top of page

Case Study: How AMAZEMET Streamlined Welding Automation with RocketWelder Platform.

AMAZEMET is a deep-tech company specializing in metal additive manufacturing and advanced ultrasonic atomization technologies. One of its flagship innovations is rePowder, an ultrasonic metal atomizer system that converts various metal feedstocks into fine powders. The rePowder process uses a high-temperature plasma arc torch to melt the feedstock on a vibrating sonotrode, enabling ultrasonic atomization of the molten metal into droplets. In pursuit of greater efficiency and consistency, AMAZEMET sought to automate the torch control in rePowder using an AI-driven vision system. This case study examines how AMAZEMET integrated the RocketWelder AI welding platform to achieve real-time, vision-guided torch control in rePowder, reducing manual intervention and improving production reliability.

 

AMAZEMET’s rePowder ultrasonic atomization platform. The system (illustrated below) uses an electric arc/plasma torch (left) directed at an ultrasonically vibrating sonotrode (center) to melt and atomize metal feedstock into fine powder. The device is highly versatile – it can process a range of feedstock forms and supports different modules (e.g. plasma or induction melting)

Challenge

AMAZEMET’s goal was to implement AI-guided torch control for the rePowder’s plasma melting stage, but this came with two major challenges:

  • Real-Time Vision with Low Latency: The control system needed a fast vision pipeline to observe the intense plasma arc and molten pool in real time. To effectively adjust the torch on the fly, end-to-end latency had to be on the order of only a few hundred milliseconds (targeting ~120 ms). Achieving this low latency was critical for responsive control.

  • Minimal Custom Integration Effort: Integrating a traditional welding camera into the system would typically require significant programming and low-level interface work. AMAZEMET wanted to avoid a heavy software development burden for camera connectivity and image processing. The challenge was to find a solution that offered an integrated vision+control pipeline out-of-the-box, without the need to write extensive custom code.

Solution – AI-Driven Torch Control with RocketWelder

To address these challenges, AMAZEMET adopted the RocketWelder platform, which provides an end-to-end environment for vision-guided welding automation. The implementation included Rocket Retina industrial cameras paired with a Rocket Neuron edge computing unit, all managed by RocketWelder’s integrated software pipeline. This setup provided the following key features:

  • Integrated Vision Pipeline: The RocketWelder platform combines industrial weld cameras, high-speed processing, and control software in one package. AMAZEMET installed a Rocket Retina camera overlooking the rePowder’s melt zone. This camera—designed for harsh welding environments—captures high dynamic range video of the bright plasma process, enabling the AI to “see” the molten pool clearly. The camera feed is processed through RocketWelder’s pipeline in real time (with an input latency of ~100–150 ms). Crucially, RocketWelder’s built-in pipeline designer and vision software handled image processing and feature detection without requiring low-level programming by AMAZEMET. This plug-and-play vision system saved the team from writing custom code to interface a camera or filter weld images, as the RocketWelder environment provided those capabilities out-of-the-box.

  • AI Model Deployment on the Edge: AMAZEMET developed a custom machine vision model to interpret the plasma torch’s alignment and the melt pool status. Using RocketWelder’s tools, they deployed their own AI model directly on the Rocket Neuron edge unit (an industrial GPU/TPU compute module). Running the model on the edge hardware (at the torch site) ensured millisecond-level responsiveness and removed any network latency. It also meant that AMAZEMET’s proprietary model and process data remain on-premise, maintaining their IP control. The Rocket Neuron unit reads the Retina camera’s video, runs the inference to detect key features (e.g. torch position relative to the melt pool), and outputs control signals to adjust the torch position or parameters in real time. RocketWelder’s platform "translates" the vision data into torch steering outputs that then are processed by PLC, effectively guiding the torch along the desired path or maintaining it at the optimal spot on the sonotrode. This closed-loop control happens continuously during atomization, eliminating the need for an operator to manually correct the torch. In other words, the torch becomes self-guiding: the AI system adapts to any variation in the melt pool or feedstock flow on the fly, just as a skilled human operator would – but now autonomously.

Results & Benefits

By integrating RocketWelder into rePowder, AMAZEMET successfully automated the plasma torch control and realized several benefits:

  • Fully Automated Torch Operation: The AI-driven system continuously controls the torch position and alignment, removing the need for a human operator to monitor and adjust the torch during ultrasonic atomization. This automation not only reduces labor requirements but also improves safety (operators no longer need to work close to the high-temperature process). The torch guidance is now consistent and repeatable, leading to more uniform powder production.

  • Real-Time Responsiveness: The vision pipeline achieves roughly 120 ms latency from camera capture to actuation, meeting AMAZEMET’s real-time requirement. Such low latency allows the system to react almost instantly to process changes – for example, correcting torch aim if the melt pool moves or adjusting parameters if the feed rate fluctuates. This responsiveness helps maintain optimal melting conditions and powder quality throughout the run.

  • Simplified Integration: Using RocketWelder greatly simplified the integration effort compared to conventional high-end welding camera solutions. All necessary hardware and software came pre-integrated in the RocketWelder platform. Production engineers at AMAZEMET could set up the camera, train the vision model, and define torch control logic. This ease of deployment shortened the development cycle and reduced complexity, as opposed to dealing with standalone welding cameras that often require low-level API programming and custom vision code.

  • IP Retention and Flexibility: Because AMAZEMET deployed their own AI model on RocketWelder’s edge device, they retained full ownership of the solution’s intellectual property – the expertise in recognizing and controlling the melt pool remains in-house. They are not locked into a proprietary black-box camera system; instead, they can update or retrain their model as needed using their data. This flexibility ensures that AMAZEMET can continue to refine the torch control AI and adapt it for new materials or process settings over time, all while keeping sensitive data local.

  • Lower Operational Costs: Automating the torch control has a direct impact on cost efficiency. With no operator needed to constantly supervise and adjust the torch, labor costs are reduced and skilled technicians can be reassigned to more value-added tasks. Moreover, the RocketWelder solution proved more cost-effective to integrate than ultra-high-end laser-based seam tracking systems or 3rd-party welding cameras. Traditional laser profilometer or premium weld camera setups can be very expensive and demand specialized integration work. In contrast, RocketWelder’s camera-first, AI-guided approach eliminated the need for such costly equipment and custom engineering, delivering a quicker return on investment.

Conclusion

By leveraging RocketWelder’s AI vision platform, AMAZEMET successfully transformed the rePowder system into a smarter and more autonomous operation. The combination of a robust vision hardware setup and on-edge AI control allowed the plasma torch to dynamically adjust itself during ultrasonic atomization, which improved process stability and throughput. This case demonstrates how a deep-tech additive manufacturing company was able to solve a complex real-time control challenge with minimal integration effort, thanks to an innovative AI-driven welding solution. AMAZEMET’s rePowder now stands as a cutting-edge example of automated metal powder production, where advanced imaging and AI work hand-in-hand to enhance manufacturing performance

©2025 by Rocket Welder

bottom of page