AI Implementation in Industrial Quality Control

Quality control in industrial surface inspection presents significant challenges for human workers, who face monotonous, exhausting tasks that demand sustained attention and precise judgment. Manual surface inspection is characterized by critical human factor issues including fatigue, subjective judgment, and high susceptibility to errors [1]. These psychological and physical limitations are compounded by hazardous working conditions, particularly in hot rolling processes where temperatures reach hostile temperatures [2], making human inspection not only impractical but unsafe.

The scale of modern industrial production amplifies these work quality challenges. Surface inspection for flat industrial materials like steel sheets requires examining massive areas while detecting microscopic defects on high-speed production lines [3]. With world steel production reaching 1.88 billion tons in 2024 [4], the volume and speed demands far exceed human capabilities. This creates an untenable situation where worker wellbeing and product quality both suffer, with quality failures contributing significantly to industrial waste generation [5, 6] and thus causing economic and environmental [7] costs.

Recent advances in AI and computer vision are transforming surface quality control through AI-driven automated surface inspection system (ASIS) systems that automatically recognize subtle defects with high accuracy and consistency [8]. However, successful implementation remains challenging due to inadequate consideration of technical, organizational, and human factors [9-11]. Organizations need clear guidance on effective system integration while ensuring user acceptance and business alignment. Thus, this article asks:

RQ: Which design requirements are critical to the successful deployment of an automated AI-enhanced surface inspection system?

The next section explores the theoretical foundation for this question, followed by the research method used, a practical use case, findings in the form of sociotechnical design requirements (STDRs), a discussion of transfer potential, and a conclusion.

Quality control, surface inspection and human-centered AI

Manufacturing output worldwide exceeds $40 trillion annually. Approximately 20% of this output is wasted due to poor quality [5]. In fact, industrial waste accounts for over 50% of waste generated globally [6]. In this context, surface inspection represents a critical quality control point where early defect detection can prevent downstream waste.

Surface inspection faces immense challenges due to production scale and quality requirements. Single coils may unroll for kilometers, requiring real-time detection of defects as small as 70 µm at speeds up to 1,000 m/min on both sides and both edges simultaneously [3]. Manual inspection fails under these demands because of severe physical constraints, most notably the ambient temperatures in hot rolling lines, which routinely exceed 1,000°C [2], not only making the environment unsafe but also impairing vision due to glow. There are also significant human psychological limitations including fatigue, subjective judgment, and error susceptibility to overcome [1].

Modern AI-driven ASIS use Deep Learning (multi-layer neural networks that learn from data) to offer solutions that automatically recognize subtle defects with consistent accuracy while operating continuously at production speeds [8]. These systems learn normal surface textures, flag anomalies, classify defects to guide process decisions (e.g., post-processing, re-melting, removal), and cross-reference defect patterns with process data to help localize root causes, enabling upstream countermeasures [12].

AI should enhance rather than replace human abilities [9] and yet AI is often implemented without considering the humans involved. The goal should be human-centered AI that augments human capabilities instead of replacing them. [11] identify human agency and augmentation as key challenges, while [9] emphasize that workplace AI should augment human capabilities without imposing additional load. [10] frame this as an “informate” strategy that augments human capabilities rather than replacing them, producing higher-quality jobs than automation alone.

In surface inspection, this means building AI that flags uncertain detections, allows operator overrides, and incorporates domain knowledge into model updates, preserving human expertise and accountability. This collaborative approach is embodied in typical ASIS design (Fig. 1), which provides separate interfaces for human-guided training and operator-supervised inspection.

Design science research method

The decision was made to utilize design science research [13]. After identifying the problem, the objectives of a solution must first be defined. To formulate them, a case study is conducted on a market-leading automated surface inspection system (ASIS). Three key stakeholders are interviewed:

1. a customer support agent, to learn about the obstacles and inquiries of companies that use ASIS,
2. a sales/product manager, to learn which features are desired by (potential) users,
3. a software developer, to learn about feedback implementation and feature development aspects.

This gave us first insights into the ASIS landscape.

To deepen the insights, nine ASIS users were interviewed (Fig. 2) from five companies on three continents. Interview partners were acquired at SIS.EUROPE and through the ASIS provider.

All interviews were transcribed and analyzed using the software MAXQDA. We applied [14]’s thematic analysis to identify patterns and themes. Initial STDRs were derived from the interview themes, then enriched through literature review and validated via iterative refinement with three employees of the ASIS manufacturer.

Findings

Twelve STDRs were identified, as seen in Figure 3. These are divided into two phases: “Pre-implementation”-DRs apply before the go-live and ensure ethical deployment, whilst “Implementation and Operation”-DRs apply afterward, ensuring that the sociotechnical system continues to function as expected.

