⚡ electrical engineering

Digital Twin Fidelity When Predicting Vessel Service Time For Roll-on Roll-off Shipping

This study demonstrates that a cost-effective, rule-based Digital Twin architecture using discrete-event simulation can robustly predict RoRo vessel service times with high accuracy, proving that significant decision-making value can be achieved without the need for expensive, high-fidelity sensor infrastructure.

Original authors: Teresa Marquardt

Published 2026-08-04

📖 3 min read☕ Coffee break read

CC BY 4.0

Original authors: Teresa Marquardt

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine a busy port as a giant, chaotic dance floor. On one side, massive ships arrive with cargo; on the other, trucks and workers wait to unload them. In a perfect world, the music stops exactly when the last box is moved, and the ship sails away instantly. But in the real world, things go wrong. A forklift breaks, a worker takes a slow break, or a crate is heavier than expected. This causes the ship to wait longer. While waiting, the ship's engine often has to stay running or be preheated to leave quickly, which burns fuel and pollutes the air. To fix this, scientists use something called a "Digital Twin." Think of this as a video game version of the real port that runs in a computer. It watches the real dance floor and tries to predict exactly when the music will stop so the ship can start its engine at the perfect moment—not too early, not too late. The big question researchers have always asked is: Does this video game need to be a hyper-realistic, expensive, super-computer simulation to work, or can a simpler, cheaper version do the job just as well?

This paper, written by Teresa Marquardt, dives into that exact question using a real-life port in Kiel, Germany, as the test subject. The author built a "pragmatic" Digital Twin—a simplified, rule-based computer model—to manage the loading and unloading of Roll-on/Roll-off (RoRo) ships. These are special ships where cars and trucks drive on and off, rather than being lifted by cranes. The study ran thousands of computer simulations to see how well this "good enough" twin could predict when a ship would be ready to leave, even when the computer model was intentionally made less perfect. The researchers tested what happened when the model didn't know the exact speed of the trucks, when it only got updates every 25 minutes instead of every second, and even when a truck in the simulation suddenly broke down.

The results were surprisingly robust. The study found that the Digital Twin didn't need to be a perfect, high-definition replica of reality to make good decisions. Even when the model was updated with "snapshots" (like checking a photo every 25 minutes) rather than live video, it still managed to predict the ship's departure time with high accuracy. The average error in predicting when to start the engine was less than 56 seconds in the best-case scenario, and even when things went wrong—like a truck breaking down or the model guessing the wrong speed—the twin still performed significantly better than older, static methods. The paper suggests that the barrier to using this technology is lower than people thought; you don't need a million-dollar sensor network to get significant value. However, the study also notes a catch: while the twin is tough against random glitches, it gets confused if it is systematically wrong about how fast things usually move. If the model consistently guesses the trucks are faster than they really are, the predictions will be off. Ultimately, the paper concludes that a simple, rule-based system that constantly re-calculates its plan is a powerful tool for keeping ports efficient and the air clean, without needing a massive infrastructure investment.

Technical Summary: Digital Twin Fidelity When Predicting Vessel Service Time For Roll-on Roll-off Shipping

Problem Statement

Roll-on Roll-off (RoRo) terminal operations are characterized by high variability due to heterogeneous cargo and a heavy reliance on human labor (tug drivers), making processing times uncertain. This uncertainty complicates the prediction of a vessel's Estimated Time of Departure (ETD). Accurate ETD prediction is critical for downstream operations, particularly for coordinating shore power usage and the precise timing of vessel engine pre-heating to minimize at-berth emissions and fuel costs.

While Digital Twins (DTs) offer a pathway to real-time decision support, their adoption in RoRo terminals faces three primary barriers:

Economic: High costs associated with installing IoT sensors on every tug and cargo unit, alongside the infrastructure required for real-time data processing.
Computational: The scheduling problem is computationally intractable to solve for optimality in real-time, especially under dynamic, stochastic conditions.
Human: The reliance on human operators introduces behavioral variability that data alone cannot perfectly predict.

Consequently, terminal operators face a trade-off between the fidelity (accuracy and data granularity) required for effective decision-making and the costs of developing and maintaining a high-fidelity system. This study investigates the minimum required fidelity for a DT to provide meaningful operational decisions in RoRo stevedoring.

Methodology

The study proposes a "pragmatic" Digital Twin architecture based on discrete-event simulation (DES) and rule-based heuristics (RBHs). The system models a RoRo terminal using real-world data from the Port of Kiel, Germany.

