The experiment addresses a structural challenge in industrial testing environments: the growing complexity, cost, and rigidity of test sequence design and management. In sectors such as hydrogen systems, battery technologies, and advanced manufacturing components, test procedures have become increasingly sophisticated. They often involve long sequences with conditional logic, loops, interdependencies between measurements, and strict safety constraints. Despite this complexity, the creation of these test sequences remains largely manual, relying on spreadsheets or custom scripts developed by highly specialised engineers.

From an operational perspective, this situation creates significant inefficiencies. Engineers spend a considerable amount of time translating technical requirements into structured sequences, validating logic, and debugging configuration errors. Each modification to a specification, standard, or customer requirement may require manual rewriting and revalidation of large parts of the sequence. This slows down project delivery, reduces responsiveness to market changes, and limits the scalability of testing activities.

Technically, the challenge lies in managing the increasing volume and interdependence of data within test procedures. Modern test benches integrate numerous sensors, actuators, and control variables that must be synchronised in real time. Ensuring coherence between parameters, thresholds, safety interlocks, and measurement conditions requires deep domain knowledge and careful validation. Manual configuration increases the risk of inconsistencies or omissions that can compromise the reliability of test results.

From a regulatory and compliance perspective, industrial testing environments,particularly in hydrogen and energy-related sectors,are subject to strict safety and quality requirements. Test procedures must be traceable, transparent, and reproducible. However, when sequences are manually defined and modified, maintaining full traceability becomes challenging. The absence of structured, standardised generation methods increases the risk of deviations and makes audits more complex.

Data management is another critical pain point. Test sequences are often defined in formats that are not fully interoperable with digital tools or documentation systems. This limits integration with broader digital transformation initiatives and prevents efficient reuse of test knowledge across projects or teams. As companies seek to adopt more digital and automated workflows, the gap between manual sequence definition and digital execution becomes increasingly problematic.

The challenge also has a strategic dimension. In fast-growing and innovation-driven sectors such as hydrogen and battery technologies, time-to-market is critical. The inability to rapidly configure and validate new test procedures can delay product qualification and industrialisation. Furthermore, the strong dependency on expert knowledge creates bottlenecks and exposes companies to operational risk if key personnel are unavailable.

Overall, the current situation reflects a mismatch between the increasing sophistication of industrial systems and the largely manual methods used to define and manage test procedures. Addressing this challenge is essential to improve efficiency, reduce costs, enhance safety and traceability, and support the digital transformation of industrial testing processes.

Objective 1: 

The first objective of the experiment is to develop and integrate a Digital Intelligent Assistant into the GEMS test bench platform in order to automate the generation of complex industrial test sequences. This objective aims to enable engineers to describe test procedures in natural language or structured documents and automatically convert them into structured, executable sequences compatible with real-time industrial environments.

Objective 2:

The second objective is to ensure safe, transparent, and reliable use of artificial intelligence in industrial testing by implementing a visual sequence viewer and human-in-the-loop validation mechanism. This objective guarantees that all automatically generated sequences can be reviewed, verified, and adjusted by engineers before execution, ensuring compliance with operational and safety constraints.

Objective 3: 

The third objective is to validate and demonstrate the solution in real industrial use cases, particularly in hydrogen and battery testing environments, and to prepare its replication through dissemination within the European Digital Innovation Hub ecosystem. This objective ensures that the developed solution is technically robust, economically relevant, and scalable for small and medium-sized enterprises across Europe.

