/aut_ai_project

This repository contains the files of my project for the "Artificial Intelligence" course at AUT (Tehran Polytechnic).

Primary LanguageJupyter NotebookMIT LicenseMIT

Defective Pump Impeller Detection Using a CNN

This repository contains my project for the Artificial Intelligence course at AUT (Tehran Polytechnic).

Project Description

This project deals with a classification problem. Since we are trying to classify images, we have decided to solve it with a convolutional neural network due to its abilities in image processing.

Dataset

The data we used are available on Kaggle.

Examples

Defective Pump Impeller

Defective Pump Impeller

Non-Defective Pump Impeller

Non-Defective Pump Impeller

The Convolutional Neural Network

We created a CNN with the following architecture:

  • Conv2D: 64x64x8
  • MaxPool: 32x32x8
  • Conv2D 32x32x8
  • MaxPool: 16x16x8
  • Flatten: 2048
  • Dense: 16
  • Dense: 16
  • Dense: 1

Total params: 33,737

The Convolutional Neural Network

Training and Validation

Accuracy

We used binary_accuracy as the performance metric of the network.

Accuracy

Loss

We used binary_crossentropy as the loss function of the network.

Loss

Results

Our CNN model can classify images of pump impeller with an accuracy of ~96%.