Intermodal Loading Units (ILUs)

• Intermodal Loading Units (ILUs) are standardized transport units designed for seamless transfer across ships, trains, and trucks, crucial for intermodal network efficiency.
• Recent research applies stochastic optimization and chance constraints to manage ILU capacities, reliability, emissions, and risk in dynamic transport systems.
• Advances in ILU identification via deep learning and in loading/unloading algorithms enhance terminal efficiency and support modular freight and specialized applications.

Intermodal Loading Units (ILUs) are transport units designed for easy transfer between different transportation modes, such as ships, trains, and trucks, without unloading or reloading their contents. In current research usage, the term encompasses standard shipping containers, interchangeable bodies (swap bodies), and semi-trailers, and it is closely tied to standardization, routing, terminal handling, identification, and capacity planning in intermodal systems (Gülsoylu et al., 22 Sep 2025). In operational models, ILUs often appear as the primary unit of demand and flow, so that costs, capacities, deadlines, emissions, and reliability are all expressed at the level of the container or analogous transferable unit (Gbadegoye et al., 21 Mar 2025).

1. Definition, scope, and standardization

The broadest contemporary definition treats ILUs as transport units built for transfer across modes without repacking. The computer-vision review on ILU identification explicitly includes standard shipping containers, interchangeable bodies, and semi-trailers, and links their interoperability to standards such as ISO 668 for physical specifications and EN 13044 for identification markings (Gülsoylu et al., 22 Sep 2025). A common misconception is that ILUs are equivalent only to maritime containers; the recent literature instead uses the term as an umbrella category for multiple standardized freight units.

A central element of ILU standardization is identification. The same review states that the globally standardized identification code is the ISO 6346 code, consisting of four letters, six digits, and one check digit, and that each ILU type must display this ISO6346-compliant ID on both sides to enable identification regardless of orientation or handling (Gülsoylu et al., 22 Sep 2025). This identification layer is not ancillary: it is operationally necessary for tracking, gate processing, terminal monitoring, and customs-related workflows.

In specialized transport domains, function can matter more than nominal classification. For commercial spent nuclear fuel and high-level waste transport, transportation casks or overpacks are described as the primary containment and transport units and are said to function in the role of ILUs because they can be transferred between rail, barge, and heavy-haul truck without repackaging the contents (Gadey et al., 2023). This does not redefine the standard freight meaning of ILUs, but it shows that the operational logic of intermodal transfer can extend to highly regulated cargoes.

2. ILUs as the unit of flow in intermodal networks

Several recent models make ILUs the explicit quantity moved through a network. In reliable road–rail routing, the intermodal network is represented as a directed graph $G=(N,A)$, where nodes include major highway intersections, major rail junctions, and intermodal terminals, and demands and flows are expressed in number of intermodal containers (Uddin et al., 2024). In that formulation, shipments travel from origins accessible only by road to intermodal terminals, transfer to rail, may transfer again at a terminal, and finally return to the road network; every intermodal route will involve at least two intermodal terminals (Uddin et al., 2024).

A complementary tactical formulation models container movement from multiple origins to designated intermodal hubs under fixed train schedules and uncertain demand and train spot capacities. In that setting, ILUs are routed from each supply location $i$ to eligible train stations $j$ for trains $n$ at time $t$, and the variables $x_{ijn\omega t}$ represent the number of containers shipped from $i$ to $j$ for train $n$ at time $t$ in scenario $i$0 (Gbadegoye et al., 21 Mar 2025). The model does not perform explicit dynamic routing optimization; truck trips follow fixed or known paths to stations, and the scheduling problem is instead to match available containers with train slots under time and capacity constraints (Gbadegoye et al., 21 Mar 2025).

The ILU abstraction also appears in staged intermodal logistics. In the logic-based Benders framework for intermodal shipment and last-mile delivery, orders are consolidated into loading compartments, which function operationally as ILUs; each order is assigned to exactly one loading compartment, and each compartment is dedicated to a specific distribution-center hub while being shippable to any satellite hub over scheduled roadway, railway, and seaway services (Avgerinos et al., 2022). In tactical railroad planning, the unit is again the intermodal container: the railcar fleet-management model focuses mainly on 40-foot and 53-foot containers, which arrive at origin terminals and must be loaded onto railcars and transported to destinations before a due date (Kienzle et al., 23 Oct 2025).

3. Optimization frameworks for ILU planning and control

The dominant analytical treatment of ILUs in recent work is mathematical programming under uncertainty. A two-stage stochastic optimization model for road–rail freight distinguishes first-stage preparation decisions from second-stage recourse. Before uncertainty is realized, the planner decides how many containers to prepare at each origin and preliminarily assigns origins to intermodal hubs or stations. After actual demand and train capacities are known, the planner can re-allocate or adjust shipped ILUs across the network, subject to capacity, with penalties for unmet demand and excess emissions (Gbadegoye et al., 21 Mar 2025).

