The inventory level was estimated to reduce by 50% and the average order fulfilment rate increased to 90% in relation to the initial CSCT system performance.

The simulation results can be used to suggest advantageous risk mitigation decisions by comparing system performance under different scenarios. To provide meaningful insights, Table 5 extracts the simulation results in the scenarios that increase the parameter values by 5%, 10%, 15% and 20%, and depicts the comparisons of the system performances of each risk mitigation scenario.



18

Page 18 of 41

Applying two different risk mitigation methods, the improvement in transportation capacity enhanced the resilience of the CSCT system, such that the effects of the hazardous events were reduced in terms of consequence severity. In the scenario of a 5% increase, the inventory level decreased by 27.62% and 28.56%, and the average order fulfilment rate increased by 22.98% and 23.31% during the simulation period. However, the rate of performance improvement gradually diminished in the

ip t

scenarios of a 10% and 15% increase. The performance of CSCT system recovered to be normal

cr

which coincided with the initial setting in the scenarios of 20% increase. In the analysis, the minimum increases that the system could perform as normal situation were addressed: 16.1% in the case of

us

transportation equipment capacity increase and increasing 15.3% of transportation equipment number. Therefore, the advantageous risk mitigation decision was suggested to deal with the risk of

an

“Breakdown in core operations” in proposed CSCT system.

The multiple modelling and simulation allows the operational performance of the CSCT system to be

M

quantified for diverse risk mitigation actions. The improvements in the transportation service flows to the financial benefits, so that the financial performance of each risk mitigation scenario should be

d

assessed in further research (Narasimhan and Talluri, 2009). A specific module is required to be

Ac ce pt e

developed to investigate the cost and benefits of the implemented risk mitigation methods, taking into account the complex interactions and dynamic feedback effects within the built system. If the damage resulting from a risk occurrence is less than the financial investment in the risk mitigation application, the suggested risk management method is not an efficient means of achieving the CSCRM objectives. By integrating economic evaluation into a developed SD model, the proposed SD-based risk management package could serve as a decision-making tool for continuously improving the system performance and optimising risk mitigation decisions in CSCRM.

4. Summary Generally, the occurrence of a hazardous event will interrupt the flow of CSCT operations and result in various negative effects. To address the risk issues, both researchers and practitioners have adapted 19

Page 19 of 41

a wide range of methods to identify, analyse and manage the risks, which are inherent in, or surrounding the CSC network. However, there remains a lack of practical methodology that takes the interactions and feedback effects among the developed models or systems into consideration. Instead of assessing the risks based on expert knowledge or historical data, this study introduces an analytical method within a changeable system to address the dynamic effects caused by hazards in the

ip t

CSCT. It combines the modelling approach for the quantification of the system performance with an

cr

interactive procedure that integrates risk scenarios and risk mitigation measures. It is particularly noteworthy that the proposed method is an intermediate platform between widely used empirical

us

study and mathematical programming. The experience of risk experts and analysts is collected by questionnaire to identify the various risk scenarios and the proposal of appropriate risk mitigation

an

measures. SD modelling method is employed to accommodate the need to describe the connections between the risks and their associated changes of system behaviour (Hirsch et al., 2007). Furthermore,

M

the SD modelling and simulation approaches (Fig. 1) can be iterated throughout the life cycle, specifically in the design and operations phases. Both the structure of the proposed model and the

d

value of a variable can be amended based on the implemented risk mitigation measures. It enhances

Ac ce pt e

the study of the interactions between the undesired risk effects and managerial activities in this system (Sterman, 2000). The changes of the system performances caused by the implemented risk mitigation methods are addressed by revealing the gap between expectation and real-time performance, so as to suggest the most beneficial CSCRM decisions. In this study, the researchers used a questionnaire survey to address the input values of various risk scenarios, but it is acknowledged that both the size of the sampling population and the subjective nature of the responses could be a source of bias. There is a requirement for a future comparative study using a more extensive data source to demonstrate the proposed method. It is hoped that the proposed SD model will encourage further research in this area. The proposed SD-based CSCRM method can be generalised and has the potential to be a useful and effective management tool in other industry sectors.

