Introduction

The current business landscape demands companies to maintain their competitiveness by adapting to rapidly evolving market conditions, satisfying increased quality and sustainability expectations, and accommodating mass customisation within condensed development cycles (Araz et al., 2020). These global trends present an unique set of challenges, but none more so than the recent COVID-19 pandemic. The outbreak has caused major disruptions to the balance of supply and demand, highlighting the vulnerability of global logistics networks (Ivanov & Dolgui, 2020) and the need for continued research and advancement in the field of supply chain resilience. Although the strong post-pandemic rebound of global trade growth has levelled off, the recovery is uneven around the world. This has been amplified by political frictions and regional agreements which are expected to reshape international trade patterns (UNCTAD, 2022).

The importance of acknowledging the coordination and interdependence between various operations when working to enhance supply chain resilience and sustainability cannot be overstated (Gurnani et al., 2012). Nonetheless, this approach is limited to conventional logistics networks, which operate independently and are dedicated to a single company (Sarraj et al., 2014) or a small group of companies in close collaboration through business arrangements such as joint ventures (Kleindorfer & Saad, 2009). These networks are shaped by long-term strategic decisions that encompass a broad range of factors, including the location and capacity of production, storage, and distribution facilities, which cannot be easily altered. As a result, these logistics networks are considered fixed and exhibit inherent structural vulnerabilities, as companies may struggle to quickly reconfigure their operations in response to disruptions (Sohrabi et al., 2012). The reliance on fixed, dedicated networks with long-term planning horizons leaves companies unable to quickly adapt to changes in demand or disruptions, resulting in inefficiencies and waste. These limitations make it imperative for companies to explore more sustainable alternatives in order to achieve a more adaptive and efficient logistics network. Even in instances where collaborating companies exhibit some level of synergistic resilience and adaptability in their network structure, their operations are still limited by the fixed nature of their logistics networks, and any disruption can have a cascading effect throughout the entire network (Sarraj et al., 2014; Peng et al., 2021). The COVID-19 pandemic was a good example of lack of resilience of traditional facilities. While restaurant distributors were stuck with food surplus, grocery shops could not cope with the demand.

Together, these global trends and structural vulnerabilities render questionable the resilience and sustainability of logistics networks. The Physical Internet paradigm seeks to tackle this by reorganising current global logistics practices. Montreuil (2011) identify thirteen symptoms of unsustainability in logistics, from an economical, environmental, and social perspective, and how the Physical Internet vision can address these challenges. The underlying principle is universal interconnectivity (Montreuil et al., 2013) of logistics networks and services which implies a reorganisation of current logistics practices (Montreuil et al., 2010). Existing practices are marked by a limited ability to effectively handle disruptions, highlighting the need for more innovative and dynamic approaches that prioritise the allocation of logistics resources. However, the physical infrastructure of the Physical Internet, which consists of modular containers and standardised material handling equipment envisioned in this paradigm (Montreuil et al., 2014), is not yet available. Henceforth, these tangible components of the Physical Internet, such as sites, facilities, and physical systems will be referred to as nodes, as described by Montreuil et al. (2010). Still, when considering the evolution of the Physical Internet as illustrated in ALICE (2020), companies could leverage existing digital technologies and existing material handling equipment to tackle with comparable efficiency the unsustainable logistics practices.

Besides the research gap on early development stages of the Physical Internet, there is little research on new logistics practices that focus on allocating logistics resources dynamically at the node level. Moreover, there is little research on how to reorganise the inter-dependencies between logistics nodes, and specifically, how to employ digital technologies in material handling systems to achieve digital interoperability between nodes. This includes comprehensive frameworks that present how the underlying features of the Physical Internet lead to the development of more sustainable and resilient global logistics practices (Treiblmaier et al., 2016, 2020; Dong & Franklin, 2021; Chen et al., 2021; Ballot et al., 2021). Also, most studies only focus on freight transportation and truck scheduling at the network level (Fazili et al., 2017; Chargui et al., 2020; Lemmens et al., 2019; Chadha et al., 2021). The node is where the routing and processing of goods takes place and therefore it requires a careful design and it is a key asset for logistics networks. In this context, Cyber-Physical Systems (CPS) could offer a promising avenue for addressing these challenges by seamlessly integrating digital technologies into material handling systems. The study of Sgarbossa et al. (2020) introduced the concept of Cloud Material Handling Systems (CMHS) and presented a prototype and its operation model. However, the study was restricted to a brief description which does not highlight the wider implications that could occur at the network level.

Research aim

The aim of this study is to explore the benefits of leveraging digital technologies and CPS applications in logistics and develop a conceptual model of a Cloud Material Handling System for efficient and resilient logistics networks. The proposed model aims to leverage the synergy between real-time location tracking and assignment of properties, metrics and constraints that could enable the dynamic resource allocation in material handling. In turn, this leads to a more levelled and continuous node operation, as well as the ability to quickly adapt to changes or disruptions in the network. This mechanism would also promote asset and infrastructure sharing between companies. The benefit of having an intelligent material handling system at the node level is the ability to monitor the node performance and introduce performance metrics for each shipment that flows through the node. Thus, the routing of shipments is done according to the capability and capacity of each node, and each order is processed by optimising the system goal, based on the requirements of all the orders in the logistics network. To manage such a dynamic environment, it is important to learn from previous states how the material handling system behaves. By using predictive analytics, the system can adapt to changes in demand or disruptions in the logistics network. In this context, the system can continuously learn from previous states, making better decisions about how to process each order and allocate tasks, and ultimately improve the efficiency of the network.

The following research questions are addressed as part of this study:

  1. 1.

    What is the current state-of-the-art in the literature regarding the application of CPS and digital technologies in logistics?

  2. 2.

    How can the application of cyber-physical systems contribute to efficient and resilient material handling practices?

  3. 3.

    What are the potential applications or practical implications of the proposed conceptual model?

The first objective is to identify what capabilities are enabled by digital technologies and CPS applications in logistics. Then, the second objective is to propose a new conceptual model that addresses the research gap and systematically describes the mechanisms involved in leveraging emerging technologies and CPS to achieve efficient and resilient material handling practices. The last objective is to provide an example of practical application in a logistics network.

Contributions

The first contribution of this study is the classification of existing studies on CPS applications in logistics based on several key capabilities extracted from the literature. This classification provides a comprehensive overview of the existing knowledge and understanding of the use of CPS in logistics.

The second key contribution of the paper is the development of a conceptual model for CMHS, presented as a potential solution for enabling interoperability in logistics networks. The authors expand and provide context to the idea by putting forward a model that leverages the use of digital technologies to capture real-time information about nodes, introduce performance metrics and constraints, and use machine learning (ML) as analytical technique, with the goal of achieving interoperability as envisioned in the Physical Internet.

The conceptual model offers several theoretical contributions. First, it provides a systematic representation of how the interactions between physical assets, material handling processes and digital technologies within a logistics network can improve resilience and efficiency. Second, it integrates theories on Physical Internet, CPS applications and use of digital technologies in logistics and introduces a new perspective considering traditional material handling equipment and emerging technologies to achieve interoperability. Last, the model can serve as framework for future research, as it provides basis for developing case studies and simulations.

