The current Generative Design Experience 

Autodesk Fusion 360 is a cloud-based 3D design and engineering platform for product development. It combines industrial and mechanical design, simulation, collaboration, and machining in a single package. Generative design is a design exploration technology available in Fusion 360. It was created to simultaneously generate multiple CAD-ready solutions based on real-world manufacturing constraints, cost evaluation and product performance requirements. The experience is derived into four phases: Switching workspaces, setting up your parameters, exploring the alternatives and consuming a selected alternative.   

Upon joining the team, it became evident that our existing solution did not meet customer and business requirements despite being in place for some time as a separate workspace in the product. Reviewing the data revealed a stark drop-off rate between each stage in the workflow.    

Understanding the problem 

To dig deeper, we ran a survey within the product and received an incredible number of responses – over 25,000! These insights gave us a fascinating understanding of what was going on.   

Themes

After conducting in-depth data analysis, three consistent themes emerged - relevance, familiarity and trust. Our findings indicated that the current Generative Design solution needed to be more suitable for our users' daily activities, and it felt unfamiliar to them. Similarly, trust was a continuous discussion as customers have a strong perspective and reluctance towards an automation system due to the unknown nature of the decision-making process.  

We took the insights from our research and translated them into clear goals that directly addressed customer needs and met both short-term and long-term business objectives.  

Persona

During the initial research survey we maintained focus on Melvin, an experienced mechanical engineer who designs industrial machinery and consumer goods for mass-market product industries. It enabled us to understand how Melvin's work fits within the landscape of these complex markets and what role he plays in finding innovative solutions in his company.  

Early Ideation

After we clustered the ideas during the early ideation phase, we represented them into single stories. 

Concept 1 (uno)

The concept involves embedding the existing generative design capabilities into the core design workspace so users have the ability to customize parameters, explore alternatives, and implement their selected designs with greater ease in the context of the design space. We were confident that this approach would significantly simplify our user's workflow and help them reach the desired outcome quicker. 

Click the arrow > to go through the concepts

Review of Concept 1 (uno)

We felt we successfully integrated the experience into the design space. However, it only reached some of our intended goals. We needed to make sure users felt comfortable with this new solution; we needed to focus more on relevance and familiarity.   

Concept 2 (dos)

What if it was as easy as simply selecting a command, specifying some inputs, and choosing an outcome from a set of generated solutions that meet our customer’s needs and expectations? What if it was that simple?  

Review of Concept 2 (dos)

Concept 2 headed in a promising direction; the idea felt more familiar and integrated well into the design workflow. However, we wanted to make it THE go-to functionality and to do this; we need to unravel further user value.    

Concept 3 (tres)

Our conversations with customers uncovered essential themes for user value that we could address to develop our idea further. At the forefront of these was a way for personas like Melvin to consider manufacturability at various stages throughout their design process - from conceptualizing an initial shape to evaluating existing parts or designs. By providing them insights in this manner, we can empower users and ensure they are factoring manufacturing requirements into their workflow.  
   

Re-design

Ivent

Review of Concept 3 (tres)

Excited by the potential, we pressed forward with this concept to explore its possibilities by running the subsequent phases of research. At the same time, the dev teams focused on developing algorithms that could produce valuable and usable outcomes.

Next rounds of research 

We first ran multiple international research protocols to validate the Concept 3 (tres) storyboards. Based on those findings, we created high-fidelity prototypes to validate the perceived value of the vision. We used various techniques, such as grounded theory analysis, to review the data to follow leads and pursue exciting patterns as they emerged. 
   

The solution

Imagine we are designing an off-road bicycle. You have already designed most components, but now we want to create a unique frame. Let’s use Automated modelling to initiate the ideation process and begin modeling.
   

Setup

To get started, you will need to define Faces to Connect and, optionally, Bodies to Avoid. That’s it. Defining performance requirements, manufacturing processes, and materials is not required to get started. We wanted to simplify the setup process making it feel more like a native modelling command (like Loft or extrude).
   

Switching between alternatives

Within minutes (generally under 1 min), you will be presented with multiple design outcomes of various levels of complexity and style to consider as solutions to move forward with. You can easily switch between outcomes, and the canvas has a direct preview, allowing you to review the outcomes in detail. We evaluated various solutions that would enable users to view a clear preview of the geometry while allowing them to change their inputs and re-run the analysis as needed.
   

Thickness control

As you review the outcomes, you also have a thickness control that will allow you to modify the nominal thickness of the shape via the sliders. Our research revealed that customers desired efficient control over the inputs, and this feature was highly sought after. The concept originated from rethinking the backend technology of the solver, where each point on the thickness slider represents a step in the algorithm's process of generating the outcome.
   

Selecting the outcome

Once the selection is made, the preview inside the canvas will transform into a functional model created from two different forms (Smooth or Prismatic). To effectively convey these shapes and help customers grasp them easily, we have labelled the thumbnail list view items with icons that mimic the native core modeling language (Sketch-based or Form-based).
   

Timeline editability

The design timeline experience in Fusion 360 is a feature that allows users to view and interact with their design feature history in a timeline format. It provides a visual representation of the design process and lets users see the changes made to the design over time. To seamlessly integrate the Automated Modeling outcome into the timeline, the selected outcome automatically creates a set of design features in the timeline history. It is presented in a group format, allowing users to expand and edit each feature created independently.
   

Co-design via the timeline

One of our main architectural challenges was incorporating the set of features the tool creates into the timeline. We derived various strategies and solutions to allow the user to go back in the timeline to modify any previous feature and edit/move any original setup bodies used to create the outcomes or any future modifications to the generated outcomes.
   

Machine Learning Engine

The goal of the machine learning engine is to provide continuous improvements to the generated outcomes the tool creates for our users. We wanted to capture the user’s intent and direction throughout their experience with the command. The critical datasets we collected to make it suitable for training the model were the following.

• Analysing their existing geometry creation process on the given project when the Automated Modeling command is activated.
• Capturing the previous type of outcome selected.
• What modifications have they made in the timeline after they selected an outcome? Have they made the part smoother, changed the thickness etc?

These datasets informed us how to modify and improve the alternatives based on their behaviour. Elements of the machine learning algorithm capture the information during single/continuous working sessions and a global capture of all users when interacting with the command to continuously improve the outcomes.

“I really appreciate the level of control I have; it gives me peace of mind. The results being in T splines is a huge advantage and a game changer.

“Having the real-time feedback is impressive; that's amazing that that's possible. It helps to make it more obvious if you're doing the right or wrong thing.”

Takeaways 

Gracias!