Food scarcity and waste are one of this century’s most threatening global problems. To minimize food losses in the supply chain, fresh fruits and vegetables (FFVs) must be transported under controlled atmospheric and temperature conditions.
Such controlled conditions are exceptionally prone to disruptions during last-mile delivery to retailers due to repeated door openings, varying customer demands, truck breakdowns, traffic jams within the city, etc., all of which are sources of operational inefficiencies and uncertainty that result in increased energy consumption and more importantly, greater food losses.
Therefore, it is desirable to devise an IoT framework that monitors the truck fleet and dynamically solves for the optimal fleet plans both before and during trips, intending to minimize energy losses, food quality deterioration, and customer dissatisfaction.
Funded by the Smart City Research Institute at AUS, this research project was a joint collaborative effort between two teams consisting of faculty, senior students, and research assistants from the Department of Computer Science & Engineering (CSE) and the Department of Industrial Engineering (INE).
INE Team:
Supervisors: Dr. Mahmoud Awad, Dr. Malick Ndiaye
Research Assistants: Ahmed Osman, Arwa Abougharib
CSE Team:
Supervisor: Dr. Ra’afat Abu-Rukba
Senior Undergraduate Students: Yousuf Islam, Muhammad Abdullah Khan, Omar Alsarookh, Abdul Hafez Usman
Several research contributions have been made as a result of this joint effort, excerpts of which are provided below.
A Generic Shelf-Life Estimation Model for Fresh Fruits and Vegetables During Transportation
Supervisors: Dr. Mahmoud Awad, Dr. Malick Ndiaye
During transportation, prediction of the Remaining Shelf-Life (RSL) of Fresh Fruits and Vegetables (FFVs) is critical for planning and quality cost estimation. The Internet of Things (IoT) enables measured environmental variables to be processed in real-time. However, there is a need for a validated, real-time computational method that translates environmental measurements to dynamic RSL estimates. Most existing generic RSL models for FFVs are qualitative, invasive, or static. This study establishes a generic RSL model for FFVs under dynamic and unplanned logistic conditions. Its implementation is non-destructive, non-invasive, and does not require accelerated shelf-life experiments before deployment. In addition, since the original model is rather computationally intensive, a surrogate model was proposed for fast, real-time applications for ‘Edge IoT.’ Experimental validation of the model using three fresh products (strawberries, apricots, and spinach) in a domestic refrigerator resulted in a maximum deviation of 1.3 days in prediction error using the original model and 2.95 days using the surrogate model. Nonetheless, the predictions made using either the original or surrogate models were statistically sound and not significantly different from the observed shelf lives of the samples, even at the 0.01 significance level.
Manuscript Availability: Publicly available here.
Data & Software Availability: The model was implemented in MATLAB and is publicly available, along with a user manual and validation data sets here.
A Dynamic Black-box Refrigeration Temperature Simulator with Limited Training Data in Simulink
Supervisors: Dr. Mahmoud Awad, Dr. Malick Ndiaye
Estimating the Remaining Shelf-Life (RSL) of chilled and frozen products in real-time during transportation is critical to quality monitoring in cold supply chains. The Internet of Things has allowed the ambient temperature to be monitored in real-time during transportation. Recently, many Shelf-Life Estimation Models (SLEMs) have been proposed which use the temperature history to estimate the RSL of a product. However, the ultimate benefit of temperature monitoring is to use it as an input to optimize the truck fleet routing. Decisions are made based on total cost minimization while incorporating cost of poor cold product quality. To aid the integration of SLEMs into optimization efforts, the ambient temperature history of a given trip must be predicted beforehand. This work produced a black-box refrigeration temperature simulator implemented in Matlab and Simulink that can generate a predicted temperature history for any refrigeration system with limited training data. The simulator was calibrated to a domestic refrigerator and was validated experimentally to acceptable accuracy, even under intermittent door opening instances, as would normally be experienced in a refrigerated truck compartment unloading at several customers.
Manuscript Availability: Currently under peer-review.
Data & Software Availability: The model was implemented in Matlab and Simulink and is publicly available, along with a user manual and validation data sets here.
An IoT Application for the Dynamic Routing of Trucks Transporting Fresh Food
For their bachelor’s capstone project, senior CSE students (Yousuf Islam, Muhammad Abdullah Khan, Omar Alsarookh, and Abdul Hafez Usman) accomplished the final objective of this research track by implementing and testing the IoT solution.
The developed IoT application collects sensor data, such as the truck’s GPS location, the temperature in the storage compartment, and the alarm sensor reading (which indicates whether the truck door is open or closed), and publishes it to the cloud. All trucks in the fleet, along with their compartment data (number of compartments, their sizes, optimal temperatures, and which sensors are assigned to each), must be configured on the cloud. Drivers can also be assigned to trucks. These configuration tasks are performed by ‘the configurer,’ one of the user types of the platform.
Another user role is ‘the planner,’ who is responsible for inputting the trip details: Customer order quantities and dates, delivery locations, and delivery time windows. Those will be the inputs to the optimization model, which will run in the background to produce the optimal trip plan. The user interface also allows setting other parameters, such as the fuel cost, the fixed cost of deploying a truck, and the penalty for missing a customer’s delivery window.
The optimization model in the backend runs first when the fleet leaves the warehouse and again whenever a pre-defined flag takes place, e.g., truck breakdown, a prolonged traffic delay, any transported products nearing expiry, etc., to re-calculate another optimal delivery plan. In the case of a truck breakdown or the predicted shelf-life of an item falling below the minimum acceptable shelf-life, the truck is routed back to the warehouse, and a new truck is sent to fulfill the unmet customer demands.
The CSE and INE teams worked closely together to formulate and implement the optimization model and re-routing flags.
A well-prepared demonstration of the IoT app in action made by the senior students is provided below.
Manuscript Availability: In preparation.
Software/Code Availability: To allow for the possible commercialization of the research findings, software or code has not been made available.
Arwa Abougharib 