⚡ electrical engineering

A Model-Based Digital Twin for Predicting Deterioration of Mogao Polychrome Sculpture

This paper presents a generic, model-driven digital twin framework that simulates and predicts the deterioration of Mogao Grottoes polychrome sculptures under various future climate scenarios by integrating five decay processes and validating them through controlled aging experiments, thereby enabling proactive preventive conservation.

Original authors: Yuan Tian, Ran Wei, Jun Zhang, Xiwan Zhang, Wenbin Hu, Le Zhu

Published 2026-07-30

📖 6 min read🧠 Deep dive

CC BY 4.0

Original authors: Yuan Tian, Ran Wei, Jun Zhang, Xiwan Zhang, Wenbin Hu, Le Zhu

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 walking into a giant, ancient cave that has been a museum for over a thousand years. Inside, there are thousands of colorful clay statues, painted with bright reds, blues, and greens. These aren't just decorations; they are fragile time capsules. But here's the tricky part: these statues are slowly getting sick. The air around them changes with the seasons, getting hotter, colder, wetter, and drier. This invisible weather inside the cave makes the clay swell and shrink, causing the paint to crack, fade, or grow mold. For a long time, experts could only guess how fast this damage would happen. They knew the statues were in danger, but they couldn't predict when a specific statue would lose its color or crack apart. It was like trying to guess when an apple would rot without knowing the temperature of the fridge.

This is where a new kind of science called "Digital Twin" technology comes in. Think of a digital twin as a ghostly, perfect clone of a real object that lives inside a computer. But this isn't just a 3D photo; it's a living simulation. It has a brain made of math that understands physics. It knows how heat makes materials expand, how humidity makes clay thirsty, and how light makes paint fade. By feeding this computer ghost real-time data about the cave's weather, scientists can run a "what-if" game. They can ask, "What happens if we turn off the air conditioning?" or "What if a storm makes the cave 10% more humid?" The computer twin then simulates the future, showing exactly how the statue would look 50 or 100 years from now under those new conditions. It turns conservation from a guessing game into a crystal ball, helping protectors save these treasures before the damage actually happens.

The Digital Ghost of the Mogao Grottoes

In the Mogao Grottoes of China, a UNESCO World Heritage site, there are over 2,400 colorful clay statues. They are masterpieces of Buddhist art, but they are slowly crumbling. The paper introduces a Model-Based Digital Twin designed to predict exactly how these statues will deteriorate. Instead of just looking at a statue and hoping for the best, the researchers built a virtual replica that acts like a time machine, showing how the statue will age under different climate conditions.

The Five Villains of Decay

The digital twin doesn't just guess; it simulates five specific ways the statues get damaged, treating each like a character in a story:

Fading: The paint loses its color.
Mold: Fuzzy green or brown patches grow in damp spots.
Salt Damage: Invisible salts inside the clay expand and push the paint off, like tiny bombs.
Cracking: The clay swells and shrinks, causing the paint to crack (craquelure).
Aging: The general wear and tear of time.

The researchers combined these five "villains" into one system. They took a real statue from Cave 328, the Kneeling Attendant Bodhisattva, scanned it into a 3D model, and then ran the simulation. The result is a video game-like interface where conservators can change the temperature and humidity settings and watch the statue's skin change in real-time. If they make the cave too humid, they can instantly see mold spreading or salt crystals forming on the statue's feet.

The "What-If" Game

The most powerful part of this tool is the counterfactual prediction. The cave where the statue originally lived is actually quite stable and safe right now. The digital twin shows that under current conditions, the statue will change very little. However, the twin's real value is answering the scary question: What if we lose control?

The researchers simulated a "Poor Storage" scenario where the climate control fails, and the cave gets hot and humid (30°C and 80% humidity). In this simulation, the digital twin predicted that the statue would age four times faster than it would in a standard museum. The composite risk score, which measures how close the statue is to disaster, would climb to a critical failure level in just about 40 years. This proves that the current climate control isn't just "nice to have"; it is the only thing keeping the statue from rotting away quickly.

