Automated Visual Inspection (AVI)

Automated Visual Inspection (AVI) is an analytical procedure that uses machine vision technology to perform a critical manufacturing step: the 100% inspection of parenteral (injectable) drug products. It functions as a final quality control test to ensure each unit is essentially free of visible particulate matter and container defects before release.

• Legacy Techniques: Manual Visual Inspection (MVI), where a trained human operator inspects each unit against black and white backgrounds. While foundational, it is subjective, slow, and fatiguing.
• Established Standard: Modern AVI systems that use high-speed cameras, controlled lighting, and computer algorithms to provide objective, reproducible, and high-throughput inspection of vials, syringes, and cartridges.
• Emerging Alternatives: AVI systems incorporating Artificial Intelligence (AI) and Machine Learning (ML) to improve defect classification, reduce false rejection rates, and adapt to challenging products like suspensions or lyophilized cakes.

AVI automates the inspection process by capturing and analyzing a series of images of each container.

1. Handling: Containers are fed into the machine and handled by servomotors. The container is spun to set any particles in motion, making them easier to detect.
2. Imaging: A series of high-resolution cameras captures multiple images of the container from different angles under various controlled lighting conditions (e.g., backlight, bottom light, Tyndall lighting).
3. Analysis: Image processing software analyzes the images in real-time. The software compares the images against pre-set parameters and acceptance criteria to identify defects.
4. Rejection: Any container that fails to meet the criteria is automatically segregated and rejected from the batch.

The system is designed to detect defects such as visible particles (fibers, glass, protein aggregates), container cracks or scratches, and closure defects like missing stoppers.

• Output: The system outputs a pass/fail decision for each unit and generates a final batch report detailing the total number of units inspected, passed, and rejected.
• Analysis: Rejection data is broken down by defect category (e.g., fiber, glass particle, cracked vial). An increase in a specific reject category from batch to batch can trigger a formal investigation into the root cause.
• Reject Verification: A critical step involves a trained human analyst manually inspecting a statistical sample of the rejected units to verify the machine's findings and classify the particles, confirming the accuracy of the AVI process.

• High Throughput: Capable of inspecting several hundred containers per minute.
• Objective & Reproducible: Replaces subjective human judgment with standardized algorithms, leading to highly consistent inspection results.
• Comprehensive Data: Generates a complete, auditable electronic record of the inspection process for each batch.
• Enhanced Detection: Can often detect defects that are difficult for the human eye to see consistently.

• Complex Method Development: "Teaching" the machine to distinguish between benign anomalies (e.g., air bubbles) and actual defects is a complex, time-consuming process.
• High False Reject Rate: A common challenge is optimizing the system to minimize the rejection of good units, which directly impacts manufacturing yield.
• Challenging Products: Opaque or viscous formulations, suspensions, and lyophilized products can be very difficult to inspect automatically.
• Not a Full Replacement for MVI: Manual inspection is still required for developing the inspection method, validating the AVI system, and investigating rejected units.

• Methodology of USP <1790>: Validation focuses on demonstrating the system's detection capabilities probabilistically, often using a test set of containers with known defects (a "Knapp Test" set).
• Detection Rate: Must demonstrate a high probability of detecting defects at and above a certain size threshold.
• False Reject Rate: Establishing an acceptable and controlled limit for the rejection of good units.
• Training Sets: The robustness of the inspection depends heavily on the quality and comprehensiveness of the image library used to train the machine's algorithms.

Automated Visual Inspection (AVI) methods must be standardized through rigorous development and validation to ensure consistent detection performance across products and sites. Reference sets of containers with known, well-characterized defects—such as the Knapp Test Set or compendial defect libraries—are critical for qualification. These serve as positive controls to challenge the system and establish detection limits for specific defect types (e.g., glass particles, fibers, cracks, stopper misplacements). Using standardized defect libraries also enables comparability between different AVI machines, supports method transfer, and provides a benchmark for periodic requalification and operator training.

• mAbs & Proteins: A primary concern is detecting proteinaceous particles, which are an indicator of aggregation and product instability. AVI is used to ensure the final product is "essentially free" of these visible aggregates.
• Lyophilized Products: AVI for lyophilized products is particularly challenging. In addition to particles, it is used to inspect for cake quality, identifying defects like collapse, meltback, or cracks in the lyo cake.
• Cell & Gene Therapies: For these high-value, often low-volume products, minimizing the false reject rate is critically important to preserve yield. As manufacturing scales up, AVI becomes a necessary replacement for manual inspection.