SCIEN selected for 'Triangle Ehime 3.0' continuation program: Advancing quality control in Ehime's paper industry

## SCIEN selected for 'Triangle Ehime 3.0' continuation program: Advancing quality control in Ehime's paper industry

SCIEN (hereafter, SCIEN) announced that its project, 'Automation of Quality Control Operations Using Paper-Industry-Specific AI Models,' has been selected for the fiscal 2026 continuation framework of 'Triangle Ehime 3.0,' a digital implementation support program promoted by Ehime Prefecture.

In this project, SCIEN aims to reduce labor, standardize, and advance quality control operations by transferring the judgment criteria accumulated by skilled inspectors to AI models. The project targets defect inspection and quality judgment tasks in the manufacturing and processing of paper, non-woven fabrics, and films in Ehime. Additionally, following selection for the 'talent development framework,' which promotes the employment of local students and the training of AI implementation personnel, the company will work on fostering next-generation talent capable of handling AI used in field operations.

### Background: Quality control challenges in Ehime's paper industry
In Ehime's paper-related industry, detecting minute defects and making quality judgments in manufacturing and processing continuous materials like paper, non-woven fabrics, release paper, and films is a crucial task. While image acquisition and alert detection via defect inspection machines have progressed, the judgment of whether a defect is acceptable or how to handle it in subsequent processes remains a domain heavily dependent on the experience and on-site knowledge of skilled inspectors.

In particular, binary pass/fail judgments often lead to lack of consensus when AI decisions deviate from field perceptions, resulting in remaining re-confirmation and correction work by inspectors. Furthermore, challenges such as labor shortages, skill transfer to younger generations, the linkage of defect information between processes, and the reduction of waste loss are overlapping. There is a demand for a mechanism that not only introduces AI as a simple judge but also allows for the continuous learning and sharing of on-site judgment criteria.

### Project Overview
Based on defect image data, field interviews, and understanding of inspection workflows accumulated up to fiscal 2025, SCIEN will work on building 'Explainable AI' and a 'data infrastructure that transfers field knowledge' in fiscal 2026. The company will develop a paper-industry-specific AI model that reflects individual company quality judgments by accumulating not just AI-determined defect images, but also metadata including reasoning for judgments, quality management rationales, and reasons for inspector corrections.

- Verbalizing not just defect images, but also the basis for judgment, inspection findings, and quality control precautions.
- Building a Human-in-the-Loop improvement cycle by recording in natural language the reasons when inspectors correct AI outputs.
- Developing a quality management infrastructure that connects image data to language data, enabling future 'Why-Why' analysis, FTA, and waste loss factor analysis.
- Refining the model into one where code, operational know-how, and data acquisition methods can be expanded horizontally while using training data specific to each implementation site.

### Building a specialized model transferring veteran knowledge
The AI model built in this project is not a generic image classification model, but a 'field knowledge transfer' model specialized for quality management in the paper and paper processing domain. It accumulates images, inspection results, judgment reasons, and correction histories—decisions that veterans implicitly make, such as 'this defect is acceptable/unacceptable,' 'needs re-confirmation in the next process,' or 'easily leads to waste loss'—and reflects them in the AI's decision-making logic and operational rules.

- **Perception**: Detects and classifies defect images, and extracts quality control characteristics.
- **Ontology**: Structuralizes defect types, processes, judgment reasons, and subsequent handling to share on-site judgment standards.
- **Orchestration**: Integrates AI output, inspector corrections, and business systems to continuously improve the model and operational rules.

Through this, SCIEN implements AI not as a 'replacement for field decisions,' but as a 'foundation for inheriting the knowledge of skilled inspectors to the next generation and increasing the reproducibility of quality judgments.'

### Selection for local student employment and AI talent development
Following selection for the AI talent development framework of Triangle Ehime, this project will also work on employing local students and young talent. SCIEN values the engineering capabilities necessary for field implementation, such as extracting data from the field, operating in secure environments, connecting to business systems, and designing evaluation metrics, in addition to creating AI models.

SCIEN aims to foster practical talent capable of carrying out AI implementation within Ehime Prefecture by having student interns and young talent experience a series of implementation processes: understanding field challenges, data acquisition, annotation, model training, evaluation, UI improvement, and maintenance. By facing incomplete data and constraints encountered in the field, rather than just handling prepared textbook data, the company will nurture talent that can contribute to solving regional industrial challenges.