The Lab Experiment: Testing the Math

To make sure their computer model wasn't just making things up, the team built a real-life test. They created replica statues using the exact same materials as the ancient ones: local Dunhuang clay, straw, hemp, and mineral pigments like vermilion (red), azurite (blue), and malachite (green).

They put these replicas in two different rooms for 61 days:

Room A: Cool (23°C), humid (40% RH), and lit with normal lab light.
Room B: Hot (40°C), very dry (10% RH), and kept in the dark.

The result was surprising. The team found that the pigments in the cool, humid, lit room actually faded 1.5 to 2 times faster than those in the hot, dry, dark room. This helped them tune the math in their digital twin. However, because the two rooms differed in temperature, humidity, and light all at once, they couldn't separate exactly how much each factor contributed. They know the cool/humid combo is worse, but they need more precise experiments to know the exact "speed limit" of the damage.

The Verdict: A Safety Net, Not a Crystal Ball

The paper is careful not to claim that this tool has solved everything. The predictions for the next 50 or 100 years are simulations, not guaranteed facts. The model is "calibrated" to match what we know about the statues' history (they are about 1,200 years old and look the way they do), but it hasn't been "validated" against a 100-year record of actual decay because we don't have that data yet.

The authors explicitly state that the model is a guide, not a prophecy. It tells us that if we keep the cave cool and dry, the statues are safe. If we let the humidity rise, the salt and mold will wake up and destroy the art. The digital twin is a powerful new way to see the invisible threats, allowing conservators to fix problems before they ever become visible cracks on the real statue. It turns the future from a mystery into a manageable plan.

Technical Summary: A Model-Based Digital Twin for Predicting Deterioration of Mogao Polychrome Sculpture

Problem Statement
The Mogao Grottoes contain over 2,400 polychrome clay sculptures whose painted surfaces are susceptible to deterioration driven by shifting temperature and humidity. While predictive models exist for flat mural paintings and other heritage materials (e.g., cellulose, oil paint, gesso), a fundamental gap persists for freestanding clay sculptures. These objects possess a heterogeneous material system (straw-bound clay, fiber-reinforced plaster, mineral pigments bound with animal glue) that responds differently to hygrothermal cycling than wall paintings. Furthermore, existing digital twin implementations in heritage contexts largely focus on structural integrity, energy performance, or static 3D records, lacking physics-based models that predict material degradation (specifically color change, mould, salt damage, cracking, and ageing) at the object level. There is no experimentally validated dose-response function linking color change ( $\Delta E^*_{ab}$ ) directly to temperature and relative humidity for mineral pigments on clay substrates.

Methodology
The authors present a model-based digital twin framework that integrates five physics-based deterioration mechanisms onto a single 3D photogrammetric mesh of a sculpture. The system is built using Model-Driven Engineering (MDE) to ensure the framework is generic and reusable.

Framework Architecture:
- Design Time: A domain-specific modelling language (Ecore/EMF metamodel) defines the entity graph (Caves, Statues, Sensors). Epsilon EGL templates (approx. 595 lines) automatically generate the backend data layer (Mongoose/Express schemas, services, controllers) for 13 entity types.
- Runtime: The system ingests live sensor telemetry (Temperature $T$ , Relative Humidity $RH$, Illuminance $I$ ) and executes five independent deterioration kernels.
- Visualization: A browser-based 3D viewer renders the results in two modes: Environmental Simulation (user-defined "what-if" scenarios) and Environment Monitoring (live data replay).
The Five Deterioration Kernels:
- Model 1 (Pigment Segmentation): A deterministic HSV-threshold classifier maps the 3D texture to eight mineralogical classes (e.g., vermilion, azurite, malachite, gold leaf) to drive per-pigment kinetics.
- Model 2 (Chemical Fading): A sorption-coupled dose-response model combining Arrhenius temperature scaling with a Paltakari–Karlsson moisture isotherm. It calculates per-pixel color decay ( $\Delta E^*_{ab}$ ) based on specific activation energies for dark and light pathways.
- Model 3 (Lifetime Multiplier): Uses Michalski's Arrhenius-based lifetime multiplier to rescale expected service life relative to a reference environment.
- Model 4 (Mould Growth): Implements the VTT Hukka–Viitanen model to predict mould index ( $M$ ) based on temperature-dependent critical humidity thresholds.
- Model 5 (Salt Crystallisation): Uses Steiger's thermodynamic framework to model crystallisation pressure of sodium sulphate. Crucially, it accounts for the Mogao cave's sub-deliquescent environment, treating salt damage as a sustained static pressure rather than wet-dry cycling, gated by a capillary availability factor ( $\alpha_s$ ).
- Model 6 (Hygro-mechanical Fatigue): Applies Basquin's law and Miner's rule to model fatigue damage ( $D$ ) from cyclic humidity amplitude ( $\Delta RH$ ).
Composite Risk Assessment:
The outputs of the five models are normalized to a 0–1 scale and aggregated into a composite index ( $R_{composite}$ ) using a maximum-risk approach. This index drives a ranked maintenance queue and visual overlays on the 3D mesh.
Experimental Pilot (Accelerated Ageing):
To ground the fading model, the authors fabricated 1:5 scale replicas of a Mogao lotus pedestal using traditional materials (chengban clay, straw, hemp, animal glue, and mineral pigments). These were aged for 61 days under two conditions: a "room" arm (23°C, 40% RH, ambient light) and a "chamber" arm (40°C, 10% RH, dark). Colorimetric measurements ( $\Delta E^*_{ab}$ ) were taken at 21 time points.

