AI AOI On-Premise for Secure Capsule Defect Detection Data

The pharmaceutical manufacturing industry, particularly in capsule production, demands extremely stringent standards for product quality and data security. WeLinkirt's DaoAI AI AOI software system, with its visual foundation model's feature recognition, 5-minute 0-code automatic programming with one good sample, APDT positive/few-shot learning (1-20 good samples), semantic false positive filtering, and SDK/API/Docker support for 100% on-premise private deployment, effectively ensures the secure compliance of sensitive production data. Simultaneously, it elevates the efficiency and accuracy of multi-defect capsule detection to a new level, reducing manual re-inspection hours by −50%.

In the pharmaceutical manufacturing industry, especially for oral solid dosage forms like capsules, quality control is paramount to patient safety and corporate compliance. Traditionally, capsule defect detection relies on manual visual inspection or rule-based AOI equipment. However, with an increasing variety of products, more complex defect types, and growing demands for data privacy and compliance, existing solutions face severe challenges. For instance, a mid-sized pharmaceutical company producing various sizes and colors of capsules needs high-precision detection for subtle defects such as micro-cracks, dents, color spots, foreign objects, and blurry printing on the capsule surface. Crucially, all detection data, including good sample images, defect images, and judgment results, must be strictly confined within the company's internal network, never uploaded to any cloud platform, to meet the data security and privacy protection requirements of regulatory bodies like the NMPA (National Medical Products Administration) and international FDA. In the current context of domestic substitution for nano-scale precision inspection in semiconductor manufacturing, localization and independent control in pharmaceutical manufacturing are equally important, especially concerning core production data processing.

Pain Points: Why This Hurdle Is Difficult to Overcome

In multi-defect capsule detection, traditional solutions face multiple dilemmas: First, high false positive rates and the risk of missed detections. Manual visual inspection is prone to fatigue and subjective judgment, with missed detection rates potentially reaching 1.5% − 2%, while rule-based AOI systems have limited capabilities in identifying complex, irregular defects (such as micron-scale cracks or irregular color spots), typically leading to false positive rates between 8% − 12%. This results in a large number of good products being misidentified, increasing unnecessary rework and re-inspection hours. Second, significant re-inspection costs. High false positive rates mean substantial human resources are required for secondary confirmation; one client reported over 2000 hours annually spent on re-inspection due to false positives, severely delaying production line rhythm and increasing operational costs. Finally, and most critically, data security and compliance risks. The pharmaceutical industry has extremely high requirements for production data confidentiality; any sensitive data leakage could lead to severe legal consequences and damage brand reputation. Traditional cloud-based AI solutions cannot meet the requirements for 100% on-premise private deployment, forcing enterprises to confront significant data compliance obstacles when adopting advanced AI technologies.

The root causes of these challenges lie in: the diversity and minuteness of capsule defect forms, making traditional threshold-based or geometric feature algorithms difficult to generalize; imaging characteristics such as capsule surface gloss and uneven coloring, which easily introduce noise interference; and the strict regulatory requirements unique to the pharmaceutical industry, imposing constraints on data flow and storage far beyond other sectors. Especially for the detection of minute defects, such as nano-scale particle adhesion or micron-scale scratches, their feature intensity is similar to background noise signals, posing extremely high demands on the robustness of imaging systems and algorithms. Traditional solutions struggle to achieve high detection rates while simultaneously reducing false positives.

Technical Principles

WeLinkirt's DaoAI AI AOI software system fundamentally solves the challenges of multi-defect capsule detection through its visual foundation model's feature recognition capabilities. This system employs a deep learning architecture, pre-trained on vast industrial image data, enabling it to autonomously learn and extract deep semantic features of various capsule surface defects, rather than simple pixel or geometric features. For example, for micro-cracks, the system can identify their unique texture patterns and edge discontinuities; for color spots, it can distinguish subtle differences in color and distribution patterns from the capsule body. This powerful feature recognition capability allows the DaoAI AI AOI software system to handle the complexity and diversity of capsule defects, accurately capturing even micron-scale defects that are difficult for the human eye to discern.

Compared to traditional rule-based AOI or manual visual inspection, WeLinkirt's DaoAI AI AOI software system offers several advantages: First, the APDT (Adaptive Positive Data Training) few-shot learning mechanism requires only 1–20 good sample images to complete model training and programming within 5 minutes, significantly shortening changeover time and reducing reliance on defect samples. Traditional rule-based AOI requires engineers to spend hours or even days manually adjusting parameters and has poor adaptability to new defect types. Second, the semantic false positive filtering function, by understanding the contextual semantics of defect areas, effectively distinguishes true defects from background noise (such as light reflections, packaging textures), reducing the false positive rate of traditional solutions by −65%. Manual inspection and rule-based AOI often struggle to differentiate these “false defects,” leading to persistently high false positive rates. Finally, and critically, the “on-premise private deployment” capability, which is the main selling point of this article, means WeLinkirt's DaoAI AI AOI software system supports SDK/API/Docker deployment options, ensuring all data runs and is stored on the customer's local servers. This fully meets the stringent compliance requirements of the pharmaceutical industry, keeping data in-house and completely eliminating the risk of cloud data breaches.