The pre-implementation phase establishes the foundation for successful AI system deployment by addressing technical and human factors before the system becomes operational. This phase is critical as it shapes user expectations, builds necessary competencies, and ensures organizational readiness.

DR04 addresses the ethical imperative of meaningful employee participation. Interview participants emphasized that excluding end-users from training creates resistance and undermines the human-centered approach. DR01 ensures transparent AI value demonstration, preventing black box deployment and supporting informed decision-making. DR06 promotes user autonomy by enabling independent problem-solving through context-sensitive support tools. Interview participants highlighted that self-service capabilities reduce frustration and dependency on external support, fostering confidence in system interaction.

The implementation and operation phase focuses on sustaining system effectiveness and user engagement after go-live. This phase addresses the ongoing challenges of maintaining user competency, system reliability, and continuous improvement in a dynamic production environment. DR07 ensures smooth onboarding through guided first-use experiences. Participants noted that step-by-step guidance significantly reduces initial adoption barriers and helps users build competency progressively, supporting the transition from novice to proficient system users.

DR09 helps to preserve human agency through experiential learning. Rather than creating passive users, this requirement ensures that operators develop expertise alongside AI capabilities, maintaining their professional growth and decision-making authority in quality control processes. This approach prevents the deskilling effects commonly associated with automation. DR10 emerged as fundamental to ethical deployment, as system unreliability erodes trust and can lead to either blind reliance or complete rejection of AI recommendations.

Transfer potential

These design requirements for ASIS demonstrate potential for cross-industry application, grounded in the Technology-Organization-Environment (TOE) framework [15]. This framework encompasses technological factors (system reliability, compatibility, usability), organizational factors (readiness, training, change management), and environmental factors (regulatory requirements, competitive pressures) [16, 17].

Our STDRs address known adoption frictions across TOE dimensions. Transparency of AI value (DR01) and reliability/usability (DR05, DR10) reduce perceived technical risk; tailored training, participation, and learning-by-doing (DR02, DR04, DR09) increase user self-efficacy and acceptance; prompt commissioning and contextual examples (DR03, DR08, DR11, DR12) accelerate time-to-value. Practitioners can track impact via time-to-go-live after training, 30/90-day active-use rates, override-rate trends for uncertain detections, model-update cadence, and support-ticket volume per shift.

Research indicates that AI adoption patterns follow similar trajectories across industries, with consistent challenges around human-AI collaboration, trust building, and organizational readiness. AI adoption rates vary between regions and sectors, ranging from 4% to 18% [18, 19]. This suggests that while fundamental implementation challenges are consistent, domain-specific factors significantly influence adoption success.

The sociotechnical nature of our requirements particularly enhances transferability. Research on human-AI collaboration across healthcare, finance, and manufacturing consistently identifies trust, transparency, and collaborative interface design as critical success factors, supporting the universal relevance of this human-centered approach [20]. Successful transfer requires attention to domain-specific regulatory variations (e.g., GDPR, FDA/CE markings) and organizational capabilities. When organizations possess adequate technological readiness, training infrastructure, and environmental support, human-centered design requirements offer a robust template for collaborative AI applications across industries.

The necessity of a sociotechnical approach

Through interviews with twelve stakeholders, this article identified twelve STDRs addressing the ethical imperatives of AI deployment in human work environments. The findings demonstrate that successful AI implementation requires explicit attention to human agency, employee participation, and responsible organizational practices, core principles of applied AI ethics.

The STDRs proposed here support ethical AI deployment by preserving and enhancing human capabilities rather than replacing them. Pre-implementation requirements ensure responsible knowledge management and meaningful employee participation, while implementation requirements promote human-centered collaboration through learning-by-doing and system reliability.

This sociotechnical approach addresses organizations’ ethical responsibility to implement AI systems that enhance work quality while establishing clear accountability frameworks. Rather than creating passive monitoring roles, our approach promotes active human-AI collaboration where operators’ domain knowledge feeds back into system improvement.

The article focuses on a single ASIS implementation, which may limit transferability to other AI systems and application areas. Also, whereas the focus here was on implementation and operation, long-term impacts need further investigation. Future research should validate these ethical STDRs across diverse industries, examine how human-centered AI maintains human agency and participatory deployment, gather additional quantitative data, and explore concrete implementation.

This research and development project is funded by the German Federal Ministry of Research, Technology and Space (BMFTR) within the “The Future of Value Creation – Research on Production, Services and Work” program (02L19C200) and managed by the Project Management Agency Karlsruhe (PTKA). The authors are responsible for the content of this publication. Declaration of AI: During the preparation of this work, Claude 3.5 was used to proofread sections of the manuscript and assist with rephrasing selected sentences.

