AI vision project optimizing quality control process
Client: Unique Biotech
Challenge
The challenge revolved around ensuring the accuracy and efficiency of the quality control process for probiotic vials aimed at improving gut health in both kids and adults. With no automation solutions available in the market for PE bottles, used in packaging these vials, Unique biotech was relying on a manual inspection process that was both time-consuming and error-prone, lacking the precision needed for detecting defects in vials.
The stakes were high as Unique Biotech sought to enhance the quality of their products, particularly crucial for probiotics. The imperative was to implement innovative AI-driven solutions that would revolutionize the entire quality control operation, ensuring the reliability and safety.
Approach
The strategic approach commenced with an in-depth research phase, delving into the distinctive attributes of packaging materials and vials. This phase involved meticulous scrutiny to define a comprehensive set of defects and to ascertain the most effective methods for their detection. Subsequently, a specialized data annotation tool was developed to facilitate the collection of a diverse and substantial sample dataset for the purpose of training the AI model.
Following this, the implementation of a Human-in-the-Loop AI vision tool was initiated. This tool was engineered to be self-learning, continuously evolving in its capabilities over time. The objective was to achieve a remarkable level of automation, ultimately striving for 100% accuracy in the detection of defects. This iterative process of improvement not only ensured efficiency but also enabled the system to adapt to evolving challenges and further enhance its precision in quality control.
Outcome
100% Automated quality control process by implementation of a robust AI vision system with the following key features:
- Defect Detection: Utilized AI algorithms to identify defects such as sediment, damage, color anomalies, and low volume in vials.
- Optimized Lighting: Experimented with different lighting techniques, including back lighting, front lighting, ring lighting, and more, to enhance defect visibility.
- Equipment Recommendations: Collaborated with vendors to procure specialized lighting equipment such as adjustable ring lights and backlights, considering factors like light intensity, color control, and adaptability to the vial-kit setup.
- Camera Setup: Configured a vision setup with adjustable cameras, ensuring accurate capture of vial defects from various angles.
- Mechanical Setup: Developed a vial-kit holder with a pulley and turner mechanism to facilitate tilting for sediment detection and precise positioning during the quality check.
- Data Collection Strategy: Outlined a systematic data collection plan, considering various lighting configurations and camera positions, ensuring a diverse dataset for model training.
- Validation and Training: Validated the AI model with a smaller dataset, progressively fine-tuning and training the model for enhanced accuracy.