1. Introduction
The Cold Supply chain plays a crucial role in supplying people around the world with their needs of food. With a growing population, climate change, depleted resources, and supply chain disturbances that may arise because of global crises (i.e., coronavirus pandemic, wars, etc.), limiting wasted food within the supply chain and increasing its shelf life is vital to help mitigate these challenges. Other than limiting waste, fuel efficiency and consumption, as well as financial benefits through cost reduction, will help achieve environmental sustainability and green logistics by feeding the world with the minimum financial and energy resources.
FAO reported that around 30% of the world’s food production is wasted, where agricultural production and fresh produce (vegetables and fruits) account for 28% and 40% of the total wasted food, respectively [
1,
2,
3]). The US Department of Agriculture conducted a study in 2010 [
1] and found that food waste accounts for up to 40% of the food supply, which is the equivalent of 133 billion pounds of produce with a total financial value of USD 161 billion worth. With this wasted food, its required production resources (water, land, bio-diversity, and financial resources), and its effects on the environment (recycling, pollution, damage to freshwater fisheries, CO
2 emissions), the US Department of Agriculture decided to work with the U.S. Environmental Protection Agency to decrease food loss by 50 percent by 2030 where 2010 is considered as a baseline at 219 pounds of food loss per capita. To reach this optimistic target (50% reduction), the supply chain (from producer to the customer) should be thoroughly investigated to identify food waste at the different supply chain stages and their corresponding levels in order to effectively address the main and most important causes of food loss.
In another study [
4], the USDA estimated that 141 trillion calories per year, or 1249 calories per capita per day, was not consumed. The top three food groups in terms of share of the total value of food loss are meat, poultry, and fish (30 percent); vegetables (19 percent); and dairy products (17 percent). The same study classified food waste at four levels: farm, farm to retail, retail, and consumer levels. When farm to retail level is considered, the identified causes include (but are not limited to): spillage and damage, such as equipment malfunction (for example, faulty cold or cool storage) or inefficiencies during harvesting, drying, milling, transporting, or processing.
The paper [
5] mentions that refrigerated transportation is required for 40% of all foods, which uses 15% of world fossil fuel energy. Furthermore, refrigerated transportation leads to significant CO
2 emissions (around 40%) originating from vehicle’s engine and refrigerant leakage. For example, in terms of CO
2 emissions, a 38-ton articulated vehicle class generates the equivalent of 58 kg CO
2/pallet-km while a multi-drop frozen medium rigid vehicle class generates a115 kg CO
2/pallet-km.
Circular economy emphasizes the production of products that can be reused and recycled as well as eliminating waste and considering the environment when managing operations. The paper [
6] that defines circular economy CE as a closed loop economy applying a materials balance model conducted a Systematic Literature Network Analysis and found eight emerging research trends in a circular economy that include the transition to CE at the micro level, transformation of wastes into valuable resources and enabling CE throughout internet technologies. In a second paper covering CE [
7], the authors conducted a systematic review of drivers and critical success factors in circular economy. Some of the drivers mentioned are economic (improving cost efficiency and profitability), environmental (climate change and global warming), and supply chain (communication and collaboration). Furthermore, another paper [
8] that developed A Maturity Model for Logistics 4.0 identified the flow of material and information among its dimensions. The Flow of material covers the Degree of automation in transportation and the Internet of things, while the flow of information covers Data-driven services, RFID, RTLS (real-time locating systems), and IT systems as areas of evaluation. In light of the above papers, this work fits the already mentioned CE research trends and success factors as well as the logistics 4.0 maturity model by investing in IoT, RFID, and RTLS to reduce cost and adverse effects on the environment. Additionally, while integrating many technologies (such as RFID, WSN, Geographic Information System, IoT, etc.), this work will emphasize automation by introducing a decision-making controller capable of recommending a course of actions and mitigation decisions using the sensing and tracking data as inputs.
