Machine Learning Course For Mechanical Engineers

Machine Learning Course For Mechanical Engineers – An example of a deep learning implementation with very simple explanations for mechanical engineers who are new to or ambitious in the field of data science and artificial intelligence. The project was organized by Pro-Match Minds, Nepal.

Artificial intelligence and machine learning seem to be the buzzwords of the moment, as everyone seems to have jumped on the bandwagon. Artificial intelligence seems to play a role in all areas of science. According to Britannica, “Artificial intelligence (AI) is broadly defined as the ability of a digital computer or computer-controlled robot to perform tasks normally associated with intelligent beings.” By intelligent beings he means mostly humans… but maybe not all humans… though

Machine Learning Course For Mechanical Engineers

Machine Learning Course For Mechanical Engineers

Artificial intelligence is a broad term that refers to machine learning. Machine learning uses statistical techniques to improve machines through experience.

New Algorithm Aces University Math Course Questions

Deep learning is a subset of machine learning that uses multi-layer neural networks that mimic the human brain and can learn incredibly difficult tasks with enough data.

We will talk about deep learning techniques and its possible role in machine engineering. Some general examples may follow

First, what is an image classifier? The famous AI that recognizes cats and dogs pictures should come to mind. Here is a link to the code for such a program. The dataset used consists of images of cats and dogs, and the algorithm learns from this and can predict with 97% accuracy whether a randomly displayed image is a cat or a dog.

We use similar code, but use nuts, bolts, washers, and locating pins as cats and dogs… because engineering.

Mechanical And Manufacturing Engineering

So how does it work? The algorithm can (efficiently) classify images using a machine learning algorithm called convolutional neural networks (CNN), a technique used in deep learning. We use a simpler version of this model, called sequential, to allow the model to divide images into the four categories of nuts, bolts, washers, and location pins. The model learns by “observing” a set of training images. After training, we will see how accurately he can predict the image (which he has not seen).

Flowchart of a machine learning algorithm trained on images of nuts and bolts using a neural network model. (Original image: pngfind)

We downloaded 238 components of 4 classes (total 238 x 4 = 952) from various component libraries available online. Then we took 8 different isometric pictures of each part. This was done to maximize the available data, as only 238 images for each segment would not be enough to train a good neural network. One class now contains 1904 images (8 isometric images of 238 sections) for a total of 7616 images. Each image is 224 x 224 pixels.

Machine Learning Course For Mechanical Engineers

Then we have labels with the numbers 0, 1, 2, 3, each number corresponds to a specific image and means it belongs to a specific category.

Mtech Mechanical Engineering 2021: Colleges, Syllabus

After training on the images above, we can see how well the model predicts an unseen random image.

The process took place in 7 stages. We’ll get to the details later. A summary is a conclusion

We have downloaded data for various nuts and bolts parts from various parts libraries on the Internet. These websites have many 3D models for standard parts from different manufacturers in different file formats. Since we use the FreeCAD API to produce the images, we downloaded the files in a neutral format (STEP).

As mentioned above, 238 segments of each of the 4 classes were loaded, for a total of 952 segments.

Is Data Science Useful For Mechanical Engineering?

We then ran the program using the FreeCAD API, which automatically took 8,224 x 224 pixel isometric screenshots of each part. FreeCAD is a free and open source general purpose parametric 3D computer design modeler written in Python.

As mentioned above, each data consists of 8 images of 224 x 224 pixels. We now have a total of 1904 images from each of the 4 classes, so a total of 7616 images. Each image is treated as separate data, even if there are 8 images from one segment.

Images are placed in separate folders according to categories. That is, we have four folders of nuts, bolts, washers, and locating pins.

Machine Learning Course For Mechanical Engineers

Next, each of these images was converted to an array containing pixel values ​​in grayscale. Pixel values ​​range from 0 (black) to 255 (white). So it’s actually 255 shades of gray.

Mechanical Engineering Ms Degree

Now each of our images becomes a 224 x 224 array. So our entire dataset is a 3D array of dimensions 7616 x 224 x 224.

Similarly, we create a label dataset by assigning the following integer values ​​for the specified classes to the corresponding index in the dataset. If our 5th (

) data in the data set (X) is the location pin, the 5th data in the character set (Y) will have a value of 0.

Since this is an image recognition problem, we use a convolutional neural network (CNN). A CNN is a type of neural network that processes image data well. A neural network is a type of machine learning algorithm that learns like the human brain.

Mechanical Engineering Vs. Computer Science

Convolutional neural network. A basic overview of how our algorithm works (original image source: Article by Adit Deshpande)

The code below looks like our CNN. Don’t worry if you don’t understand. The idea is that 224 x 224 features of each of our data passes through these networks and responds. The model adjusts its weights accordingly and can predict the class of a random image after several iterations.

Here is a YouTube video by Mark Robert (NASA Mechanical Engineer) explaining how neural networks work with very little coding.

Machine Learning Course For Mechanical Engineers

Now it’s time to train the model using a dataset of 7616 images. So our [X] is a 3D array of size 7616 x 224 x224 and our label set [y] is a 7616 x 1 array. For all training purposes, the data set should be divided into at least two parts: training and validation (test) sets (when only 2 sets are involved, test and validation are used interchangeably).

Mechanical Engineering Technology

The training set is the data that the model sees and trains on. This is the data that adjusts its weight and learns. The accuracy of our model on this set is the training accuracy. This is generally higher than inspection accuracy.

The test data usually come from the same distribution as the training set and are unsampled data. After the model is trained from the training set, it tries to predict the data from the validation set. How accurately this predicts is our verification accuracy. This is more important than the accuracy of the exercise. This shows how well the model generalizes.

In real life, it is customary to divide it into three parts. Practice, test and test.

In our case, we only split it into training and test sets. It will be an 80-20 split. 80% of images are used for training, 20% for testing. It is prepared on 6092 samples out of 7616 samples tested on 1524 samples.

Ai And Ml Syllabus

The number of epochs is a hyperparameter that defines the number of times the training algorithm runs through the entire training data set.

An epoch means that each model in the training dataset has had the opportunity to update its submodel parameters. An era consists of one or more collections.

You can think of a loop for the number of epochs each loop moves through the training data set. Inside this for loop is another nested for loop that iterates over each batch of samples, where batch is the “batch size” number of samples. [2]

Machine Learning Course For Mechanical Engineers

That is, our model runs through 7616 samples a total of 15 times (epochs) and adjusts its weights each time to make the prediction more accurate each time. In each epoch, it goes through 7616 samples, 64 samples (batch size) at a time.

Mechanical Engineering Meng

The model constantly updates its weights to minimize loss (waste), thus giving us the best possible accuracy. The cost is a measure of the model’s inaccuracy in predicting the image category. Cost functions are used to evaluate how poorly the models perform. Simply put, the cost function is a measure of how inaccurate the model is in terms of its ability to predict the relationship between X and y. [1]

After 15 epochs of training, we can see the following graph of loss and accuracy. (Cost and cost are interchangeable for our case)

As the model is trained more often, the cost is reduced. Image categorization improves with each era. The model does not significantly improve performance on the validation set.

Accuracy increased in the form of model trains for each era. It will be better at classifying images. The accuracy of the validation set is lower than the training set because it was not directly trained on it. The last value

Machine Learning Courses & Certifications [2023]

About nakamichi

Check Also

Certification Courses For Civil Engineers

Certification Courses For Civil Engineers – The BEST Certificate is a transcribed certificate combining multidisciplinary …