Projects – EFFRA

Home

Projects

Regions and DIHs alliance for AI-driven digital transformation of Euro...

AI REGIO - Regions and DIHs alliance for AI-driven digital transformation of European Manufacturing SMEs

The AI REGIO project aims at filling 3 major gaps currently preventing AI-driven DIHs from implementing fully effective digital transformation pathways for their Manufacturing SMEs:

at policy level the Regional vs. EU gap;
at technological level the Digital Manufacturing vs. Innovation Collaboration Platform gap;
at business level the Innovative AI (Industry 5.0) vs Industry 4.0 gap.

POLICY.

Regional smart specialization strategies for Efficient Sustainable Manufacturing and Digital Transformation (VANGUARD initiative for Industrial Modcernisation) are so far insufficiently coordinated and integrated at cross-regional and pan-EU level. SME-driven >AI innovations cannot scale up to become pan-EU accessible in global marketplaces as well as SME-driven experiments remain trapped into a too local dimension without achieving a large scale dimension. Regional vs. EU Gap.

TECHNOLOGY.

Digital Manufacturing Platforms DMP and Digital Innovation Hubs DIH play a fundamental role in the implementation of the Digital Single Market and Digitsing European Industry directives to SMEs, but so far such initiatives, communities, innovation actions are running in a quite independent if not siloed way, where very often Platform-related challenges are not of interest for DIHs and Socio-Business impact not of interest for DMP. DMP vs. DIH Gap.

BUSINESS.

Many Industrial Data Platforms based on IOT Data in Motion and Analytics Data at Rest have been recently developed to implement effective Industry 4.0 pilots (I4MS Phase III platforms). The AI revolution and the new relationship between autonomous systems and humans (Industry 5.0) has not been properly addressed in I4MS so far. AI I5.0 vs. I4.0 Gap.

AI REGIO is following the 4 steps for VANGUARD innovation strategy (learn-connect-demonstrate-commercialize) by constantly aligning its methods with the AI DIH Network initiative and its assets with I4MS/DIH BEinCPPS Phase II and MIDIH / L4MS Phase III projects. AI REGIO: Industry 5.0 for SMEs

Resources

Unfold all | Fold all

Technologies

Advanced material processing technologies

Engineering tools

Information and communication technologies

Mechatronics and robotics technologies

Demonstrators, pilots, prototypes

Industrial pilot or use case

A Smart Predictive Maintenance Toolbox for drawing lines of car body elements - SPMTcar

The experiment, “A Smart Predictive Maintenance Toolbox for drawing lines of car body elements SMPTcar”, carried on by AIN (Asociacion de la Industria Navarra), aimed at extending in the Predictive Maintenance of stamping presses for the car industry a condition monitoring system based on the measure of vibrations, currently in use on rotating machines, by introducing Artificial Intelligence techniques.

Condition monitoring of machinery and industrial installations by vibration measuring and analysis is widely used in the industry on machines in which the consequences, in case of failure, can lead to large production losses and repair costs. Historically it has been applied to rotating machines. However, its use in non-rotating machinery such as stamping presses, and in particular those used in the automobile industry, is very limited. On the one hand, the acquisition of data for its analysis is complex due to the impact that occurs in each stamping cycle. This causes the need to treat the vibration signals so that only the valid information related to the condition of the drive components (shafts, gears, bearings ...) remains. On the other hand, fault analysis and diagnosis, although technically feasible, is extremely laborious, making it practically unfeasible from an economic point of view due to the need for expert personnel in signal analysis and high-end dynamic signal analyzers. This fact acts as a barrier to the access of stamping companies to the advantages of Predictive Maintenance.

For these reasons, the possibility of introducing automated analysis and diagnosis techniques, using Artificial Intelligence, is of the utmost interest since

It would improve the reliability and availability of stamping facilities by reducing downtime due to unforeseen breakdowns.
It would reduce the analysis times of specialized personnel.
It would reduce the costs of the Predictive Maintenance service for stamping companies, facilitating the access of SMEs to the advantages of failure prevention technologies
It would improve the ability to failure anticipation by making possible a much faster analysis
It would improve the competitiveness of the stamping sector by reducing production costs due to unforeseen failures

The experiment: A Smart Predictive Maintenance Toolbox for drawing lines of car body elements SMPTcar is an experiment in predictive maintenance, and consists of developing an AI-based system for measuring and analysing vibrations on metal sheet stamping presses in the automotive industry, in order to improve failure prediction and management of maintenance shutdowns.

