A user vision for visual inspection: A view from the industry brings together experts from leading biopharmaceutical companies to define a future vision for inspection systems that are more intelligent, flexible and data-driven, and highlights several technologies that are expected to shape the future of visual inspection.
Abstract
Visual inspection (VI) remains a cornerstone of pharmaceutical quality assurance, helping manufacturers detect defects, protect product integrity and safeguard patient safety. However, as the industry advances towards increasingly complex biologics, personalised medicines and smaller production batches, inspection strategies must evolve to meet new manufacturing and regulatory challenges.
In A user vision for visual inspection: A view from the industry, BioPhorum brings together experts from leading biopharmaceutical companies to define a future vision for inspection systems that are more intelligent, flexible and data-driven. The publication advocates a hybrid approach in which manual visual inspection (MVI) and automated visual inspection (AVI) work together, combining human judgement with the consistency and scalability of advanced technologies.
The report highlights several technologies that are expected to shape the future of visual inspection, including artificial intelligence (AI), robotics, advanced imaging systems and digital twins. These innovations have the potential to improve defect detection, reduce false rejects, accelerate training and validation, and support predictive quality control through enhanced data analysis.
Importantly, the vision recognises that human expertise remains essential. While automation can improve efficiency and reproducibility, manual inspection continues to play a critical role in evaluating novel or ambiguous defects, validating AI-driven systems and providing contextual decision-making.
Developed through BioPhorum’s collaborative model, this consensus-based report reflects the collective expertise of global manufacturers and industry specialists. It provides a practical roadmap for organizations seeking to modernise visual inspection while maintaining the highest standards of quality, compliance and patient safety.
The adoption of Industry 4.0 technologies has intensified the demand for reliable quality assurance in industrial packaging processes, particularly in areas where manual visual inspection is still prevalent. In an investigation of the production environment, recurrent packaging errors, including missing or duplicated accessories, led to increased rework and quality-related costs. This study presents a semiautomated scene-level packaging inspection approach based on computer vision and artificial intelligence. The proposed solution integrates a deep learning-based visual inspection module with an electromechanical Poka-Yoke system, forming a closed-loop quality assurance mechanism that physically prevents nonconforming packages from advancing along the production line. Experimental results evaluating the visual inspection module showed that a YOLO-based detector achieved a scene-level exact-match accuracy of 99.35% under varying illumination, object configurations, and camera viewpoints, outperforming an RF-DETR baseline. These results demonstrate the suitability of the proposed approach for inline industrial inspection within the production takt-time constraints.
D. Cardozo, A. Loureiro, Leonardo Camelo et al.· IEEE Sensors Letters· 0 citations
There is an increasing demand for reliable and automated visual inspection methods for large civil and industrial structures such as bridges, dams and transportation infrastructure. Traditional visual inspection methods for large structures are struggling with accessibility, scale, and data consistency. Digital twins (DTs) are transforming structural health monitoring. DTs address these limitations with the help of drone or robotic imagery, integration of multi-modal data and repeatable inspections, giving inspectors a virtual walk-through of the structure and its condition. We examine the state of the art in visual inspection DTs that detect and predict surface damage in large structures. We study the maturity levels of 112 DT deployments from perspectives of Business, Usage, Functional and Implementation Viewpoints. The DTs are further classified into five categories: Supervisory, Operational, Simulation, Intelligent and Autonomous, based on the levels of their implementation across the full DT pipeline of data acquisition, data exchange, DT model creation, visualisation and use of Artificial Intelligence (AI). This paper also highlights various design considerations these implementations take into account to overcome challenges observed in earlier visual inspection DTs. It also examines the extent to which AI is integrated into these implementations and the capability of these DTs to provide predictive insights and maintenance planning. The paper ends with a brief description of key findings and research gaps.