That model is explicitly risk-aware through Conditional Value-at-Risk. With random total system cost $i$1 and confidence level $i$2, the paper uses

$i$3

where $i$4 (Gbadegoye et al., 21 Mar 2025). The objective trades off expected cost and risk through $i$5, with $i$6 corresponding to risk neutrality and $i$7 to risk aversion, while also penalizing emission-limit violations (Gbadegoye et al., 21 Mar 2025).

Reliable routing under disruption adopts a different uncertainty formalism. In the road–rail model, highway-link, rail-link, and terminal capacities are random variables, and chance constraints are used so that flows will not exceed realized capacities with a user-specified confidence level (Uddin et al., 2024). The paper derives distribution-free planned capacity reductions from the reliability target $i$8 and uncertainty parameter $i$9, yielding

$j$0

This framework allows decision makers to determine the amount of capacity reduction to consider in planning routes to obtain a user-specified reliability level (Uddin et al., 2024).

Decomposition methods have been particularly important when ILU planning is coupled to downstream operations. In the logic-based Benders model, the master problem handles ILU assignment, consolidation, and movement through scheduled intermodal services, while the subproblems solve last-mile routing after ILUs arrive at satellite hubs (Avgerinos et al., 2022). This staged formulation exploits the fact that intermodal shipment and last-mile delivery are tightly coupled but structurally distinct.

4. Capacity, reliability, emissions, and resource realism

The recent literature emphasizes that ILU planning is constrained simultaneously by capacity uncertainty, service reliability, emissions, and equipment realism. In the two-stage stochastic road–rail model, selected constraints include supply limits, train-capacity limits, time windows for catching fixed departures, demand satisfaction, and an emissions cap with penalty: $j$1 The case study reports that transportation costs dominate at 58% of aggregates, while supply or handling costs are also significant at approximately 35%; higher $j$2 values reduce exposure to worst-case costs but increase average cost, and increasing train capacity strongly reduces both total cost and unmet demand up to a threshold beyond which benefits plateau (Gbadegoye et al., 21 Mar 2025).

In reliability-oriented routing, terminals emerge as especially critical ILU-handling nodes. The model finds that disruption at intermodal terminals has a greater impact on network performance and cost than disruptions to rail or road links of similar disruption size; when all terminals are disrupted, intermodal transport is not possible for containers and only direct road shipment remains (Uddin et al., 2024). The same study also finds that total system cost increases with the level of capacity uncertainty and with increased confidence levels for disruptions at links, nodes, and intermodal terminals (Uddin et al., 2024).

Railroad tactical planning adds another layer of realism by integrating loading constraints and heterogeneous railcar fleets. The scheduled service network design with resource management formulation shows that ignoring railcar fleet management or container loading constraints can lead to severe underestimation of required capacity: 2–6% when empty moves are ignored, and 17–27% when loading rules are also ignored (Kienzle et al., 23 Oct 2025). The same experiments show that multi-platform railcars improve overall capacity utilization and benefit the network, even if they can locally lead to less efficient loading as measured by terminal-level slot utilization performance indicators (Kienzle et al., 23 Oct 2025). A related misconception is therefore that terminal-level slot utilization is a sufficient indicator of ILU efficiency; the railroad study states the opposite.

5. Identification, perception, and benchmarking

Identification is a major bottleneck in ILU-intensive terminals. The scoping review of automatic ILU identification covers 63 empirical studies from 1990 to 2025 and states that standardization of ILUs has revolutionised global trade, yet their efficient and robust identification remains a critical bottleneck in high-throughput ports and terminals (Gülsoylu et al., 22 Sep 2025). Major ports handle tens of millions of ILUs each year, ILUs are often visually similar, and only the ISO 6346 identifier distinguishes them operationally (Gülsoylu et al., 22 Sep 2025).

Methodologically, the field evolved from digital image processing and traditional machine learning to deep learning. Early work relied on thresholding, denoising, contrast normalization, edge detection, morphological operations, geometric transforms, and template-based or early neural recognition; hybrid approaches then combined handcrafted features with SVMs, Random Forests, Decision Trees, or $j$3-NN; recent work is dominated by end-to-end deep learning pipelines using CNNs, scene-text detection, text recognition, and transfer learning, with over 90% of recent articles using deep-learning approaches (Gülsoylu et al., 22 Sep 2025). The review distinguishes fixed-camera gate scenarios from mobile or moving-camera scenarios, where cluttered scenes and variable viewpoints require scene-text spotting rather than document-style OCR (Gülsoylu et al., 22 Sep 2025).