Methodology

The methodology used to develop the conceptual model involved examining related work on the topics of digital interoperability and applications of CPS in logistics. The choice is motivated by the fact that the end goal is to achieve interoperability by leveraging the use of digital technologies embedded with physical processes. Moreover, the scope was set to cover applications in logistics in order to investigate similar concepts not only in material handling systems, but also in manufacturing and production-logistics environments.

Paper structure

The remainder of this paper is structured as follows: Section “Literature review” begins by laying out the theoretical foundations on digital interoperability in the Physical Internet and applications of (CPS) in logistics, then structures the findings based on several identified capabilities in order to position the proposed conceptual model along the same criteria. Section “Conceptual model of the cloud material handling system” presents the conceptual model, focusing on describing each functional layer. Section “Practical applications in the Physical Internet” examines the application of the proposed concept in the context of the Physical Internet. Lastly, Section “Conclusions” discusses the implications and limitations of the study and highlights promising future research avenues.

Literature review

The methodology employed in the literature review was a qualitative, non-systematic analysis of papers published between 2012 and 2022. The search was conducted using the Scopus database and relevant combinations of keywords such as “cyber-physical system”, “digital interoperability”, “Physical Internet”, “logistic networks”, “real-time information and analytics”, “cloud computing”, “dynamic resource allocation”, “machine learning”, and “smart material handling system”. The literature review was conducted to identify the most relevant papers on these topics. In order to ensure comprehensive coverage, the review employed snowballing to discover additional sources in the identified papers, leading to a more thorough analysis of the literature. After the content analysis, 52 papers were included in the final analysis. The study aimed to understand the depth of knowledge and the papers were classified, according to their scope, in application at Node Level and/or Network Level, and according to the environment in Production and/or Material Handling. From the analysis, eight capabilities, enabled by digital technologies and CPS applications in logistics, were identified. Then, the publications were categorised based on the characteristics of each model or framework presented in the studies. The key capabilities identified in the literature served as starting point for developing the CMHS conceptual model, which was created using a function-based layered architecture that describes the inter-dependencies between physical and digital assets, with the aim to address the problem statement in Section “Introduction”.

Digital interoperability in the Physical Internet

The concept of the Physical Internet, as defined by Montreuil (2011), is a new logistical approach aimed at reorganising supply chain networks to improve reliability and resilience. Unlike traditional supply chain networks, which are fixed and limited to a single company or small group of collaborating companies (Sohrabi et al., 2012; Yang et al., 2017), the Physical Internet allows for dynamic access to the most appropriate facilities across a global network. By breaking free from the constraints of dedicated supply chain networks, companies can now respond dynamically to unexpected events and changes in their business environment through access to the best facilities worldwide. This shift enhances the reliability and resilience of logistics networks, leading to improved overall supply chain performance.

Despite its newfound relevance in both industry and academia, the Physical Internet is still a relatively new and complex concept, incorporating multiple emerging technologies and having a wide range of applications. A number of literature reviews have attempted to bring structure to the scattered research on the Physical Internet. For instance, Sternberg and Norrman (2017) emphasised the benefits of innovative Physical Internet business models and an organisation’s readiness to adopt the Physical Internet based on their technological advancement. Treiblmaier et al. (2016), on the other hand, aimed to outline the main areas of research in the Physical Internet, focusing on real-world implementation, optimisation of hub operations, and future research challenges. Additionally, Treiblmaier et al. (2020) conducted a systematic literature review and developed a comprehensive framework that organised the Physical Internet literature.

The Physical Internet was envisioned as a network of interconnected logistics networks, akin to the Digital Internet. The goal is to establish standardised and universal interconnections between logistics services and networks. The Physical Internet draws inspiration from the Digital Internet’s principles of data transmission and infrastructure, utilising universal and standardised protocols and interfaces for both digital and physical connectivity across the global logistics network. The flow of goods through the Physical Internet is facilitated by open, visible and shared resources such as vehicles and facilities, as well as data, which enables seamless interoperability (Ballot et al., 2021). It is worth noting the distinction from the Internet of Things (IoT), which involves connecting physical entities to the Digital Internet. In this sense, the IoT acts as an enabling factor for the Physical Internet, extending visibility and openness beyond a company’s information systems. Other enablers include the Internet of Services for logistics services (Schroth & Janner, 2007), and cloud computing to store, process and transfer data (Armbrust et al., 2010).

Several papers formulate definitions for the Physical Internet (see Montreuil et al. (2013); Ballot et al. (2014); Treiblmaier (2019). Notably, resilience is not mentioned as a key performance indicator (KPI) besides efficiency and sustainability. Therefore, in a broader sense, the Physical Internet can be defined as an open global logistics system that is founded on physical, digital and operational interconnectivity, and aimed at improving the efficiency, sustainability and resilience of supply chain management operations. This system is driven by technological, infrastructural and business innovation, and facilitated by standardised and optimised physical components that exchange information through encapsulation, interfaces and protocols.

A simple graph-theoretic model put forward by Dong and Franklin (2021) serves as a straightforward illustration of the workings of the Physical Internet. The model depicts a directed graph in which the vertices and edges represent nodes and transportation segments, respectively. To each edge, a weight is assigned, represented by a vector of multiple elements which correspond to performance metrics for a given route (e.g., cost, lead time, emissions, capacity) at a given time. Similarly for every vertex, there is a corresponding weight vector which represents the performance metrics of a node’s operations. The weights for every vertex (infrastructure element) and edge are updated in real time. Therefore, the system must calculate the best routing decision for every shipment in the Physical Internet given various objective functions (Dong & Franklin, 2021). The complexity of such an undertaking would entail addressing various routing, re-consolidation, storage, and transportation mode decisions for all material flows in the network, using information gathered in real-time from the infrastructure.

Pan et al. (2021a) investigated how digital interoperability is a key enabler of the Physical Internet, and defined it as “the ability to achieve quick, seamless, secure, and reliable data and information exchange between computing devices (viz. devices being able to transfer data), between information systems (of organisations, infrastructures, logistics networks), or between devices and systems, for the aim of enhancing cooperation or coopetition of independent logistics parties or networks”.

Given recent advancements in digitalisation, the fundamental characteristics of digital interoperability could be achieved through implementations of CPS, which have seen increased attention in the literature for their application in logistics. Additionally, cloud computing is another complementary area for digital interoperability, through which real-time information processing can be achieved using scalable, on-demand computing resources (Borangiu et al., 2019). In turn, vast computational capabilities make room for predictive analytics (Zhong et al., 2016), federated learning and privacy preserving approaches for inter-organisational data sharing (Rodríguez-Barroso et al., 2020). Ultimately, these advancements influence design principles of existing supply chains. To this end, Dolgui et al. (2020) put forward a comprehensive framework of a re-configurable supply chain or X-Network which, at the structural level, is strongly defined by the digital infrastructure composed of CPS and IoT.