In fact, many works assessed the use of new technologies (IoT, RFID, WSN, GP, etc.) and their potential for helping decision makers achieve sustainability in transportation operations. Nonetheless, as far as we are aware, there is no comprehensive approach in the literature that addresses different sustainability dimensions that includes both cost efficiency and transportation effects on the environment, as well as the adoption of logistics 4.0 maturity model dimensions. In other words, current research still lacks a comprehensive framework that can be utilized by decision makers (such as producers, truckers, and retailers) to assess the benefits of investing in such technologies. Thus, the purpose of this paper is to fill the gap in the literature by introducing a real-time IoT-based monitoring and control system framework for green logistics decision-making. With this purpose, this work will help perishable products supply chain companies achieve economic and environmental sustainability through cost reduction, energy consumption minimization, and waste control. Additionally, this work targets several decision makers that include producers who control their transportation operations through their own trucks fleet and trucking companies that offer logistics services to producers and retailers. Consequently, the aforementioned decision makers will be able to assess the proposed system hardware (RFID, WSN, and GPS systems, etc.), its maintenance and cloud services, and the cost savings and environmental benefits (CO2 emissions, reduced food waste, energy consumption). The targeted decision makers will also be able to link cost and environmental benefits to their intrinsic supply chain configurations parameters such as cost components, failure probabilities, wasted shipment, etc., which will help them decide whether to adopt the proposed system or not.
As a matter of fact, this work helps answer the following questions: what is the architecture and the different components of the proposed system? How can the benefits (financially and environmentally) of such a system be quantified in terms of the system variables and parameters? The proposed system is implemented through a case study where different scenarios are compared in terms of their financial and environmental benefits.
In the following section, the different technologies used in the proposed system, as well as cold supply chain monitoring systems in the literature, will be reviewed.
3. Methods
The methods section consists of the following three different parts:
First, the architecture of the IoT cloud computing system is introduced, which covers its network of things and the cloud computing levels. The components of both levels, their capabilities as well as the shared data between the two levels are described.
Second, the assumptions and annotations that are used for the different simulation models are introduced. This includes all variables ad parameters, such as the cost components, failure probabilities, different timing annotations, etc.
Third, the three simulation models, as well as their intrinsic scenarios, are explained in this section. For each simulation model, the set of considered decisions, as well as each scenario probability and cost, is calculated.
Fourth, at the end of the Methods part, the case study data that cover transporting produce in the intercontinental United States are introduced. Data include produce and transportation costs, distances from supplier to the different destinations, etc.
3.1. Proposed IoT Cloud Computing Architecture
Figure 3 presents the two levels of architecture of the proposed IoT cloud computing system where the network of things provides sensing, identification, and location awareness data to the cloud computing unit that is responsible for computation, visualization, analytics, and control.
The things in a transportation network are the three echelons supply chain network which are the producer, the trucks, and the customers. The trucks are equipped with WSN to monitor the transportation conditions of the goods, and the transported products are equipped with RFID for identification purposes. The Geographic information system tracks the location of the traveling goods and calculates the distances and travel times between the trucks and the different destinations.
The cloud computing level comprises the controller responsible for decision making and interfering with mitigation plans whenever necessary. It also offers storage, computation, analytics, and visualization of real-time data received from the network of things such as transportation conditions (temperature, humidity, etc.) and real-time distances between the trucks and the different destination. The controller preloaded with decision making modules can automatically make transportation-related decisions without human interference.
3.2. Simulation Models Assumptions and Annotations
In the following built simulation models, a producer fulfilling the nTj periodic demand dj of NC customers Cj’s is considered. In case of failure, the following annotations are adopted:
SnTj = failure time;
FT = the time it takes for the products to perish just right after failure;
tnTj = SnTj + FT is the spoilage time;
Similar to the works [
24,
26], and without loss of generality, the following cost components will be considered:
Products Cost: if vj is the product unit cost, then the product cost corresponding to customer Cj is equal to vj ∗ dj.
Transportation cost: If rj is the transportation cost per time and product unit, and the transportation time corresponding to customer Cj is tj, then the transportation cost is dj ∗ rj ∗ tj.
Penalty: If products fail to reach customer Cj and Lj is the penalty per product unit, then the penalty incurred is equal to Lj ∗ dj.