Key Results

Pilot Kinetics: The accelerated ageing campaign measured fade-rate contrasts between the two arms. Vermilion, azurite, and malachite aged 1.5–2.0 times faster in the cooler, humid, lit room arm compared to the hot, dry, dark chamber arm. Due to the confounding variables (light, temperature, and humidity varying simultaneously), the study reports these rate ratios rather than deconvolved activation energies.
Case Study Application: The framework was applied to the Kneeling Attendant Bodhisattva (Mogao Cave 328).
- Current Climate: Under the monitored Mogao cave climate (mean 12°C, RH < 80%), the sculpture is in a benign regime. Mould growth is dormant (index = 0), and hygro-mechanical fatigue is negligible. Salt crystallisation exerts sustained static pressure but is spatially limited to the capillary-fed base.
- Counterfactual Scenarios: The twin demonstrates that if climate control is lost (e.g., RH rises to 85–96% and temperature increases), salt crystallisation and mould risk activate rapidly. A "Poor Storage" scenario projects the composite risk index rising to failure levels ( $R=1.0$ ) within four decades, whereas active buffering maintains the low-risk baseline.
Visualisation: The system successfully renders distinct deterioration signatures (e.g., non-uniform pigment fading, salt efflorescence at the base, craquelure networks) directly onto the 3D mesh, allowing conservators to inspect mechanisms individually or in composite.

Significance and Claims
The paper claims four primary contributions:

Generic Framework: It introduces a Model-Driven Engineering approach where the backend is generated from a domain-specific model, making the digital twin adaptable to different heritage objects without rewriting the core data layer.
What-If Rendering: It enables interactive, physics-driven visualization of deterioration signatures on a 3D mesh, allowing conservators to simulate environmental trajectories and observe real-time texture changes.
Multi-Mechanism Integration: It operationalizes five distinct physical degradation kernels on a single object, making heterogeneous processes commensurable and reducing them to a ranked, physics-derived maintenance queue.
Pilot Kinetics: It provides the first controlled fade-rate contrasts for Mogao polychrome clay systems, establishing that cooler, humid, lit conditions can be more damaging to specific pigments than hot, dry, dark conditions, even if absolute color change is small over short periods.

Limitations and Modesty
The authors explicitly state that the current pilot does not provide independent, deconvolved activation energies due to the confounding of light and temperature in the two-arm design. The long-horizon projections (e.g., 200-year simulations) are calibrated to match archaeological degradation patterns (approx. 48% color loss over 1200 years) rather than validated against independent multi-decade records. The authors acknowledge that the linear rate extrapolation from the 61-day pilot shows poor out-of-sample performance ( $Q^2 = -16.7$ ), necessitating the use of a saturating model for long-term projections. They identify the need for matched-light factorial designs and external validation against archival data as priority future work.