Emergent Behaviors in a Resilient Logistics Supply Chain
Rob Bemthuis is a PhD student in the department Pervasive Systems. Promotors are prof.dr. P.J.M. Havinga from the faculty of Electrical Engineering, Mathematics and Computer Science and prof.dr. M.E. Iacob from the faculty of Behavioural, Management and Social Sciences.
This PhD dissertation explores the susceptibility of supply chains to disruptions, such as pandemics, natural disasters, and geopolitical issues. These disturbances influence logistics operations, compelling stakeholders to adjust to shifting behaviors, operational conditions, and the challenge of managing incomplete information. Such disruptions can trigger a domino effect, impacting multiple entities. The dynamic nature of these networks, often perceived as complex socio-technical systems akin to integrated electric grids and transportation networks, can intensify these challenges.
Resilience, a dynamic and evolving process influenced by numerous factors, is a key characteristic that enables stakeholders in the logistics sector to endure and potentially prosper amidst these challenging circumstances. Nevertheless, the design and enhancement of these socio-technical systems for resilience pose challenging due to their complexity, dynamic nature, interconnectedness, heterogeneity, and the emergence of unpredictable behaviors.
Part I (Chapter 1) discusses advancements in smart logistics, encompassing the design of resilient supply chains, the incorporation of Internet of Things (IoT) technologies for real-time decision-making at the network’s periphery, and the adoption of principles derived from the Physical Internet vision. These advancements facilitate rapid information acquisition about unforeseen events, recovery from disruptions, and response to emergent behaviors.
A practical example that highlights the contributions within this dissertation is the concept of ‘smart pallets’. These are business objects, equipped with sensing and computational capabilities, and are typically distributed near the network’s edge. These pallets adhere to a “business logic”, which is the course of action implemented on these smart business objects. This logic outlines policies for real-time decision-making in real-world scenarios. Smart pallets provide valuable insights into the movement and condition of goods during transportation and can enable new services based on real-time monitoring, detection, analysis, and prediction of emergent patterns.
This dissertation investigates three approaches to real-time decision-making: centralized, decentralized, and hybrid. The centralized approach involves a central entity that regulates all realized business logic (e.g., business rules), ensuring uniformity but potentially overlooking local nuances and facing resistance due to perceived infringement on individual freedoms. An example could be a single supply chain partner who controls the entire network of smart pallets. In contrast, the decentralized approach allows each entity to make its own decision based on real-time data, allowing for local adaptability but could lead to disparities in effectiveness due to variations in resources and compliance. For instance, smart pallets could be individually fine-tuned based on their traversed paths and faced conditions, while not necessarily favoring network-wide opportunities and optimizations. The hybrid approach combines elements of both, with a form of central government that sets broad guidelines and restrictions for the network, while allowing individual entities to make autonomous decisions within those boundaries. This approach balances network-wide coordination and local adaptability, potentially leading to acceptable solutions without undermining effectiveness. For example, smart pallets could learn from and adapt to changing conditions based on knowledge obtained through a central coordination mechanism.
This PhD dissertation is motivated by the challenges in supply chain logistics, with a focus on real-time decision-making to augment the resilience of logistics organizations. The research uses smart business objects, capable of capturing and anticipating emergent behaviors near locations of disruptive events. These objects are exemplified by smart pallets. Therefore, the objective of this dissertation is:
How can smart distributed business objects, which capture business logic and facilitate the discovery, guidance, and anticipation of emergent behaviors, augment the management of resilience in real-time decision-making for supply chain logistics organizations?
The term 'augment' is used to imply an augmentation rather than a direct improvement. This perspective shifts from quantifying resilience through Key Performance Indicators (KPIs) to actively managing it. The unpredictable nature of emergent behaviors and the multidimensional aspects of resilience are taken into account. Particularly, the focus is on three stages in addressing a disruption: robustness/resistance, stability/recovery, and adaptation/benefit.