Applications of cyber-physical systems in logistics

Cyber-physical systems are embedded computers and networks integrated with physical systems, which monitor and control the physical processes with feedback loops that create inter-dependencies between physical processes and computations (Lee, 2006). This technology is capable of complex interactions with the physical world and has seen advanced applications in almost all areas (Lee & Seshia, 2016).

The academic literature on applications of CPS distinguishes three different areas – manufacturing, material handling, or an integrated approach for production-logistics systems. While there is a large volume of published studies on this topic for manufacturing (Yao et al., 2019) or production-logistics systems (Monostori et al., 2016), a relatively small body of literature that is concerned with material handling systems.

Lee et al. (2015) proposed a now prominent CPS architecture, namely the 5C architecture, which provides a detailed and practical structure for implementation in manufacturing, and could potentially be applied in other areas, like distribution and material handling. The architecture clearly defines five functional requirements of a CPS: smart connection for data acquisition, data conversion to meaningful performance indicators, data analytics at cyber level, decision support, and self-configuration for corrective and preventive actions. Building on this work, Redelinghuys et al. (2020) presented a six-layer digital twin architecture and a case study implementation that show how data can be exchanged between cyberspace and the physical twin. Similarly, Ding et al. (2019) introduced a Digital Twin-based cyber-physical production system for real-time and data-driven operations control in manufacturing. The framework leverages the interconnection between parts and resources which have smart and autonomous capabilities. CPS are also used in smart manufacturing for job shop scheduling. Fang et al. (2019) introduced an architecture and working principle of a novel job shop scheduling method by leveraging the detection and analysis of dynamic events, as well as the feedback loop between the physical workshop and its Digital Twin. Zhong et al. (2017) investigated the use of Big Data Analytics in manufacturing shop floors under the Physical Internet paradigm and presented an overall framework for the use of Big Data collected by the Physical Internet environment. In order to maximise the use of production resources and deal with disturbances effectively, Zhang et al. (2016) proposed a CPS for manufacturing shop floor based on the multi-agent technology.

A number of authors have considered the implications of real-time data from manufacturing and material handling in production-logistics systems. Zhang et al. (2015a) proposed a mechanism to capture real-time information from the manufacturing environment and optimise the material handling tasks. Similarly, Li and Huang (2021) developed an implementation framework for production-intralogistics synchronisation that leverages real-time information to support decision-making and synchronous operations in flexible assembly lines. Other authors (see Zhang et al. (2015b, 2020)) draw on the idea of real-time information capturing to build a layered architecture of a CPS that integrates data from physical resources, as well as various production subsystems to provide visible and traceable processes through insightful analysis of manufacturing big data. Building on the framework by Ren et al. (2019), Wang et al. (2020) proposed an architecture of a proactive material handling method which integrates predictive analytics and the ability to identify the real-time status of manufacturing and logistics resources in order to allocate material handling tasks based on a predicted future status of the manufacturing system.

Deployment of cloud computing for real-time information capturing and analytics

Other researchers have gone further and investigated the use of cloud computing for real-time information capturing and analytics in the application of CPS.

Cloud computing is the on-demand access to a shared pool of flexible, easily deployable and configurable computing resources that can be managed with little management work or interaction. This is made possible by the ability to deploy vast computing capabilities when needed, without human interaction with the service provider. These capabilities can be quickly scaled up or down and are available over the network through standard mechanisms. Organisations can use cloud infrastructures that are provisioned for private use, collaboratively by a group of companies that have shared goals and interests, or as a hybrid infrastructure. The latter offers a flexible environment for different types of workloads with specific portability, security, privacy and resilience requirements and purpose (Mell & Grance, 2011).

The adoption of cloud computing technologies marked the era of smart manufacturing by introducing new and more efficient business practices. Moreover, it led to the emergence of the Cloud Manufacturing paradigm, which is the manufacturing equivalent of cloud computing. Thus, manufacturing resources are encapsulated and provided as cloud services across all stages of a product life cycle (Xu, 2012). Several studies have developed architectures for cloud manufacturing service systems (see Tao et al. (2011, 2014a); Wang et al. (2021)). Luo et al. (2022) proposed a cloud simulation architecture for solving the dynamic resource allocation problem in flexible manufacturing lines. Rojas and Rauch (2019) proposed a conceptual framework and cloud-enabled CPS architecture for smart manufacturing control. Mourtzis and Vlachou (2018) proposed a system that uses a cloud platform for storage and data visualisation. The resulted insights are used for adaptive scheduling and condition-based maintenance. Similarly, but at the network level, Tao et al. (2014b) proposed an architecture describing how intelligent identification, monitoring, and management of manufacturing resources can be achieved, aggregated and used across enterprises in order to efficiently allocate manufacturing resources through the use of a cloud platform. While the Cloud Manufacturing paradigm has been well-established in the literature, attention has only recently shifted towards a more integrative supply chain vision. Ivanov et al. (2022) have introduced the concept of the “cloud supply chain”, which aims to optimise the design and management of a supply chain network by leveraging the sharing of both physical and digital assets.

In the context of material handling, most studies scope smart factories, focusing on the integration of real-time machine data with material handling activities. Guo et al. (2020) proposed a self-adaptive collaborative control mechanism for integrating production and logistics resources in a manufacturing environment. Pan et al. (2021b) built on this work to develop a digital twin enabled by cloud computing for the real-time monitoring and control of a production-logistics system. Qu et al. (2016) developed a production-logistics synchronisation system in a cloud manufacturing environment which enables the use of the real-time data capturing and dynamic resource management. Similarly, Guo et al. (2017) developed an integrated framework for self-adaptive production-logistics systems based on a cloud service platform that enables the monitoring and matching of key production and material handling tasks. Taking advantage of the vast computing resources available on the cloud, Luo et al. (2017) investigated the use of a cloud-based decision model to address the simultaneous optimisation of production and material handling scheduling. Wan et al. (2017) analysed a cloud-based architecture for context-aware cloud robotics for material handling in smart factories, where tasks are scheduled using traditional algorithms. Fewer studies focus solely on material handling activities enabled by cloud technology. Zhao et al. (2017) introduced the concept of cloud forklifts which are characterised by smart identification, communication, and information sharing on a cloud-based management platform. The result is a decision support system for warehousing activities that improves visibility and traceability in warehouse management. Zhang (2018) presented the advantages of a cloud computing platform for the collection, storage, and sharing of information from a logistics system to enable data analytics and traceability.

Use of machine learning for dynamic allocation of tasks

Material handling systems often operate in stochastic environments with both internal and external variability which may affect decisions at the strategic, tactical, or operational level. In practice, companies must address daily challenges arising from the need for real-time response on order information updates (Gong & De Koster, 2008) or internal variance in waiting time, routing, or capacity (Gong & de Koster, 2011). This adds another layer of complexity as the number of material handling tasks increases Mařžk and Lažanský (2007). Traditionally, predefined dispatching rules have been used for controlling the material handling movement. However, their performance depends on the environment state at a given time (Le-Anh & De Koster, 2005), since there is no single rule that supports all possible states of a stochastic environment (Priore et al., 2014). Instead, dispatching should be done dynamically, in the presence of real-time events (Ouelhadj & Petrovic, 2009).

