From Needs to Infrastructure: Insights from AI Quality Inspection

The Challenge: Quality, Productivity, and Waste

Quality inspection plays a central role in laminate production, directly affecting yield, customer satisfaction, and material waste. At Skultuna Induflex, increasing quality demands are of strategic importance while managing production speed and waste reduction. Today, manual inspection remains essential, but its limitations are widely recognized in the process. It is difficult to scale, sensitive to human variation, and challenging to apply consistently in high-throughput environments.

From a research and development perspective, challenges like this are common across manufacturing. Improving inline inspection in production processes is rarely just about adding new technology. It requires understanding how quality decisions are made, how information flows, and how production constraints shape daily work. This framed the challenge not as an AI problem, but as a broader question of how quality inspection fits into the production system and what digital infrastructure would be required to support future improvements. 

While quality inspection is one concrete use case, the underlying questions about digital infrastructure are relevant across many industrial applications.

Why Digitalisation Was Not the First Step

Studies indicate that technology should not be primary focus in industrial digitalisation and that human factors play an equally crucial role in digital transformation initiatives [Lykourentzou et al. 2025]. Experience from earlier research projects shows that technology-driven programs often struggle when they are not grounded in real operational needs. Although existing frameworks address operational needs to some degree, their complexity makes them difficult and time consuming to use for planning and design of a digital infrastructure.

This insight was a deliberate starting point for the STRIDE project. Instead of starting with AI models and imaging systems, the focus was placed on understanding workflows, roles, and information needs.  By viewing quality inspection as a socio‑technical process – where people, decisions, and systems continuously interact – we uncovered insights that a technology-first approach would have missed. This approach helped avoid premature technical choices and ensured that later solutions could be integrated and scaled effectively.

The STRIDE Approach: From Needs to Infrastructure

The primary objective of STRIDE project was to design and develop a method/service that would serve as guidance for infrastructure decisions. As a result, the STRIDE Guidance Service was shaped as a lightweight, practical, and use case‑driven method that fills the gap between heavy architecture frameworks and the fast pace of industrial digitalisation.

The approach is stakeholder focused, drawing insights from relevant stakeholders connected to the use case in question. These insights are then systematically translated into requirements for systems and digital infrastructure. Iterative refinement steps allow assumptions to be tested and adjusted as understanding improves. 

The Skultuna Induflex case provided a concrete context to apply and validate this approach, linking early needs exploration to concrete infrastructure requirements and later technical validation. The researchers worked together with operators, quality engineers, and technical specialists to map information flows, decision points, and constraints. 

Proof-of-Concept: Validating What Matters

The Proof-of-Concept (PoC) played a specific role in the project. Its aim was not to build a full production system, but to test and validate key assumptions that emerged from the earlier needs-driven work in the current quality control process.

Although Mitsubishi Electric’s line-scan imaging technology has been proven in similar industrial contexts, validation was necessary to confirm that it could handle the specific defect types and inspecting characteristics of Skultuna Induflex’s products. . Real defect samples were therefore collected and scanned to determine whether relevant features could be reliably captured.

Once this was confirmed, an AI model was trained by RISE team using a limited but representative dataset and evaluated on new set of samples. Despite the small number of training examples, the model showed promising performance. This also serves to illustrate state-of-the-art of AI model capabilities when deploying in production settings with limited training. Just as importantly, the PoC also outlined what was intentionally left out of scope, such as parameters of a real-world deployment and production integration, to ensure the approach targeted the identification of these conditions rather than defaulting to generic or premature solutions.

Why Infrastructure Matters More Than the AI Model

In industrial AI projects, attention often centers on model performance. Experience from both research and practice shows that this is rarely the key success factor. More often, success depends on whether the surrounding digital infrastructure and organisational conditions are in place. Without them, AI solutions tend to remain in pilot phase.

For the Skultuna Induflex use case, discussions started from questions about data handling, system integration, cybersecurity, and operational responsibility, and later moved towards the implementation of AI models. This approach, informed by the STRIDE Guidance Service, proved central for understanding what would be required to move towards and beyond a pilot. Without infrastructure that supports transparency, traceability, and secure data handling, even technically sound AI solutions are difficult to deploy and sustain.

From this perspective, the utilization of AI model becomes one component within a larger system. It is the infrastructure that determines whether solutions can be scaled, adapted, and reused over time. This insight is a key foundation of the Guidance Service, which explicitly focuses on the broader conditions that enable responsible and resilient use of AI in industry.

What the Industry Gained: Beyond a Single Use Case

Although the quality inspection use case was grounded in Skultuna Induflex’s production environment, the insights extend well beyond a single company. The infrastructure questions addressed in STRIDE are shared across many industrial digitalisation initiatives.

For Skultuna Induflex, the project resulted in clearer and validated requirements for a future inspection system and greater confidence in future investment steps. At the same time, this enables partners such as Hitachi Energy to address similar challenges across different use cases, including condition monitoring and data-driven decision support. From this perspective, the Guidance Service offers a reusable approach for exploring digital infrastructure needs across different applications in the industry.

From an industrial automation standpoint, Mitsubishi Electric’s involvement highlighted the importance of validating even proven technologies in specific operational contexts. Product variants, defect types, and integration constraints vary between different processes and environments, reinforcing the need for structured collaboration between technology providers and the end users.

Across the consortium, the work contributed to a more holistic understanding of how digital infrastructure, cybersecurity, regulation, and organisational readiness must be addressed together. While use cases vary, the underlying approach is broadly applicable across industry.

From a Single Use Case to a Broader Method

The Skultuna Induflex case illustrates how a concrete industrial challenge can be used to explore broader questions about digital infrastructure and AI adoption. While the technical details are specific, the underlying issues are widely shared across manufacturing.

The STRIDE project addressed these challenges through the STRIDE Guidance Service, a lightweight and structured approach for defining digital infrastructure needs based on real operational requirements. The method helps organisations ask the right questions early in the process and make informed decisions before major investments are made.

As AI and digital technologies continue to evolve, starting with people, workflows, and infrastructure provides a stronger foundation for responsible, resilient, effective and scalable digitalisation. The results from STRIDE are intended as a starting point for continued collaboration and application across industrial contexts.

To conclude, if your organisation is exploring new digitalisation or AI‑enabled improvements, we welcome the opportunity to discuss your use case. The STRIDE approach is designed to help industry clarify infrastructure needs early and reduce uncertainty before major investments.

The STRIDE project was funded by Vinnova, Sweden’s Innovation Agency, under the Advanced Digitalisation program.