Inventory Cost: if a shipment aimed at customer Cj is rerouted to another customer Cj’, and stored in inventory for a period Tjj’ before being used to fulfill customer Cj’ periodic demand, then an inventory cost equal to Ij’ ∗ dj ∗ Tjj’is incurred, where Ij’ is the inventory cost per product unit per time period.
Failure probability pntj: the Failure probability during period n corresponding to Cj.
Spoilage probability psntj: the probability that products spoil before reaching Cj during nTj.
3.3. Simulation Models
For all simulation models, P(SMi-Scj) and C(SMi-Scj) denote the probability of scenario j (under simulation model i) and its corresponding cost, respectively.
Simulation Model 1 (SM1): SM1 represents the baseline where the RFID-WSN-GPS Decision Support System is not deployed. Actually, if a failure occurs (snTj is the failure time), then the producer will not be notified until the delivery of products. All SM1 Related scenarios are depicted in
Figure 4 below.
The three different scenarios corresponding to SM1 as well as their incurred costs are described below:
SM1-Sc1: No failure will happen, and products will reach customer
j in good condition
SM1-Sc2: a failure will happen; products will be spoiled before they reach customer Cj. In this case, other than Products and transportation costs, a penalty will be incurred.
SM1-Sc3: a failure will happen; however, products will not be spoiled by the time they reach customer
j, and transportation will continue.
As a result, the mean cost per shipment corresponding to customer Cj is equal to:
For SM2, stopping transportation in the case of failure is considered. In fact, when the nTj shipment corresponding to customer Cj cannot be delivered before the spoilage time t
nTj, the Decision Unit will stop transportation. SM2 scenarios are depicted in
Figure 5.
The three different scenarios corresponding to SM2, as well as incurred costs, are described below:
SM2-Sc1: No failure will happen, and products will reach customer
j in good condition
SM2-Sc2: a failure will happen; products will be spoiled before they reach customer Cj. In this case, transportation will stop, and the goods will be disposed of.
SM2-Sc3: a failure will happen; however, products will not be spoiled by the time they reach customer
j, and transportation will continue.
As a result, the mean cost per shipment corresponding to customer Cj is equal to:
Under Simulation Model 3 (SM3), when customer Cj periodic (nTj) shipment cannot be delivered before the spoilage time tnTj, the following two decisions are considered by the decision unit:
- 🗸
Reroute the shipment to a different customer Cj’, in case nTjj’, the delivery time to customer Cj’, precedes the spoilage time tnTj = snTj + FT.
- 🗸
If rerouting is not possible, then the decision unit will stop transportation.
The four different scenarios corresponding to SM3 (depicted in
Figure 6), as well as their corresponding incurred costs, are described below:
In this case, other than Products and transportation costs, a Lost sales cost will be incurred.
SM3-Sc4: a failure will happen, and products will be spoiled before they reach customer
j; however, the shipment can be rerouted to another customer Cj’ without being perished if the arrival time nTjj’ to customer Cj’ precedes the spoilage time t
nTj.
As a result, the mean cost per shipment corresponding to customer Cj is equal to:
3.4. Case Study
Because of the stochastic environment originating from the failure probability and the failure time, the case study consists of a Monte Carlo simulation corresponding to the three Models introduced. Actually, Monte Carlo simulation selects the value of the considered random variable from the defined probability distribution and uses it in the computer trials. Some simulation parameters (such as distances, prices, trucking costs, and product costs) are real data retrieved from different sources such as databases and reports. On the other side, some other parameters that include confidential data cannot be accessed through public databases, or they are product and equipment related, assumptions were made. For example, the penalty imposed by a customer for not receiving their shipment is usually specified in a supplier contract. The list of assumptions includes the failure probability distribution, failure time, spoilage time, penalty, and inventory-related cost (as percentages of the product’s value). However, the assumptions made do not result in a loss of generality as many discussed results (financial benefits, savings, environmental benefits, etc.) are expressed analytically in terms of the simulation parameters.