A major shortcoming of dynamic dispatching is the NP-hard nature of the problem which cannot yield a feasible solution within a reasonable time (Qin et al., 2021). Alternatively, meta-heuristics are used for dispatching problems to achieve a sufficiently good solution, but they require extensive knowledge and human intervention (Wang et al., 2019). The emergence of Industry 4.0 has prompted much interest in the use of predictive analytics to address such problems, with numerous studies on the applications of ML.

Machine learning is the part of artificial intelligence devoted to building computational methods that leverage data to improve performance and make predictions or decisions without explicit, predefined rules (Murphy, 2012). Deep learning is a ML method that uses an artificial neural network to learn a function by transforming a set of inputs into a set of outputs (Deng & Yu, 2014). Reinforcement learning is a learning framework that enables an agent to learn the optimal behaviour given an objective through interaction with the environment, similarly to a trial-and-error approach (Sutton & Barto, 1998). The use of deep learning and reinforcement learning methods is a well-established approach in achieving human-level or super-human artificial intelligence systems for a wide range of applications. The DRL method integrates reinforcement- and deep learning using artificial neural networks for function approximation. This approach can be used in many applications for decision making under uncertainty (Matsuo et al., 2022), being a good choice in complex and dynamic environments where the DRL agent can learn from experience and can handle high-dimensional inputs. In dynamic resource allocation, the agent learns to allocate resources based on the current state of the logistics networks, the orders that need to be fulfilled, and the priorities associated to these orders. Once the agent is trained, it can be deployed in a real-time environment and remain relevant even when the optimal solution may change over time. A particularly suitable area for this approach is supply chain management. According to Rolf et al. (2022), publishing trends have shown a sharp increase in the number of publications on reinforcement learning in supply chain management, particularly to address inventory management, vehicle routing and scheduling problems.

Jeong et al. (2021) proposed an architecture for smart production-logistics and analysed the use of reinforcement learning for material handling in manufacturing environments. The paper presented a Q-learning algorithm for determining the optimal path of Automated Guided Vehicles (AGVs). Similarly, Flores-García et al. (2021) developed a CPS architecture integrating ML for the dynamic scheduling and execution of material handling in smart production-logistics and described how reinforcement learning can be used for path planning. A mixed rule dispatching approach based on a deep Q-network was developed by Hu et al. (2020) to minimise the makespan and delay ratio of AGVs by almost 10% compared to the benchmarks. Hu et al. (2021) proposed a self-adaptive traffic control model using reinforcement learning for dispatching material handling tasks for AGVs and enhance collision avoidance. Zhang et al. (2018) suggested the potential applications of ML for self-organising production-logistics where manufacturing resources can actively respond to disturbances. Li et al. (2022b) addressed the dynamic flexible job shop scheduling problem using DRL to minimise the makespan and total energy consumption.

Lee et al. (2018) proposed an IoT-based warehouse management system that uses ML as predictive analytics for information processing and decision support. Malus et al. (2020) proposed a real-time task dispatching using reinforcement learning that enables dynamic dispatching of Autonomous Mobile Robots (AMRs) according to the layout and order arrivals. Feldkamp et al. (2020) presented an approach for using simulation with DRL in a modular production system to improve dispatching of tasks to AGVs and minimise lead times. Using simulation modelling and DRL, Mikkelsen and Dahl (2021) investigated the benefits of implementing a CMHS in a manufacturing environment. The study comprised of several scenarios depicting different stochastic activities in the shop floor and tested the performance of a policy-based dispatching compared to traditional dispatching rules. The CMHS outperformed traditional dispatching rules in terms of throughput and utilisation of material handling equipment.

A summary of the main findings from the literature review, together with the identified capabilities enabled by digital technologies and CPS applications in logistics, is provided in the next section.

A summary of literature findings

The studies presented thus far provide evidence that there is a large and growing body of literature investigating the applications of CPS in logistics. However, most studies remain narrow in focus, dealing only with local optimisations. The literature section began by looking at studies on the Physical Internet and digital interoperability. It went on to highlight some fundamental characteristics of digital interoperability and the emergence of CPS in logistics as a mean to achieve it. Three application areas emerge from the studies: manufacturing, material handling, and an integrated approach for production-logistics systems. As was pointed out in the previous section, few studies are concerned solely with CPS applications in material handling systems, much of the literature focusing particularly on the integration of manufacturing resources with material handling activities. Therefore, the focus is mostly on integrating within a node, and not externally with other companies. In order to integrate outwards, eight capabilities enabled by digital technologies and CPS in logistics have been identified in the literature, namely real-time monitoring and control, data visualisation and analytics, predictive analytics, digital interoperability, dynamic resource allocation, node and network KPIs, resource pooling, and agile and re-configurable networks. These capabilities should be considered collectively to achieve interoperability in material handling systems. Additionally, they could be utilised to map the development stage in the adoption and integration of digital technologies and CPS in logistics, as shown in Fig. 1. This approach is also corroborated by findings from Schuh et al. (2020) that describe the natural development stages of Industry 4.0, but in that case with less granularity than here.

Fig. 1
figure 1

Development Stages in Achieving High Integration and Interoperability in a Logistics Network. Distribution of Research Papers by Identified Capabilities

Full size image

As companies start capturing information by locally integrating computers and networks with their physical assets, they unlock the possibility to monitor and control them, as well as extract meaningful insight through data visualisation and analytics. Data can be used for predictive analytics in decision support systems and to dynamically allocate production and material handling resources. Digital interoperability can thus be achieved locally, and between nodes, as companies integrate their information systems outwards and enhance cooperation in the network. At this stage, not only can organisations extract and share meaningful network KPIs, but material handling resources can be pooled and dynamically allocated between nodes. Ultimately, companies can react quicker and logistics networks can be reconfigured by dynamically allocating capacity to flexibly accommodate changes in customer demand or hedge against major disruption events.

Collectively, these studies outline a critical role for CPS applications in logistics, but their focus and scope are rather limited to leveraging real-time information locally to dynamically allocate resources within the node, as seen in the number of papers in Fig. 1. A holistic view over the implications of leveraging real-time information is needed in order to achieve high interoperability and consequently more autonomous material handling and distribution.

Table 1 Selection of relevant identified papers from the literature - classification by scope, environment, and capabilities
Full size table

While the capabilities identified in these papers do not always build on each other in the literature, the model presented in this study attempts to incorporate all of them in order to achieve a high level of integration and interoperability. For comparison, Table 1 samples the most relevant papers from the literature review, as described in Section “Introduction”, based on scope, the application environment, and the capabilities identified in the literature.

In this section, the analysed body of literature was classified according to the application areas and identified capabilities. Also, the limitations of the existing literature were emphasised by assessing the level of adoption and integration of CPS in logistics. The following section moves on to describe in greater detail the structure and functions of the proposed conceptual model.

