Process-Knowledge-Guided Latent Diffusion for Generative Design of Porous Architectures: A Case Study in Aerosol Jet Printing
Organizations: School of Manufacturing Systems and Networks, Ira A. Fulton Schools of Engineering, Arizona State University, Mesa, AZ 85212, United States
Abstract
Printed porous layouts interleave functional material with open channels, and the layout that uses the material best depends on how the process settings shape the deposit. Generative models propose layouts quickly, but a generator that does not receive the process setting samples from the same distribution at every setting, even when the process-dependent score ranks candidate layouts differently. This paper examines whether process knowledge can steer a latent diffusion generator toward layouts suited to a given process setting in a prescribed design domain. We formulate a process-dependent utilization score in which a process model sets the length over which the utility of printed material decays with distance from a channel. The generator combines a vector-quantized variational autoencoder (VQ-VAE) latent representation, a topology U-Net, a ControlNet domain branch, and a separately trained regressor that predicts the score from noised latents; the domain branch conditions each denoising step and the regressor gradient steers it. The method is applied to aerosol jet printing, where the atomizer and sheath gas flows enter the score through the flow-focusing relation. Across seven modeled flow settings, the lowest-scoring candidate layout of a domain differed between settings in 191 of 500 domains. In a sampling study, the mean directly computed score on three validation domains and one test domain was 0.642 with domain conditioning alone, 0.636 at a weight of 300, and 0.450 at 12,000 (112 scores per weight; lower is better). The results support process-knowledge-guided generation as an approach to process-specific porous design under the stated assumptions.
Figures & tables
| Demonstrated scope in prior work | Element of the present architecture | Evidence in this paper |
|---|---|---|
| Latent denoising conditioned on an encoded strain-energy field [ 9 ] ; penalties for floating material, insufficient material at load sites, and volume-fraction deviation added to the autoencoder loss [ 20 ] . | The topology prior is retained unchanged; a domain branch conditions denoising and a separately trained regressor steers the latent trajectory at sampling time. | Direct scores of decoded designs with and without guidance on training and held-out domains (Section 4.3.2 ). |
| Guidance of a graph-based latent diffusion over vector-quantized unit cells toward elastic targets with a separate property predictor [ 40 ] ; editing of a partially noised reference design with a characteristic loss on the estimated clean latent [ 3 ] . | A regressor predicts the process-dependent utilization score from the noisy design latent, the encoded domain mask, and the process setting. | Fixed-geometry audit of the process-dependent score across seven settings (Section 4.3.1 ). |
| ControlNet spatial control [ 42 ] combined with timestep-aware regressor guidance in image synthesis [ 7 ] . | A ControlNet domain branch and a regressor without a diffusion-time input, applied to a VQ-VAE-derived layout representation with a porous-design evaluator. | Guidance-weight sweep, start of guidance, and regressor accuracy by noise level (Section 4.4 ). |
| Domain 379 | Domain 213 | |||||||
|---|---|---|---|---|---|---|---|---|
| #26573 | #26540 | #26589 | #14958 | #14944 | #14921 | |||
| (15, 15) | 0.44 | 0.401 | 0.437 | 0.414 | 0.455 | 0.427 | 0.413 | |
| (15, 45) | 0.65 | 0.438 | 0.486 | 0.458 | 0.487 | 0.471 | 0.466 | |
| (15, 75) | 0.80 | 0.464 | 0.519 | 0.486 | 0.508 | 0.500 | 0.501 | |
| (30, 30) | 0.88 | 0.475 | 0.533 | 0.499 | 0.517 | 0.513 | 0.516 | |
| (30, 45) | 1.00 | 0.493 | 0.555 | 0.519 | 0.532 | 0.533 | 0.540 | |
| Training domains | Held-out domains | |||
|---|---|---|---|---|
| Guidance weight | Mean | Change | Mean | Change |
| 0 (domain-conditioned) | 0.6889 | – | 0.6416 | – |
| 300 | 0.6863 | 0.6359 | ||
| 8,000 | 0.5276 | 0.5215 | ||
| 12,000 | 0.4611 | 0.4499 | ||
Diffusion Model Guidance
Diffusion model guidance steers denoising toward desired conditions, rewards, constraints, or away from unwanted concepts.
See all 397 papers