GOAL

To develop a model that answers a question

2025-08-18_v1.0

Algorithms & Machine Learning

Applied Machine Learning with Catapults

Breakout Development Team

2

JAMES VENDITTO

College: University of Notre Dame

Major: Electrical Engineering

JOANNA CAUDLE

College: Georgia Institute of Technology

Major: Mechanical Engineering

Industry Experience: Bechtel Power Corporation

3

PART 1

Conceptual Basics

4

What is Machine �Learning

WITH THANKS TO AI ADVOCATE AT GOOGLE, MR. LAURENCE MARONEY

5

How Do Computers Do This?

The Machine Learning algorithm creates a mathematical model that can answer questions like “Is that a Dog or a Cat?” or “What number is that?”.

The mathematical model learns using data and aims to get better with the more data �it uses.

6

Can you think of some other examples where machine learning might be a very useful tool?

Any text here?

7

Computers Develop Algorithms

A machine learning algorithm is given a “teaching set” of data, then asked to use that data to answer a question.

Algorithm (n): A process or set of rules to be followed in calculations or other problem-solving operations, especially by a computer.

“THE TOP 10 AI AND MACHINE LEARNING USE CASES EVERYONE SHOULD KNOW ABOUT” – FORBES SEPT 30, 2016

Teaching Data

New Data

Machine Learning Algorithm

Model

Prediction

8

Can a Computer Recognize Drawn Numbers?

Try this simple machine learning experiment by clicking here.

IN THE LEFT HAND BOX, DRAW A DIGIT FROM 0 TO 9 AND SEE IF YOUR COMPUTER CAN RECOGNIZE IT.

9

Let’s Go Back to Number Recognition

10

This Is The Number 9!

11

Training and Using the Models �of the Digits

​

​

12

Do you think there are things a computer can learn to do that we can’t do?

Any text here?

Follow up info here?

And conversely, can you think of things humans can do that computers can’t do?

​

13

PART 2

Lab Activity

14

Today’s Machine �Learning Lab … �The Catapult

DESIGN CHALLENGE:

Fundamentals of mathematical model building:

• Factors and Responses
• Developing a Model that relates these two variables

Gather catapult data set, model the response, and use your model to hit a target

Determine the catapult settings needed to hit a specific target

15

Now you might be wondering:

You are asking us to make a catapult - what does that have to do with machine learning?

We know that machine learning is an extremely complex process. We understand that is not a realistic ask with the tools and time that you all have.

Instead, you will be working with a catapult. You are doing the machine learning by examining variables with the catapult as your tool.

16

Today’s Machine Learning Lab … �The Catapult

STEP 1

Build a catapult of your own design

STEP 2

Test various input factors and collect data

​

STEP 3

Analyze your data graphically and develop a model

STEP 4

Make a prediction about how to set up your catapult to hit a target

Teaching Data

New Data

Machine Learning Algorithm

Model

Prediction

17

The Engineering �Design Process

Imagine you are a team of mechanical engineers building a catapult.

How would you use the Engineering Design Process to create a catapult that can accurately and reliably hit a fixed target?

18

Catapult Terms

LEVER

A simple machine used to easily lift or move objects

FULCRUM

The support that a lever rests on and pivots about

PROJECTILE (WEIGHT/PAYLOAD)

The object that a catapult launches

ARM (LEVER)

The lever that a catapult uses to launch a projectile

STRUCTURE

Includes the BASE, FRAME, and STOP

POTENTIAL ENERGY

Energy that is stored to be used later

KINETIC ENERGY

Energy of a moving object

BASE

19

Lab Overview

1. DEFINE THE PROBLEM: What problem or challenge are you trying to solve? (Create a catapult you can set to repeatedly hit a specified target)
2. RESEARCH: What design could you use? What materials will you need? (Note: you have some in your Engineer’s Toolkit (sticks, �rubber band, ball) but can use other things you have at home)
3. DESIGN YOUR SOLUTION: Draw out a diagram of your team’s design.
4. BUILD YOUR FIRST PROTOTYPE: Create your first prototype of your bridge.
5. TEST: Place your catapult on the floor or table and practice launching the catapult a few times. Are there things you should change about your design before you begin collecting data? If so, iterate and improve your design. When you are satisfied with your design, you are ready to collect your data for a series of pull back angles. Be sure to fill out the chart in your student workbook in order to organize your data. (See example on the next slide.)
6. ANALYZE YOUR RESULTS: Use your observations and data from step 5 to answer the questions in your student workbook.

