Background

Glass vials are essential in various industries, particularly pharmaceuticals, where they are used to store and transport delicate substances. Ensuring the integrity and cleanliness of these vials is critical, as even minor defects such as fractures can compromise the product's safety and effectiveness. However, the transparent and reflective nature of glass presents significant challenges for traditional inspection methods.

Problem Statement

The primary challenge in inspecting glass vials lies in differentiating between genuine defects (such as fractures) and benign irregularities (such as dirt or reflections). Traditional rule-based machine vision systems often struggle to make these distinctions, especially when defects or dirt are located on different layers of the glass (front or back). The high variability in lighting conditions and the inherent transparency of glass exacerbate these difficulties.

Solution: ViDi Detect

ViDi Detect, an AI-based deep learning tool, was implemented to address these challenges. Unlike traditional machine vision systems, ViDi Detect can be trained to recognize and distinguish subtle variations in the glass surface, enabling it to accurately identify defects even under challenging conditions.

Implementation

The process began by creating a reference set of “good” vials, capturing images under various lighting conditions and with different amounts of natural dirt. This dataset was then used to train the ViDi Detect system, enabling it to distinguish between acceptable variations and true anomalies. Through this training, the system learned to detect fractures in the glass by recognizing their specific patterns and characteristics, while differentiating them from other marks or reflections on the surface. It was also trained to identify dirt on the glass by understanding how it appears in different forms and lighting conditions, allowing accurate separation of dirt particles from actual defects such as fractures. Despite the challenges posed by transparent materials, ViDi Detect’s deep learning capabilities made it possible to reliably inspect the glass vials, effectively handling issues like reflections and refractions. Additionally, because the system was trained under varying lighting environments, it maintained consistent accuracy even in the presence of strong fluctuations in light reflections, which are common when inspecting transparent and reflective surfaces like glass.

ROI Calculator

Discover the potential cost savings of robotic automation over a 20-year system life

This calculator compares your current manual labor costs against the total cost of owning and operating a robotic system over its 20-year lifespan.

EXPLORE TODAY

Results

After training, ViDi Detect demonstrated a remarkable ability to accurately detect fractures and dirt on glass vials and ensured that only genuinely defective vials were flagged for further inspection.

The implementation of ViDi Detect proved to be a game-changer. By leveraging deep learning, it overcomes the limitations of traditional machine vision, providing a reliable and efficient solution for ensuring the quality of glass vials.

How do you know if you need deep learning in your machine vision application? Download the Vision AI White Paper for a guideline.