The experiment is positioned within the advanced manufacturing and energy technology sectors, with a primary focus on hydrogen systems and battery technologies. These sectors are characterised by rapid technological evolution, strict safety requirements, and increasing pressure to accelerate qualification and validation processes. Industrial actors operating in these domains rely heavily on complex laboratory and pre-production test benches to validate components such as fuel cells, electrolyzers, battery cells, and fluidic systems before integration into larger systems or market deployment.
The experiment will be carried out in France, primarily at the premises of Gemesis, where test benches and the GEMS software platform are developed and validated. The operational environment includes laboratory and pilot-scale test systems used for component validation and system qualification. Demonstrations will take place in real testing conditions, where sequences are executed on physical test benches equipped with sensors, actuators, safety interlocks, and data acquisition systems. The experiment therefore operates in a realistic industrial setting rather than a purely simulated environment.
The target users of the solution are test engineers, laboratory operators, automation specialists, and research and development teams working in hydrogen, battery, and related advanced manufacturing sectors. Additional stakeholders include system integrators, industrial SMEs, and research laboratories that require flexible and reliable test automation tools. The Digital Innovation Hub CARA will support engagement with these stakeholders, ensuring alignment with real industrial needs and facilitating broader replication within European manufacturing ecosystems.
Several operational and technical constraints shape the context of the experiment. Hydrogen and battery testing environments are subject to strict safety standards due to high pressures, reactive gases, thermal risks, and electrical hazards. Test procedures must therefore comply with internal safety rules, traceability requirements, and, where applicable, certification frameworks relevant to energy and mobility applications. Any automated generation of test sequences must preserve full transparency and allow human validation before execution.
Integration constraints are also significant. The solution must be compatible with existing test infrastructure, including programmable logic controllers, data acquisition systems, and industrial communication protocols. It must support interoperability with existing software environments and data formats used in industrial testing. Access to data is controlled and limited to operational test parameters and documentation provided within secure environments, ensuring compliance with data protection and confidentiality requirements.
Overall, the experiment takes place in a demanding industrial context where safety, reliability, interoperability, and regulatory compliance are critical. This real-world setting ensures that the developed solution is robust, applicable, and directly relevant to industrial testing challenges faced by European small and medium-sized enterprises.

Before implementing the MINDIA experiment, Cromic operates its manufacturing processes using largely manual and fragmented workflows where production-related information is distributed across multiple systems, physical documents, and informal communication channels. This fragmentation limits real-time visibility on the shop floor and makes timely decision-making difficult because there are no integrated tools to support the operators. Consequently, process deviations or equipment issues are often detected only after manual inspections or routine checks, resulting in slow response times, increased downtime, and a higher likelihood of scrap generation before any corrective actions can be taken.

Operators currently rely heavily on printed manuals or the verbal experience of colleagues to interpret faults, which makes identifying the root cause of an issue time-consuming and dependent on individual expertise. Accessing operational knowledge is inefficient because the necessary information is typically stored in static documents dispersed across different systems, making it difficult to retrieve during active production. As a result, operators must frequently interrupt their manual tasks to search for guidance, which increases their cognitive load and heightens the risk of human error.

From a sustainability perspective, the lack of structured and digital reporting for material usage and scrap events hinders the company’s ability to implement timely corrective measures. Communication with the logistics department regarding mold and color changes is entirely manual, leading to significant delays in scrap collection and recycling coordination. These systemic challenges negatively impact production costs and resource utilization, highlighting a critical need for a digital solution that can transform scattered experience and static documentation into an accessible, intelligent interface.

Objective 1: 

The primary technical objective of the MINDIA experiment is to develop and validate a multimodal Digital Intelligent Assistant (DIA) that integrates real-time machine monitoring, voice-based interaction, and augmented reality (AR) visualization to support manufacturing operations. By leveraging Open Voice OS (OVOS) and Large Language Model (LLM) reasoning, the project aims to turn scattered production data and operational experience into an accessible, intelligent interface that provides operators with hands-free, context-aware guidance. This integration is designed to bridge the gap between theoretical plans and field execution, significantly reducing operator cognitive load while accelerating response times to production deviations by an estimated 30%.

Objective 2: 

The second objective is to enhance sustainability and operational efficiency within the plastics manufacturing sector through optimized resource management and improved process transparency. MINDIA targets measurable improvements in material usage, specifically aiming for a 10% reduction in raw material consumption and a 5% increase in the recovery of recyclable materials through structured scrap reporting and real-time alerts. By enabling the early detection of inefficiencies and fostering human-AI collaboration, the experiment promotes a more sustainable and productive shop-floor environment that maintains full human oversight in line with Industry 5.0 principles.