Typical Application Scenarios

• **Detection of Micro-cracks and Damage on Capsule Surfaces:** During capsule formation and transport, minute cracks or edge damage, imperceptible to the naked eye, may occur. These defects can affect drug stability and packaging integrity. WeLinkirt's DaoAI AI AOI software system, through high-resolution image acquisition and deep learning algorithms, precisely identifies these micron-scale structural defects. The challenge lies in the subtlety of cracks and distinguishing them from normal textures.
• **Detection of Uneven Capsule Coloration, Color Spots, and Foreign Objects:** Capsule color consistency is a critical quality indicator. During production, uneven coloration, localized color spots, or the inclusion of foreign objects like black particles or fibers may occur. The system learns the normal color distribution model of capsules and identifies any abnormal color patches or attached materials. Challenges include color calibration under different batches and lighting conditions, and the diverse morphology of foreign objects.
• **Detection of Blurry Printing and Missing Characters:** Batch numbers, expiry dates, and other information are often printed on capsule surfaces, and print quality directly impacts product traceability. The system recognizes and compares printed characters, detecting defects such as blurriness, breaks, omissions, or double images. Challenges involve font diversity, printing position deviation, and image stability during high-speed production.
• **Detection of Capsule Deformation and Size Anomalies:** The geometric shape and size of capsules must conform to standards; deformation or size anomalies can affect packaging and administration. Through precise contour extraction and size measurement, the system identifies deformations such as ellipticity, dents, and bulges, ensuring product appearance consistency. The challenge lies in precise edge extraction at high speeds and quantifying the degree of deformation.

Implementation Case Study

A leading domestic pharmaceutical enterprise, operating a high-speed capsule production line, faced bottlenecks in multi-defect detection efficiency and data compliance pressures. Their original solution relied on imported rule-based AOI equipment combined with partial manual spot checks, processing millions of capsules daily. However, this approach had a false positive rate as high as 9%, requiring at least 3 additional quality inspectors to perform 8 hours of re-inspection work daily, severely impacting production line rhythm and labor costs. More critically, the enterprise strictly prohibited any production data from being uploaded to external cloud platforms, limiting its ability to adopt advanced AI vision technologies. After evaluating multiple vendors, the company ultimately chose WeLinkirt's DaoAI AI AOI software system due to its powerful defect recognition capabilities and commitment to 100% on-premise private deployment. During implementation, the WeLinkirt team assisted the client in deploying the DaoAI AI AOI software system on their local servers. Using the APDT few-shot learning function, they completed model training and line integration for 8 common capsule defects in just 3 days, using only 15 good sample images. After system go-live, the false positive rate significantly decreased by −65%, falling below 3.5%, while the detection rate for micro-cracks and foreign objects improved to 99.4%.

“WeLinkirt’s DaoAI AI AOI software system not only solved our most pressing data security issues but also significantly boosted detection efficiency, allowing us to focus more on core production.” — Quality Director, a leading pharmaceutical enterprise

WeLinkirt Solution and Products

The core solution WeLinkirt provided to this client is based on the DaoAI AI AOI software system. This system, through its powerful visual foundation model, achieves adaptive learning and precise identification of various capsule defects. In practical implementation, we first seamlessly integrated the DaoAI AI AOI software system into the client's existing production line control system and image acquisition hardware (such as high-speed industrial cameras) via SDK/API interfaces. For changeovers involving different batches and colors of capsules, operators simply import 1-20 good sample images through the system interface, utilizing the APDT 0-code automatic programming function to complete new model training and deployment within 5 minutes, without requiring professional AI engineers. This greatly simplifies the operation process and improves changeover efficiency. Concurrently, the system's built-in semantic false positive filtering module intelligently identifies and filters out pseudo-defects caused by factors like lighting changes or background reflections, ensuring the accuracy and reliability of detection results. All defect data, good sample data, and detection logs are stored in encrypted form on the client's local server and accessed through strict permission management, ensuring data security and compliance. WeLinkirt also provides continuous model optimization and remote support services to ensure long-term stable operation of the system. If the production line has higher requirements for 3D morphology, the DaoAI 2D/3D AI AOI equipment, utilizing its self-developed 3D camera for micron-level morphology detection, can also be considered.

By deploying WeLinkirt's DaoAI AI AOI software system, the client achieved significant business value and quantifiable results. First, the false positive rate decreased from 9% to below 3.5%, a reduction of −65%, leading to a substantial −50% reduction in re-inspection hours caused by misjudgments, saving nearly a thousand hours of labor costs annually. Second, the detection rate for micro-defects improved to 99.4%, effectively reducing the risk of missed detections and ensuring product quality and patient safety. Furthermore, the 5-minute quick changeover capability allows the production line to more flexibly respond to multi-variety, small-batch production demands, enhancing production efficiency. Most importantly, 100% on-premise private deployment completely eliminated the risk of data breaches, ensuring that the enterprise fully complies with stringent pharmaceutical industry regulations while benefiting from AI-driven efficiency improvements, providing robust data security assurance for the company's long-term development.

This article was generated by AI. Customer cases are simulated scenarios based on real product capabilities and figures are illustrative; see product pages for official benchmarks.