Supply Chain Resilience through Small and Medium-Sized Enterprises (SMEs): Representation, Partnership Prediction, and Standardization using Graph Databases

Restricted (Penn State Only)
- Author:
- Kumar, Vishnu
- Graduate Program:
- Industrial Engineering
- Degree:
- Doctor of Philosophy
- Document Type:
- Dissertation
- Date of Defense:
- September 29, 2023
- Committee Members:
- Steven Landry, Program Head/Chair
Soundar Kumara, Chair & Dissertation Advisor
Timothy Simpson, Outside Field Member
Paul Griffin, Major Field Member
Kamesh Madduri, Outside Unit Member
Vijay Srinivasan, Special Member - Keywords:
- Supply Chain
Graph Database
Knowledge Graph
Machine Learning
Link Prediction
RDF
Manufacturing
BioPharma - Abstract:
- As globalization continues to shape the world, supply chains have become increasingly complex, driving critical sectors such as manufacturing, food, and healthcare. At the same time, the global business disruptions triggered by recent events such the COVID-19 pandemic, wars and economic sanctions, have clearly shown the need for manufacturing enterprises to be self-sufficient, resilient, and less dependent on the global supply chains. In the race to recover from these disruptions and to build a resilient supply chain, manufacturing companies are now moving towards a localized ecosystem involving Small and Medium sized Enterprises (SMEs). However, the large enterprises are often devoid of the knowledge and mapping of local or regional SMEs that can be integrated into the supply chain ecosystem. This has created a barrier for effective localization. SMEs on the other hand struggle to integrate themselves into the supply chain, get discovered and enter into partnership with large enterprises due to the lack of resources and effective marketing services. To bridge the research gap in the domain and facilitate the establishment of resilient supply chains, this dissertation aims to address the following fundamental questions: 1. What is the significance of SMEs in the context of localization endeavors, and what opportunities do they present? 2. What computational framework needs to be developed to effectively integrate and represent SMEs in a supply chain ? 3. What database schema is required to search for and filter SMEs within the supply chain representation? 4. What methodologies should be developed to help SMEs predict and select potential partners? 5. How can the framework for an SME supply chain be standardized to enhance interoperability? The first question is comprehensively examined through a case study centered on the innovative biotechnology platform for COVID-19 vaccine production. This case study not only sheds light on the localization initiatives undertaken by pharmaceutical giants but also underscores the pivotal role played by SMEs in the vaccine manufacturing process. Furthermore, this research delves into the specific contributions and significance of SMEs within the PA BioPharma domain, facilitating a precise understanding of the challenges faced by pharmaceutical giants and highlighting the opportunities available to SMEs within a localized supply chain. To address the second and third problems, this dissertation starts with the development of an innovative and robust framework developed using Neo4j Graph database application to effectively represent SMEs and integrate them with the supply chain. This establishes a well-defined method to capture both the company details and their relationship information. To make full use of the graph database as a decision-making tool, the graph database is enriched with semantics to form a Knowledge Graph. The Knowledge Graph representation enables the addition of relationship types and definitions and allows for the development of an effective query-based search mechanism within the supply chain ecosystem. Common questions and specific information, that the enterprises are likely to seek from the knowledge graph, can be programmed as “human readable” dynamic search phrases into the Neo4j database and these can be invoked, as necessary. The performance of this search mechanism built using graph database was evaluated using a variety of simple and complex multi-conditional query prompts and the results were found to be accurate and superior to some of the existing solutions. In response to the fourth problem, the dissertation then develops a machine learning model using XGBoost classifier to predict potential partnership opportunities for enterprises. The proposed model represents partnership prediction as a link prediction problem and is robust enough to use any existing information and engineer new features based on network topology to predict new partners. The performance of this model is evaluated using various metrics and on comparison, was found to perform better than the existing solutions. Additionally, this work uses SHapley Additive exPlanations (SHAP) to enhance the interpretability of the model by highlighting the impact of each input feature on the prediction. Finally, by using Resource Description Framework (RDF), a World Wide Web Consortium (W3C) standard, along with the addition of widely used Schema.org vocabulary, this dissertation presents a novel methodology to standardize the supply chain framework to enhance its interoperability, thereby addressing the final problem. By examining BioPharma SMEs in Pennsylvania (PA) as a compelling use case, this dissertation presents a proof-of-principle implementation that tackles all the above problems. This will be the first step towards building a more resilient supply chain involving SMEs.