⚡ electrical engineering

Multi-Physics Coupled Modeling and Predictive Control of the DLP Photocuring Printing Process

This study proposes an integrated framework combining a nonlinear multiphysics model, parameter identification via UV-DSC, and an MLP-assisted nonlinear model predictive control strategy to significantly improve curing uniformity, prediction accuracy, and warpage control in DLP-based vat photopolymerization additive manufacturing.

Original authors: Pengbing Zhao, Mengyao Chen, Zheng Gao, Yize Chen, Jin Huang, Jianjun Wang, Jie Zhang, Hongxiao Gong

Published 2026-09-10

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Pengbing Zhao, Mengyao Chen, Zheng Gao, Yize Chen, Jin Huang, Jianjun Wang, Jie Zhang, Hongxiao Gong

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 factory floor where liquid plastic is transformed into solid, intricate objects not by a mold, but by a beam of light. This is the world of digital light processing, a high-precision form of 3D printing used to create everything from medical implants to aerospace components. In this process, a projector flashes ultraviolet light onto a vat of photosensitive resin, causing the liquid to harden layer by layer. While the final result looks seamless, the journey from liquid to solid is a chaotic internal battle. As the light penetrates the resin, it triggers a chemical reaction that releases heat and causes the material to shrink. These physical forces—light, chemistry, heat, and mechanical stress—do not act alone; they are locked in a tight, nonlinear dance where a change in one instantly ripples through the others. If this internal tug-of-war is not managed perfectly, the printed object can warp, crack, or develop weak spots, ruining the precision that makes the technology valuable. For decades, engineers have struggled to control this complexity, often relying on trial and error or fixed settings that cannot adapt to the shifting conditions inside the vat.

A team of researchers at Xidian University has tackled this challenge by building a new way to see and steer the printing process. Instead of guessing how the resin will behave, they created a detailed digital twin of the entire system. This model acts like a sophisticated weather forecast for the microscopic world inside the resin, tracking how light fades as it travels deeper, how fast the chemical bonds form, how heat builds up, and how the material pulls itself together as it shrinks. To ensure their digital map matched reality, the scientists first ran a series of physical experiments. They exposed the resin to different light intensities and measured exactly how deep the light could penetrate and how much heat was generated during the reaction. They even used a specialized instrument to measure the heat flow of the curing resin, using those signals to fine-tune their equations until the computer model predicted the physical behavior with near-perfect accuracy, matching real-world measurements with an error of only about one percent.

With this reliable model in hand, the researchers developed a smart control system designed to manage the printing process in real time. Traditional printers often use a steady, unchanging beam of light, which works well for simple shapes but fails when the resin gets too hot or the bottom layers cure too slowly compared to the top. The new system, however, acts like a conductor directing an orchestra, constantly adjusting the intensity of the light beam as the print progresses. It uses a mathematical strategy that predicts what will happen in the next few seconds and adjusts the light before problems arise, rather than waiting to fix them afterward. Because these calculations are complex and must happen instantly, the team also trained a computer neural network to learn the control patterns, allowing the system to make split-second decisions without getting bogged down by heavy calculations.

The results of this approach were striking. When the researchers tested their variable light strategy against the traditional method of using a constant beam, the difference was clear. In the standard approach, the bottom of the printed layer often remained under-cured because the light weakened as it traveled through the resin, leading to a significant gap in quality between the top and bottom surfaces. By dynamically adjusting the light intensity, the new system delivered just the right amount of energy to every part of the layer, regardless of depth. This improved the uniformity of the cure by nearly twenty-two percent, ensuring that the entire object hardened evenly. Furthermore, the system successfully predicted and prevented the warping that usually plagues these prints, keeping the final shape true to the design.

This work represents a significant step forward in moving 3D printing from a craft of fixed settings to a science of adaptive control. By proving that a model can accurately predict the chaotic interplay of heat, light, and chemistry, and by showing that a smart controller can use that model to improve the final product, the researchers have provided a blueprint for higher-quality manufacturing. While the current experiments focused on single layers and specific materials, the framework they established offers a path toward printing larger, more complex structures with the same level of precision. The findings suggest that the future of additive manufacturing lies not just in brighter lights or faster printers, but in the ability to understand and guide the invisible forces that shape the material as it comes to life.

Technical Summary: Multi-Physics Coupled Modeling and Predictive Control of the DLP Photocuring Printing Process

Problem Statement
Digital Light Processing (DLP) based vat photopolymerization is a high-resolution additive manufacturing technique capable of fabricating complex geometries. However, the process is intrinsically a strongly coupled multiphysics phenomenon involving optical attenuation, polymerization kinetics, transient heat conduction, and dynamic stress evolution. These nonlinear interactions lead to spatiotemporal nonuniformity in the curing process, resulting in warpage deformation and inconsistent mechanical properties. Current industrial practices rely heavily on empirical, open-loop parameter settings, which fail to address uncertainties from material batch variations and environmental disturbances. Furthermore, key state variables such as internal temperature, curing gradients, and internal stress are difficult to measure directly in situ, creating a significant barrier to achieving high-precision shape and property control.