The two random variables considered in this Monte Carlo simulation are:
The failure probability pnTj: for all shipments, a fixed failure probability pnTj will be considered. Then the Monte Carlo process will consist of generating a random value (for example R), by sampling from the uniform distribution UNIFORM [0,1]. If 0 ≤ R ≤ pnTj, then a failure will occur (the failure Variable F = 1); otherwise, there will be no failure (F = 0). For example, if pnTj = 0.1 and R = 0.05, then there will be a failure and F = 1. However, if pnTj = 0.1 and R = 0.2, then there will be no failure, and F = 0.
The Failure Time: for each shipment, values for the random variable, which is the failure time, will be generated according to the assumed probability distribution. From a real-life perspective, the probability distribution can be estimated from historical failure data retrieved from the transportation company. In our case, for each shipment, a Uniform Distribution [0, Arrival Time to Destination = nTj] will be considered.
In the conducted case study, a fresh produce (grapes) supply chain consisting of transporting a full truckload FTL shipment of 40,000lb from California to several destinations across the US is considered. Actually, two valleys in California (San Joaquin and Coachella valleys) are the biggest grape growers in the US, according to the statistics company Statista [
27]. For example, in 2021, 5.76 million tons of grapes were grown in San Joaquin and Coachella valleys compared to only 325 thousand tons by the second biggest producer, Washington. The considered transportation destinations are several Terminal Markets across the US where the selling price of grapes is tracked by the Agricultural Marketing Service, US Department of Agriculture [
28]. For the discussed case study in the next section, the following data and assumptions are considered:
Google Maps [
29] is used to calculate the distance (in miles) from the San Joaquin Valley to each one of the terminal markets (depicted in
Table 1).
The trucking cost per mile is calculated using data from ATRI, the American Transportation Research Institute [
30]. The trucking cost per mile from San Joaquin valley to each one of the terminal markets is calculated based on the operational trucking cost per region (depicted in
Table 2). For instance, if the Dallas terminal market is considered, then the transportation route falls in the west-southwest regions, and the transportation cost per mile will be the average of the two regions’ costs. The different terminal markets’ transportation costs per mile are depicted in
Table 2.
The different terminal markets produce (grapes) costs per lb. considered are retrieved from the Agricultural Marketing Service, US Department of Agriculture [
28]. A yearly average price is calculated based on monthly prices of 18lb containers corresponding to the different markets. Average prices corresponding to the different terminal markets are presented in
Table 3.
For each shipment, there will be a fixed failure rate pnTj, and a Monte Carlo simulation about failure/No failure will be run.
For each shipment, if a failure happens, the spoilage time tnTj will depend on the remaining traveling time. If the remaining traveling time is greater than a certain timing FT in hours in the distance (which is the amount of time it takes for the products to spoil after failure), then products will spoil. Then, the spoilage time tnTj is equal to the failure time plus FT, tnTj= snTj+ FT.
In our case study, the spoilage will happen within 200 miles of failure time.
Rerouting will only happen if the arrival time to the new destination tnTjj’ precedes the spoilage time.
For each of the destinations, the rerouting will only happen if the new destination is within a 200miles distance from the failure location.
Unless mentioned otherwise, the considered failure probability is 0.01 for all discussed results.
Google Maps [
29] is used to calculate the traveled distance and possible rerouting options (terminal markets) whenever a failure happens, and rerouting is considered. For example,
Figure 7 presents the route from the source (San Joaquin valley) to the destination New York, which passes by Chicago. In fact, for each destination, the route was checked on Google Maps, and a failure range representing the range at which a failure should happen in order to reroute the shipment to the new destination was built. Actually, two different types of destinations (
Table 4) emerged as the following:
A Destination with No Rerouting option: Atlanta, Chicago, and San Francisco: even if a failure happens in these destination routes, the rerouting option is not possible because the shipment cannot make it to any of the possible destinations before the spoilage time.
A Destination with a Rerouting option: Dallas, New York (
Figure 4), Miami, Toronto, and Los Angeles. If a failure happens during the shipment transportation to one of these destinations, then the shipment can be rerouted to at least another destination. However, this will depend on the failure time and FT. Therefore, for each potential rerouting option, a failure range was built: if the failure happens within that range, then the shipment can be rerouted to that new destination because the arrival time nTjj’ precedes the spoilage time, which is tnTj = snTj + FT; otherwise, the shipment cannot be rerouted.