IDEAS:

INTRODUCTION TO AI:

Applied Machine Learning with Catapults

21

Prototyping, Testing, and Analyzing the Results

CATAPULT INPUT DATA

PROTOTYPE #1

PERSPECTIVE SHOWS�

Tension of the rubber bands will differ��# of rubber bands

PROTOTYPE #2

22

Engineering Process: Test

Your team will fill out a chart like this:

23

A Catapult For Data Gathering Example

​

​

USING PHYSICS TO PREDICT FLIGHT DISTANCE WOULD BE VERY DIFFICULT… WHY?

CATAPULT SETTING AT 2CM

CATAPULT SETTING AT 5CM

24

A Catapult For Data Gathering Example

ARM SHOULD BE PULLED ALL THE WAY BACK TO THE FLOOR WITH ONE HAND, AND THE CATAPULT SUPPORTED WITH OTHER HAND

25

A Catapult For Data Gathering Example

​

​

Plotted data along with a hand-drawn (red) line of the model

What catapult setting should you use to fly 200cm?

26

OPTIONAL EXTENSION

Multiple Factor Model

27

Adding a Second Parameter - Number of Sticks

4 STICKS AT CATAPULT SETTING OF 3CM

2 STICKS AT CATAPULT SETTING OF 3CM

28

Adding a Second Parameter - Number of Sticks

​

​

Plotted data along with a hand-drawn (red) line of the model for varying number of sticks

This is now a model for 2 variables (Catapult setting and number of sticks)

29

Optional: Create a Spreadsheet

​

​

Help Videos:

• Input the Data
• Add the scatter plot and trendline
• Predicting distance

Spreadsheet

30

Conclusion: You Are Machine Learning Using a Catapult

Input Data�Information (+ Answers)

Output�Optimum Model

CATAPULT INPUT

Pull back angle, height, number of rubber bands, etc.

CATAPULT OUTPUT

Distance ball predicted to travel shown by Trendline on Excel graph (Model)

Machine Learning

Relationships�Patterns�Dependencies�Hidden Structures

Algorithms + Techniques

31

Optional Extensions:

• Create a Virtual AI Sorter with Google’s Teachable Machine
• Play with a Neural Network Simulator in TensorFlow

This section will provide an overview of the extension lab(s) and/or optional activity(s).

​

• These activities are opportunities for students to dive deeper and ideate.
• The materials associated with the extension labs may not provide as many detailed instructions as the main lab activity.

Create a Virtual AI Sorter with Google’s Teachable Machine

What you’ll be doing:

• Designing a AI sorter using pictures of various materials that you find

Ideal option if you:

• Want to use the engineering design process to test technology.

​

Play with a Neural Network Simulator in TensorFlow

What you’ll be doing:

• Tinker with a simple neural network in TensorFlow Playground

Ideal option if you:

• Want a better understanding of how a neural network is constructed.

32

Optional Extension Activities

Any text here?

Teachable Machine Sorter

Neural Networks

Teachable Machine Sorter

• In 2018, the total generation of municipal solid waste (MSW) was 292.4 million tons.
• Approximately 69 million tons were recycled and 25 million tons were composted, which is equivalent to a 32.1% recycling and composting rate.
• Nearly 35 million tons of MSW were combusted and more than 146 million tons of MSW were landfill.

Research on the Problem

HOWEVER, if there was better technology available to sort through this waste, the recycling and composting rate could be increased!

• Glass
• Wood
• Metal
• Paper
• Plastic
• Rubber
• Cardboard

Design Considerations: Materials

What kind of materials/products can be identified to make recycling more efficient?

• E-Waste (electronics, batteries, etc.)
• Compost (Food, garden trimmings, etc.)
• Hazardous waste (paint cans, pesticides, etc.)
• Construction waste (concrete, roof shingles, etc.)

How to use Teachable Machine:

​

LINK: https://teachablemachine.withgoogle.com

​

GATHER: Gather and group your example into classes, or categories, that you want the computer to learn.

​

TRAIN: Train your model, then instantly test it out to see whether it can correctly classify new examples.

