Lecture 04 – MECH 4403:

Artificial Intelligence and Generative

Design for Mechanical Systems

Mehdi Ataei, Alidad Amirfazli

Oct. 1, 2024

25
Brain Storming / Ideation

 Formal methods to engaged in this activity

1. Design Thinking

2. The SCAMPER method

3. Mind Mapping

4. Experience

26
 Visualization method/process
to organize ideas
Mind Mapping

27
 After Generation of Ideas SOLUTION
1. Narrow down your list of ideas
Average human beings cannot grasp a multifactored issues
fully in mind by considering all factors simultaneously and
2. Choose the best option to pursue their importance

A design selection is conceptually and practically a

multifactored problem

A system is needed to make this process possible

▪ Question: How to do the above?

▪ Use judgment/experince to narrow the list (some wild ideas – illigal ideas – etc. -
may be deselected from further consideration… very obvious ones only….)
▫ Class activity… can you remember some of the ideas about Toronto transportation?
▪ For remaining ideas: Use a decision matrix to choose a final solution for desing
analysis
 STEP
1. Determine the 3: DEVELOP
important criteria to A SOLUTION DECISION MATRIX
be considered
A tool used to compare ideas against one another using
a. Issues related to original “pain specific criteria
point” BUT
b. Economic issues
c. Timing issues
d. Regulatory/Environmental issues
e. Manufacturing issues
f. Desing language issues
g. Etc.
2. Determine the importance of each
criterion (its Weight)
3. Decide on the “goodness” of each of
the ideas generated against each of
the criterion determined (its Rating)
4. Calculate the Weighted Score for
reach criterion (WS = W*R)
5. Find Total Score by adding up all WS
from each of the criterion
Subjective/experience  numbers are not absolute
 DECISION MATRIX
Only an Engineering Design tool?
Subjective/experience  numbers are not absolute

Is UofT absolutely better than UBC??

STEP 3: DEVELOP A SOLUTION
We have our winner idea, what’s next?
• Analysis of how the chosen solution can work!
• Overall systems analysis
• Subsystem analysis
• Etc.

• How to do this analysis?

www.shutterstock.com
Engineering Analysis of the Chosen Solution
▪ Escalate the level of Analysis
1. Back of the envelope estimations (guestimation involved)
2. Calculations based on first principle methods
a. First principles methods rely on analyzing the design using fundamental axioms in the given
field, thermal design, mechanical strength, flow interactions, etc.
● Not violate any fundamental laws.
b. Design Software, CFD, Stress-strain analysis, thermal
i. Multiphysics numerical methods
○ Combination of various fields, e.g. thermal stress for re-entry of space vehicles (Fluid mechanics,
Heat transfer, strength of materials, etc.)
3. AI assisted tools
 Class activity: what AI tools you think are available to you?

▪ Good engineering analysis is usually enhanced by the experience.

Back of the Envelope (BoE) estimations in design
▪ BoE estimation is useful during the initial stages of problem-
solving/design
▫ Quick
▫ Rough estimates
▫ Feasibility assessments
▫ Risk identification
▫ Communication purposes
▫ Resource allocation

Famous examples of BoE calculations:

1. Fermi estimating the yield of an atomic bomb by dropping bits of paper and measuring the distance
they were blown by the blast wave (he estimated 10K tons of TNT, the actual result was 18.6K tons).
2.
3. Also, his famous ‘Fermi problem’ of ‘how many piano tuners are there in Chicago?
 Class Activity: You have come out with a new frame
design for an electric bike that is stiffer than what is
Methods for BoE estimations in the market. What should be the target weight of
the frame?

• Order of Magnitude or Benchmarking: Approximate quantities to the

nearest power of 10, or to the existing solutions. For OM analysis, physical
laws should be considered
• use round numbers, simplified calculations, identifying similarities and differences.

• Sizing by Analogy: Extrapolating from known solutions, to estimate

parameters for the new system.

• Dimensional Analysis: Examine the dimensions of quantities involved in the

problem to derive relationships between variables.
• Can help in identifying dependencies and estimating unknown quantities.

