Design tools such as topology optimization and generative design perform analyses of a given design space optimizing the material distribution for performance criteria such as maximum stress or compliance of a design. Designers using these tools can explore options beyond what a human designer could potentially have conceived. Additive manufacturing increases the feasibility to fabricate the complex, high performing designs generated by these tools not previously possible with subtractive manufacturing methods. As additive manufacturing further advances its capabilities, the manufactured designs are being used in distinguishable and visible areas for various products due to their effectiveness. As such, the appearance and aesthetics of the design are now a factor of the overall performance, a criterion not yet implemented within commercial topology optimization and generative design tools. To satisfy the desired appearance, designers are then required to modify the optimized result to suit their given design preferences. This requires additional time spent designing, increasing the associated costs and possibly reducing the performance of the final product due to the redesigns.

In this work, we present a novel topology optimization method using neural style transfer to simultaneously optimize both the structural performance criteria and the appearance to match a given reference for a design space. Performing the optimization with the knowledge of the desired appearance then reduces the amount of redesign necessary after optimization to match the designer’s preferences. The neural style transfer quantifies the characteristic features of a reference design and the design space being optimized using the convolutional layer activations of a pretrained classifier neural network for each reference design and the design space. The optimization analysis is evaluated as a multi-objective function, with the ability for the designer to determine the influence of the neural style transfer objective with the topology optimization performance objective. The designer can explore how matching the desired appearance affects the performance and tailor the influence of the neural style transfer accordingly. Non-linear optimizers developed for machine learning perform the analysis, with the neural style transfer minimally impacting the optimizer’s performance compared with only topology optimization. The efficiency of the method allows for design exploration without the need for large computational resources. The optimized designs exhibit influences from the reference used for the neural style transfer without significantly impacting the structural performance compared with the standard topology optimization result. Two case studies are presented demonstrating the capabilities of this method with different design spaces, reference designs, and influence from the neural style transfer. The benefits and limitations of the method are discussed with additional future work outlined.


Appearance Constrained Topology Optimization Using Neural Style Transfer

Division / Sponsors: 
Aerospace, Bioengineering, Applied Mechanics, Design Engineering Division, Advanced Energy Systems, Manufacturing Engineering Division, Noise Control and Acoustics, Nondestructive Evaluation, Diagnosis, and Prognosis (NDPD), Nuclear Engineering Division, Process Industries Division, Fluids Engineering, Heat Transfer, Materials, Microelectromechanical Systems, Safety Engineering and Risk Analysis, Computers & Information in Engineering


Paper Proceedings: https://asmedigitalcollection.asme.org/IMECE/IMECE2020/volume/84478

IMECE2020 - Virtual Conference

Supporting the need among researchers worldwide for additional online and multimedia content, ASME now offers the ASME Conference Video Collection. This platform will collect and provide long-term access to conference presentation videos, encompassing all technology areas within ASME's technical conference and events program. About 30,000-40,000 engineers across all mechanical engineering and related disciplines participate in ASME conferences each year. This includes authors of approximately 6,000 to 8,000 technical papers published by ASME, the majority of whom also submit video presentations for our virtual conferences. The video collection will be available as an add-on to existing product packages or as a separate subscription.