About
Robotic welding innovation is at the heart of the AITOOLS1 project, (AI Toolbox for Lot-Size-One Robot Welding) is a national collaboration between leading Finnish companies and research institutions, launched to tackle the complex challenges of robotic welding for customized, small-batch production. Funded by Business Finland, the project brings together expertise from Wärtsilä, Kemppi, HT Laser, Cavitar, Visual Components, Tampere University, and VTT. Over the past two years, these partners have worked together to explore, prototype, and refine intelligent welding solutions — and the results are already showing promise.
Authors
The Original publication was in Finnish language in Hitsaustekniikka magazine 1/2025 and written collaboratively by leading experts involved in the AITOOLS1 project, representing key Finnish companies and research institutions driving innovation in robotic welding and AI-based quality assurance. The contributors include Raimo Mäki-Reini (Wärtsilä Finland Oy), Antti Kahri (Kemppi Oy), Juha Kytöharju (Visual Components Oy), Niina Koivuniemi (HT Laser Oy), Aku Savolainen (Cavitar Oy), Jari Kuosmanen (Tampere University), and Mika Sirén (VTT Oy).
Translation: Pouya Eghbali (Cavitar Oy)
Project Overview: Robotic Welding Innovation
For the past two years, we have been studying the challenges of robotic welding in collaboration with key companies through the AITOOLS1 project (AI Toolbox for Lot-Size-One Robot Welding), funded by Business Finland. Solutions are already emerging!
Robotic welding has been practiced in Finland for decades and has become a standard in countless machine shops. Robots can produce high-quality welds on straight and clearly defined sections. However, without proper control, they can also result in poor-quality welds. Often, post-welding corrections and finishing work are required, reducing arc time efficiency. For instance, tack welds can cause the welding torch to miss the welding groove, fillet welds may have insufficient throat thickness, or defects may appear at the starting points of welds (Figure 1).
Wärtsilä Optimizes Design to Improve Robotic Welding Quality
For the mock-up structure of the project, we incorporated diverse challenges for robots by defining different welding positions and factors reflecting assembly inaccuracies. The welding sequences were carefully determined so that the results of residual stress and deformation analyses could be compared with the actual structure.
The design included as many features as possible found in Wärtsilä’s engine baseframes. So far, five prototype structures have been welded, with improvements made to each new iteration.
Key optimizations included:
- Ensuring adequate space for the robot using welding simulations.
- Reducing stress concentrations at the end plate weld terminations, lowering quality demands on edge fillet welds.
- Adding positioning pins to the end plates and corresponding holes in the base plate, speeding up assembly and improving dimensional accuracy.
- Challenges in tack weld quality were addressed by beveling the end plates.
- These changes were made cost-effectively via laser cutting, resulting in significant time and cost savings during assembly.
Kemppi Addresses Robotic Welding Issues with Power Supply Technology & Quality Control Tools
Kemppi’s robotics team has been researching and developing solutions using cutting-edge power supply technology and quality management tools. Through extensive welding tests, optimal solutions have been identified for common welding errors such as:
- Weld terminations
- Welding over tack welds
- Welding over air gaps
By properly selecting specialized functions, welding parameters, robot motion paths, and speeds, time-consuming manual finishing work like grinding can be avoided.
One of the most critical factors in minimizing the need for manual control in robotic welding is reliable seam tracking. During the project, Kemppi improved arc-based tracking (TAST) to prevent welding path deviations, even under challenging and variable conditions (Figure 1).
An additional power supply module was developed, enabling 10-100 kHz readout frequencies for welding parameters during test welds.
Visual Components Enhances Robotic Welding Programming & Simulation
Achieving full autonomy requires seamless integration of design and manufacturing data. When design requirements can be directly transferred into the product’s 3D model (Model-Based Definition, MBD) and integrated into robot programming software, programming errors can be minimized (Figure 2).
During the AITOOLS1 project, Visual Components has made significant progress in this research.
HT Laser Develops Solutions to Improve Weld Detail Reproducibility
Weld quality begins with clean and precisely cut materials. HT Laser uses laser technology within the AITOOLS1 project to enhance weld repeatability.
Key improvements:
- Laser-cut positioning guides and bevels for tack welds reduced preparation time.
