Data scarcity is holding back AI innovation in the industrial sector. Our partner, NETZSCH, a forward-thinking company exploring new business models in the industrial sector via their corporate venturing unit NEDGEX, joined us in a venture to push the boundaries of AI data generation in machinery industry.
This collaboration aimed to tackle a common challenge in AI projects: The lack of training data. That said, we sought to develop a solution that would automate the generation of AI training data using CAD files as well as training an AI fully end to end starting from a CAD model.
Challenge
Training AI models requires vast amounts of accurate data, yet many industrial companies face the challenge of producing these datasets efficiently and cost-effectively. Although CAD files for parts and machinery are typically available, the manual process of generating training data—such as taking photos and labeling them—is time-consuming and costly, often disproportionate to the outcomes achieved. This inefficiency creates a significant barrier to fast and affordable AI development, limiting the potential for innovation.
We aimed to solve this issue by automating the process of generating training data from CAD files, allowing AI models to be trained without requiring access to real-world data. The vision was clear: upload CAD files to a platform, generate the necessary training data, and automate the AI training process, all from a single interface.
Solution
We led the technical development of this solution by building a system that allows users to upload CAD files, which are then processed through a pipeline to create high-quality training data for AI models.
The architecture was designed for scalability and ease of use. After the CAD files are uploaded, they are normalized into a common format and then passed through a series of steps. These steps include using Blender to simulate and render images with different materials, textures, angles, lighting, and zoom levels. The final result is a standardized dataset used to train AI models directly within the platform.
We leveraged Docker for containerization, ensuring the system could efficiently manage tasks both sequentially and in parallel, depending on infrastructure needs. By preparing the backend for orchestration tools like Kubernetes, HashiCorp Nomad, and Docker Swarm, we made sure the system could scale as demand grew. The entire process—from data normalization to image rendering and AI training—runs automatically, without requiring human intervention.
Outcome
This innovative approach led to the creation of a fully functional prototype, proving that AI training through automated data generation is not only feasible but highly efficient. Users can now upload CAD files, generate datasets, and receive fully trained AI models seamlessly, all without the need for real-world data.
By automating data generation and AI training, the project demonstrated the potential to overcome data scarcity—a major barrier in the industrial sector. As a result, the groundwork is set for scaling similar solutions across industries where access to training data is a critical bottleneck.
If your business is looking to explore innovative AI solutions and overcome data-related challenges, NeuroForge is ready to bring your ideas to life. Contact us to learn how we can help you harness the power of AI with custom-built solutions tailored to your needs.