Similar systems already exist for rotating machines, but in the case of presses, data acquisition is made much more difficult by the vibrations caused by the impact during stamping: these must be eliminated in order to have clean data referring to the elements actually being monitored (shafts, gears, bearings).

The experiment involves the introduction of AI-based analysis and diagnosis techniques, thereby improving the following aspects of maintenance: fault prediction, downtime reduction, reduction in staff deployment, overall efficiency increase, cost reduction.

The outcome of the experiment is a system that is able to autonomously anticipate failure and send a notification to maintainers.

AI-based Process Control c/o ARMAC bv

Radboud University (together with Radboudumc part of Stichting Katholieke Universiteit) located in Nijmegen, the Netherlands, is a leading university in the Netherlands. Radboud Dept. of Analytical Chemistry is the oldest and largest academic research group exclusively focussing on data science and AI in an (analytical) chemistry context. It has a long history in finding solutions for industry regarding the analysis, control and monitoring, and optimization of their production processes.

ARMAC B.V. is an SME that deals with automation systems for large and small users, on different sectors. SVP is a heat distribution company that manages the central production and distribution of heat in the town of Purmerend; it operates 3 plants, one biomass (main) and two gas (back-up and for peak demands).

The objective in this experiment was to test AI solutions for a customer of ARMAC B.V. and to implement this solution in the operational control environment delivered and maintained by ARMAC B.V. In this way ARMAC gains experience in applying AI solutions for their customers. The customer is SVP, which is a district heating (DH) system of the city of Purmerend. The test involved the optimization of the district heating (DH) system through the application of AI using weather forecast information at the city of Purmerend; currently the weather is used by operators to estimate the heat demand. The system is not driven by the supply, but the demand of customers for heat. The heat demand is estimated from the operator expertise, that use previous experience, the weather conditions, the day of week and the season, as basis for decision making. However, this kind of control is non-optimal, and might be inaccurate, mainly because the weather conditions can vary greatly in a short period of time, affecting the heating demand, and leading to heat loss. The adoption of AI comes to overcome this problem. The AI has been applied to forecast the heat demand based on the weather conditions and forecast information at the city of Purmerend. IThe energy efficiency has increased after the adoption of AI, helping the operators along the decision making on heating demand, saving energy and keeping the system more efficient.

The experiment was conducted by the Radboud University as the experiment leader, Armac as the technology provider and the DH at Purmerend as the experimental site.

The historical data available is from about 10 years of operation. It includes the temperature, pressures, heating demand and weather conditions from 35000 dwellings. To better forecast the heat demand, this project has applied AI technologies using weather conditions and weather forecast information at the city of Purmerend. The following technologies have been employed: deep learning, supervised learning, and expert systems. The experiment was divided in four main phases, data collection, data modelling, model learning, model validation and deployment.

AI-enabled Operators Assistance through AR

AI-enhanced control strategy for production environment

The experiment, “AI-enhanced control strategy for production environment”, conducted by ART-ER in Bologna, is an answer to the need of a comprehensive controlling strategy for the production lines, a strategic need for manufacturing SMEs.

This need is expressed in terms of different goals to be achieved: process efficiency (e.g. in terms of production time, or energy efficiency), quality control, market request readiness. The need is stronger for those SMEs operating in highly dynamic market sectors. Production line controlling strategy has always been an human-centered activity, and this is due to the need for a clear and wide view and understanding of the potential issues coming from the different areas of the production environment, starting from the warehouse management to the final delivery. This activity then strongly depends on the previous SME experience. AI technology allows for an extension of this experience. One of the strongest qualities of AI is that it can help humans preventing events on a real-time basis. This means AI can look forward in a higher-dimension space than humans: different heterogeneous data can drive AI through the suggestion of specific activities to avoid, in example, shortage of raw material due to a sudden higher request on the market, potential damages to machines, of unexpected wear due to incorrect usage for a long time. New technologies and sensors help in collection of data that were never collected before: data from machines and operators that provide a lot of information. Processing information is extremely time-consuming, and not all the SMEs are interested in spending personnel efforts in elaborating data and interpreting them. Artificial Intelligence can help in this task, providing a fast interpretation of data, and also a deep level learning strategy from them. Training AI algorithms on previously collected data from production line will provide them a baseline of how operations are made. Improving AI knowledge by mean of other data coming from new sensors will allow for the detection of fail points in the actual controlling strategy and the definition of a wider and far-in-the-future new strategy.