​

Testing: Using Teachable Machine

Create an AI sorter that sorts between one of the following options:

​

• Plastic vs Metal
• Paper vs Compost
• Cardboard vs Plastic

​

​

​

​

​

NOTE: Think about the variety of materials you have in your class to use as samples (ex. Plastic→bottle, container, grocery bag)

Testing: Option 1

Create an AI sorter that sorts between at least two materials.

​

Be creative, and think about what would make recycling more efficient now!

​

​

​

If you need help thinking of materials, look back at slides 33 & 36.

​

You can also conduct research on what materials are commonly found at landfills/recycling plants and create an AI sorter that identifies materials that concern you most.

Testing: Option 2

39

Testing Efficiency

What is a sample?

A sample is how many examples the machine has been given of a specific category during the “train” portion (material in this lab).

You will then compare the efficiencies of these trials.

Users can hold down the train button instead of clicking it to get a stream of photos (similar to a video).

Whether you choose to do Option 1 or 2, try two separate trials where Trial 1 has less samples than Trial 2.

TRIAL 1:

TRIAL 2:

40

Engineering Process Step : Trials

Create a Virtual AI Sorter with Google’s Teachable Machine

What you’ll be doing:

• Designing a AI sorter using pictures of various materials that you find

Ideal option if you:

• Want to use the engineering design process to test technology.

Play with a Neural Network Simulator in TensorFlow

What you’ll be doing:

• Tinker with a simple neural network in TensorFlow Playground

Ideal option if you:

• Want a better understanding of how a neural network is constructed.

41

Optional Extension Activities

Any text here?

Teachable Machine Sorter

Neural Networks

Build a Neural Network

Extension

43

Extension Development Team

MARY BRUSCO

College: University of Notre Dame ‘26

Major: Computer Science & Engineering

MARY JARRATT

College: Virginia Tech, Ph.D. student ‘24

Major: Engineering & Robotics Education

JOANNA CAUDLE

College: Georgia Institute of Technology

Major: Mechanical Engineering

Industry Experience: Bechtel Power Corporation

JASUN BURDICK

College: University of Central Florida

Major: Industrial Engineering

Industry Experience: Building Automation, Robotics

In the last activity, you used the Engineering Design Process (EDP) to build a catapult and model its behavior. You learned about the problem using data, developed a model for your catapult, and iteratively tested your design to make your catapult more accurate and precise, all without step-by-step instructions!

​

We mentioned that machine learning is a way for computers to do the same: learn from data, build a model for solving a problem, and then iteratively test that model to make it more accurate and precise.

​

There’s just one problem: How do we program a computer to build a model on its own?

Introduction

There’s something about your brain and the way it’s structured that allows it to analyze data, come up with solutions to problems, and learn from its mistakes to build even better solutions. Your brain is an example of a powerful computer that can use EDP to solve problems on its own!

​

So, since your brain can do what we want our computer to do, let’s give our computer a brain.

​

Specifically, instead of writing step-by-step programs for each problem we want our computer to solve, we’ll make a program that mimics the structure of your brain to hopefully give our machine the ability to “learn.” This “machine brain” is what we call a neural network.

​

NEURAL NETWORK

• A type of computer program modeled after the human brain that allows computers to make decisions about data on their own, without step-by-step instructions.

Solution: Your brain!

IN THIS EXTENSION…

​

You will explore neural networks by building one in a virtual playground called TensorFlow. By the end, you’ll have built a neural network that can classify points in a data set all by itself!

46

What You’ll Do In This Extension

1. Introduction to Neural Networks with TensorFlow
2. Connection to the Catapult Lab
3. Tinker with the TensorFlow Playground
4. Learn How the Model Works
1. Data Input
2. Neurons and Weighting
3. Error Reduction
5. The Challenge - Build a Simple Model

47

Part I: TensorFlow

TensorFlow is a free and open-source software library for machine learning and artificial intelligence. It can be used across a range of tasks but has a particular focus on training and inference using neural networks.

TensorFlow has a playground that lets you expand your understanding of neural networks by developing a simple model to sort a collection of data points.