Conceptual model of the cloud material handling system

CMHS is defined as a CPS that facilitates the movement, storage, protection and control of materials by dynamically allocating logistics resources to material flows using cloud computing, machine learning algorithms and real-time information about the material flow, and its requirements and constraints. The objective is to enable operation in a Physical Internet context by achieving digital interoperability and providing open, visible, resilient, sustainable and re-configurable node services, as well as more levelled and continuous node operation. Not only is the CMHS enabled by cloud technology, but it must also be regarded as an analogy to cloud computing, i.e., the material handling version of cloud computing, since the proposed mechanism effectively enables the use of material handling as a service. Several essential characteristics should be present in a CMHS, namely the cloud material handling should be regarded as an on-demand service available throughout a logistics network, where resources are pooled to serve all participating actors, are dynamically allocated, and are measured to automatically optimise and balance capacity.

Figure 2 shows the high-level architecture of the system, consisting of five functional layers with specific blocks representing groups of elements at each level, as well as the information flows between them. These elements are represented based on their scope, as either node-specific if their function is limited to a given node, or network-specific if they are visible and operate at every node in the Physical Internet. The Physical Layer encompasses material handling equipment and unit loads in a node. The Communication Layer is comprised of the communication technologies used to connect the Physical Layer with the Digital Layer, facilitating real-time location tracking and condition monitoring. The Digital Layer has the function of gathering and integrating information from the logistics network infrastructure, at different aggregation levels. The Analytics Layer uses the data collected in the Digital Layer to determine the optimal allocation of resources in each node within the network, as well as the routing of goods through the network given the desired performance metrics. It provides real-time insights on node and network KPIs for decision support and visibility. Lastly, the Solution Layer presents this concept as an innovative logistics and supply chain management solution, offering advantages like dynamic resource allocation, improved decision-making, visible and levelled node operation, as well as improved adaptability within the logistics network. This approach aligns with the results of the literature review, showcasing how digital technologies and CPS enable capabilities such as real-time monitoring, predictive analytics, dynamic resource allocation, and enhanced cooperation. The model aims to achieve a more sustainable, resilient, agile, and digitally inter-operable material handling system capable of meeting diverse logistics network demands and constraints. Each layer is further described in the next sub-sections.

Fig. 2
figure 2

Architecture of the cloud material handling system

Full size image

Physical layer

This layer includes all the material handling equipment in a node, as well as the unit loads flowing through the node.

Material handling equipment

The material handling equipment refers to any logistics resources used to move and store goods within the node. It spans a spectrum of options based on flexibility and automation levels, covering everything from manual operator with manual racks, to forklifts, AGVs and AMRs.

Unit loads

The unit loads can be any standardised or widely accepted material handling entities, like pallets, totes, bins, containers, moving racks, which can be handled with existing material handling equipment (Sgarbossa et al., 2020). The location of each unit load can be indirectly determined by monitoring the real-time location of the material handling equipment using an indoor positioning system, and by triggering events on pick-up, release, or when passing through specific checkpoints, gates, or other locations that affect the status of the unit load. Alternatively, their real-time location can be directly monitored through the indoor positioning system. However, the latter involves strict management of the antennas and unique identifiers and associated metrics allocated to each unit load, in scenarios where reconsolidation is required.

A unit load has a unique identifier based on a standard protocol within the system. Consequently, useful properties, requirements, constraints and metrics (Dong & Franklin, 2021) can be associated, like:

  • Type of equipment that can handle the unit load

  • Origin and destination points, both inside the node and at the network level

  • Decision variables in the form of weighted key performance indicators (KPIs)

Communication layer

The Communication Layer is comprised of the communication technologies used to connect the Physical Layer with the Digital Layer, specifically for the indoor positioning system and condition monitoring system. By tracking the location of each material handling equipment and unit load in real time using an indoor positioning system (IPS), more useful data can be utilised for meaningful insights about the system, thus enabling digital interoperability and dynamic resource allocation within the node.

Communication technologies have seen rapid development, and wider adoption will make them even more cost-efficient in the future (Farahsari et al., 2022). These technologies can be classified in short- and long-range (Brena et al., 2017), the former being most suitable for indoor positioning systems (Sadowski & Spachos, 2018). In logistics environments, Radio-frequency identification (RFID), Bluetooth Low Energy (BLE) and Ultra-wideband (UWB) are the most common communication technologies for localisation (Karaagac et al., 2017). Wi-Fi is also a viable option in cases where high accuracy is not needed (Yan et al., 2019).

RFID is a widely studied communication technology with multiple tags that emit data and communicate with a reader. Although it cannot support real-time localisation, RFID technology can work with UWB-based indoor positioning to reduce costs (Huang et al., 2020). UWB technology is not susceptible to interference from other signals, making it particularly suitable for indoor positioning. It provides high accuracy for moderate power consumption (up to 10–20 cm) (Djosic et al., 2021), but relatively short range. However, the signal can penetrate walls and a variety of materials, and the system can be scaled by increasing the number of receivers inside the facility. Because it requires additional hardware to operate, it bears a higher cost compared to other solutions (Zafari et al., 2019). BLE is the preferred alternative for IoT applications because of its low energy consumption, high transfer rate and wide coverage area (Zafari et al., 2015). However, it has lower overall localisation accuracy. It performs better than WiFi, but only for short distances, making it less suitable for large areas in industrial applications (Yun et al., 2018). WiFi is widely available on most gadgets and portable user devices, making it one of the most investigated technologies for positioning applications in the literature. Nevertheless, WiFi is mainly used for communication rather than positioning systems and thus more efficient algorithms are needed to improve its accuracy (Zafari et al., 2019).

Digital layer

The Digital Layer has the function of gathering and integrating information from the Physical Internet infrastructure, at different aggregation levels. It consists of real-time information about the indoor positioning and condition monitoring systems on a given node, as well as the performance metrics, requirements and constraints of the nodes and network. The purpose is to leverage cloud computing to use this information for advanced analytics and ultimately reach not only a higher level of autonomous information and goods transfer through the Physical Internet, but also efficient resource allocation at the node and network levels.

Indoor positioning system

The real-time location of each material handling equipment (resource) and unit load within a node can be accurately tracked using an IPS. At the node level, knowing the position of each resource and unit load in real time implicitly determines the relative distances between them and enables the system to dynamically allocate resources following the most appropriate policy, like the closest resource to a given unit load (Sgarbossa et al., 2020). Also, in environments with high uncertainty and variability of material handling activities, real-time position monitoring generates insights into where the areas with higher traffic are in the facility, such that available resources can be more efficiently positioned to respond quickly to new tasks.

For an IPS to be effective, several evaluation metrics are proposed in the literature, like availability, cost, energy efficiency, reception range, localisation accuracy, latency, scalability, robustness (Zafari et al., 2019; Farahsari et al., 2022), as well as security and privacy (Yassin et al., 2016; Alarifi et al., 2016).

Condition monitoring system

The proposed architecture takes into account useful information that can be retrieved from a condition monitoring system used to measure the performance of material handling resources inside a node.

