Thin[gk]athon Distributed AI for Predictive Maintenance

Challenge "Distributed AI for Predictive Maintenance in Semiconductor Manufacturing": Edge-to-Cloud AI Pipeline with the Infineon AI Kit

powered by Infineon, ZEISS Digital Innovation & Smart Systems Hub –
3 days of ideation & prototyping

ABOUT THE THIN[GK]ATHON

The Thin[gk]athon is more than a classic hackathon. It is an intensive co-innovation sprint in which interdisciplinary teams work on real industrial challenges.

Within less than 3 days, teams from software development, data science, embedded engineering and industry develop working prototypes and technical architecture concepts together.

The focus is on:

• the integration of hardware and software

• working with real data and Technologies

• the development of solutions under practical conditions

The Thin[gk]athon offers a space in which new technical approaches can be quickly tested, evaluated and further developed.

THE CHALLENGE: DISTRIBUTED AI FOR PREDICTIVE MAINTENANCE

Within less than 3 days, teams from software development, data science, Embedded engineering and industry jointly functioning prototypes and technical architecture concepts.

The focus is on:

• the integration of hardware and software

• working with real data and technologies

• the development of solutions under practical conditions

The Thin[gk]athon thus offers a space in which new technical approaches can be quickly tested, evaluated and further developed.

In Modern production facilities in the semiconductor industry continuously generate sensor data from machines and systems. This data contains valuable information from which the condition of a machine can be derived - for example with regard to wear, anomalies or impending malfunctions.

In practice, however, this data is often not used sufficiently to make consistent and well-founded maintenance decisions. In addition, the data required for predictive maintenance is often missing because it has not yet been systematically recorded - even in the highly automated semiconductor industry, where availability, stability and precision have a direct impact on production. Companies face several challenges:

• Large amounts of data are created directly on site, but due to limited resources cannot be fully processed there for all required use cases
• Purely cloud-based processing causes latencies for time-critical decisions
• At the same time, suitable or economically sensible AI infrastructures are often missing locally

However, quick and reliable decisions are crucial for availability, stability and production performance This is exactly the area of tension addresses Distributed AI: the intelligent linking of local AI execution on edge systems with central model training and management in the cloud.

THE MISSION

Develops a functioning edge-to-cloud AI pipeline for industrial use that uses sensor data to automatically detect and visualize error states and draw conclusions about the maintenance requirements of a line enable.

The starting point is a real test environment at Infineon: A test track with a conveyor system forms a practical application where sensor data is recorded directly on transport containers. The solutions developed can therefore not only be designed conceptually, but also checked and validated directly on the real object.

The aim is to show how distributed AI can be implemented in industrial environments - from data acquisition on the system, through training and deployment, to local detection and cloud-based evaluation.

The teams are developing an end-to-end architecture that:

• Sensor data on the transport containers of the test route records

• selects and processes relevant data for status detection

• develops a concept for the appropriate data and system architecture

• technically implements data recording, status description and error detection

• makes training data usable for known and new error states

• the most efficient training possible for new error states enables

• Machine learning models are trained and optionally made available automatically for deployment on edge systems

• Detects error states locally on the edge device

• The detected states and results are made available and visualized in a suitable form

The focus here is on a predictive maintenance scenario: How can you recognize early on when maintenance is required on the line - for example Abnormalities in mechanical components such as bearings or other elements susceptible to wear.

The solution space includes modern industrial and cloud-native technologies, especially in the AWS and CNCF ecosystem. The use of GenAI for solution finding and prototyping is explicitly desired.

The Thin[gk]athon serves to develop a solution that not only works technically, but also makes the added value of Distributed AI for industrial error detection and predictive maintenance visible in a real application.

BENEFITS:

• direct collaboration with Industry partners

• Access to real hardware and industrial use cases

• Support from mentors and technology experts

• Intense collaboration in interdisciplinary teams

• Networking with industry, start-ups and research

PARTICIPANT PROFILE

We are looking for participants with different backgrounds Perspectives and skills.
Typical roles are:

• Software Developers
• Data Engineers / ML Engineers
• Embedded Engineers
• System Architects
• Industry Experts
• Creative Minds / IoT Enthusiasts

CHALLENGE OWNER

ZEISS Digital Innovation develops individual software solutions for complex production environments. Thanks to a high level of industry expertise and a broad understanding of technological innovations as well as an agile mindset, innovative solution ideas are implemented quickly and creatively.

The company combines expertise in:

• Industrial Software Engineering

• Data & AI

• Edge and Cloud architectures

• Cybersecurity

Infineon is one of the world's leading companies in the semiconductor industry. The Infineon Dresden manufacturing site was founded in 1994 - back then as part of Siemens.

Today, Dresden is one of the most modern and largest locations for manufacturing, technology and product development and now employs more than 4,000 people at two locations. This makes Infineon Dresden one of the most important industrial employers in the region.

Infineon Dresden manufactures over 400 different products based on 200mm and 300mm wafers for all four business areas of the group. With the Smart Power Fab, Infineon is opening in Dresden one of the most modern and at the same time most flexible 300mm semiconductor production facilities in Europe with one of the largest technology spectrums in the world.

In the Thin[gk]athon, ZEISS Digital Innovation and Infineon Technologies bring a real challenge from the context of Industrial AI and Distributed AI.

PARTNERS
Smart Systems Hub

Infineon Technologies

ZEISS Digital Innovation


RESULTS AND RECOGNITION
The best teams get the opportunity:

• to further develop their idea together with industry partners

• to make their results visible in the Smart Systems Hub network.