Business logic is articulated in terms of business rules that are interpretable by humans. This research contemplates three viewpoints on business rules: the evaluation of predefined business rules, the mining of business rules from performance outcomes, and the use of these evaluations and extracted rules to iteratively develop new ones. We put forth several approaches that concentrate on resilience stages as well as business rule perspectives.
Part II (Chapter 2) introduces a reference architecture that integrates distributed business logic with centralized control mechanisms. This architecture is designed to identify, monitor, and respond to emergent behaviors. A systematic literature review was conducted to comprehend the existing architectural practices for managing emergence. The insights gleaned from this review guided the design of the architectural blueprint, which serves three main objectives: facilitating the application of the proposed architecture across diverse settings, providing a framework that integrates specialized proposed methodologies into a broader enterprise vision, and offering a practical context for incorporating and expanding the proposed techniques into an enterprise.
Chapter 2 also aligns the solution implementations in subsequent chapters with this reference architecture. This alignment, along with additional contributions, addresses multiple facets of solutions related to managing emergent behaviors in enterprise environments. Our IT-artifacts also contribute to augmenting resilience management, with a focus on real-time decision-making within supply chain logistics.
Part III focuses on evaluating the effectiveness of business rules. In Chapter 3, an architecture is instantiated within a logistics case study to assess the efficacy of business rules in real-time decision-making across distributed business objects. The study examines the equilibrium between centralized decision-making and decentralized autonomy, regulated by business rules. Simulations evaluate the impact of disruptions and scrutinize various mitigation strategies at a decentralized level. The results indicate that disruptions can be partially mitigated by implementing reactive and proactive strategies, but improvements in performance are marginal. The implementation of business rules allows the system to adapt to new patterns, guiding the intended emergence while preserving the autonomy of individual decision-making processes.
Chapter 4 diverges from Chapter 3 by focusing on the design of specific components associated with the evaluation of business rules in the context of the reference architecture. An architecture that uses process mining techniques and a multi-agent system is proposed to facilitate real-time decision-making in dynamic environments. The study illustrates how modifications to business rules affect the system's emergent behavior, highlighting the system's robustness and adaptability. By analyzing event logs collected at the individual agent level, process models are mined that disclose how the collective behaviors of agents affect system-wide dynamics. The application of this architecture is demonstrated with a case study of a job-shop factory equipped with automated guided vehicles. The study investigates the effects of diverse agent-design choices using process mining algorithms and examines how the decisions of the agents impact the overall system performance.
Part IV presents two chapters on a posteriori methods for mining business rules based on performance outcomes. Chapter 5 uses a decision tree machine learning approach to mine business rules from IoT device data, as demonstrated in a case study on smart pallets. The method is accurate, flexible, and produces human-interpretable decision trees. Chapter 6 addresses the extraction of business rules from complex, multi-agent systems. It proposes a method for mining models that encapsulate individual agent behaviors from data fragments, or "event logs". This method is applied to Schelling's model of segregation as a case study in logistics scenarios. The research indicates that examining these data fragments enables effective extraction of rules that define agent behavior, enhancing our comprehension of emergence in agent systems.
Part V (Chapter 7) discusses an iterative adaptation of business rules, using a framework that leverages emergence insights from physical-level sensing technologies. It employs a functional decomposition approach, incorporating three types of agents: cyber-physical controller, business rule management, and emergent behavior detection. The approach examines emergence through discovered business processes and uses insights from process models to augment agents' decision-making capabilities. A simulation study in cold chain logistics illustrates that agents can learn and adapt decisions based on knowledge of macro-behaviors, acquired through process mining techniques.
Part VI (Chapter 8) synthesizes this research and offers key managerial insights. These include the adaptability of business rules, the context-dependent nature of resilience, the necessity for IT architecture to evolve alongside organizations and technologies, the importance of continuous evaluations and modifications in reference architecture, and the benefits of integrating process mining with multi-agent systems.