This approach aligns with current practices in the implementation of CPS in smart factories, where machines and other objects are monitored (Lee et al., 2015). Condition monitoring systems support operators in performing the most suitable maintenance actions at the appropriate time (Tang et al., 2013). These insights help reduce the risk of downtimes by monitoring the performance of each asset (Cao et al., 2019). Therefore, such insights from condition monitoring systems could be used in the dynamic resource allocation process, so that these job allocation decisions take into account capacity reductions from a potential breakdown or scheduled downtime.

Performance metrics

One of the most important features of a CMHS is the use of performance metrics for node and network operations. Dong and Franklin (2021) introduced a graph-based model to describe the flow of goods in the Physical Internet according to user preferences and performance of each node of the system. To elaborate along this line, the performance metrics used in the CMHS can be classified into user metrics and infrastructure metrics.

The user metrics refer to the performance requirements of each user (i.e., cargo owner) in the Physical Internet. Based on their preference, each user finds an acceptable performance level for different logistics metrics like cost, lead time, emissions. This is usually formulated as weights and should be ranked accordingly.

The infrastructure metrics are related to the performance of logistics resources within a node in the network, as well as that of transportation segments between nodes, which together define the performance of the Physical Internet. Relevant metrics include the real-time capacity of each node and the status of the logistics resources. Therefore, while the user metrics are static and would not be changed, the infrastructure metrics are constantly updated as the unit loads flow through the system, influencing how resources are allocated.

The performance metrics in both categories will serve as input in the Analytics Layer, for the optimisation of the network for the common goal, rather than focusing on local or short-term gains. The DRL model should find the optimal solution based on the performance of the network as a whole, given the individual preferences in terms of logistics metrics for each user in the Physical Internet. The resources of each node would then be dynamically allocated according to these metrics, and the capacity of the entire network could be utilised in a more balanced way.

Requirements and constraints

Similar to the performance metrics, certain requirements and constraints must be defined for the dynamic allocation decisions. These requirements and constraints refer to the limits imposed by the functional and physical characteristics of unit loads and their compatibility with the material handling equipment. They serve as input in the Analytics Layer and should be continuously updated reflect any changes in the material handling system and to ensure no misalignment occurs between the material handling equipment and unit loads, and how the latter are processed through the node. The constraints should follow standard protocols across the network, such that the CMHS operates with consistency in each node.

Hybrid cloud computing service platform

The CMHS is enabled by the use of cloud computing. A hybrid cloud computing service platform, together with the ML models deployed on the platform, would be the foundation of the CMHS and a way for the participating companies in the network to achieve digital interoperability while preserving the desired portability, security and privacy. Given the low latency requirements of real-time localisation and the dynamic allocation of resources by incorporating user and infrastructure metrics, the application of a CMHS and the ML models it relies on for dynamic scheduling require significant computing power and increased bandwidth. Edge computing can be used, thus reducing the need to continuously transfer large amounts of data between servers. While operation at a single, isolated node, requires edge computing on-premise, the operation of an entire network must capitalise on a cloud computing platform, hence the hybrid cloud architecture of the entire system. A valid approach for a CMHS implemented at a network level could be to train ML models at the edge and use the cloud to manage and centralise the models for federated learning. This is further described in Section “Machine learning models”.

Analytics layer

The data collected in the digital layer are used as input in the ML model to determine the optimal allocation of resources in each node within the network, as well as the routing of goods through the network given the desired performance metrics. Therefore, the Analytics Layer has the function of data visualisation and analytics, providing real-time insights on relevant node and network KPIs for decision support and visibility throughout the network.

Machine learning models

As described in Section “Literature review”, decisions at the strategic, tactical, or operational level can sometimes be influenced by the unpredictable conditions in which material handling systems frequently operate. Consequently, task dispatching should react to real-time events (Ouelhadj & Petrovic, 2009). Meta-heuristics are commonly used to solve such problems, but they become increasingly unreliable when the number of parameters and constraints increase (Qin et al., 2021; Wang et al., 2019). Recent advancements in the field of artificial intelligence have prompted researches to turn their attention to the DRL framework for its ability to receive large inputs corresponding to a large number of possible actions for each state in a decision-making problem. To date, it is the most adequate ML approach to solve the task at hand. The choice for this learning framework is motivated by the increased complexity of the resource allocation problem which stems from the high number of metrics and constraints for each user in the Physical Internet network.

In the context of a Cloud Material Handling System using DRL for order processing and task allocation at node level, some key elements can be broken down, as follows. The state is defined as the situation comprising of the relevant information about the node for the system to make decisions (i.e., take actions), including the state of resources, unit loads, and orders. The model would take in inputs from several sources. First, the decision to allocate a material handling task to a resource is driven by their physical constraints, position and performance, as output by the indoor positioning and condition monitoring systems. At the same time, the user- and infrastructure metrics drive the decisions for both the order handling and individual material handling tasks within a node. The actions would involve selecting the next order to process and allocating specific tasks to appropriate resources. Each action for material- and order handling would yield a reward based on its appropriateness in a given situation. The rewards should be designed to reflect the performance objectives of the logistics network, such as assembling orders efficiently, optimising resource utilisation, minimising delays, all while meeting the user metrics. The system would learn the approximate optimal policy that maximises the cumulative reward over time, reflected by the aforementioned performance objectives.

By using federated learning (Yang et al., 2019), the ML models can be trained at the node level without having to share sensitive information between nodes (when they belong to different organisations). Therefore, this decentralised approach allows the whole logistics network to build the same ML models while maintaining data privacy and security (Li et al., 2020). The global ML model would take inputs from each node, such as the node and network KPIs and performance metrics to decide the routing of shipments between nodes in order to level the capacity of the network. While federated learning can be both collaborative and adversarial, in this context we discuss the former, while the latter is a matter of cyber-security which is valid any time Industry 4.0 topics are discussed. However, these aspects are not addressed in this paper. Despite potential malicious attacks, current research show promising results in increasing robustness of federated learning methods (Li et al., 2022a; Zhang et al., 2023).

Material handling policies

The dynamic allocation process consists of choosing the most appropriate action given the current state, that would maximise the system goal. The material handling policies are the result of the state-action pairs (see Section “Machine learning models”) of the DRL algorithm and make up the optimal strategy for task allocation to each material handling equipment. Also, the idle policy for each state-action pair should indicate the next action from a list of predefined possible areas to return to when a resource is idle, while waiting for the next material handling task.

Order handling policies

Similar to the material handling policies, the optimal policy drives order handling. Each order has a priority which is driven by the performance metrics (both infrastructure metrics and user preferences). This information is part of the state at a given time. Different mechanisms can be implemented for computing order priorities. Besides the performance metrics, the model can use a function that would increase the priority of an order the longer that order has waited to be fulfilled. This would prevent some orders with low priority from never being fulfilled.

While at task level the performance of a system might not see considerable improvement, it has the potential to reorganise the way orders are fulfilled by enabling resource pooling and interoperability between nodes, since the infrastructure metrics provide insights on the inbound and outbound flows at each node.