20

Page 20 of 41

Acknowledgements The research is partially funded by the FP7 Marie Curie International Research Staff Exchange Scheme (IRSES) of the European Union (Grant No. 314836) and NSSFC of China government (Grant No. 14LBG133). The authors would like to thank Prof. Jin Wang, Prof. Xinping Yan and Dr. Di

ip t

Zhang for their support in chemical supply chain data collection and chemical supply chain system dynamics modelling.

cr

References

supply chain. Ind Eng Chem Res 2010; 49: 9917-9931.

us

Adhitya A, Srinivasan R. Dynamic simulation and decision Support for multisite specialty chemicals

an

Angerhofer BJ, Angelides MC. System dynamics modelling in supply chain management: research review. Proceedings of the 2000 Winter Simulation Conference; 2000 Dec 10-13; Orlando, USA.

M

New York: Springer; 2000.

Barlas Y. Formal aspects of model validity and validation in system dynamics. System Dynamics

d

Reviews 1996; 12(3): 183-210.

Ac ce pt e

Barilas Y, Gunduz, B. Demand forecasting and sharing strategies to reduce fluctuations and the bullwhip effect in supply chains. J Oper Res Soc 2011; 62(3): 458-473. Bonvicini S, Leonelli P, Spadoni G. Risk analysis of hazardous materials transportation: evaluating uncertainty by means of fuzzy logic. J of Hazard Mater 1998; 62(1988): 59-74. Bouloiz H, Garbolino E, Tkiouat M, Guarnieri F. A system dynamics model for behavioural analysis of safety conditions in a chemical storage unit. Safety Sci 2013; 58(2013): 32-40. Campuzano F, Mula J, Peidro D. Fuzzy estimations and system dynamics for improving supply chains, Fuzzy Sets and Syst 2010; 161(11): 1530-1542. Campzano F, Mula J. Supply Chain Simulation: A System Dynamics Approach for Improving Performance. London: Springe-Verlag Limited, 2011.

21

Page 21 of 41

Cai T. Robust optimization for smart manufacturing planning and supply chain design in chemical industry. In: Luo Z, editor. Smart Manufacturing Innovation and Transformation: Interconnection and Intelligence. Hershey: Business Science Reference, 2014. P. 21-27. Carneiro MC, Ribas GP, Hamacher S. Risk Management in the Oil Supply Chain: A CVaR Approach. Ind Eng Chem Res 2010; 49(7): 3286-3294.

ip t

Chen A, Yang H., Lo HK, Tang W. Capacity reliability of a road network: an assessment

cr

methodology and numerical result. Transport Res Part B-Meth 2002; 36(3): 225–252.

Chen A, Kasikitwiwat P. Modelling capacity flexibility of transportation networks. Transport Res Part

us

A-Pol 2011; 45(2011): 105-117.

Das D, Dutta P. A system dynamics framework for integrated reverse supply chain with three way

an

recovery and product exchange policy. Comput Ind Eng 2013; 66(2013): 720-733. Erera AL, Morales JC, Savelsbergh MS. Global intermodal tank container management for the

M

chemical industry. Transport Res Part E-Log 2005; 41(6): 551-566.

Fabiano B, Currò F, Reverberi AP, Pastorino R. Dangerous good transportation by road: from risk

d

analysis to emergency planning. J Loss Prevention in the Process Ind 2005; 18: 403-413.

Ac ce pt e

Ferrio J, Wassick J. Chemical supply chain network optimization. Comput Chem Eng 2008; 32(2008): 2481-2504.

Fisk P. Chemical Risk Assessment: A Manual for REACH. Chichester: John Wiley and Sons Ltd, 2014.

Forrester JW. Industrial Dynamics. Cambridge: The MIT Press, 1961. Forrester JW, Senge P. Tests for building confidence in system dynamic models. TIMS Studies in the Manag Sci 1980; 14(1980): 209-228.