Objective 3: Trustworthy and privacy-first AI deployment

The final objective is to ensure the scalability and market-readiness of the developed solution by packaging MINDIA as a reusable, open-source asset for distribution via the WASABI White Label Shop. The experiment validates how a domain-specific manufacturing assistant can be containerized and successfully deployed in real industrial conditions, providing a replicable blueprint for other SMEs to adopt with minimal integration effort. By demonstrating the commercial viability and technological maturity of open-source components within the WASABI ecosystem, the project supports the broader goal of fostering human-centered digital transformation and digital sovereignty across European manufacturing.

The MINDIA experiment is situated within the plastics manufacturing sector, a field that increasingly requires high levels of process awareness and efficient material management to remain competitive. The experiment will be conducted at the industrial facilities of Cromic Plastik in Eskişehir, Turkey, which serves as the primary pilot site. While initial development and functional testing take place in the controlled laboratory environment of IFARLAB-EDIH, the core of the experiment involves a real-world pilot deployment on the production floor. This setting allows the system to be validated under actual operating conditions, focusing on specific production lines where material waste and process deviations have the highest impact.

The target users for this digital assistant are shop-floor operators and production supervisors who require instant, hands-free access to machine data and technical documentation during their daily tasks. Key stakeholders include the MINDIA consortium partners—Cromic Plastik as the industrial coordinator, Lider Teknoloji Geliştirme (LTG) as the technology provider, and IFARLAB-EDIH as the scientific advisor, as well as the broader WASABI consortium, which provides the underlying architectural framework. These stakeholders are collectively invested in proving that AI-driven tools can enhance operator performance while maintaining human-centered control in an industrial environment.

Operating in a real manufacturing environment introduces several critical constraints and requirements that the MINDIA solution must address. From a safety perspective, the system is designed for fully hands-free voice interaction to ensure that operators can receive guidance without diverting their attention from dangerous machinery or interrupting manual tasks. Technical development follows rigorous engineering standards, including IEEE 12207 and AQAP2210, while the AI deployment is governed by the EU AI Act and GDPR to ensure ethical and trustworthy operation. Furthermore, the solution must seamlessly integrate with existing factory systems via OPC-UA and MQTT protocols and operate within the secure, containerized WASABI Docker Compose stack to maintain data integrity and role-based access control.

The pilot is carried out together with our manufacturing partner who is experienced in regulated medical device production environments. The experiment focuses on an assembly process involving wearable devices and sensitive electronic components. It examines how a conversational Digital Intelligent Assistant (DIA) can be introduced as a supportive layer within existing workflows. Particular attention is given to privacy-by-design principles, user acceptance and maintaining a non-intrusive interaction model.

Primary users are assembly workers performing detailed tasks, while stakeholders include technical teams and organisational decision-makers interested in practical approaches to human-centred digitalisation. The exploration seeks to better understand how such assistants are perceived in everyday work contexts and what conditions support meaningful adoption.

Currently, SMEs largely depend on manual supervision and legacy systems with limited real-time data access, despite significant efforts towards improving efficiency and sustainability in the food manufacturing domain. This makes it difficult to optimize production processes and energy usage, since multiple tasks such as product counting, registering batches, and machine monitoring are often manual, resulting in fragmented data, delayed reporting, and higher operational costs. Existing systems for packaging control and inventory management mostly rely on complex high-cost infrastructure, such as dedicated computer vision solutions, barcode/RFID tracking systems, and advanced warehouse management platforms. Such solutions require substantial capital investment and specialized expertise, placing them beyond the practical reach of most food manufacturing SMEs.

Challenge 1: Limited real-time data access in food manufacturing.

SMEs often face inventory management inefficiencies, product batch handling issues, and higher production line costs, as the integration of data is often delayed and/or fragmented.

Challenge 2: Manual activities can lead to inefficiencies and wasted resources.

Existing manual processes adopted by SMEs, such as counting and verifying packaged products, slow production and potentially introduce errors, resulting in overproduction and energy management issues (e.g., opening fridges often to check inventory), and even expired stock.

Challenge 3: Low adoption of intuitive and human-centered digital tools.