Node and network KPIs

The output of the ML models act as a feedback loop on the initial Performance Metrics that are used as input. Given the dynamic nature of such a system, the initial Performance Metrics are updated, and also fed into the Analytics Layer for visibility and decision support, hence the Node and Network KPIs block in this layer. In other words, the Node and Network KPIs are an updated version of the previous Performance Metrics which they will replace at each feedback loop. For example, in a distribution centre, real-time information from the Digital Layer can be used to measure KPIs such as resource utilisation rate, resource downtime, order cycle time, or product throughput, as well as real-time capacity and material flow visibility in this node. This would allow for better control of shipments in the network, as they can be routed according to changing conditions or disruptions.

Solution layer

As a solution, this concept has the potential to bring about innovation in logistics and supply chain management. Its implementation offers several advantages over traditional systems, such as dynamic resource allocation, increased efficiency in decision-making, a more levelled and continuous node operation, as well as improved visibility and re-purposing of node services. The integration of real-time location tracking, cloud computing and ML algorithms could result in more optimised decision-making processes and a material handling system that is capable of meeting the demands and constraints of different actors within a logistics network. The combination of these features can result in a more sustainable, resilient, agile and re-configurable system that is capable of achieving digital interoperability and operating within a Physical Internet context.

Dynamic resource allocation

The integration of real-time monitoring and control of physical assets through CMHS can lead to a significant improvement in the dynamic resource allocation of material handling resources. By capturing and integrating real-time information from physical assets, companies can optimise their material handling resources and dynamically allocate them based on current demand and network KPIs. This results in a more sustainable and cost-effective use of resources, which ultimately improves the overall performance of the network. Available information about the state of a node enables resources to be dynamically allocated to different areas and tasks in the facility, especially in highly uncertain environments. At network level, capacity gaps or excess capacity can be identified and adjustments can be made accordingly. This enables a redistribution of resources and optimised capacity allocation based on the network’s current needs, through rerouting of shipments based on real-time needs and demand patterns and will lead to a more levelled node operation, as described further in Section “Levelled / continuous node operation”.

Visible/open node services

The architecture of this system would allow for visibility into every node in the Physical Internet network through the up-to-date node and network KPIs. The material handling capacity at each node should be regarded as an on-demand service available throughout a logistics network, where resources are pooled to serve all participating actors and are measured to automatically optimise and balance capacity. Such visibility enables the routing of goods through any suitable node according to the initial performance metrics and constraints of each shipment, with possible re-consolidation where this would yield higher gains for the system, allowing nodes to serve shipments for any user in the Physical Internet. By monitoring and controlling the physical assets and pooling the material handling resources, the information systems can extract meaningful network KPIs and react quicker to changes in customer demand or major disruption events.

Levelled/continuous node operation

The Physical Internet implies an inherit balancing of the network, and this can specifically be achieved through a CMHS. Traditional allocation of resources based on fixed policies, and limited visibility between actors, lead to “bulky” flows of goods through the logistics network, and consequently, seasonal demand on each node in the network. The dynamic allocation of resources through real-time data could lead to a more levelled and continuous node operation, where shipments arrive and leave nodes in a more uniform manner, allowing networks to better respond to disruptions or sudden changes in demand. This results in a more levelled and continuous operation of nodes, as well as more open and visible node services. While distributing the load across competing actors could sound unfeasible, competition should be a key factor in shaping the infrastructure of the logistics network and in the way material handling as a service is made available. First, it ensures that a node can handle as many goods as possible without congestion. This can help improve the overall resilience and efficiency of the network. Secondly, it can improve the utilisation levels of the material handling equipment without over-investment in capacity. Lastly, it could prompt logistics service providers to offer better, more reliable services.

Re-purposing of node function

Conventional logistics networks function as standalone entities and are usually owned and operated by one or a few closely collaborating companies. These networks are constructed based on strategic decisions with long-term outlooks and are not easily altered, making them rigid and inflexible. As a result, organisations may struggle to adapt and reconfigure their networks in response to unforeseen events. Nodes with CMHS can be re-purposed to quickly adapt to changes or disruptions in the network, enabling similar interoperability in the Physical Internet, but with traditional material handling equipment and unit loads. This function is mainly enabled by moving away from a model of ownership to a model of using material handling as a service, which offers several benefits. First, it reduces the need for large capital expenditures on equipment and technology. This allows companies to allocate their resources more effectively and focus on their core competencies. Additionally, it provides them with better services, as providers are incentivised to constantly develop their offerings to remain competitive.

The re-purposing of material handling capabilities refers to finding alternative uses or applications for existing equipment in times with excess capacity or major disruptions in one industry. The Analytics Layer offers valuable insights into the node operations and can support process optimisation by identifying bottlenecks, areas of improvement and re-configuring layouts. Moreover, companies can identify alternative/temporary use of excess material handling capacity for cross-functional use, beyond their original purpose, by handling, for example, different types of goods. A distribution centre that uses forklifts to primarily move pallets can be used to handle barrels, large containers, components or other irregularly shaped items. Conveyor belts that are used for transportation can be re-purposed for sorting or assembly processes. Material handling resources can be used for implementing reverse logistics or parcel flows. Other re-purposing examples include investigating opportunities to share material handling resources with other actors or adapting equipment to suit other industries. For example, equipment originally used for food supply chains can be used to handle manufacturing or healthcare goods.

Agile and re-configurable networks

CMHS could have significant benefits and implications at the network level because it enables a more efficient and flexible logistics network, as resources can be quickly reallocated in response to changes in customer demand or disruptions. The visible and open node services allow for better collaboration and cooperation between nodes, leading to the emergence of agile and re-configurable logistics networks. This means that the network can more easily adapt to changes in customer demand and hedge against major disruptions, which results in a more reliable and efficient network.

Decision support

The dynamic resource allocation at the node level, enabled by CMHS, has far-reaching benefits and implications for companies and entire logistics networks. At node level, managers can extract meaningful insight from their real-time material handling operations, such as their overall service score, utilisation and down-times, and base their tactical and strategic decisions on, with respect to scaling capacity or deploying improvement projects. At network level, companies can react better to changes in business environment and adjust their operations by taking more informed decisions at tactical and strategic level about resource allocation, risk management and deployment or decommissioning of nodes.

Practical applications in the Physical Internet

This concept has the potential to bring benefits to distribution networks following principles of the Physical Internet. Focusing on the node level, CMHS could have practical applications at distribution centres, which handle tasks such as receiving, put-away, replenishment, order picking, packing, sorting, unitising, cross-docking, and shipping. It should be noted that the inherent ability to adapt in a timely and controlled manner to patterns in demand and disruptions distinguishes this model from traditional operations and its working principle could be implemented regardless of the handling technology.

The implementation of CMHS in a distribution centre environment can significantly impact operations and efficiency. The real-time tracking of product, operator and equipment locations, which comprise the Physical Layer of a node, enhances order fulfilment by allowing the distribution centre to allocate tasks more efficiently. In practice, each pallet, tote, bin, or even single units are typically tagged with RFID. These unit loads, together with each material handling equipment (pallet jacks, forklifts, manual trolleys, AGVs, AMRs), storage equipment (automated storage and retrieval systems, vertical lift modules, simple racks), picking, labelling, controlling, and packaging stations, as well as the loading and unloading docks and operators are identified and tracked via indoor positioning systems in the Digital Layer of the node. This information allows the access to a more insightful overview of the performance of the distribution centre, such as the throughput time, resource utilisation, waiting time, congestion, or availability.

