top of page
Bild_Fertigung.jpg

Use Cases

The 4 SoliDAIR use cases are proposed directly by mature EU industries (Brose, CIE, Bosch, AUTFORCE), and represent real challenges of automation in quality inspection and control processes. By solving them, the industry partners achieve improved​

  • process efficiency (e.g. reduction of time, cost and resources)

  • sustainability (reduced emissions)

  • working conditions (replacing tedious manual tasks, and retraining staff to operate AI and robotic systems)

  • competitiveness (reduced defect rates, improved OEE)

In all cases, the most important innovation is not (only) the technical performance of the individual technologies themselves (e.g. AI, robotics), but rather how to ensure the successful acceptance and adoption of the integrated solutions.

Ãœbersicht_use_cases_neu.png

Use case 1

AI-enabled optical quality control system for non-AI-experts

Automotive door module assembly

Production process improvement targets


Reduced development and maintenance effort for AVI (Automated Visual Inspection) while reaching better decision performance​.

Main challenges

Technical:

  • Data challenges (How to deal with a lack of data when using AI in highly optimized production settings?)

  • AI training challenges (How to make the process of training AI modules transparent?)

  • Validation challenges (How to develop and define metrics to validate trained AI modules?)

  • Lack of regulations (How to initiate, validate and operate AI Systems in practice?)

Non- technical:

  • Explainability (Today most AI solution are black boxes)

  • Trust (Lack of transparency of AI modules à humans trust rule-based solutions more, despite inferior performance compared to AI systems)

  • Collaboration with humans (Due to missing trust, humans refuse to accept AI solutions in production)

  • Security/ Safety (For security and safety critical applications, transparency and trustworthiness are mandatory

Main partners involved

7_UG_Systems_transparent.png

Use case 2

Robotics & AI enabled automated visual inspection and manufacturing efficiency optimisation

Aluminium high-pressure die casting

Production process improvement targets

 

  • Replace human-led inspection with robotic / AI inspection​

​

  • Optimise manufacturing process through the root-cause analysis AI and human intervention​

Bild_1.png

Main challenges

Technical:

  • AI training challenge- how to train AI models in an automotive industrial sector with very unbalanced data due to high quality demands ( defects measured on parts per million)

  • Q-control cycle time- in-line AV Q-system should be compatible with the production cell , while controlling a high amount of characteristics on several faces of the parts

  • AV control reliability- will the AV Q-system be able to replace the human manual control on so complex parts with so many characteristics?

Non- technical:

  • Collaboration with humans- operators and Q-staff refuse to accept AI solutions at production sites

  • Explainability- AI models should address where the problem is and provide reasoning

  • Trustworthiness- due to very high volume production programs, and the necessity to fully load the machines, the solution should be robust and trustworthy.

Main partners involved

Use case 3

Robust AI quality prediction in matured, high rate and high volume production

Components, high-precision

Production process improvement targets


Eliminate bottleneck inspection step in the process, by replacing it with AI predictive quality control

Main challenges

​Technical

  • Data challenges: heavily imbalanced data distribution between OK / NOK parts (NOK parts are 100 times fewer than the OK parts) resulting of the highly optimized production process;

  • AI precision challenges: The current AI approaches have not resulted in sufficient prediction quality (too many false positives);

  • Validation challenges: AI Uncertainty-determining models will be used to understand whether an OK/NOK prediction of the model comes with high certainty or not. In the case of uncertain predictions, this can be an indicator of a possible false positive. Further optimization of the model, or physical testing of part might be needed

Non- technical

  • Explainability: prediction of an AI model should address where the problem is and give reasoning which parameter affecting results more at the deviation observed;

  • Trustworthiness: Solution should be robust and trustworthy as the level of real functional measurements;

  • Collaboration with humans: The parameters relevant for decisions are defined by people during the design process, but the decisions themselves are delegated to the AI product. The application allows those affected by the decision to appeal for review. AI will be worked as human in the loop system.

Main partners involved

Use case 4

Predictive quality control of a multi-step assembly process integrating human operator

​

Automotive gear box assembly

Production process improvement targets

 

  • Eliminate the end-of-line inspection step in the process for 50% of products coming through, by replacing it with AI/ data driven predictive quality control​

​

  • Optimise manufacturing process through the root-cause analysis AI and human intervention​

Bild_Getriebe.png

Main challenges

Technical

​

  • imbalanced datasets that significantly impair the required training of today’s AI solution

  • development of a reliable assessment approach of suitability of the individual AI algorithm/ data driven approach for application in manufacturing environment

  • How to derive valuable knowledge/ insights

  • How to systematically relate AI findings to root causes of defects and actionable correction measures

Non- technical

​

  • Explainability

  • Trust

  • Collaboration with humans

Main partners involved

bottom of page