Industrial Imaging & Machine Vision
Kritikal’s Deep-learning based technology platform provides fast and accurate solutions for AI-based Defect Inspections like Barcode Reading, 360° Inspection, 3D Imaging, Robotic Guidance, and Surface Inspection. Our best-in-class AI-enabled Machine Vision technology with accurate defect classification capabilities requires significantly less training time as compared to rule-based Machine Vision systems and support decision-making on the fly.
KritiKal's Defect Inspection System
Kritikal’s Deep-learning based defect inspection system contains four modules that allow for Objection detection, localization of key issues in the object, and classification of defects into categories. It also provides an AI-enabled OCR engine that analyses the text in the image and provides a digital output.
This figure shows how our system works in a real-life scenario. Starting with the problem statement that includes an understanding of defect profile, speed, and accuracy requirements followed by getting images for error classification, testing, and learning purpose to resolve any error before deploying the system online.
POC, Proposal Development & Go-Ahead
Sample data is collected from the client and tested on our deep-learning system to get accuracy and defect detection rates as per the client requirements.
This step helps us in proposal development and deciding the implementation approach, Cost, and Project timeline.
Data Gathering &
Ground Truth Preparation
Based on the client’s requirement, camera and lighting systems are implemented at the client site to collect data for our deep-learning algorithm.
In-depth knowledge of the project helps us in deciding the required data and creating a standard for the image to identify all border-line cases.
Good – Bad Parts
At this stage, the images are analyzed and tested on our Deep-learning based platform to detect & classify good and bad samples.
Continuously training the system against the set parameters and result comparisons help us in optimizing and achieving the best possible accuracy rate.
Once the system is ready to classify Good & Bad images, the images get further disaggregated based on defect type classification standards.
Data from 3-5 production lines is sufficient to build and achieve acceptable defect classification rates before deploying the system.
Community Tools Supported
The AI based Machine Vision practice at KS comprises of experts who have worked on various open source and proprietary business intelligence tools, some of which are outlined below:
When to Use AI Enabled System
In places where the defect possibility is high Al-based Machine Vision is recommended rather than rule-based MV.
To detect defects that cannot be easily characterized based on quantification, shape, or size.
If defect classification is important for process control and improvement.
e.g. Contamination of grease, dust and chips varying in size, shape, and location.
To detect cracks or cosmetics defects that can come anywhere in the product or inspection of shiny or low contrast items.
To identify defects in components with different background/pattern or challenging feature extraction.
Possible Use Cases
360-degree Contamination Inspection
Contamination in the Form of Grease, Dust, Hair,
Black Particles, Fly, Chips etc.
Cosmetic Defects Inspection
Cosmetic defect in the form of Scratch, Dent, Burr,
Cracks, Pores, Bubbles, Blisters etc.
Complex Barcode Reading
Difficult to read Barcode, Shiny surfaces,
Challenging backgrounds etc.
OCR or Print Inspection on Difficult Surfaces
VIN (vehicle identification) inspection - Deep-learning based OCR in difficult to read background or objects.
Shiny Polished or Transparent Surface Inspection
Engine block inspection, Scratches
inspection on piston rings
Confusing & Challenging backgrounds
where multiple parts are involved.