• Rule of Thumb: Rules of thumb are simplified guidelines or heuristics

commonly used in engineering and other fields.
Analysis based on first principle methods
▪ All components of the selected concept, their shapes, and dimensions are defined.
▫ Material for components are selected
▫ Manufacturing process is considered
▫ 3D CAD models created
▪ Analysis by way of calculations to verify that the proposed solution meets all
requirements and specifications
▫ Can involve manual or
▫ Computer assisted calculations
▪ Modify and/or replace components/systems that do not meet the specifications
Class Activity: Think of a school or otherwise project that these steps were followed, reflect and report back to class:
1. What was the design problem?
2. How you did your analysis?
3. What methods or tools you used?
4. What do you think of the success of the method you used? Success and improvements?
 First principle
methods

Adapted from: https://www.geeksforgeeks.org/back-of-the-envelope-estimation-in-system-design/

 Watch the video here for PINN

AI assisted tools
▪ TensorFlow
▫ TensorFlow is flexible machine learning platform. This open-source library, developed
by Google, is designed to help engineers develop and train ML models.
▪ PyTorch
▫ Deep learning and complex computations can be facilitated. Supports development in
computer vision, NLP and more.
▪ Physics Informed Neural Networks (PINNs)
▪ PINNs is a machine learning technique that can learn the governing equations
of fluid flow from data.
▫ It incorporates the fundamental principles of fluid mechanics into the neural network
structure,
▫ PINNs can solve inverse problems such as reconstructing the velocity field of a flow
from pressure measurements, which was previously impossible with traditional CFD
approaches.
▫ One of the key benefits of PINNs is that they can handle noisy data and incomplete
information.
▫ PINNs can learn from a small amount of data and generate accurate predictions, which
can save time and resources in the design process.
 Machine learning can be used to have fast rendering and find
more complex geometries optimized for a specific set of
physical constraints. This is called generative design. With deep
learning we can obtain same or increased stiffness for the
desired component while being lighter and more efficient with
Neural networks can also help for accurate and adaptive
material use. From Rami Salem
mesh generation. The mesh refinement allows for more
precise simulation without excessive increase in
computational cost.

Many techniques used in machine learning for image science is

directly transferable. For example, it can improve the resolution
of your results if the computation was done on a coarse grid for
faster simulation and filling in the details by upsampling. These
techniques can also help to extract data from experimental
studies with noisy signals [7,8].
https://ai.googleblog.com/2016/11/enhance-raisr-sharp-
images-with-machine.html
[8] https://arxiv.org/pdf/1902.07358.pdf
 Materials selections
Identifying the criteria that the material 3. Environmental
must meet a. Process to obtain the material
1. Functional b. Source of the material (country or process
a. Strength
to get it)
b. Stiffness
c. Toughness c. Distance for material to be transported
d. corrosion resistance (CO2 footprint)
e. Weight 4. Economic
f. etc. a. Cost to purchase
b. Cost to work with
2. Aesthetic c. Cost to recycle
a. Shape
d. Country/region for sourcing
b. Color
5. Manufacturing Method
c. Tactile
d. wear
▪ Class Activity: Select the materials for your new frame design for an
electric bike that is stiffer than what is in the market?
 ▪ Class Activity: Select the materials for your new frame design for an
electric bike that is stiffer than what is in the market?

▪ Template
Fun Fact
▪ Around 70% of the manufacturing costs of a product can be derived
from design decisions like materials and manufacturing method.

▪ The remaining 30% of the costs make up production decisions like

process planning and tool selection.

▪ Focusing on design optimization reduces the cost of manufacturing.

From: markforged.com
Guidelines for Design for Manufacturing According to me!

▪ How many needed

▪ Materials involved
▪ Product/part complexity
▪ Time to market
▪ Equipment/expertise available
▪ Waste
▪ Quality/precision
▪ Flexibility
▪ Risk
▫ Such as supply chain disruptions, equipment malfunctions, and quality control
issues.
Guidelines for Design for Manufacturing According to me!

▪ Simple
▪ Off the shelf – integrate rather than build -standardized
▪ Less parts
▪ Less joining
▪ More joining Benefits:
▪ More cutting 1. Lower costs
2. Faster production
▪ More drilling
3. Improved quality
▪ Less Milling
▪ Less assembly
▪ Less 3D printing 
Designing for
Manufacturing and
Assembly (DFM/DFMA)

• Desing an apparatus to measure

the dust lift from a surface using a
sticky pad (need to have ability to
measure force & control the
distance before making contact
with the dust on a surface)
Desing 2 versus Desing 1