Furuhama A, Hasunuma K, Aoki Y, Yoshioka Y, Shiraishi H. Application of chemical reaction mechanistic domains to an ecotoxicity QSAR model, the KAshinhou Tool for ecotoxicity (KATE). SAR and QSAR in Environ Res 2011; 22(5-6): 505-523. Gao Y, Shang Y, Kokossis A. Agent-based intelligent system development for decision support in chemical process industry. Expert Syst Appl 2009; 36(8): 11099-11107. 22

Page 22 of 41

Garbolino R, Chery JP, Guarnieri F. Dynamic system modelling to improve risk analysis in the context of Seveso industries. Chem Eng Transactions 2009; 17: 373-378. Garbolino E, Chery JP, Guarnieri F. Dynamic risk assessment in the process industries: Proposal of a methodology based on the system dynamics approach. ESREL. Rhodes: Greece 5-9 September 2010. García-Flores R, Wang XZ. A multi-agent system for chemical supply chain simulation and

ip t

management support. OR Spectrum 2002; 24(3): 343-370.

cr

Ge Y, Yang JB, Proudlove N, Spring M. System dynamics modelling for supply-chain management: A case study on a supermarket chain in the UK, Int Trans in Oper Res 2004; 11(5): 495-509.

us

Hirsch BG, Levine R, Miller LR. Using system dynamics modelling to understand the impact of social change initiatives. Am J of Commun Psychol 2007; 39(3-4): 239-253.

an

Janamanchi B, Burns JR. Reducing bullwhip oscillation in a supply chain: a system dynamics modelbased study, Int J of Inf Syst and Change Management 2007; 2(4): 350-371.

M

Kaggwa M. Modelling South Africa’s Incentives Under The Motor Industry Development Programme, 2005-2008. Ph.D. thesis, University of Pretoria.

Ac ce pt e

53-68.

d

Kleindorfer PR, Saad GH. Managing disruption risks in supply chains. Prod Oper Manag 2005; 14(1):

Kumar S, Yamaoka T. System dynamics study of the Japanese automotive industry closed loop supply chain. J of Manuf Technol Manag 2007; 18(2): 115-138. Laínez JM, Puigjaner L, Reklaitis GV. Financial and financial engineering considerations in supply chain and product development pipeline management. Comput Chem Eng 2009; 33(12): 1999-2011. Laínez JM, Puigjaner L. Prospective and perspective review in integrated supply chain modelling for the chemical process industry. Current Opinion in Chem Eng 2012; 1: 430-445. Leonelli P, Bonvicini S, Spadoni G. Hazardous materials transportation: a risk-analysis-based routing methodology. J Hazardous Materials 2000; 71(1-3): 283-300. Lertpattarapong C. Applying System Dynamics Approach to the Supply Chain Management Problems. Msc. Thesis, Massachusetts Institute of Technology, 2002. Leveson N. A new accident model for engineering safety systems. Safety Sci 2004; 42(4): 237-270. 23

Page 23 of 41

Liu L, Li S, Fan T, Chang X. Transportation risk assessment of chemical industry supply chain based on a dual model. Procedia Environ Sci 2011; 11(Part A): 393-397. Liu S, Papageorgiou LG. Multiobjective optimisation of production, distribution and capacity planning of global supply chains in the process industry. Omega 2013; 41(2): 369-382. Manuj I, Mentzer JT. Global supply chain risk management. J of Bus Log 2008; 29(1): 133-155.

ip t

Mokhtari K, Ren J, Roberts C, Wang J. Decision support framework for risk management on sea ports

cr

and terminals using fuzzy set theory and evidential reasoning approach. Expert Syst Appli 2012; 39(5): 5087-5103.

us

Mullai A. Risk management system-A conceptual model. In: Zsidisin GA, Ritchie B, editors. Supply Chain Risk: A Handbook of Assessment, Management, and Performance. New York: Springer US;

an

2009, p. 83-101.

Narasimhan, R. and Talluri, S. Perspectives on risk management in supply chains. J Oper Manag 2009,

M

27(2), pp.114-118.

Oehmen J, Ziegenbein A, Alard R, Schönsleben P. System-oriented supply chain risk management.

d

Prod Plan Control 2010; 20(4): 343-361.

Ac ce pt e

Oliveira F, Hamacher S. Optimization of the Petroleum Product Supply Chain under Uncertainty: A Case Study in Northern Brazil. Ind Eng Chem Res 2012; 51(11): 4279-4287. Oliveira F, Gupta V, Hamacher S, Grossmann IE. A Lagrangean decomposition approach for oil supply chain investment planning under uncertainty with risk considerations. Comput Chem Eng 2013; 50(2013): 184-195.

Papageorgiou LG. Supply chain optimisation for the process industries: Advances and opportunities. Comput Chem Eng 2009; 33(12): 1931-1938.

