Creating the Recipe (AI Model)

All we hear about anymore is AI with other frightening yet somehow awesome names like machine learning and computer vision thrown in for good measure. But before we begin imagining the Terminator or the Matrix lets take a moment and try to make sense of what the computer people mean when they use these terms.

First off, AI is simply algorithms, mathematical equations and lines of code written by humans telling the computers how to perform a task. The AI takes the instructions, applies it to some data, and then spits out the results. If the results are what was desired, then great. If not, then the algorithms are refined and put to work on more data, over and over, until we get the result we wanted. These algorithms are called a model. Once the model generates the desired results then its time to feed some data in it and let it start crunching. Eventually as the algorithms identify more and more patterns they are able to refine themselves to better deliver the desired results.

That's a very simple overview. Lets roll up our sleeves and dive a little deeper. But to help visualize how this works, I'm gonna use a fantastic metaphor: Cooking!

We're going to compare building an AI model to creating a recipe. You have ingredients and you know the dish you want to make, but you need to figure out how all the parts of dish preparation fit together. Just Like you have data and you know you want to create a digital twin of your bench. Its just figuring out the steps along the way...

Step 1: Readying Your Ingredients

a. Selecting the Right Ingredients = Selecting the Method of Data Collection

The first step in creating any successful recipe is gathering and prepping your ingredients. For AI, your ingredients are your data. You will need to gather the proper data for your recipe. Like with food, the better the ingredients, the better the dish. While you can't squeeze a drone image like you would an avocado, at least try to make sure it isn't rotten. After all, Garbage in = Garbage out.

Depending on what dish you are trying to make, you need to select the right type of data- does your recipe call for satellite data, manned aircraft imagery, laser scanners, drones? They each have different characteristics that could dramatically effect the outcome of your dish. Just like you can't substitute tofu for a pot roast, you can't use satellite data to generate a 3D model of a bench.

For information on which data capturing method you should choose to measure burden check out this awesome White Paper by Dr. Paul Worsey from the Missouri University of Science and Technology

b. Preparing the Ingredients = Data Cleanup & Preparation

Now that you've gathered your ingredients its time to prepare them, i.e. wash, dry, and chop. When you buy ingredients (or pick them) they may come with dirt, chemicals, or bad spots that could mess up your dish (Nothing worse than some gritty pasta because someone didn't feel like washing the basil).

It's the same with data, remove any extreme outliers, make sure all the data is in the same scale, double check all your files are the correct ones, try to get as complete and accurate a data set as possible. Make sure your data is quality, comprehensive, and applicable!

Drone Data Collection Best Practices

c. Measuring the Ingredients = Data Organization

Then measure all of your ingredients to make sure you have the right quantities in the right proportions. Do you have enough or too little?

Was your drone flying at the proper height and angle with sufficient overlap? A bad flight plan with insufficient overlap or image gaps means bad data which generates a bad model. Images that are too large can slow down processing but don't necessarily add any value to the model. Images that are too small may not contain enough data.

Step 2: Applying Heat = Putting the Training Data into the Early Model

Once your ingredients are gathered, washed, chopped, and measured its time to begin applying heat. This is similar to putting the data into the AI model. First you try browning the meat, sauteing the onions and adding the spices. Next, mix in some diced tomatoes until you have a nice paste. Then add the bigger veggies with some water (or other liquid) and bring to a boil. Finally, turn down the heat and let it simmer for a while.

Once the training data is added to the model the algorithms get applied one after another, each changing the original raw data into something new until its ready to taste the results!

Step 3: Taste Testing = Algorithm Training & Testing

Mamma mia, that's a spicy meat ball!

Somethings not quite right. Perhaps it boiled too long and the veggies are mushy? Maybe there's not enough salt, maybe we don't need to saute the onions? Time to adjust the recipe.

This is the training part of creating the perfect AI dish. The result was not quite what we had hoped for, so we need to identify what went wrong. After all, the algorithms were only doing what we told them to. We need to adjust the algorithms and try again.

Sometimes it takes many tries and a lot of adjustment to get the model right. The algorithms get adjusted repeatedly, perhaps the data we use to train the AI is adjusted as well. The training and validation data sets are run through the updated model to see the new results. We repeat this adjustment process until we get the result we want. Then we add fresh test data to the model to see how well it performs on the new input.

Step 4: Refining the Recipe = Model Evaluation & Selection

You love your dish. Its excellent. You want to share it with your friends. But you find out one of your friends has given up beef. So now it's time to test the recipe on some new ingredients. Will it taste as good if prepared with pork? chicken? lamb? tofu? What can it handle? How will it need to be adjusted?

And with each new data set put into the recipe, does the AI model still generate accurate results? Does it yield better or worse outcomes? The correlation analysis begins.

The model's identified that using a particular blast design at a particular location will generate a particular fragmentation. But what if we change the geology? Its not limestone anymore, now its granite. How does the same blast plan's fragmentation change if the rock mass changed?

Step 5: Bon Appetite, Time to Eat! = Deployment & Modeling

Your Recipe is ready! Time to post it on the internet and share it with the world...

Once the Model has sufficient Data it is able to identify patterns in the data. It can sift through mountains of data in seconds and from the patterns it identifies it is able to make predictions based on the data it encounters.

You know exactly what your recipe will produce if you change the ingredients. You know what will happen if you leave out the onions, switch beef for chicken, or double the basil. If the dish is gritty, you realize you skipped washing the veggies washing phase.

The same is true with your AI model- eventually the AI will be able to "predict" the results based on the data you give it. It will know what will happen if you leave out the onions. It will see early on that you forgot to wash the spinach and you will get a gritty dish. The AI has "learned" based on repeated performance to identify the patterns in the creation of the dish.

Now the AI can warn you, "you forgot to wash the veggies" or "if you use chicken, remember to add some oil" or "if you boil it at high heat reduce the time by half and stir frequently or it will burn."

The same is true with your model. If you use this shot plan on this geology, you will likely get this muckpile shape and this fragmentation. If you use this timing and loading sequence you will likely get this vibration.

Hungry for more?
Change is coming fast, and it can be overwhelming. But those who can adapt gain huge competitive advantages. Technologies like AI make true use of your data, give you predictions and enable affirmative action.

Using the data available to you you'll be back in control of your site.

Time to dig in.

Did You Save Room for Dessert?
AI is huge and we love writing about it.

Check out our 2 Free E-books on AI applications for the drilling, blasting, and mining industries to see all the amazing advances that are available.
AI Guide for Drilling and Blasting 
AI Guide for Mining

TLDR? Watch some videos instead:
YouTube

Another great comparison of AI to cooking is this one