Hybrid AI for Supply Chain Risk Identification in Industry 4.0

Abstract

Over the years, the complexity of supply chains has increased due to globalization, out-sourcing, and the integration of digital technologies, which has introduced different types of risks and uncertainties. The process of Supply Chain Risk Management (SCRM) provides a systematic approach that involves the identification, assessment, mitigation, and monitoring of supply chain risks. To develop effective SCRM strategies, organizations should utilize relevant data, information, and expert knowledge from both internal and external sources. The integration of Industry 4.0 technologies has improved the collection, storage, and management of large volumes of data. However, the effective utilization of this data and knowledge remains challenging due to high volume, distribution into multiple fragments, and heterogeneous formats. Previously, various analytical approaches such as mathematical programming, operational research, and simulation have been used in SCRM, however, these methods require structured numerical data and are limited in their ability to process unstructured information or expert knowledge. On the other hand, Artificial Intelligence (AI) offers advantages over traditional analytical approaches by addressing their limitations. AI can be broadly categorized into statistical and symbolic AI approaches. Statistical AI techniques, such as machine learning and deep learning, are effective in identifying patterns from large datasets. Symbolic AI techniques, such as ontologies and knowledge-based systems, support domain knowledge representation and offer explainability. However, both approaches face challenges when applied individually. Statistical AI techniques require large amounts of historical data for modelling and lack explainability, while symbolic AI suffers from knowledge acquisition and scalability challenges. This thesis explores the application of Hybrid AI, which combines statistical and symbolic AI approaches to support supply chain risk management. The first contribution of this research is a systematic literature review that analyses the existing application of statistical and symbolic AI techniques in SCRM. The review focuses on the types of risks and the risk management tasks supported by AI. The second contribution proposes a semi-automated ontology development approach using academic literature. This study involves the integration of Large Language Models with expert supervision for supplier risk management. The final contribution presents a hybrid AI-based conceptual framework that combines a transformer model with a knowledge graph to support contextualized and explainable risk identification in supply chains. In order to validate the proposed framework, a case study on the semiconductor chip shortage has been carried out which demonstrates how early warning signs from news sources can be transformed into actionable risk insights for supply chain managers.