Peng M, Peng Y, Chen H. Post-seismic supply chain risk management: A system dynamics disruption analysis approach for inventory and logistics planning. Computs Oper Res 2014; 42(2014): 14-24. Pet-Armacost JJ, Sepulveda J, Sakude M. Monte Carlo sensitivity analysis of unknown parameters in hazardous material transportation risk assessment. Risk Anal 1999; 19(6): 1173-1184.

24

Page 24 of 41

Peterson DW, Eberlein RL. Reality check: A bridge between system thinking and system dynamics. System Dynamic Review 1994; 10(2-3): 159-174. Qudrat-Ullah H, Seong BS. How to do structural validity of a system dynamics type simulation model: The case of an energy policy model. Energ Policy 2010; 38(5): 2216-2224. Rao S, Goldsby TJ. Supply chain risks: a review and typology. Int J Log Manag 2009; 20(1): 97-123.

ip t

Rausand M. Risk Assessment: Theory, Methods, and Applications. Hoboken: John Wiley & Sons, Inc;

cr

2013.

Reiskin ED, White AL, Johnson JK. Servicizing the chemical supply chain. J Ind Ecol 1999; 3(2-3):

us

19-31.

Ruiz-Femenia R, Guillén-Gosálbez G, Caballero JA. Multi-objective optimization of environmentally

an

conscious chemical supply chains under demand uncertainty. Chem Eng Sci 2013; 95(2013): 1-11. Ren J, Wang J, Jenkinson I, Xu DL, Yang JB. An offshore risk analysis method using fuzzy Bayesian

M

network. J Offshore Mech Arct 2000; 131(4), pp. 1-12.

Schreckenberg M, Pottmeier A, Hafstein SF, Chrobok R, Wahle J. Simulation of the autobahn traffic

d

in North Rhine-Westphalia. In: Kitamura R, Kuwahara M, editors. Simulation approaches in

Ac ce pt e

transportation analysis. New York: Springer; 2005. p. 205-233. Scruggs CE, Ortolano L, Schwarzman MR, Wilson MP. The role of chemical policy in improving supply chain knowledge and product safety. J Environ Stud Sci 2014; 4(2): 132-141. Sterman JD. Business Dynamics: System Thinking and Modelling for a Complex World. New York: McGraw-Hill Higher Education.

Tako AA, Robinson S. The application of discrete event simulation and system dynamics in the logistics and supply chain context. Decis Support Syst 2012; 52(4): 802-815. Tatano H, Tsuchiya S. A framework for economic loss estimation due to seismic transportation network disruption: a spatial computable general equilibrium approach. Nat Hazards 2008; 44: 253265. Tong K, Feng Y, Rong G. Planning under Demand and Yield Uncertainties in an Oil Supply Chain. Ind Eng Chem Res 2011; 51(2): 814-834. 25

Page 25 of 41

Tse YK. Supply Chain Quality Risk Management: An Empirical Study of its Dimensions and Impact on Firm Performance, 2009-2012. Ph.D. thesis, University of Nottingham. Tsiakis P, Papageorgiou LG. Optimal production allocation and distribution supply chain networks. Int J Prod Econ 2008; 111(2): 468-483. Waters C. Supply Chain Risk Management: Vulnerability and Resilience in Logistics. Norfolk:

ip t

Biddles Limited; 2011.

cr

Wilson MC. The impact of transportation disruptions on supply chain performance. Transport Res Part E: Log and Transport Review 2007; 43(4): 295-320.

us

Yeo G, Pak J, Yang Z. Analysis of dynamic effects on seaports adopting port security policy. Transport Res Part A-Pol 2013; 49(2013): 285-301.

an

You F, Wassick JM, Grossmann IE. Risk Management for a Global Supply Chain Planning Under

M

Uncertainty: Models and Algorithms. Am Inst Chem Eng J 2009; 55(4): 931-946.

Table 1

Definition

Ac ce pt e

Variable Name

d

Definition and role of major variables used to model the risk affected CSCT system.

