top of page
Search
Raghu Adla
Apr 25, 2018

Augmenting Human Creativity with AI

At Paninian we constantly think about Augmenting Human Creativity in a holistic way, valuing the intrinsic Intelligence of our own species and seeking to Augment it with extrinsic forms of Intelligence to aid you in your Creative Endeavors.

Our soon to be launched Product Offering "AirVoxels 1.0" aims to help you create your own designs which can be readily fabricated using state-of-the-art Additive Manufacturing processes.

This includes packaging the following forms of Intelligence into our Scalable Cloud Platform to deliver the promise of an Intuitive Design Interface which can adapt to your skill level and gracefully lead you through a seemingly complex workflow which also adapts to your own intent to iterate between various steps in the Design Process.

6 Core Components of Extrinsic Intelligence

At Paninian our predominant focus is on excelling in 6 key forms of Intelligence which interleave one another to defy the conventional norms of Sketching, CAD/CAE/CAM or 3D Printing Software Markets.

How ever we believe this sort of convergence is inevitable and also being projected by Market Analysts as the future or evolving trend aiming to cater to the unment expectations of Makers, Designers and Manufacturers alike.

Traditional CAD Software Market Segmenation

Future is all about Convergence and Non-Linear Workflow

Additional Reference:

Aesthetic Intelligence

Our focus today is primarily on understanding how Aesthetic, Geometric, Material,Machine and Artificial Intelligence all converge to empower you to express your intent which allows you custom design your own stuff, be it a pair of cool signature sneakers which fit you well or a bucket list of other items which you wanted which is yet are yet to be made.

How do you express complex inspirations or moods which translate into design. How do you translate the inspiration of fall color, an ocean wave, a birds beak, including the aesthetic elements of your own favorite brand?

Our research and some inspirational work to integrate these elements into an Intuitive and Augmented Design Interface can assure you of simplifying these abstract concepts which can morphed into a real world product with the right mix of "Form and Function" just the way you would desire.

Geometric Intelligence

Shaping your ideas confidently on a Digital Canvas without having to worry about ability to sketch with perfection or spend weeks learning complex 3D Design Tools or understand complex Science or Math which helps us to accomplish "Topology Optimization" allowing these designs to fuse more parts into less and acheive "Lightweighting", is a key bottleneck for most Makers.

The challenge here is to allow users to retain Creative Control of their Design Process while leveraging state of the art "Generative Design" techniques which can take of your Parametric Design Goals.

Material Intelligence

Makers often wonder what material or combination of base materials to choose from if the possibilities are potentially endless. This indeed plays a crucial role in influencing your design choices down to the Microstructure at Micron or even NanoScale.

What kind of Microstructures or patterns would support your material choice or how can we enable you to retain your Creative Form which bending possiblities to accomplish the desired outcome?

Humanity has long been engineering Materials to extract desirable properties to accomlish seemingly impossible things.

Be it ancient Steam Bending technique of Wood or Modern Polymer Material Mixtures or Flexible Agents to attain optimal functionality by tweaking Elextro-Mechanical, Chemical, Optical Properties apart from Color and Texture.

Some conventional Digital Catalogues for Material Intelligence include Granta Design and Material Connexion.

The former one allows Designers to follow a parametric approach, it still limits one to make manual choices which may not be feasible at fine grain level that is possible and needed in the case of Voxel based printing.

High Level Classificiation of 3D Printing Materials

We believe that Intelligent Material Mapping to suit desired properties in different regions of the object need to be dynamically suggested by AI and it is an integral part of DfAM process rather being an afterthought in the workflow.

Additional References:

Machine Intelligence

The evolution of CAD and CAE has inevitably lead to the development of Computer Aided Manufacturing ( CAM ) and methods for attaining high precision in Computer Numeric Control ( CNC ).

However since to the advent of 3D Printing technology, a plethora of approaches are available to users to fabricated components including Fused Deposit Modelling, Selective Laser Sintering, Stereolithography and Multi-Jet Fusion Technology and more recently 4D printing or more exotic Adaptive Fabrication of Bent Wood Assemblies.

These disparate Build Processes themselves have a huge impact on the Design Process which has to be factored in by the Maker at an early stage rather than as an after thought to be fixed in the workflow.

Generation of optimal support structures which either uses similar materials or wax fillings to aid the additive manufacturing build process.

What makes a seemingly Robust Design fail, is often lack of proper understanding or factoring of the Build Process and various machine parameters like Temperature, Viscous Drag Effect of Resin Liquid Pool on account of rising Platform or lack of sufficient layer thickness and other tolerances in case of Multi-Jet Fusion.

Though best practices and rule of thumb precautions and post processing techniques are often published by manufacturers, it's literally beyond the scope of a creative designer or an enthusiastic maker to truely accomplish success at one go.

We believe that this problem should be delegated to the software and also be made design specific to address issues beyond file repair, 2D,3D Collision Detection G-Code Slicing, Misalignment and Thread Mishmatch Checks, Tolerance Analysis for production.

The Maker's Pain Point

Use Case: We get the opportunity to analyze hundreds of prototypes and a wide range scenarios in 3D printing ranging from Components, Jewellery, Metal and Polymer MultiJet Fusion components.

