Smart “Design for Manufacture” – integrated design tool
Topic(s) : Industrial applications
Co-authors :
Jakub KUCERA (UNITED KINGDOM), Tim NEWMAN , Bryn GOODHEAD , Jon TAYLORAbstract :
The aerospace industry relies on a wealth of expertise when designing any new components, with many written and unwritten rules on how to approach design. The design of each component is inherently a multidisciplinary task. The component needs to sustain loads throughout its lifecycle. Manufacturing costs should be minimized. Compliance with aerodynamic requirements is needed. Assembly needs to work with full range of manufacturing tolerances, just to name a few. For composite components this list increases significantly, by having to account for ply drop-offs, fibre orientations, or galvanic corrosion, etc. All this expertise is either contained in lengthy design documents, or worse, in the heads of experienced engineers.
A Smart DFM (Design for Manufacture) tool is being developed under the CoSinC (Composite Smart Industrial Control) programme at the National Composites Center (NCC). The purpose of the tool is to provide immediate feedback to the design engineer based on a list of design rules. Design rules have been collected from aerospace design documents and from expert knowledge capture. The aim of the tool is to ensure a robust transfer of knowledge between engineers of various grades, in place of the more ad-hoc approach commonly used. From business perspective, this can minimise the workload of the more expensive engineers; allowing automation to reduce the burden of lower-level knowledge transfer. Similarly, the tool can be used by experienced engineers as a quick but broad self-check.
The end-product will be an integrated button in your CAD system that provides immediate feedback on the design, based on all implemented rules and design suggestions.
In expert systems all knowledge considered about the world, or problem at hand, are called “facts”. The inference engine in the tool compares the ruleset against the facts known about the part. This is currently done using Python and the Pydantic library, where “facts” are stored as overarching class of design variables.
With reference to the CAD part, in our system “facts” are automatically gathered from .stp file and layup file using variety of tools/modules. These include graph neural network (GNN) for feature identification, step interrogation scripts for feature detail extraction, highly interoperable layup-definition method, and various smaller specialized scripts. These design interrogation tools are highly interconnected and rely on execution order being adjusted based on available information. This is also supported by the Pydantic library.
Up to date we have collected over 200 rules, not all of which are suitable for the automated tool. We would like to present the functionality of a demonstration tool. Approximately 30 rules have been implemented up to date as a proof of concept and tested against a few simple parts that are specifically designed to trigger some of the rules. We will also outline the current challenges and our planned way forward.
A Smart DFM (Design for Manufacture) tool is being developed under the CoSinC (Composite Smart Industrial Control) programme at the National Composites Center (NCC). The purpose of the tool is to provide immediate feedback to the design engineer based on a list of design rules. Design rules have been collected from aerospace design documents and from expert knowledge capture. The aim of the tool is to ensure a robust transfer of knowledge between engineers of various grades, in place of the more ad-hoc approach commonly used. From business perspective, this can minimise the workload of the more expensive engineers; allowing automation to reduce the burden of lower-level knowledge transfer. Similarly, the tool can be used by experienced engineers as a quick but broad self-check.
The end-product will be an integrated button in your CAD system that provides immediate feedback on the design, based on all implemented rules and design suggestions.
In expert systems all knowledge considered about the world, or problem at hand, are called “facts”. The inference engine in the tool compares the ruleset against the facts known about the part. This is currently done using Python and the Pydantic library, where “facts” are stored as overarching class of design variables.
With reference to the CAD part, in our system “facts” are automatically gathered from .stp file and layup file using variety of tools/modules. These include graph neural network (GNN) for feature identification, step interrogation scripts for feature detail extraction, highly interoperable layup-definition method, and various smaller specialized scripts. These design interrogation tools are highly interconnected and rely on execution order being adjusted based on available information. This is also supported by the Pydantic library.
Up to date we have collected over 200 rules, not all of which are suitable for the automated tool. We would like to present the functionality of a demonstration tool. Approximately 30 rules have been implemented up to date as a proof of concept and tested against a few simple parts that are specifically designed to trigger some of the rules. We will also outline the current challenges and our planned way forward.