A high-volume IV bag line with variable printing faced ongoing false rejects, prompting deployment of an AI-powered automated visual inspection (AVI) system from concept to production floor. The AVI system learns to distinguish subtle print artifacts from true particulates in real time, preserving print legibility while reducing misclassifications caused by printing variability, bag deformation, and lighting. The program progressed from concept through design, proof-of-concept testing, scale-up, FAT, SAT, and production qualification and validation. The AVI architecture blends deep learning defect detection with high-quality image capture and integrates with existing MES/packaging controls to gate good product and flag anomalies without slowing throughput. Practical design questions were addressed: camera placement, illumination strategies, data management and labeling, and model training and validation. Regulatory alignment (IQ/OQ/PQ, change control, GMP) underpins lifecycle management for AI-enabled inspection, including retraining and performance monitoring. Real-world learnings cover key design choices, maintenance strategies, and deployment milestones to sustain performance in continuous production. Expected outcomes include lower false rejects, improved detection of particulates on challenging flexible bag formats, and enhanced line efficiency while maintaining compliance.
Francesco Brazzarola, Luca Vescovi, B. Balboni et al.· PDA journal of pharmaceutica...· 0 citations
Automated visual inspection in manufacturing aims to replace slow and inconsistent manual checks, but its economic value depends on whether its decisions can be trusted enough to automate routine inspection while reserving human expertise for ambiguous cases. In production-line settings, defective samples are scarce, since the process is optimized to produce good parts, which limits any learning-based inspector trained on real data alone. Compounding this, defect decisions emitted as hard labels with no confidence estimate carry an asymmetric cost: a false reject wastes a good product, while a false accept may increase the risk of undetected defects progressing through the production process. We address both problems by mitigating data scarcity through the generation of synthetic defective samples with a diffusion model, and meeting the need for confidence-aware decisions with a Bayesian classifier that defers ambiguous units to human review rather than misclassifying them. These components are embedded in a staged pipeline of successive, complementary checks. We evaluate how synthetic augmentation affects classification and localization on a test set of real defects, and examine the system's trustworthiness at three points: the decision, the synthetic data, and the pipeline structure. This work-in-progress reports preliminary results suggesting that diffusion-generated defects, combined with uncertainty-aware classification, can lower the cost of reaching a trustworthy, deployable inspection model under data scarcity.
Panagiotis Sapoutzoglou, Jessy Ribaira, Martin Kanounnikoff et al.· 0 citations
This survey categorizes AI-based vision systems by their functional objectives rather than algorithmic classification, and organizes prior work by functional roles and practical deployment considerations, rather than algorithm-centric evaluations, to provide practitioners with actionable insights for industrial adoption.
Minjung Kim, Hwan-Sik Yoon· AI for Engineering· 0 citations
The recent progress in imaging sensors, artificial intelligence, real-time processing hardware has led to a change in industrial quality inspection, where machine vision systems have become a key idea. The conventional manual inspection processes can be quite tedious, irregular and inapplicable to high production throughput. Machine vision systems offer fully automated, objective and repeatable inspection systems with benefits that improve the quality of the products, minimizes the cost of operations and acquires the industry with high standards. The paper is a thorough examination of machine vision systems used in industrial quality inspection especially in the areas of system architecture, image acquisition, preprocessing, feature extraction, defect detection and decision making processes. The incorporation of classical computer vision methodology with the latest deep learning systems like convolutional neural networks has achieved a high rate of quality defect detection, high resilience, and scalability in various manufacturing industries such as the automobile, electronics, pharmaceutical, and food processing industries. Moreover, the paper discusses the issues of inconsistency in lighting, real-time, imbalance in the data set, and system integration with industrial automation systems. Recent studies experimental tests have indicated that machine vision based inspection systems have the capability of having accuracy above 98% which is far much higher when compared to traditional rule based method. At the end of the paper, some future trends have been mentioned like edge-based vision system, explainable artificial intelligence, and Industry 4.0 integration which will likely characterize the future of intelligent inspection systems.
M. Bianchi· International Journal of Mod...· 0 citations