Model Predictive Control in Laser 3D Printing

THE CHALLENGE

Laser powder bed fusion (LPBF) manufacturing faces significant challenges due to its traditional reliance on empirically optimized, constant processing parameters that do not adequately address layer‐wise variations in thermal history, leading to inconsistent microstructures, porosity, poor surface finishes, and geometric distortions. Additionally, conventional empirical build-and-test strategies are inefficient and expensive, as they do not effectively capture the complex, layer-specific thermal fluctuations. As a result, challenges persist in achieving the precise control needed to mitigate the formation of structural flaws, ultimately limiting the overall performance and reliability of the manufactured parts.

OUR SOLUTION

The technology integrates a digital twin framework with physics-based rapid thermal simulations and real-time in-situ sensor data to predict and control critical LPBF part properties. It uses a computation model to estimate part-scale thermal history and combines optical and thermal sensor signatures with a k-nearest neighbor machine learning approach to accurately predict porosity, meltpool depth, grain size, and microhardness. Additionally, a mesh-free, graph theory-based thermal simulation supports a model predictive control strategy that dynamically adjusts laser power on a layer-by-layer basis, effectively reducing variations in microstructure, geometric integrity, and surface finish. What sets this approach apart is its ability to execute layer-wise, feedforward process control, thereby eliminating the need for costly, empirical build-and-test methods. By fusing real-time data with robust, physics-guided simulations, the system rapidly tailors processing parameters to meet a target thermal profile. This enables precise control of thermal gradients and cooling rates, substantially mitigating defects common in traditional LPBF processing while achieving superior consistency in complex geometries and material properties.

Overview of the model predictive control approach. Shown here is an example of the cone geometry studied in this work. The approach has four steps: (1) Predict, (2) Identify, (3) Parse, and (4) Select. Steps 3 and 4 are iterative.

(a) Schematic, and (b) Picture of the experimental LPBF setup with longwave infrared (LWIR) thermal camera installed outside the build chamber.

Advantages:

• Enables precise, layer-wise control of thermal history to optimize microstructure and material properties.
• Dynamically adjusts laser power to reduce common defects such as porosity and inconsistent meltpool dimensions.
• Enhances geometric accuracy and surface finish in complex and overhanging part features.
• Eliminates the need for costly empirical build-and-test approaches by using simulation-driven, feedforward process control.
• Supports efficient, support-free manufacturing, reducing material usage and post-processing time.

Potential Application:

• Physics-based Process Control
• Predictive defect management
• Adaptive laser tuning
• Digital twin simulation
• Quality-enhanced LPBF production