In this use case I wish to share the experience of a Fashion Jewellery Maker at the Academy of Arts trying to make a fancy bracelet using a SLA technique.

Inspite of following best practices recommended by Printer guidelines, the maker failed to accomplish a fine print in the first 4 attempts.

Standard workflow for printing STL Files using SLA

Additional Reference:

A seemingly simple and joyful effort of a Jewellry Maker can soon lead to disappointment and frustration, which can be directly attributed to the lack of convergence of Geometric, Material, Machine and Artificial Intelligence as against current Sequential and Siloed Workflows.

Assuming one has access to all the commercial state of the art CAD,CAE,CAM and OEM and 3rd party 3D printing software available in the market, this Jewellry Maker would find it extremely challenging to resolve this issue for each unique design she might create.

How do we accomplish a feasible relationship between elemental building blocks called "Voxels" which are Digital Representation of material along with its properties and the innocent pencil strokes via Geometric Primitives, Splines and 3D representation as Nurbs or Mesh Representations, Point Cloud Representations for Simulation and Analysis ( FEA, CFD etc ).

How we explore such Design Space using conventional DoE approaches like Taguchi and Factorial Design among others?

Additional Reference:

It is interesting to note that these seemingly complex computational techniques can yield poor results if the user makes wrong assumptions due to lack sufficient expertise and experience in modeling these experiments and simulations to match real world expectations.

Here's a list of parameters she would require to analyze and then apply specific corrections to her file.

Even then one cannot rule out the possibility of the designer loosing control of her creative process while fixing the functional and feasible aspects of the manufacturable design.

This is where our AI steps in to Augment your effort to find a feasible solution for the problem.

Note: Though in case the printer used happens to be a particular brand ( Formlabs ) we do not imply in any way that failure is attributed to the printer.

Rather it is the lack of appropriate tool chain which more often causes the failure.

Intelligent Data Representation - Printing Data

Ability to apply Deep Learning, Rule Engines, Simulation and Intelligent Versioning requires an Intelligent Data Representation ( Data Structure) for iteratively modifying an Object design in the true Convergent DfAM Paradigm.

It's about representing the interdepency of disparate phenomenon effecting disparate properties and their relationships.

In this sense we take the approach of simply Capturing, Analyzing, Discovering optimal solutions for various problems and ultimately the Physical Object is just a manifestation of 'Printing Data'.

This Intelligent Internal Data Representation remains one core aspect of our Innovation. The impact of the run time inference of our AI also depends on this.

3D Printing File Formats and Other Challenges Impeding the Convergence​

One key impediment to the convergence of this DfAM process is the choice of File Formats and flow of information across various vendor software landscape.

STL and OBJ( including MTL ), which have been the most popular file formats so far in the industry are not capable of capturing and passing all relevant information at each stage as well as to the 3D Printer OEM Software, making it difficult to pass on the Geometric Intelligence, Material Intelligence and to leverage the Machine Intelligence or the build process specific to a particular Design being realized.

It is imminent that in future more modern formats like AMF and 3MF to capture additional valuabel information including Complex Geometry like Curved Surfaces, NURBS and Microstructures with high Geometric Fidelity, Color & Texture Mapping, Graded & Composite Materials with support for Mutli-Properties and Constellation.

Further to this we also witness the emergence of custom file formats for Voxelized Representations ( HP and GrabCad Voxel File ) though we have popular opensource formats like OpenVDB ( .VDB ).

We strongly believe that extending proper open file support by 3D Printer as well as other Spatial Computing ( AR Devices ) Hardware OEMs holds the key to unleash the potential of DfAM and aid in the Convergence of the painful and patchy workflows as they exist today.

Efforts to accelerate the adoption of 3MF/AMF Format could possibly spur more innovation in this area.

Addtional References:

The Progress of Aritificial Intelligence in Augmenting 3D World

Overview of Machine Learning Techniques used of DfAM ( Design for Additive Manufacturing)

3D Shape and Object representations in the Digital World:

Rasterized Form

  1. Multi-view RGB(D) images

  2. Volumetric

Geometric Form

  1. Polygonal mesh

  2. Point cloud

  3. Primitive-based CAD models

Problem Classification

  1. 3D Scene Understanding

  2. 3D Object Classification

  3. 3D Object Recognition

  4. 3D Shape Retrieval

  5. Variational Shape Learning

  6. 3D Geometry Analysis

  7. Generative Design for Parametric Optimization

  8. Multi-Material Optimization

————————————————————————————————————————

3D Deep Learning Algorithms based on Inputs

• Deep learning on regular structures • Deep learning on meshes • Deep learning on point cloud and parametric models

————————————————————————————————————————

3D CNNs are also used to better exploit the 3D geometrical information of objects.

These methods can be divided into two major classes:

Extrinsic Methods

1.Multi-view CNNs

2. Volumetric CNNs.

Intrinsic Methods

1. Non-Euclidean CNNs ( Geometric Deep Learning )

————————————————————————————————————————

Augmenting the Generative Design Process:

Stay Tuned.....


bottom of page