The experiment provides end-users (SMEs) an instrument to implement AI and AI-related technologies into their production environment. Benefits for end-users can be achieved in terms of time and costs reduction, final product quality and general better production plant management, with the possibility to allow interaction between the production line management and other services (e.g. predictive maintenance, digital twins...). A better production line management also allows for productivity enhancement and reduction of dead costs, such as warehouse management and waste management. Thanks to AI enhanced predictions, activities inside the production environment are fine-tuned, allowing for a better efficiency and cost reduction. More, the AI-enhanced management tool provides support in terms of energy efficiency and machine allocation.

The solution developed is a service provided to SMEs in order to assess their needs and identify the kind of AI-enhanced innovation that they can bring into their production lines, and gain benefit from its application. The final solution makes use of a leverage on novel Ai-related technologies (AR/VR, …).

Training algorithms on different datasets from different experiences allow the acquisition of methodologies from other production sectors, fostering cross-contamination.

Automatic Capability Matchmaking For Re-Configurable Robotics Platform

The experiment, “Automatic Capability Matchmaking for Reconfigurable Robotics Platform”, has been performed by Tampere University and involved the development of an AI-based system to support the design and reconfiguration of production lines in the event of product changeovers.

During the Horizon2020-funded project ReCaM, TAU developed an automatic capability matchmaking system, which intends to ease up the manufacturing system design and reconfiguration planning procedure by automatically suggesting alternative resource combinations for specific product requirements. The approach relies on formal description of resources and products, providing a foundation for rapid creation of new system configurations through capability-based matchmaking of product requirements and manufacturing resource offerings. The matchmaking system allows the designer to search through large resource catalogues to find feasible resource combination alternatives without manual effort.

The new experiment aimed to show how the Capability Matchmaking System can:

Automatise the search and filtering of feasible resources and resource combinations for specific product requirements from large search spaces.
Automatise the identification of required reconfiguration actions on the current system layout.
Let the designer to concentrate on final resource selection and layout planning, instead of search and filtering.
Reduce the time used for system design and reconfiguration planning activity
Increase the amount of alternative resource solutions considered, leading potentially to more efficient and fitting for purpose production system configurations.

This AI REGIO experiment applies the Capability Matchmaking system in a context of Reconfigurable robotics laboratory, with the purpose to extend the old resource catalogue with a large set of new resource descriptions in order to be able to test and validate the system with larger search spaces. Furthermore, several new case products are being utilized to test the system with different requirements. Due to new kind of resources and processes, the Capability Model needs to be extended with new capabilities. Also, the rule base needs to be extended to enable the combined capability calculation and capability matchmaking with these new capabilities. The efforts during the experiment has been focused on creating a large resource catalogue, which can be used, as such, as a search space for the experimentation service offered through the DIH network. The case products that are used for the experiment implementation, are real products from real companies.

Idss For Predictive Quality Assurance

The experiment, conducted by COMET, is about IDSS FOR PREDICTIVE QUALITY ASSURANCE.

COMET decided to focus its experiment on quality control. Advanced technology enabled quality control can be applied in various industries by leveraging e.g. big data analytics and smart sensors and, of course, AI. Such predictive quality system performs cause-effect analysis, big data monitoring and analytics, and effective decision-making in real time.

The underpinning idea is to provide local manufacturing companies with an use case, a pilot experiment, easily replicable and scalable, that might demonstrate the early identification of defects and failures in larger series production. Early estimating the potential faultiness could help local manufacturing companies on planning, controlling and executing productive activities in an optimized and predictive manner. The scope of investigation is of particular relevance in large-series production e.g. for local Tier 2 companies in the automotive sector, steel producing companies and metal industry suppliers, manufacturing enterprises producing anchors and wall plugs, etc... In this context, the idea was to focus on an intelligent system that, leveraging on machine learning, could be of help to the decision maker (i.e. the quality assurance manager) to identify and detect the defects, anomalies and outliers (i.e. faulty pieces) in an environment similar to the real one.

One local manufacturing SME producing components in big lots was involved in the experiment, providing the faulty pieces that will be the object of the investigation. Large amounts of data have been the raw material on which the system relied on, as AI technology make it possible to identify regular patterns in available data and to use these patterns for a wide variety of analysis, decision-making, and prediction purposes.

COMET belief is that an intelligent learning approach such as Machine Learning should be investigated and compared for manufacturing quality prediction. The use case will leverage on an intelligent approach, part of a broader family of learning methods based on artificial neural networks with representation learning. In the age of industry 4.0, machine and deep learning has attracted increasing interest for various research applications. In recent years, such AI models have been extensively implemented in fault detection and diagnosis systems. The machine architecture’s automated feature learning process offers great potential to solve problems with traditional fault detection and diagnosis systems.

