Machine Learning Based Industrial Sorting System using Neural Network

Industry 4.0 is already demonstrating its value by enabling manufacturers to reach this goal more successfully than ever, and one of the core technologies driving this new wave of ultra-automation is Industrial Artificial Intelligence AI and Machin

Project Title

Machine Learning Based Industrial Sorting System using Neural Network

Project Area of Specialization

Artificial Intelligence

Project Summary

Industry 4.0 is already demonstrating its value by enabling manufacturers to reach this goal more successfully than ever, and one of the core technologies driving this new wave of ultra-automation is Industrial Artificial Intelligence AI and Machine Learning. The machine learning for image recognition allows a computer to learn from a training data set what the important "features" of the images are. By using a hierarchy of numerous artificial neurons, machine learning can automatically classify images with a high degree of accuracy. Thus, neural networks can recognize different species of cats or models of cars or airplanes from images. Sometimes neural networks can exceed the performance of the human eye for certain applications. So we aim to design a sorting system that uses machine learning and process a huge amount of information and make decisions based on logic, enabling the system to learn and to complete the task without any further programming. The system uses PLC as the main controller and Raspberry Pi 3 to take images of the Lego bricks with a camera, and in a first phase, runs a small-scale neural network on TensorFlow to detect whether or not the image is of a Lego brick. It then forwards the image to a larger TensorFlow neural network running on a Linux server to perform a more detailed classification.

One of the current challenges with machine learning is that you need to have a large number of training datasets. To train the model, we will feed the feeder one by one with the same type of Lego brick to capture each and every possible angle of the Lego brick we will repeat this multiple times. we might take 5000+ images, the more images the better will be the result. We would use the Inception V3 model to classify the 11 brick classes. We would run the training on a GPU TensorFlow library that leverages my desktop’s CUDA enabled NVIDIA GPU. After training the model, it would be ready for the test. The procedure of the sorting system would be as the Automatic feeder would feed the Lego bricks into the Conveyor Belt-1, the servo motor of the automatic feeder and DC Gear motor of Conveyor Belt-1 will be controlled by the PLC, and we would design a mechanical system that would separate the Lego bricks and would shift Lego Bricks to Conveyor Belt-2 whose motor is also controlled by the PLC. There would be an IR Beam sensor that detects the Lego brick and the PLC will stop the Conveyor Belt-2 at the same time the PLC will give the command to Raspberry Pi for capturing the picture. The Raspberry Pi will capture the picture through a camera and then the small-scale neural network on TensorFlow would run to detect whether the image is of Lego Brick or not. It then forwards the image to a large TensorFlow neural network running on a Linux server to perform the more detailed classification. When the system would identify the class the Raspberry Pi/PLC would control the servo motor to drop the Lego brick to the respective bin.

Project Objectives

The aim of this project is to build a machine that can reliably sort between 10–20 types of Lego bricks reliably without manual feeding using machine learning and an image-based neural network classification model and helps in increasing the efficiency of the existing sorting system.

The goal of the project is to introduce machine learning in the sorting system, to build a neural network to perform classification, automation of sorting system and the interfacing of PLC with Raspberry Pi through RS-485 or RS-232, to increase efficiency, to reduce the limitations

The existing sorting system that uses image processing has some limitations such as the angle from where the image was captured during training and if the object has placed other than the captured position the system would fail to recognize the object. This project would address these limitations through machine learning and neural network.

Project Implementation Method

When it comes to implementation of the proposed Machine Learning based Industrial Sorting System there would be many challenging tasks because we plan to introduce machine learning in the system that uses PLC using Raspberry Pi.

We would start first by developing the mechanical design for the Lego sorter. The mechanical system would have two conveyor belts and a Lego brick feeder and sorting mechanism and the bins.

Figure 2 shows the approximate mechanical design of our project. The developing of mechanical design would be a very challenging task because the feeder would drop multiple blocks on the conveyor so we will design a Lego bricks separation system after being separated the position of Lego brick is very essential because camera should be able to capture the image of Lego brick as the whole sorting system is depending on the image of Lego brick so we would develop a mechanism which will address this challenge and will align the Lego brick in the range of camera.

After developing hardware of the system the model will be ready for training. As machine learning requires a large number of datasets so we will provide sufficient datasets of each type of Lego brick. First, we will manually feed the feeder with the same type of Lego brick in order to get every possible picture of the Lego brick in this way we will train our machine for each and every type of the Lego brick.

There are different types of machine learning solutions for image classification. But the most suitable and the most accurate one is CNN -Convolution Neural Network CNN applies filters to detect certain features in the image. The way the convolutional neural network will work fully relies on the type of the applied filter. So, when applying machine learning solutions to image classification, we should provide the network with as many different features as possible. It will then analyze their values upon training.

When the system is trained we will run the first test of the system, we will feed the feeder with random Lego bricks, and then feeder will feed the Lego bricks on conveyor 1 where the Lego brick would be separated and move to conveyor 2.

The speed of conveyor 1 will be slower than conveyor 2 as we plan to implement the Lego separation system on conveyor 1.

When the system would have sorted all the Lego bricks we will check the results whether it is correct or have some errors. If the system contains errors then we will provide more data to our system because the more data yield a more accurate result. After providing more data we will run the second test and will check the result and would take further decisions according to outcomes of the machine.

The project then is finally tested and will determine the separation effectiveness, Classification accuracy, and sorted accuracy.

Benefits of the Project

Artificial Intelligence (AI) is a cognitive science to enables humans to explore many intelligent ways to model our sensing and reasoning processes. Industrial AI is a systematic discipline to enable engineers to systematically develop and deploy AI algorithms with repeating and consistent successes. We believed that this project will work as a guideline and roadmap for industries towards the real-world implementation of Industrial AI.

The proposed project Machine Learning based Industrial Sorting system using Neural Network is addressing the limitations of the existing sorting system. There are many automatic sorters on the market, but they have limitations in terms of performance and cost, and small industries don't tend to use them. So this project would overcome these limitations and would provide more accurate results.

It would be giving a machine the ability to learn, it lets the system make predictions and also improve the algorithm on their own.

It is versatile it can sort any type of object such as nut bolts, defected products, etc. we just have to train our model according to the object.

It would increase the efficiency and precision of the system.

Technical Details of Final Deliverable

The final deliverables would consist of a Hardware system along with the trained model of the system. The hardware system would consist of PLC, Raspberry Pi, Two conveyors, Camera along with backlight and IR beam sensor, and the sorter system.

We would deliver the trained model of the system on which the system would make predictions.

Final Deliverable of the Project

HW/SW integrated system

Core Industry

Manufacturing

Other Industries

Others

Core Technology

Artificial Intelligence(AI)

Other Technologies

Artificial Intelligence(AI)

Sustainable Development Goals

Industry, Innovation and Infrastructure, Responsible Consumption and Production

Required Resources

If you need this project, please contact me on contact@adikhanofficial.com