Downstream Order

A probabilistic input

Function Variable assuming a kind of uncertainty

Upstream Fulfil Customer

DELAY FIXED (input Downstream

Returns the value of the input

Demand

Order, delay time Upstream Lead-time,

delayed by the delay time

0)

Inventory Level

Backlogged Orders

Net Inventory Level

INTEG (entry of Upstream Fulfil

Variable representing the volume

Demand – exit of Products Transported)

of products need to be transported

INTEG (entry of Products Required to be

Variable representing an

Transported – exit of Product

accumulation of backlogged

Transported

products

Equal: Inventory level – Backlogged

Variable representing the volume

Orders

of the products

26

Page 26 of 41

Transportation Capacity

Min (Transportation Capacity can be

The products transported with the

Used

used, Products Require Transport)

maximum capability

Infrastructure Capacity

Depends on the selected route condition

Variable affecting the value of transportation capacity

Depends on the operational capacity of

Variable partly determining the

Capacity

transporter

value of transportation capacity

Order Fulfil Rate

Equal: Product Received / Order needs to

Variable representing the rate of

be fulfilled

order completion

RANDOM UNIFORM (0, 1, 365)

Generating uniformly distributed

cr

Random Number

ip t

The Available Transporter

us

random varieties on the closed interval [0, 1]

IF THEN ELSE (Random Number >

Occurrence

Likelihood of Risky Event Occurrence,

Variable active when the value of

an

Likelihood of Risky Event

M

0, 1)

it exceeds Random Number

IF THEN ELSE (Random Number >

Variable active when the value of

Risky Event Affects

Probability of Occurred Risky Event

it exceeds Random Number

Variable

Affects Variable, 0, 1)

Ac ce pt e

d

Probability of Occurred

Consequence Severity

Consequence Probability

Variable Value Reduce

Depends on the effects of risky event

Variable representing the impact of risky event

IF THEN ELSE (Random Number >

Variable active when the value of

Probability of Consequence, 0, 1)

it exceeds Random Number

INTEG (entry of Risky event affected by

Variable representing the level of

risky event – exit of Variable value

variable value affected by risky

recover rate)

event

Table 2 The descriptions of SD simulation results of thirteen risk scenarios. Transportation