The empirical base remains fragmented. The review reports that 85.71% of datasets used in surveyed studies are not public, that only 4.76% are publicly available, that 74.60% involve only fixed cameras, and that dataset size has a median of 1050 images with a range from 30 to 34,000 (Gülsoylu et al., 22 Sep 2025). Reported end-to-end accuracy ranges from 5% to 96%, and evaluations variously use end-to-end accuracy, precision, recall, $j$4-score, mAP, AP, Character Error Rate, FPS, and GFLOP (Gülsoylu et al., 22 Sep 2025). The review therefore calls for standardised terminology, open-access datasets, shared source code, and contextless text recognition optimised for ISO6346 codes (Gülsoylu et al., 22 Sep 2025).

6. Loading, unloading, and physical train effects

ILU research also includes the physical act of loading and unloading. The algorithmic study of balanced dynamic loading and unloading models heavy objects such as containers as one-dimensional intervals and seeks to minimize maximal motion of the center of gravity during the entire process (Fekete et al., 2017). For unloading, positions are fixed and the task is to find an order that minimizes center-of-gravity deviation; this variant is NP-complete, and the paper provides a polynomial-time $j$5-approximation algorithm (Fekete et al., 2017). For loading, where both positions and order are to be determined, the paper gives optimal approaches for several variants including plane, stacking, connectedness, and heterogeneous-length scenarios that model loading processes that may arise in the loading of a transport ship with containers (Fekete et al., 2017).

A more recent physical analysis addresses intermodal container loading patterns on freight trains passing through tunnels. In that paper, intermodal container loading patterns mean the spatial arrangement of ISO-standard shipping containers on flatbed wagons, including fully loaded, partially loaded, and unloaded states, with gap lengths varying from 0 to the full wagon length (Liu et al., 11 Sep 2025). The study examines 40-foot and 20-foot ISO containers on $j$6 flatbed wagons and shows that each discontinuity between loaded and unloaded regions generates sub-pressure waves, producing a step-wise pressure history not present in fully loaded or passenger trains (Liu et al., 11 Sep 2025).

The paper extends a 1D numerical programme by implementing a new mesh system and boundary conditions to represent discontinuities created by intermodal container loading patterns, and validates the improved model through Large Eddy Simulation results (Liu et al., 11 Sep 2025). It also proposes a parameterisation for the effective blockage area in a gap,

$j$7

showing that larger gaps reduce effective blockage area and increase the magnitude of pressure steps through larger separation bubbles and stronger area changes (Liu et al., 11 Sep 2025). This line of work treats ILUs not merely as units of flow but as geometric bodies whose arrangement affects infrastructure loads and operational safety.

7. Specialized applications and emerging architectures

Beyond conventional freight, ILU-like abstractions appear in specialized and emerging systems. In spent nuclear fuel transport, casks or overpacks serve as transferable units moved between rail, barge, and heavy-haul truck, especially when reactor sites lack direct rail access (Gadey et al., 2023). The operational scenarios include barge-to-transload-site, barge plus short-distance heavy-haul truck, and heavy-haul truck to transload site, with standard shipment assumed to be five casks per movement for rail or barge (Gadey et al., 2023). The modeling stack combines the Next Generation System Analysis Model, the Java Transportation Operations Model, and START, and explicitly represents site handling time, site transfer time, mode transfer time, and operating hours (Gadey et al., 2023).

Another extension appears in Physical Internet logistics, where PI-containers are treated as analogues for ILUs and can be split into smaller modules at PI-hubs functioning similarly to cross-docks (Gallo et al., 2024). The centralized mathematical programming model allocates modules across truck, train, and direct truck services and evaluates four Key Performance Indicators: direct truck usage, total delivery time, total transportation cost, and delivery gap between first and last module arrival (Gallo et al., 2024). Global Sensitivity Analysis in that study finds that the number of modules per PI-container is by far the most influential parameter for most KPIs, whereas processing time at individual PI-hubs has very low first-order sensitivity (Gallo et al., 2024). This suggests that modularization policy can be a more important design lever than local hub-processing variability.

A distinct but related strand uses the term “loading units” for homogeneous modular units in passenger–freight integration. In that system, units can be coupled into vehicles, dynamically assigned to passenger or freight service, and reconfigured at stations, with loading and unloading coordinated with docking and undocking and jointly optimized on a space-time-state network under stochastic demand (Chen et al., 30 Jun 2026). A plausible implication is that the abstraction of standardized, reusable transferable units is expanding beyond classical road–rail freight into broader integrated mobility systems.