Lj is the penalty charged by the customer in case of non-delivery (based on supplier-customer contract) or can be calculated based on the shipping time and cost in addition to the cost of the lost product and may also include reputation and sales opportunity costs. In the simulation model, it is equal to 1% of the production value.
Inventory cost: this includes refrigeration, handling and storage, and inventory management-related costs. In the simulation model, it is equal to 1% of the production value.
A 95% confidence interval is built for each configuration’s total supply chain cost as well as each destination-related cost. Therefore, we can be 95% confident that the mean cost is in the confidence interval. If s is the standard deviation and n is the sample size, then the 95% confidence limits formulas are as follows:
Upper limit = mean cost + 1.96(s/
Lower limit = mean cost − 1.96(s/
To narrow the confidence interval for the supply chain mean cost, n = 20,000 trials (shipments) are considered for each customer.
5. Conclusions
This work introduces an IoT Real-Time Decision Support System for a Green supply chain. The system will interfere in case of failure during the shipment route with mitigation plans such as stopping transportation and rerouting to minimize financial losses. The proposed IoT framework is two levels composed of the network of things and cloud computing and combines RFID, WSN, and GIS.
Three different supply chain simulation models that consider the major cost components of any supply chain, such as transportation, shipment value, lost sales, and inventory cost, were built:
A baseline denoted as SM1 where the system is not deployed;
SM2, where stopping transportation is considered;
SM3, where stopping transportation and rerouting decisions are considered.
The financial and environmental benefits corresponding to the different simulation models are expressed in terms of cost components (transportation cost, product cost, lost sales cost, etc.), failure probabilities, and other system decision variables and parameters. This will enable the IoT-based system adopters to assess its financial and environmental benefits according to their business models and decide whether to adopt it or not. In fact, the proposed system considers stopping transportation and rerouting shipments in real time, which makes cost savings higher compared to a monitoring and control system with data checked only at certain checkpoints.
The deployed decision support system shows cost reductions under SM2 and SM3 compared to the baseline SM1. For instance, under SM2, the mean cost per shipment corresponding to Atlanta is equal to 0.34%. The savings under SM3 exceed savings under SM2 because of the rerouting option. For example, Chicago-related mean cost per shipment decreased by 3% under SM3.
In addition, the effect of failure probability on cost savings was investigated. As expected, under SM2, the mean cost will decrease by 0.04% and 0.34% for failure probabilities pnTj equal to 0.01 and 0.1, respectively.
Other than financial benefits, environmental benefits were quantified for the different supply chain models. For example, SM3 shows a decrease in the average traveled distance per period (for all shipments). Actually, the average traveled distance per shipment decreased by 2.5% and 5% for failure probabilities pnTj equal to 0.05 and 0.1, respectively. This decrease in the total traveled distance per period results in a decrease in the CO2 emissions; accounting for 2.47% (1,216,259 grams of CO2) and 4.9% (2,408,372 grams of CO2) corresponding to failure probabilities pnTj equal to 0.05 and 0.1, respectively.
Also, significant quantities of produce were saved thanks to the rerouting option. In fact, for a failure probability pnTj equal to 0.1, the mean total quantity of rerouted produce (that would be disposed of without the rerouting option) is equal to 4162 lb per period, which represents 1.3% of the total shipped produce (all destinations) per period. For instance, 4.75% of the shipped product to Los Angeles was saved.
For future work, the following should be considered:
Expand the list of decisions that the controller can take.
Consider simulation models with different supply chain configurations, such as considering many suppliers or considering a three-echelon supply chain. In this case, other decision variables, such as the optimal quantities that should be transported from each supplier to each destination, should be considered. In addition, supplier capacity, as well as customer demand, should be included.
Investigate the implementation cost of the suggested system.
In addition to cost efficiency and sustainability metrics (such as CO2 emissions and saved produce), other supply chain indicators, such as on-time delivery and truck usage, must be considered in the analysis to better show the utility of the proposed IoT monitoring and control system.