Client: Global Top 10 Pharma Manufacturer – US Manufacturing Site
The client faced persistent risks in their secondary bottle packaging process, including:
- Packing mix-ups and mislabeling due to manual errors.
- Reintroduction of rejected products into the process stream.
- An average of 11 monthly deviations, requiring significant QA time for video review.
- Safety non-compliance due to improper use of PPE (e.g., missing cut-resistant gloves).
- Manual review of full video footage during cleaning validation, consuming high-value management time.
These issues not only impacted operational efficiency but posed serious risks to product integrity and workplace safety.
Client: Global Top 10 Pharma Manufacturer – US Manufacturing Site
The client faced persistent risks in their secondary bottle packaging process, including:
- Packing mix-ups and mislabeling due to manual errors.
- Reintroduction of rejected products into the process stream.
- An average of 11 monthly deviations, requiring significant QA time for video review.
- Safety non-compliance due to improper use of PPE (e.g., missing cut-resistant gloves).
- Manual review of full video footage during cleaning validation, consuming high-value management time.
These issues not only impacted operational efficiency but posed serious risks to product integrity and workplace safety.
Objective
To evaluate whether AI-powered video analytics modules could automatically detect critical deviations and unsafe behaviors in real time — without producing excessive false alerts — and be integrated into existing business and video infrastructure.
Solution
DEEPEYES deployed its AI Video Analytics modules, integrated with Client’s existing CCTV infrastructure at the POC site. Key capabilities included:
- Real-time detection of human interaction with the product on the assembly line.
- Prevention of unauthorized product reintroduction into the line.
- Monitoring of eject stations to ensure proper product disposal.
- Identification of PPE non-compliance (e.g., missing gloves).
- Logging and visualization of alerts via the QlikView interface for analysis and training.
The solution was hosted locally on Client hardware to meet performance and data security requirements. Remote access for algorithm tuning was securely managed through a solution for remote access with a site firewall proxy setup.
Implementation
- Duration: 4 months
- Total Cost including Client Staff Resources: $180,430 USD (including hardware, development, IT and business resources, and travel)
- Vendors Involved: DEEPEYES (AI Video Technology), XXXX (CCTV Vendor)
- Infrastructure: AI server with GPU acceleration and real-time integration into business networks and dashboards.
Results
Client: Global Top 10 Pharma Manufacturer – US Manufacturing Site
The PoC met all defined success criteria:
- 100% detection of human interactions with the product in the monitored area.
- All reintroductions of product to the line were accurately flagged.
- All ejects monitored and verified as scrapped.
- Non-compliance alerts (e.g., missing gloves) were generated in real time.
- Minimal false positives, ensuring user trust.
- Alerts logged in Management Dashboard QlikView, enabling use in training and continuous improvement.
Key Learnings
- Training AI models to distinguish between compliant and non-compliant processes requires significant development time.
- For full value realization, process integration and operator acceptance are critical next steps. Full blurring will help.
- Real-time video alerts provided insights into previously undetected process anomalies, offering new opportunities for continuous improvement.
Next Steps
- Begin technology integration into standard business processes.
- Conduct site-wide adoption assessment by the digital transformation team.
- Launch global RFP process to standardize technology selection for broader deployment across other manufacturing sites.