Transportation

Inventory

Order

capacity

time

level

fulfilment rate

27

Page 27 of 41

90.00

Max

4.60

Max

320.74

Max

1.00

Min

7.92

Min

2.00

Min

0.00

Min

0.00

Ave

88.29

Ave

2.14

Ave

96.03

Ave

0.66

Improper service provider

Max

90.00

Max

3.00

Max

139.63

Max

1.00

selection

Min

70.02

Min

2.00

Min

0.00

Min

0.00

Ave

89.59

Ave

2.03

Ave

80.76

Ave

0.95

Improper transportation

Max

90.00

Max

3.40

Max

139.63

Max

1.00

route selection

Min

70.02

Min

2.00

Min

0.00

Min

0.00

Ave

89.64

Ave

2.04

Ave

80.71

Ave

0.95

Inadequate transportation

Max

90.00

Max

3.94

Max

123.70

1.00

capacity

Min

62.20

Min

2.00

Min

0.00

Min

0.00

Ave

89.63

Ave

us

Max

2.16

Ave

80.34

Ave

0.95

Max

90.00

Max

2.91

Max

125.68

Max

1.00

Min

76.99

Min

2.00

Min

0.00

Min

0.00

Ave

89.79

Ave

2.03

Ave

80.09

Ave

0.95

Max

90.00

Max

3.00

Max

125.68

Max

1.00

Min

Min

Min

0.00

Min

0.00

cr

an

d

Complexity of product types

M

High level of process variation

ip t

Max

Breakdown in core operations

2.00

Ave

89.81

Ave

2.04

Ave

80.07

Ave

0.95

Lack of/inappropriate

Max

90.00

Max

3.00

Max

131.71

Max

1.00

inventory management

Min

73.98

Min

2.00

Min

0.00

Min

0.00

Ave

89.72

Ave

2.03

Ave

80.33

Ave

0.95

Lack of/inappropriate

Max

90.00

Max

2.73

Max

131.71

Max

1.00

container management

Min

73.98

Min

2.00

Min

0.00

Min

0.00

Ave

89.71

Ave

2.02

Ave

80.35

Ave

0.95

Max

90.00

Max

2.91

Max

123.70

Max

1.00

Min

71.10

Min

2.00

Min

0.00

Min

0.00

Ave

89.75

Ave

2.03

Ave

80.12

Ave

0.95

The challenge of technology

Max

90.00

Max

2.44

Max

123.70

Max

1.00

innovation

Min

77.98

Min

2.00

Min

0.00

Min

0.00

Ac ce pt e

75.60

Lack of qualified Labour

28

Page 28 of 41

Ave

89.84

Ave

2.01

Ave

79.96

Ave

0.96

Information sharing delay

Max

90.00

Max

2.58

Max

116.68

Max

1.00

Min

81.99

Min

2.00

Min

0.00

Min

0.00

Ave

89.93

Ave

2.02

Ave

79.75

Ave

0.96

Information sharing

Max

90.00

Max

2.82

Max

118.66

Max

1.00

inaccuracy

Min

80.51

Min

2.00

Min

0.00

Min

0.00

Ave

89.92

Ave

2.04

Ave

79.79

Ave

0.96

Max

90.00

Max

2.91

Max

121.63

Max

1.00

Min

72.00

Min

2.00

Min

0.00

Min

0.00

Ave

89.86

Ave

2.02

Ave

79.86

Ave

0.96

cr

us

an

Financial problems

ip t

innovation

Table 3

Transportation

Transportation

Inventory

Order

capacity

time

level

fulfilment rate

d

M

The comparisons of the risk scenarios simulation results with the initial system performance.

7.00%

20.61%

-14.65%

-0.46%

1.5%

1.43%

-1.66%

-0.40%

1.9%

1.37%

-1.35%

-0.41%

8.05%

0.90%

-1.25%

High level of process variation

-0.23%

1.55%

0.59%

-0.83%

Complexity of product types

-0.21%

2.00%

0.57%

-0.94%

Lack of/inappropriate inventory

-0.31%

1.50%

0.89%

-1.35%

-0.32%

1.05%

0.92%

-1.45%

-1.90%

Ac ce pt e

Breakdown in core operations Improper service provider selection

Improper transportation route selection

Inadequate transportation capacity

management Lack of/inappropriate container management

29

Page 29 of 41

Lack of qualified Labour

-0.28%

1.35%

0.63%

-0.83%

The challenge of technology

-0.18%

0.40%

0.43%

-0.52%

Information sharing delay

-0.08%

0.95%

0.16%

-0.32%

Information sharing inaccuracy

-0.09%

1.85%

0.21%

-0.52%

Financial problems

-0.16%

1.10%

0.30%

-0.43%

cr

ip t

innovation

Case study conditions of suggested risk mitigation methods.

us

Table 4

Variable value set

Case 1: Effect of transportation

Current situation

20 Units/device

Capacity increase

Degree of increase: 5%, 10%, 15% and 20% of current situation

Current situation

20 devices

Number increase

Degree of increase: 5%, 10%, 15% and

Ac ce pt e

equipment number increase

M

Case 2: Effect of transportation

d

equipment capacity increase

Description

an

Case study

20% of current situation

Table 5

Effects of implemented risk mitigation methods. Risk mitigation method

Degree of increase

Effect of

5%

10%

15%

20%

transportation

Average inventory level

-27.62%

-29.64%

-32.94%

-33.15%

equipment capacity

Average order fulfilment

22.98%

33.76%

37.45%

38.24%

increase

rate

5%

10%

15%

20%

Effect of transportation

30

Page 30 of 41

Average inventory level

-28.56%

-31.21%

-33.08%

-33.15%

equipment number

Average order fulfilment

23.31%

35.03%

38.24%

38.24%

increase

rate

Ac ce pt e

d

M

an

us

cr

ip t

transportation

Highlights •

Review chemical supply chain transportation risk management theory and application. 31

Page 31 of 41

Propose a system dynamics based method to identify, assess and mitigate risks.

•

Explore the complex interactions and dynamic feedback effects among the system.

•

Assess and benchmark the effects of various risks through conducting simulation.

•

Provide a transparent decision support tool to risk management.

Ac ce pt e

d

M

an

us

cr

ip t

•

32

Page 32 of 41

Backlogged order Upstream supplier inventory

-

Product shipped to customer +

+ -

+ +

+

us

Order to supplier

cr

Received order from supplier +

Order fulfilment

Inventory level

+

ip t

Net inventory level

Ac

ce pt

ed

M

an

Fig. 3. Causal loop diagram of the CSCT inventory system.

