BLOG

October 14, 2020

HARNESSING AI AND DEEP LEARNING FOR VISION INSPECTION

Rapid development in vision inspection technology with AI and Deep learning have pitched manufacturing industry at the tipping point of a new industrial age. The new age technology is poised to make a major impact on manufacturing industries. AI and Deep Learning algorithms will analyse, process and classify images to assess product quality, guide machines, read texts, ensure traceability, and deliver valuable data for optimizing operations in industrial manufacturing.

Deep learning Technology

Deep learning technology is based on artificial neural networks which mimic neural network of human brain. The advanced technology assists automated vision inspection system with cognitive abilities. Like the neural network in the human brain comprising of several interconnected neurons which relay response after receiving inputs, the artificial neural network is a virtual neural network, comprising of a collection of codes running statistical regressions or models which based on the input provide an output

Artificial neural network architecture of a Deep learning based vision inspection model consists of interconnected input layer, hidden layer, and output layer which assist in understanding complex data. A vast amount of data is fed to train the neural network architecture for it to learn to analyse, detect and classify images. For training a neural network a set of virtual neurons assigned with different weights are mapped out, the weights are continually adjusted until training data with the same label, consistently recognizes the correct patterns in images.

Self-Organization is the Key

A neural network does not need to be programmed overtly to perform a complex task, it has a unique ability of self-learning from a vast amount of data, to detect trends and find patterns. Just like the human brain process current inputs along with inputs from past experience, Deep learning vision inspection system with a vast amount of data along with a collection of artificial neural network enables the machine to identify, process or classify images quickly and consistently. Deep learning based vision inspection can handle multiple variables like variation in parts and defect size on various backgrounds. The technology is used when there are several variables, where simple rule based machine vision will not be able to give reliable results, for example for doing surface inspection for electronic devices, where defects such as scratches, dents, digs, discoloration appear on multiple locations and in different combinations. The defects can be present on screen, on the curved sides, or the back which will require deep learning based vision inspection to capture all defects. With numerous combinations of different variations, Deep learning model keeps on continuously learning and after sufficient training, can differentiate a good part (without defects) from a bad part with enhanced accuracy over a period of time

For developing the Deep Learning algorithm for predictive inspection, one goes through the following steps: A. Gather training data:
    1. Capture images
    2. Perform image annotation
B. Train the model:
    1. Pre-processing using the GPU
    2. Model training using GPU
C. Image prediction:
    1. Deploy the model at customer site and capture new images
    2. Image prediction using the deployed model and inspection results for customer

Data is the fuel for AI

It is very important to note that, for AI based machine learning and deep learning inspection models to function well, it is essential to have diverse data set. The AI enabled inspection model with a vast set of data enables the system to respond to changes in variables and environments. Deep learning based models continuously learn itself, and overtime as the model receives more and more data, it improves in accuracy. Lack of annotated and labelled data is the major barrier for adopting deep learning based inspection models. A robust good quality data set without systemic errors is essential for an accurate deep learning inspection system.

Deep learning models excel at addressing complex vision inspection application with multiple variables. The technology works best for surface quality inspection to check for scratches, dents, and digs on shiny or rough backgrounds, it is also used for complex OCR and associated inspection projects. Deep learning inspection models are highly flexible and can be retrained for new image data on the factory floor with ease and precision.

With advent of high performance image processing techniques, GPU (graphic processing unit), ASIC (application specific integrated circuit) and cloud storage, vision technology with AI will increasingly be adopted by manufacturers to improve productivity, product quality and worker safety. With detail specific industry knowledge, we at Griffyn Robotech, have introduced intelligent inspection solution-OPTIVITY®,

which offer high resolution, vision inspection systems to provide complete inspection coverage for large and small components with ease and precision. At Griffyn Robotech, we provide customized, fully automated, high speed vision inspection systems using leading edge technology with robotics, machine vision and AI.

Connect with us today, to know how vision inspection technology can help your manufacturing facility be Industry 4.0 ready

For more information mail to

and visit

Leave a Comment

Your email address will not be published. Required fields are marked *