Methodology
The study proposes an integrated framework comprising multiphysics modeling, parameter identification, and advanced control strategies.

Multiphysics Modeling:
- A modular approach decomposes the system into four sub-models: irradiation ( $\Sigma_I$ ), reaction kinetics ( $\Sigma_p$ ), heat transfer ( $\Sigma_T$ ), and mechanical evolution ( $\Sigma_M$ ).
- Optics: Light attenuation is modeled using the Beer–Lambert law, neglecting photobleaching for simplification.
- Kinetics: A reaction kinetics model relates the degree of cure ( $p$ ) to light intensity and temperature, utilizing an $n$ -order kinetic equation.
- Thermal: A one-dimensional transient heat conduction equation accounts for convective boundary conditions and the exothermic heat of polymerization.
- Mechanical: A simplified 3D mechanical model characterizes stress-strain evolution, decomposing total strain into elastic, thermal, and chemical shrinkage components. The Young's modulus is modeled as a function of the degree of cure using a fifth-order polynomial approximation.
- State-Space Formulation: The coupled partial differential equations (PDEs) are discretized in space using the finite difference method (FDM) to form a unified nonlinear state-space model suitable for controller design.
Parameter Identification and Validation:
- Experiments were conducted using a self-developed bottom-up DLP system and an ultraviolet differential scanning calorimeter (UV-DSC).
- Penetration depth ( $D_p$ ) and critical exposure energy ( $E_c$ ) were identified via layer thickness measurements.
- Kinetic parameters ( $r, n, b$ ) and reaction enthalpy ( $\Delta H$ ) were derived from isothermal UV-DSC data.
- An indirect validation method was proposed using a "capture factor" ( $\eta$ ) to map theoretical heat generation to experimental DSC heat-flow signals, verifying the thermodynamic coupling terms.
Control Strategy:
- Nonlinear Model Predictive Control (NMPC): An NMPC strategy was designed to track reference degree-of-cure trajectories while minimizing energy consumption and respecting physical constraints (e.g., light intensity limits).
- State Observation: Since internal states (stress, strain, internal temperature) are unmeasurable, a Sliding Mode Observer (SMO) with a continuous hyperbolic tangent switching term was integrated to estimate full states and reject disturbances.
- Real-Time Acceleration: To address the computational burden of online nonlinear programming, a Multilayer Perceptron (MLP) neural network was trained offline on NMPC data to serve as a surrogate predictive controller. The trained function was embedded into Simulink to validate the logical structure and performance of the learned control law.

Key Results

Model Accuracy: The proposed multiphysics model demonstrated high fidelity when compared against COMSOL Multiphysics simulations. Prediction errors for the degree of cure and temperature were approximately 1% across different layers. The model also effectively predicted warpage trends and residual stress distributions.
Control Performance:
- The NMPC strategy outperformed conventional LQR and GS-PID controllers, improving degree-of-cure tracking accuracy by 16.38% and exhibiting superior predictive behavior to mitigate delays.
- The SMO achieved high-precision state reconstruction, with a relative error of only 0.85% for residual stress estimation under noisy conditions.
- The MLP surrogate controller successfully replicated NMPC trajectories in the Simulink environment, demonstrating its potential for real-time implementation.
Experimental Validation:
- Fourier Transform Infrared Spectroscopy (FTIR) experiments confirmed that the proposed variable light-intensity control strategy could be implemented on physical hardware.
- Due to limitations in the current experimental setup, an "offline optimization–open-loop implementation–offline characterization" validation framework was adopted. The optimal light-intensity time sequence generated by the controller was discretized and applied to the actual DLP platform.
- Compared to a constant light-intensity strategy with equivalent total energy, the proposed variable intensity strategy increased the bottom-layer degree of cure from 68.22% to 72.80%.
- Crucially, the interlayer curing difference was reduced from 19.40% to 15.17%, representing a 21.80% improvement in curing uniformity along the thickness direction.

Significance and Claims
The paper claims to provide a critical theoretical and practical framework for intelligent photopolymerization manufacturing. By integrating physics-based modeling with advanced control theory, the study addresses the challenge of unmeasurable states and strong multiphysics coupling. The primary significance lies in demonstrating that a closed-loop framework, even when validated through an "offline optimization–open-loop implementation" approach due to current hardware limitations, can significantly enhance forming quality and interlayer consistency. The work establishes a pathway toward achieving precise shape and property control in DLP printing, moving beyond empirical trial-and-error toward model-driven, adaptive regulation. The authors note that while the current study focuses on single-layer curing of pure resin, the framework lays the groundwork for future extensions to multi-layer continuous printing and complex composite slurries.

More like this