48

Part I: TensorFlow Playground

• The playground simulation attempts to build a neural network model consisting of Features and layers of Neurons that predicts where the colored dots are
• Features are inputs to the model. Neurons that have particular shapes are what the computer program uses to build a model for classifying the data.
• By weighting and summing connections into neurons you can build more advanced detection

Part II: Connection to our Catapult Activity

Think of the student as the machine/computer

The more input data you give it the more accurate the model is getting.

50

Your Catapult is Machine Learning

CATAPULT INPUT

Pull back angle, height, number of rubber bands, etc.

CATAPULT OUTPUT

Distance ball predicted to travel shown by trendline on Excel graph

Think of the features of the neural network the same way you think of the different features of your catapult.

Intuitively, the student learns that an important feature of the catapult is the force you push down on it to make it go farther. Similarly, the computer model learns by guessing and checking which features to emphasize more. The model shows which features are more important to the output by assigning a heavier weight (we see this through the thickness of the line coming from a feature or neuron).

Think of the importance you give each feature in the catapult as the weight of importance the computer program gives each feature in the network.

53

Part III: Introduction to the Playground

WATCH ME

54

Part III: Tinker in the Playground

• Go to the playground and try running the model a few times by pressing the play and reset buttons
• Watch the model learn to reduce error
• Change between data sets and see how it changes.
• Look at the Training Loss Graph after each data set. With the sample setup, the model is only about 50% correct on some of the data sets, which is just like guessing or flipping a coin.

55

Part IV: How does the model work?

The TensorFlow Playground has 4 data sets comprising the shapes on the left.

• Circle - Top left
• Exclusive OR - Top right
• Gaussian - Bottom left
• Spiral - Bottom right

Data is broken up into training and testing (validation)

These data sets are similar to the written number data set you saw in the handwriting example in the Machine Learning introductory video.

56

Neural Network: How does it work?

How do you write code that classifies whether a data point is orange or blue? Perhaps you could draw an arbitrary diagonal line between the two groups like that shown on the next slide. You could define a threshold to determine in which group each data point belongs.

​

Let's look at a simple classification problem. Imagine you have a dataset such as the one below. There are two groups of data (blue and orange) that the model is searching for. Each data point has two values X₁ (horizontal axis) and X₂ (vertical axis).

57

Neural Network: How does the program work?

If x₁ + x₂ > some threshold value → Then the dot is blue

​

But the thing is, a programmer has to find appropriate values for variables and instruct the computer how to classify the data points. That’s A LOT of steps of code!

​

A neural network computer program uses artificial neurons and trains itself to classify the data.

58

Neural Network: Neurons and Weights

• Neurons

Neurons are formed by combining the weighted inputs of features.

In a simple neural network model, the mathematical equation might look something like this:

If w₁x₁ + w₂x₂ > some threshold value, ��Then the dot is blue, where w₁ and w₂ are the weights given to each feature

​

• Layers

Adding layers allows the program to combine previous neurons to make a more complex model to classify data

59

Part V: Challenge: Build Your Own Neural Network

Requirements: Build the simplest model that has the lowest error rate.

Plan: Try selecting data sets, features, layers, and neurons

Run the model for short bursts (< 1,000 epoch)

Record the testing loss (error rate)

Improve: Change an input and rerun

60

Challenge

As you gain understanding of how the model is working try this challenge.

Build a model with 5 items: features + neurons. (below has 4)

Test your model against the 4 data sets without altering it

Record your testing loss for each at < 1,000 epoch.

Sum your total error = ______

​

​

​

Link to the: Playground

61

Next Steps: How is machine learning used in business?

Machine learning is being used in the banking, healthcare and music and entertainment industries to name a few.

​

Check out this case �study from Netflix �about how machine �learning is used �in their organization.

​

​

​

62

Next Steps: Want to learn more?

Here are some concepts you may be interested in and also a great video that further explains neural networks:

• Neural Networks
• Linear Algebra
• Calculus

​

​

63

Continue to Explore

If you liked today’s breakout, you may be interested in these topics:

• AI
• Machine Learning
• Computer Algorithms
• Statistics - Regression

Types of engineering relevant to the AI:

• Mechanical Engineering
• Computer Scientist
• Mechatronics
• Operations Research
• Information Engineering

Types of engineering relevant to Machine Learning:

• Computer Science
• Computer Engineering
• System Engineering
• Operations Research

Thank you!

Any text here?

Follow up info here?