Food manufacturing facilities either do not integrate digital tools to aid workers or adopt tools with complex dashboards that discourage engagement and require specialized training.

Challenge 4: Lack of openness and interoperability across digital solutions.

SMEs often adopt proprietary or isolated digital tools, which limits interoperability and knowledge transfer. This fragmentation slows digital transformation and increases scaling costs.

Objective 1: 

Develop, and demonstrate a DIA for inventory management, production line efficiency, and packaged product quality assurance in food manufacturing facilities.

A task-oriented, conversational assistant will be developed, integrated with the Closed-Circuit Television (CCTV) system and energy-metering infrastructure at HELIOS Bakery. The solution will automatically count packaged items, monitor machinery activity, log production batches, and provide real-time natural-language insights to workers, and it will be deployed on the operational packaging line at HELIOS with real-time testing and iterative feedback from operators to validate usability, reliability, and sustainability. Through this approach, the assistant will enhance production efficiency, improve packaging-line quality assurance, and reduce manual workload and human errors in stock handling and process supervision.

Objective 2: 

Promote sustainability, resilience, and human-AI collaboration in food manufacturing and ensure transparency and interoperability with open, modular, and replicable technologies.

Integration of energy meters for packaging-line machinery will enable the monitoring and analysis of power consumption, and the assistant will correlate energy data with production throughput, helping operators identify periods of unnecessary consumption or anomalies and supporting optimization of energy usage and insights for equipment efficiency. The solution will also integrate open components from the WASABI ecosystem (such as OVOS, RASA, and open LLM models (e.g., Llama) for task-oriented dialogue, rely on open data exchange formats, and be deployed with Docker, while complying with GDPR and AI Act principles.

Objective 3: 

Distribute the developed DIA via a WASABI marketplace instance through deployment of WWLS instance hosting the developed DIA as a reusable skill for other SMEs in the food manufacturing sector. The marketplace instance will include documentation and demonstration material, enabling straightforward replication and adoption across the food manufacturing domain.

ELECTRA is carried out in the food manufacturing sector, focusing on the packaging quality assurance and inventory management of a food manufacturing SME. The experiment targets day‑to‑day shop‑floor needs such as inventory tracking, packaging‑line supervision, and quality assurance, by introducing a DIA that combines real‑time CCTV feeds and energy‑meter data from packaging machinery to generate actionable insights through a conversational AI interface.

The experiment will be carried out at HELIOS’s facilities in Spata (Attica), Greece, in real-life food manufacturing conditions. The ELECTRA solution will be deployed on the operational packaging line at HELIOS, with real‑time testing and iterative feedback from operators to validate usability in a real food manufacturing environment. HELIOS provides the pilot site and access to a real production environment with already deployed hardware (IoT sensors, energy meters, CCTV) that will be used as a basis for the DIA integration and validation.

The target users are employees in the facility who interact with packaging and inventory processes, i.e., workers/operators and supervisors who will use the assistant through natural, task‑oriented dialogue (through intuitive voice or text commands) to ask for information (e.g., inventory levels, production needs, energy consumption) and receive real‑time answers without need for technical expertise or manual data entry. Beyond the pilot, the solution is intended for other SMEs in the food manufacturing sector via distribution through a WASABI marketplace instance.

ELECTRA does not introduce any additional safety, regulatory, or technical constraints beyond those already present in the pilot environment. The solution operates purely at software level and builds upon already installed and operational infrastructure, including energy meters and CCTV systems at HELIOS, without requiring new physical installations, electrical modifications, or changes to certified equipment. Therefore, it does not pose any additional safety risks nor does it require new certifications, as it neither replaces nor alters existing certified components but functions as a data-driven analytical and conversational assistance layer. Access to the necessary operational data is already ensured through the established collaboration between HELIOS, as consortium leader, and PLEGMA, with existing data governance and technical pathways in place. Furthermore, integration is limited to the systems already described in the experiment documentation (energy meters and cameras). As such, all relevant constraints regarding safety standards, certification, data access, and system interoperability have already been addressed within the current framework.  In terms of data flow, the DIA will integrate with HELIOS data sources and the WASABI ecosystem using open and standardized data exchange formats, i.e., data from cameras and energy meters will be transmitted via open protocols such as RTSP, HTTP, and MQTT, while the energy data will be using formats, such as JSON and XML, exposed through through RESTful APIs.