- Better consistency in welding precision and product quality.
- Easy repeatability of robot welding programs.
HT Laser also utilized ISO 3834-2:2021 quality requirements and SFS-EN 5817 weld class C standards. Quality assurance included 3D scanning for deformation analysis and visual & ultrasonic weld inspections (Figure 3).
Cavitar Explores Real-Time Weld Quality Assurance with Welding Cameras
Cavitar has developed welding camera technology utilizing laser illumination (Figure 4).
Advantages of laser-based welding cameras:
- Clearly visualize the molten weld pool, even through the arc.
- Capture precise surface topography data for quality assessment.
- Ensure weldability before welding by detecting dimensional deviations.
- Measure real-time quality features near the weld pool.
Using AI-driven image analysis, Cavitar is working on weld documentation, quality assurance, and production control.
During the project, image analytics methods were developed to extract weld topography information for measuring throat thickness, correcting robot paths, and analyzing the molten pool. Early results are promising, and Cavitar will continue developing the solution after AITOOLS1.
Tampere University (TAU) Develops Weld & Inspection Imaging via Data Analysis
TAU has developed an AI-based molten pool recognition system for real-time quality control.
Breakthroughs:
- 90% accuracy in molten pool recognition (Figure 5c).
- Tack welds detected via structured light patterns and edge detection algorithms (Figures 5a & 5b).
- Tested on a microcontroller platform, eliminating the need to transfer camera data to a computer for analysis.
- Achieved 80 ms latency in real-time processing.
The system also supports pre- and post-weld scanning, directly integrating into robotic programming (Figure 4).
VTT Enhances Welding Quality with Multi-Modal Data & AI
Traditionally, weld quality assessment relies on quality management systems and manual inspections. However, robotic welding generates real-time sensor and control data, which can be combined with AI and machine learning for new quality control methods.
Key AI Innovations:
Multi-Modal Variational Autoencoder (MMVAE)
Learns from real-time welding camera & power supply data.
Encodes welding data into a compressed format for easier anomaly detection.
Can reconstruct visual data from power supply signals alone, enabling defect detection without a camera (Figure 6).
Hyperspectral Imaging (HSI)
Characterizes molten pool shape.
Detects process changes and disturbances.
Maps the weld temperature profile during welding.
VTT is also coordinating the AITOOLS1 project.
Traditionally, weld quality assessment relies on quality management systems and manual inspections. However, robotic welding generates real-time sensor and control data, which can be combined with AI and machine learning for new quality control methods.
Key AI Innovations:
- Multi-Modal Variational Autoencoder (MMVAE)
- Learns from real-time welding camera & power supply data.
- Encodes welding data into a compressed format for easier anomaly detection.
- Can reconstruct visual data from power supply signals alone, enabling defect detection without a camera (Figure 6).
- Hyperspectral Imaging (HSI)
- Characterizes molten pool shape.
- Detects process changes and disturbances.
- Maps the weld temperature profile during welding.
VTT is also coordinating the AITOOLS1 project.
In Short
Status of the AITOOLS1 Project
- 70% of Business Finland’s funding has been used.
- The project continues until the end of 2025.
- After completion, developed solutions will be commercialized to enhance efficiency, control, and quality in robotic welding.
- Solving 20% of the most critical challenges can address up to 80% of issues.
- Autonomy metrics will be defined, allowing companies to integrate relevant parts into their own solutions.
Goal:
To create the most efficient and high-quality robotic welding system for producing single-piece, high-requirement welds.
Contact Info
Raimo Mäki-Reini (Wärtsilä Finland Oy) — raimo.maki-reini@wartsila.com
Artturi Salmela (Kemppi Oy) — artturi.salmela@kemppi.com
Juha Kytöharju (Visual Components Oy) — juha.kytoharju@visualcomponents.com
Niina Koivuniemi (HT Laser Oy) — niina.koivuniemi@htlaser.fi
Aku Savolainen (Cavitar Oy) — aku.savolainen@cavitar.com
Jari Kuosmanen (Tampere University) — jari.kuosmanen@tuni.fi
Mika Sirén (VTT Oy) — mika.siren@vtt.fi