Page 33 of 41

Variable Factor

Risky event factor

ip t

+ + The probability of risky The severity of + The consequence event influence the selected consequence probability variable + + Variable recover + + The occurance rate + Variable influenced likelihood of risky event by risky event + recover Effective ability Variable damage + Variable damage + stack + rate Variable value loss -

cr

Variable value

Ac

ce pt

ed

M

an

us

Fig. 2. Causal loop diagram of the variable affected by a hazardous event.

Page 34 of 41

Problem definition

Step 1

Problem definition

Step 2

Causal loop diagram creation

Step 2

Causal interdependencies formulation between the flow of CSC operations and risk factors

Step 3

Conceptual design of model

Step 3

Causal loop diagram development

Step 4

Stock and flow diagram creation

Step 4

System dynamics model development

Step 5

Data collection

Step 6

Step 7

Logical modelling phase

Model quantification phase

Model validation

Model validation

Step 6

Model test

Model test

Step 7

Step 8

System Dynamics Modelling Process

us

Model application phase

Risk mitigation scenarios simulation

System Dynamics based CSCRM Modelling Process (Source: Drawn by Author)

ed

(Source: Yeo et al., 2013)

Risk scenarios simulation

an

Model application

M

Step 8

Model quantification phase

cr

Step 5

Model application phase

Logical modelling phase

ip t

Step 1

Ac

ce pt

Fig. 1. Framework of SD-based CSCRM modelling process.

Page 35 of 41

Downstream Order

Inventory Level

1

100

1

1

1

200

1

1

1 1

1

1

150

1

1

Unit

1 1

50

1

100

1

1

1 1

25

0

1

73

146

219

292

365

1

1

1

1

1

0

73

146

Time (Day) Downstream Order : Normal

1

1 1

219

292

365

Time (Day)

1

1

1

1

1

1

Inventory Level : Normal

1

1

Backlogged order

1

1

1

1

ip t

0

1

1

1

1

50

0

1

1

1

1

1

1

1

1

1

Order Fulfil Rate 1

1

1

1

1

1

1

1

1

1

1

cr

20

1

1

10 1

us

0.75 Dmnl

15

0.5

1

5

0.25

1 1 1 1

0

1

1

1

1

1

73

1

1

146

1

1

219

292

0

365

Time (Day) 1

1

1

1

1

1

1

1

1

1

1

0

73

Order Fulfil Rate : Normal

1

146

1

219

292

365

Time (Day) 1

1

1

1

1

1

1

1

1

M

Backlogged order : Normal

an

0

ce pt

ed

Fig. 6. Initial performance of CSCT system.

Ac

Unit/Day

75

Page 36 of 41

Order Lead-time Upstream Lead-time Average Order Fulfill Rate

Products Requried to be Transported Upstream Fulfilment Customer Demand

Downstream Order

Accumulative Order Fulfill Rate

Products Reecived

Order Fulfil Rate Downstream Order Needs to be Fulfilled

Initial Transportation Time

Backlogged Order

Transportation Time

Backlogger Order Transported

ip t

CSCT system

Block Transportation Time

Inventory Level

Net inventory level

Products Transported

Variable Value Reduce Rate

Variable Value Recover Rate

Consequence Severity Variable affected by risky event Consequence Severity Lookup

The Available Transporter Capacity

us

Variable Value Reduce

Random Number 2 Probability of Occurred Risky Event Affects Variable

M

Random Number 1

an

Risk factor

Consequence Probabilitye Lookup

Assigned Probabilities of Hazardous Event Occurrence Likelihood Probability of Occurred Consequence Risky Event Affects Variable Probability Lookup Likelihood of Risky Event Likelihood of Risky Occurrence Lookup Event Occurrence

ed

cr

Transportation Capacity Used

Infrastructure Capacity

Labour Recovery

Available Equipment Number

Equipment Recovery

Loading per Equipment

Number of Available Labour

Transportation Backword Time Transportation Capability

Total Number of Labour

Labour Start to Work

Required Labour per Equipment Number of Available Equipment



Equipment Used Total Number of Equipment

Ac

ce pt

Fig. 5. Stock and flow diagram of risk affected CSCT system.

Page 37 of 41

Infrastructure capacity

Inventory on-hand

+

Products from upstream

-

+ + Transporter capacity

+

+

+

Product shipped

+

Transporter Capacity Used

+ Total available transporter capacity

+ Transportation time +

Number of available labours Labour recovery

+

+

+

+

-

Vehicle recovery

Vehicles start to work

+

an

Labours start to work

-

Number of available vehcles

cr

+

us

+

ip t

Loading per vehicle

Ac

ce pt

ed

M

Fig. 4. Causal loop diagram of the dynamic transportation capacity.