The company involved in the experiment was an Italian enterprise, dealing with cold moulding of metal wires for the production of small steel components to be employed in in the automotive and furniture sectors.

The use case implementation is based on different steps:

first of all, a number of sensors were installed on the first part of the printer-line, in order to detect, since the beginning, possible defects and stopping the production of a low-quality component;
then, a MES was installed and it was the basis of the AI-based decision support system developed by COMET, but the installation was not part of the Experiment and it was performed by a third party;
finally, data collected were exploited to implement an AI-based system for fault detection.

Machine Vision for Warehouse Optimization

The experiment was carried on by Gualini Lamiere SpA.Gualini experiment was about “Machine Vision for Warehouse Optimization”, and the warehouse of Gualini was the problem and the core of the Pilot itself.

Gualini’s warehouse is made of piles of heavy steel sheets, which, for this reason, need to be placed on the ground. The sheets have different dimension, thickness, structure and weight. The responsible of the warehouse, who receive punctual inputs from the production line, need to know exactly where to find the sheet they need, where it is located within the piles and in which position exactly.

The problems in Gualini’s warehouse was related to labelling, recognition and handling of the steel sheet, and was extendable to many manufacturing companies. Labelling is often inefficient due to the cover of the label made by the pile structure; besides, the frequent handling of the sheets exposes the label to a physics degradation; the bar codes attached on each sheet in the pile can wear out, tear and get lost during the handling which is frequent, each day.

The focus of the experiment was the implementation of a system able to support operators in the warehouse to identify exactly the unique sheet they need at that time for further delivery to production line.

The gain of the Pilot in Gualini has been to “recognize” exactly the characteristic of each steel sheet and to know exactly the position of the needed steel sheets within the piles and inside the pile, so as to reduce the handling, that has a high impact on men efforts, time and cost.

The idea of the experiment was to use AI, machine vision and a vision system to identify in real-time the sheets inside piles in the warehouse (storage area) and efficiently and fast moving them to forward to the production line.

Machine Vision is needed to automatically analyse the dimension of the “incoming” steel sheet and recognize the correctness between the physical element and the digital data. After which, track the position of every single sheet in every pile.

Artificial Intelligence is needed to track the material available and every handling in the warehouse, identify possible human mistakes and advice, and automatically analyse the data and be aware of the position of every steel sheet in the Wearhouse.

Expert system is needed because the system must improve its ability during time and allow a perfect and impeccable handling and warehouse management.

The project tested this recognition and traceability technique; in this experiment the technology has been applied in the warehouse in order to easily trace exactly the sheet needed by the production line.

The Royal Eijkelkamp BV Exp.09

Royal Eijkelkamp BV provides a complete suite of solutions of sensors and telemetry for soil and water data collection. They provide a solution for groundwater level monitoring that comprises the setup of the entire monitoring network, which is made up of multiple monitoring wells. The Royal Eijkelkamp BV seeked to improve its services to end-user customers, optimize the maintenance process, and improve the quality control of its measurements.

Each monitoring well includes sensors (pressure and temperature), the telemetry system (data logger), a database, and a visualization portal to manage the data. The values of groundwater level, battery state, and other parameters are logged by their data logger system and sent at pre-programmed intervals to the remote server. There is additional storage of maintenance procedures in addition to measurement data.

They also provide a corrective and periodic maintenance plan for the network. Their system relies on an automatic validation system to check for measurement errors due to many measurements. This validation system executes a univariate analysis to check for inconsistency in the collected data. The monitoring well requires maintenance if the quality check indicates a fault (i.e., malfunction or breakdown).

Due to the large amount of data collected, a manual inspection of the collected data is complex. Thus, inevitably it requires an automatic procedure to perform the quality control tests. The AI model drives the information extraction from the correlation between the measurement and the maintenance data to perform automatic quality control and failure detection system. The AI system assists the maintenance staff by reducing the maintenance effort and increasing trust in the measurement platform.
The Radboud University leads the experiment, and the Royal Eijkelkamp BV is the end-user.

This experiment aimed to provide an AI-powered quality control that will relate the measurement and maintenance data to deliver a grade (score) of the quality measurement, detect system faults, and detect their root cause. The system has used expert systems, multivariate data analysis, and deep learning models as the solution’s core.

This quality control system assigns a grade to the quality of the measurements and detect system failure. Many gains are expected from this solution, as this AI quality-control system will assist the maintenance staff in identifying the root cause of fault more quickly and reducing time spent on-site. Also, the company can understand the multiple causes that affect the measurement, allowing quality improvement of their products and solutions.

Water leakage detection