This is an original article. The German translation can be accessed via DOI: 10.30844/I4SD.26.1.120

You might also be interested in

Generative AI in Organizational Knowledge Management AI literacy as a prerequisite for augmentation Uta Wilkens , Valentin Langholf , Niklas Obermann

Generative Artificial Intelligence (GenAI) offers new opportunities for organizational knowledge management, particularly when it comes to learning processes at the interface between explicit, firm-specific, and tacit knowledge. Its use is therefore of particular interest for application areas such as industrial maintenance. Based on a mechanical engineering case study, this article demonstrates that augmenting both processes and employees with GenAI requires AI literacy combined with professional skills.

Industry 4.0 Science | Volume 42 | 2026 | Edition 5 | Pages 118-126 | DOI 10.30844/I4SE.26.5.14

AI-Based Identification of Knowledge Transfer Situations An adaptive multi-agent system for agile product development in engineering Georg David Ritterbusch , Ravil Goetzke, Norbert Gronau

Although it has been demonstrated that organizational knowledge transfer can be improved in principle, there is still no automated approach for identifying patterns in complex, context-dependent, and domain-specific knowledge transfer situations. This conceptual article therefore examines and characterizes knowledge transfer situations using the real-world example of product development in engineering. Furthermore, a concept for an adaptive, AI-based, real-time multi-agent system uses data to recognize recurring patterns in knowledge transfer situations and enables context-sensitive interventions. Finally, an outlook is provided on AI-based learning mechanisms (reinforcement learning) that can be used to adapt interventions for higher effectiveness in the long term.

Industry 4.0 Science | Volume 42 | 2026 | Edition 5 | Pages 110-116 | DOI 10.30844/I4SE.26.5.13

Systematically Planning Vertical Factories Developing factory types as a foundation for intelligent planning support Nikolaus Kremslehner , Sebastian Schlund , Wilfried Sihn

Vertical factories reduce land use and harness the advantages of urban production. However, their multi-story structure significantly increases planning complexity. To counter such effects, this paper presents a typology that combines proven solutions for recurring challenges in the planning and operation of vertical factories. Relevant characteristics were examined using real-world case studies and structured within a morphological framework. This framework was consolidated into factory types that provide guidance for conceptual planning and serve as the foundation for intelligent planning support.

Industry 4.0 Science | Volume 42 | 2026 | Edition 5 | Pages 68-76 | DOI 10.30844/I4SE.26.5.6

AI-Based Building Inspection for Large Structures A new approach to construction progress monitoring Jan Sender, Konrad Jagusch, Michael Geist , David Jericho , Christian Scharr

Monitoring construction progress, as required in the one-off production of large structures, is very time- and labor-intensive due to a high level of complexity and individuality. The goal of this article is to develop a sensor-based approach for capturing and evaluating multiple inspection characteristics. The use of machine learning models to detect objects and derive relevant information forms the basis for linking current condition to construction schedule. This enables a significant increase in efficiency during construction progress monitoring and a well-founded assessment of progress.

Industry 4.0 Science | Volume 42 | 2026 | Edition 5 | Pages 78-84 | DOI 10.30844/I4SE.26.5.9

Experiential Knowledge Powered by AI  Practical insights from industry Martin Schmauder , Gritt Ott , Bianca Windisch

The use of experiential knowledge is a key success factor for companies. Based on four corporate case studies, this article analyzes the technical, organizational, and personnel challenges associated with the use of retrieval-augmented generation (RAG) systems. The results show that the success of such systems depends on the strategic development and maintenance of the knowledge base, as well as on employee engagement.

Industry 4.0 Science | Volume 42 | 2026 | Edition 5 | Pages 128-135 | DOI 10.30844/I4SE.26.5.15

Inclusive Work System Design Automation, standardization, and adaptability Sebastian Schlund

The design of inclusive work systems is gaining importance due to demographic change and the increasing digital penetration of value creation processes. While traditional ergonomic approaches are based primarily on percentile logic and thus address only a portion of the user population, the integration of digital technologies opens up new possibilities for the dynamic and individualized adaptation of work systems. This article presents a conceptual framework for inclusive work system design that integrates standardization, automation, and adaptability. Methodologically, the article is based on a conceptual analysis of existing approaches from ergonomics and human-centered design. The article makes a theoretical contribution to the systematization of inclusive work system design and identifies areas of focus for further research and industrial practice.

Industry 4.0 Science | Volume 42 | 2026 | Edition 5 | Pages 70-76 | DOI 10.30844/I4SE.26.5.8