System Architecture

The framework consists of two interacting components:

Analog Twin (AT): A stochastic, discrete-event simulation representing the physical terminal (ground truth). It simulates tug movements, cargo handling, and precedence constraints.
Digital Twin (DT): A simulated agent running a "simulation-in-simulation." It receives state updates from the AT, plans operations, and predicts outcomes.

Core Modules

Planning Module: The DT generates feasible operation sequences using RBHs (e.g., row-wise or diagonal unloading/loading). It assigns cargo to tugs using a round-robin approach to balance workload. To account for uncertainty, the DT runs $n=51$ stochastic simulations per schedule, varying processing and travel times via Monte Carlo methods. The optimal schedule is selected based on a weighted objective function minimizing both average makespan and the average of the worst-case (top 5%) makespans.
Prediction & Decision Module: The DT predicts the ETD by analyzing the distribution of simulated makespans. It determines the optimal time to switch on the vessel's engines (triggered 15 minutes prior to departure) using a $k$ -quantile proxy. The timing of the next planning iteration is governed by a $p$ -of- $q$ rule, which dynamically adjusts the wait time based on the remaining process duration.

Experimental Design

The study systematically degrades the DT's fidelity to test robustness across various scenarios:

Validity Reduction: Introducing systematic errors in expected processing times (over/underestimation), manipulating the assumed level of randomness (variance), and altering process logic (allowing tugs to skip blocked operations).
Granularity Reduction: Replacing real-time updates with periodic snapshots (every 25 minutes) and analyzing the impact of delayed replanning.
Structural Disruptions: Simulating tug breakdowns at different stages of the process (80 and 95 minutes) to test recovery capabilities.

These scenarios were compared against two benchmarks:

First ETD: A static plan made before stevedoring begins with no replanning.
16 Cargo Units Left: A rule-based heuristic triggering engine pre-heating when 16 cargo units remain.

Key Results

The computational study, based on 100 generated instances, yielded the following findings:

Baseline Performance: The DT significantly outperformed static planning. The mean absolute error (MAE) in pre-heating time was 56 seconds for the DT, compared to 5 minutes 13 seconds for the First ETD benchmark and 2 minutes 8 seconds for the "16 Cargo Units" rule. The DT also reduced the average Vessel Service Time (VST) by approximately 4 minutes through iterative replanning.
Robustness to Validity Errors: The DT remained robust even when assumptions about processing times were systematically faulty (e.g., assuming times were 150% or 70% of actual). While errors increased, the iterative replanning mechanism compensated for most deviations, limiting the impact to the final phase of the process.
Robustness to Granularity Loss: When updated via snapshots rather than real-time, the DT's performance degraded slightly (MAE increased to ~1:20 min) but remained superior to benchmarks. The system effectively interpolated missing data, though frequent assumptions limited the scope of replanning.
Disruption Management: In the event of a tug breakdown, the DT successfully adapted by reassigning remaining tasks to the remaining tugs. The MAE increased only marginally (by seconds) compared to the baseline, whereas benchmarks failed to adapt, resulting in errors exceeding 10 minutes.
Combined Stress Test: Even when combining tug breakdowns with snapshot updates, the DT maintained an MAE (1:25 min) lower than the benchmarks operating under ideal conditions.

Significance and Contributions

The paper argues that the barrier to adopting Digital Twins in RoRo terminals is lower than often assumed. The primary contributions are:

Pragmatic Architecture: The study demonstrates that a simple, rule-based DT combined with stochastic simulation can provide significant value without massive infrastructure investment or complex optimization algorithms.
Fidelity Trade-off Analysis: It provides empirical evidence that high-fidelity (real-time, perfect data) is not strictly necessary for effective decision-making. A system with reduced validity and granularity still outperforms static planning and simple heuristics.
Resilience: The proposed DT architecture proves resilient to structural disruptions (tug failures) and data latency, making it suitable for the dynamic and uncertain environment of RoRo operations.

Limitations and Future Work

The author acknowledges several limitations:

The model assumes all cargo is handled by tugs, excluding driver-handled units (trucks).
Stowage and yard plans are treated as static, whereas real operations involve dynamic reshuffling.
Human behavior is modeled as compliant, neglecting "algorithm aversion."
The study does not benchmark against sophisticated optimization methods like reinforcement learning.

Future research is suggested to focus on adaptive DTs that self-calibrate parameters using historical data to correct systematic errors and to explore the interaction between the DT and human operators. Additionally, a hybrid approach combining rule-based heuristics for speed with sophisticated optimization for later stages of the process is proposed.