The experiment addresses several interrelated operational and technical challenges within a small manufacturing environment where production coordination and quality assurance processes are largely experience-driven and manually executed. Production orders are managed through human supervision, while visual inspections depend heavily on skilled personnel. This creates bottlenecks, variability in inspection consistency and limited structured data for performance analysis. The challenge is not only technological but also organisational, as any introduced system must integrate into existing workflows without disrupting productivity.

From an operational perspective, task coordination across multiple concurrent orders requires continuous managerial oversight. In environments with limited staff, this can lead to suboptimal task sequencing and idle periods between production steps. Introducing a digital assistant must therefore balance automation with human control, ensuring that the system supports rather than overrides managerial decision-making.

Technically, the reliability of voice interaction in an industrial environment presents a significant challenge. Background noise from printers and machinery, short or ambiguous commands, worker accents and potential multi-device interference may reduce speech recognition accuracy. Ensuring robust wake-word activation, reliable intent detection and effective clarification mechanisms is essential to avoid operational disruption caused by misinterpreted commands.

Similarly, computer vision-based quality assurance introduces its own complexity. Variability in lighting conditions, camera positioning and image alignment can directly affect defect detection performance. The system must ensure stable hardware setup, controlled lighting, calibration procedures and threshold tuning to minimise false positives and false negatives while maintaining practical inspection speed.

Deployment within the factory environment presents additional constraints. The introduction of tablets, cameras and embedded computing equipment must not interfere with established production flows or workspace ergonomics. Hardware positioning and mounting structures must therefore be carefully designed to prevent operational disturbance.

Finally, organisational adoption and user trust represent critical challenges. Workers and managers must perceive the system as supportive rather than intrusive. Ensuring intuitive interaction, gradual integration and adequate training is essential to achieving sustainable adoption and long-term impact.

Currently, finding the right support tools for these specific environments is difficult for many SMEs. Traditional monitoring systems are often costly, technically complex or difficult to integrate into agile SME environments, while standard training materials and safety instructions tend to be static and provide no contextual or real-time support during work. Furthermore, generic voice assistants often rely on cloud processing, which raises concerns regarding latency, reliability and data privacy. 

As a result, there is a lack of accessible, privacy-first technologies that align with operational realities and support worker wellbeing without compromising established processes or data protection expectations.

Objective 1: 

Develop and integrate a fully operational Digital Intelligent Assistant that combines voice interaction, task coordination, computer vision-based quality assurance and analytics into a unified system, ensuring functional coherence and technical correctness across all modules.

Objective 2: 

Evaluate and validate the Digital Intelligent Assistant within the real production environment of the manufacturing SME, executing pilot scenarios, measuring performance against defined indicators and assessing its operational feasibility and user acceptance.

Objective 3: 

Prepare, publish and validate the Digital Intelligent Assistant within the WASABI marketplace ecosystem by packaging the solution according to marketplace requirements, integrating Web3-enabled capabilities and completing the necessary interoperability and compliance evaluations.

The experiment is positioned within the industrial electronics manufacturing sector, specifically focusing on membrane switch production. The primary goal is to integrate and test the TERIYAKI Digital Intelligent Assistant directly within real manufacturing conditions at the SKOUPAS production premises in Greece. The validation will take place under normal operating conditions, including screen printing, assembly and quality control, ensuring that the solution is assessed in an authentic industrial environment rather than a laboratory or simulated setting.

The main users involved in the experiment include the screen printing specialist, responsible for operating the printing process and interacting with the quality assurance system during production; the Production Manager, who oversees workflow coordination, scheduling, and prioritization across multiple production orders; and a hybrid professional role combining technical understanding with operational responsibilities, acting as a bridge between production staff and the digital solution deployment. These users will interact with the system through workflow coordination tools, computer-vision-assisted quality inspection, and voice-enabled task support.