Page 38 of 41

Inventory Level 400

Unit

300

3

3 3

200

1

3 1

3

1

2 3 1

1

1 2

2 3

3

2

3

2

0

1

12

2

1

12 3

1 2

2

23 1

23

2

73

146

1

2

1

1

0

1

2

3 1

2

cr

100

3

3

ip t

3

219

292

365

us

Time (Day)

1 1 Average 1 1 Order 1 1Fulfill 1 1 1 1 1 1 1 1 Rate Inventory Level : 2 2 2 2 2 2 2 2 2 2 2 2 2 2 Inventory Level : 1 1 3 32 32 3 3 3 Inventory Level : Breakdown in core opeations 3 2 23 23 2 2

1

.75

2 3

3 1

1

2

2

2

2

2

3 2

3

3

3

ce pt

1 2

3 3

1

1

2

1

3

3

3

3 3

M

.5

0

3

1

3

ed

Dmnl

3

0

1

3

1

.25

1

1

1

1

1

1

1

an

1 2

73

146

219 Time (Day) 13 1

Average Order Fulfill Rate : 2 1 1 3 1 1 Breakdown in core operations: Order Fulfill Rate : 1 2 2 2 2 2 Effect ofAverage transportation equipment number increase 5%: Average Order Fulfill Rate : Breakdown in core opeations 3 Effect of transportation equipment capacity increase 5%:

1

2

2 3

2 3 1

292 13 2

1

1 2

3

1 2

3

365

23 1

1 2

1 2

3

1 2

23

3

1

Ac

Fig. 9. The comparative simulation results of two risk mitigation cases.

Page 39 of 41

Probability of consequence Lookup Consequence severity on the variable Probability of consequence

Consequence severity on the variable Lookup

Reduced value of affected variable

0 0

Random Number 1

Hazardous event occruance likelihood

73

146

219

292

365

Time (Day)

Hazard Event Impact on Infrastructure Capacity : Breakdown in core opeations

M

Hazardous event occruance likelihood Lookup

.075

us

Assigned Probabilities of Hazardous Event Occurrence Likelihood

ip t

Probability of the variable affected by the hazardous event Lookup

.15

an

Probability of the variable affected by the hazardous event

.225

cr

The effect of hazardous event

Dmnl/Day

Random Number 2

Hazard Event Impact on Infrastructure Capacity .3

Ac

ce pt

ed

Fig. 8. An illustrative example the generation of the risk of “Breakdown in core operations”.

Page 40 of 41

Downstream Order 200

2

1 1

2

2

0

2

12

2

1

1

1

2

2

1

2

1

2

2

2

1

2

50

1

1 1

1

2

1

0

5

10

15

20

25 30 Time (Day)

35

Downstream Order : 5% increase 1 1 1 Downstream Order : Initial performance of CSCT system

1 1 1 1 Backlogged 2 2orders 2 2

1

40

45

1

1

2

2

50

1

1

2

2

2

us

40

2

1

ip t

100

cr

Unit/Day

150

30

20

an

Units

1

2

1

10

1

1

1

1

2

12

0

12

12

5

12

10

M

2

0

12

2

15

2

12

20

1

25 30 Time (Day)

35

Backlogged orders : 5% increase 1 1 1 Backlogged orders : Initial performance of CSCT system

1 1 1 Order 1Fulfil 2 2 Rate 2 2

2

1

2

2

1

2

ed

1

12

12

2

1

2

1

12

2

40

1

1

1

2

2

50

1

1

2

2

2

2

2

2

45

1

2

2

1

2

1

Dmnl

ce pt

0.75

1

1

2 1 1

0.5

0.25

0

12

Ac

0

12

5

1

10

15

20

25 30 Time (Day)

35

Orderunder Fulfil Rate : 5%increase increase 1of “Downstream 1 1 1 1 1 1 CSCT performance 5% order”: Order Fulfil Rate : Initial performance of CSCT system 2 2 2 Initial CSCT performance: 2 2 2

40 1

12 2

45

1

1 2

50 1

2

12

2

1 2

Fig.7. CSCT system performance under 5% increase of “Downstream order”.

Page 41 of 41