The Hybrid Cloud Computing Service Platform in the Digital Layer will pool together the information of the distribution centre and other nodes in the network. Locally, this information can be used for monitoring and control, and will be integrated with the Enterprise Resource Planning (ERP) system, Warehouse Management System (WMS) or Manufacturing Execution System (MES) where the orders and tasks are dispatched. The ML approach (i.e., DRL) on the platform allocates resources dynamically, leading to operational and strategical benefits. Fluctuations in material and order flows can be smoothed out through the day by reallocating resources and operators, leading to a more flexible operation. The tracking of inbound and outbound trucks also provides valuable information for the DRL algorithm, improving the loading and unloading process. For example, by reallocating resources to the receiving- or storage area throughout the day, based on the expected arrivals of inbound trucks, the unloading and storage process times could improve substantially. Similarly, the workload can be levelled by reallocating operators from picking activities to replenishment so that products are available for picking, thus avoiding any waiting time due to product unavailability. The DRL algorithm takes into account relevant real-time information about the distribution centre, including the state of the resources, unit loads, order priorities, inbound/outbound flows, and assigns tasks accordingly. For example, the DRL algorithm can assign a task to a nearby worker, reducing fulfilment time. Moreover, real-time tracking of product locations optimises storage and retrieval, leading to a more efficient order fulfilment process. Operators can quickly determine product availability and spend less time searching, improving the overall picking efficiency.

This flexible operation of the distribution centre allows for open and visible services and provides network-level feedback, enabling cost-effective routing and decision-making. The Digital and Analytics Layers cut across the information silos of isolated nodes and provide visibility into the entire network, as illustrated in Fig. 3. At tactical level, the node KPIs can be used to have flexible capacity and capabilities in terms of material handling. Managers can extract meaningful insight from the real-time operations and make tactical and strategic decisions on scaling capacity or deploying improvement projects. Having a network-level feedback through open and visible node services, goods can flow through different distribution centres in the network in order to keep a more levelled and continuous operation. This leverages the holistic view of the entire network and the capabilities of its nodes given the cost-benefit trade-off of re-routing and re-consolidation of shipments between nodes. In the event of major disruptions, the ability to reconfigure the network becomes critical. The function of the distribution centre can be more quickly reconfigured to take some of the rerouted shipments in order to ensure that the network can still operate. Managers can use this for decision support to minimise the impact of disruptions on their operations by collectively implementing contingency plans. Additionally, any excess capacity due to disruptions or demand fluctuations can be re-purposed to serve other industries or processes, as described in Section “Re-purposing of node function”, enabling similar capabilities as the Physical Internet.

Fig. 3
figure 3

Three-dimensional map of nodes with CMHS within a Physical Internet network. The Digital and Analytics Layers cut across the information silos of isolated nodes to achieve digital interoperability

Full size image

Conclusions

This study set out to address the challenge of inefficient material handling and information silos in logistics networks that lack reconfigurability and agility, and are therefore vulnerable to disruptions. The aim was to investigate how CPS could enable interoperability within logistics networks, specifically in material handling systems, by exploring the inter-dependencies between physical and digital asses.

The study provided a comprehensive overview of the existing knowledge on CPS applications in logistics. The literature review suggested that although there is a large and growing body of literature on this area, most studies remain narrow in focus, dealing only with local optimisations. Few studies are concerned solely with CPS applications in material handling systems, and therefore, this study focused on integrating within a node and externally with other companies. To achieve a high level of integration and interoperability, the study highlighted the importance of considering eight key capabilities which include real-time monitoring and control, data visualisation and analytics, predictive analytics, digital interoperability, dynamic resource allocation, node and network KPIs, resource pooling, and agile and re-configurable networks. These served as the foundation for the development of the CMHS model which aims to enable the dynamic resource allocation in material handling and promote asset and infrastructure sharing between companies, leading to a more levelled and continuous node operation, as well as the ability to quickly adapt to changes or disruptions in the network. Lastly, the paper explored practical applications of the conceptual model in the context of a distribution centre. The adoption of the proposed model could lead to more efficient, resilient and sustainable logistics practices, and promote asset and infrastructure sharing between companies.

Limitations

It must be noted that this paper has certain limitations that prevent it from offering a complete overview of all the existing literature on the use of CPS in logistics. The authors had to consider several practical constraints, such as scope and time, which limited the depth and breadth of the review. However, the study provides a robust and insightful examination of the most relevant and recent advancements in this field. Despite these limitations, the authors remain confident in the ability of this paper to provide valuable insights.

The reader should bear in mind that the paper presents a contribution by offering a conceptual model. It is important to acknowledge that conceptual models, while useful in providing a theoretical framework, have inherent limitations in terms of accuracy and applicability to real-world situations. These limitations stem from the simplified nature of conceptual models, which often overlook or abstract certain complexities and uncertainties in the system being modelled. Additionally, the assumptions and premises used to develop the model may not always hold true in practice. Despite these limitations, this model serves as a valuable tool for promoting future research on the topic.

Future research

The development of advanced ML algorithms is a crucial area of future research for the CMHS conceptual model. The implementation of such algorithms could significantly enhance the decision-making processes of the CMHS and lead to improved overall performance. This line of research would benefit from a systematic approach, incorporating simulation modelling, to evaluate the efficacy of various algorithms in real-world logistics scenarios. The results of such studies would provide valuable insights into the design and implementation of CMHS systems, ultimately contributing to the realisation of a more efficient and effective logistics network.

Another potential research avenue that could be explored in relation to CMHS is the development of different mechanisms for real-time monitoring and control. Given the importance of accurate real-time information for effective material flow management, there is a need to find innovative ways to collect, process, and utilise this information. This could involve exploring the use of various indoor positioning systems at each facility in the logistics network to track the movement of materials and ensure they are being handled efficiently. Additionally, establishing standard protocols for these systems could help to ensure compatibility and improve overall network performance. Researchers could explore the integration of CMHS with block-chain technology. This integration could bring a number of benefits, such as increased transparency and security in the management of material flows, as well as improved efficiency in terms of real-time monitoring and control. Additionally, integrating with block-chain could also help to establish standard protocols and streamline the communication between different facilities in the logistics network.

Lastly, evaluating CMHS in real-world scenarios is crucial for understanding its effectiveness. This can be done through pilot programs and case studies, validating the hypotheses in the conceptual model while assessing its performance against established metrics such as total throughput of the system, resource utilisation, response time to disruptions, and recovery metrics. The research could also assess the improvement in visibility and traceability achieved by CMHS, which is important for supply chain management and ensuring the smooth flow of materials. The research could also evaluate the interoperability enabled by CMHS in a Physical Internet context, assessing how well it enables different actors in a logistics network to work together.