Relevant stakeholders include company management evaluating operational efficiency improvements, technical partners responsible for solution development and integration, digital innovation support organizations facilitating the experiment, and potentially end customers interested in improved product consistency and traceability. These stakeholders contribute to defining requirements, validating outcomes, and assessing the broader applicability of the solution within manufacturing environments.

Several operational and technical constraints must be carefully addressed during the experiment. The computer-vision-supported quality assurance system operates in-process during screen printing and therefore requires high sensitivity and stability in image acquisition. Multiple variables may affect consistency and defect detection accuracy, including lighting conditions, camera alignment, surface reflections, vibration, print variability, and environmental factors. Ensuring reliable defect identification under these dynamic production conditions represents a critical technical challenge. In parallel, the voice-enabled interaction must function within a real factory environment characterized by background noise from printers and other machinery. Potential issues such as wake-word false positives or false negatives, worker accents, short command structures, and possible interference between multiple devices must be considered to maintain usability and operator trust. These constraints require robust calibration and careful system integration to ensure that TERIYAKI performs reliably without disrupting production continuity or operator workflow.

High-precision assembly environments require sustained concentration, fine motor control, and prolonged static postures. Workers may experience physical strain, fatigue or reduced ergonomic comfort during repetitive tasks, while maintaining strict quality requirements. Supporting worker wellbeing without disrupting established workflows remains a significant challenge.

Currently, finding the right support tools for these specific environments is difficult for many SMEs. Traditional monitoring systems are often costly, technically complex or difficult to integrate into agile SME environments, while standard training materials and safety instructions tend to be static and provide no contextual or real-time support during work. Furthermore, generic voice assistants often rely on cloud processing, which raises concerns regarding latency, reliability and data privacy. 

As a result, there is a lack of accessible, privacy-first technologies that align with operational realities and support worker wellbeing without compromising established processes or data protection expectations.

Objective 1: Human-centred multi-channel worker support

The project aims to explore how a conversational digital assistant can support workers’ well-being and situational awareness during precision assembly tasks. The assistant combines voice-based interaction with selected vision-supported features to enable touch-free use, accessible feedback and context-aware guidance. It is designed to integrate naturally into existing workflows without interrupting the working process. The system will follow a modular architecture that allows flexible configuration and future extension, while enabling adaptation to different workplace needs without changing the core interaction experience.

Objective 2: Scalability and long-term sustainability

This objective focuses on preparing the WorkWell OVOS Skill for broader adoption and distribution through the WASABI ecosystem. It will be packaged as a modular deployment so that it can be integrated and adapted by other European SMEs according to their specific needs. The approach supports long-term sustainability by providing a structured framework that allows future extensions without requiring fundamental redesign.

Objective 3: Trustworthy and privacy-first AI deployment

The project places strong emphasis on privacy, transparency and ethical AI practices. The assistant is designed to operate primarily on local edge devices, reducing dependency on cloud processing and supporting GDPR-aligned data handling. This objective focuses on demonstrating that conversational and multi-channel AI systems can be implemented in a way that respects user trust while remaining technically practical.

The experiment takes place in the context of high-precision manufacturing. These environments require consistent procedures, careful handling of materials, and adherence to strict quality and safety requirements. Workers often perform repetitive or fine-motor tasks that demand sustained concentration and static postures. In many cases, the use of protective equipment or specialised work attire further limits conventional interaction methods such as touch-based interfaces, reinforcing the need for accessible and non-intrusive interaction modalities.

The pilot is carried out together with our manufacturing partner who is experienced in regulated medical device production environments. The experiment focuses on an assembly process involving wearable devices and sensitive electronic components. It examines how a conversational Digital Intelligent Assistant (DIA) can be introduced as a supportive layer within existing workflows. Particular attention is given to privacy-by-design principles, user acceptance and maintaining a non-intrusive interaction model.

Primary users are assembly workers performing detailed tasks, while stakeholders include technical teams and organisational decision-makers interested in practical approaches to human-centred digitalisation. The exploration seeks to better understand how such assistants are perceived in everyday work contexts and what conditions support meaningful adoption.