Ph.D. Thesis Colloquium of
Mr. Prithvirajan D
Research Supervisor: Dr. M. Mathirajan
Date: 12th March 2025 [Wednesday]
Time: 04:00 PM
Venue: Seminar Hall [Management Studies]
Title:
“Development of Analytical Solution Methodologies for a Class of Problems in Last Mile Delivery (LMD) of Liquefied Petroleum Gas Cylinders (LPG-C) in India”
Abstract:
Last Mile Delivery (LMD) is often the most expensive and time-consuming part of logistics, particularly in service sectors like e-commerce, food delivery, and courier services. In India, the distribution of Liquefied Petroleum Gas (LPG) Cylinders (LPG-C), is one of the prominent LMD business. LPG is an essential commodity in India and is widely used for cooking and industrial purposes. The LMD of LPG-C (LMD-LPG-C) is managed by distributors selected by public and private oil marketing companies. Considering the operations related challenges / issues faced by the distributors, in this study, the following 3 decision problems related to LMD-LPG-C are considered:
1. Optimal / Efficient delivery sequencing (DS) of customers’ orders of each of the delivery agents in LMD-LPG-C.
2. Minimizing the restocking distance by proposing Dynamic Temporary Re-Stocking locations (DTRSL) for re-stocking LPG-C.
3. Developing a Prototype Delivery Authentication System (P-DAS) to have a robust delivery authentication system and to prevent black-market deviations as well as improve delivery accountability.
The first problem on DS decision is approached as a Multiple Traveling Salesman Problem and a (0-1) Integer Linear Programming model is developed. Due to computational intractability in getting optimal DS decision using exact method, alternate non-conventional optimization algorithm: Meta-Heuristic Algorithms (MHA): Tabu Search Algorithm (TSA), Simulated Annealing Algorithm (SAA), Genetic Algorithm (GA) and Ant Colony Optimization (ACO) are considered for determining efficient DS decision. The performance of the selected MHA is evaluated in comparisons with the existing DS decision of the delivery agents through a series of empirical and statistical analyses. From the performance analyses, it is observed that the DS decision by the MHA: TSA is giving on an average better solution and is identified as the best performing one.
The second study prescribes Dynamic Temporary Re-Stocking Location (DTRSL) to station the mini trucks and for the delivery agents to collect / restock LPG-C for their routine delivery to customers which would result efficiency of multiple restocking tripes by each of the delivery agents. The customers historic order booking dates and their location details are obtained. Using these data, multiple forecasting methods such as Linear Regression, Decision Tree Regression, and Random Forest Regression have applied to predict the Number of Days between a Customer-Last Order and Expected New Order (ND-CLO-ENO). Out of various forecasting methods applied Random Forest Regression method emerging as the best. Using the ND-CLO-ENO, MAE of the Random Forest Regression and the last ordering date of each of the customers, the Expected Order Date Range (EODR) for each of the customers is determined. Customers whose EODR falling partially or completely within the study period are segregated into three planning period (ten days per planning period). Planning period wise, the location details of the segregated customers and the number of mini trucks operated by the distributor are provided as input to the K-Means clustering algorithm and segregated the customers as cluster of customers and the centroid location of each of the cluster of customers is also obtained and prescribed as DTRSL. The efficiency of the prescribed DTRSL is analysed considering the DS method: TSA in comparisons with actual customer orders received during the study period and empirically showed that DTRSL as re-stocking location offers significant reduction in the total distance travelled by each of the delivery agents.
The third decision problem tackles the lack of the robustness of the current delivery authentication system. Accordingly, the proposed P-DAS system introduces a two-stage authentication process by integrating an Image Classification Module (ICM) and an Image Metadata Extraction and Verification Module (IMEVM). The ICM utilizes advanced image recognition techniques to distinguish LPG-C from similar objects and verifies if the input image is an LPG-C or not. ICM is trained on an image dataset – LPG-C and other similar objects images – using techniques like data augmentation, transfer learning, and CUDA computing, and its performance is assessed with accuracy and f1-score. The IMEVM is a structured logic which is developed to extract the metadata from the input image and verifies if the image is captured within 10 meters from the customer’s delivery location or not. If any of the two stage authentication processes fails, the P-DAS does not authenticate the delivery. The performance of the P-DAS is analysed using case study organisation data and the results highlighted the robustness towards the adaptability of the P-DAS in LMD-LPG-C.
Finally, the study presents the main contributions with managerial implications in applying the proposed analytical solution methodologies for each of the 3 class of problems, a standard operating procedure to implement and adopt the proposed analytical solution methodologies for each of the 3 class of problems, the limitations of the study and future research direction.
ALL ARE WELCOME