ESA: Quantum-Inspired Quality Control for Rocket Manufacturing

The Challenge 

Despite significant investment in digitalization, quality assurance – particularly for highly complex aerospace components – often remains a bottleneck. 

At the heart of our project is the analysis of image and sensor data from a particularly safety-critical manufacturing process: so-called shot peen forming. In this process, the dome segments of rocket tanks are reshaped by blasting them with small metal spheres. Detecting defects at an early stage is essential. Ideally, quality assurance could be integrated directly into the production process – automated, precise, and in real time. 

But this is where current systems reach their limits. The enormous volumes of data generated during modern manufacturing processes – from optical systems, sensors, or machine data – can often not be processed or analyzed in real time. The risk: relevant anomalies are detected too late. The result: defective areas lead to high costs because they can no longer be repaired. 

Our Approach 

At the core of the ESA project is planqc’s quantum-inspired technology: advanced tensor network methods, originally developed for the simulation of quantum many-body systems, enable up to exponential compression of data structures by filtering out unrealized correlations. 

planqc adapts these algorithms to detect defects in image and sensor data from the shot peen forming of Ariane 6 rocket components. Unlike deep learning models, which require large datasets and energy-intensive training, tensor networks achieve comparable accuracy with drastically lower computational cost—making them ideally suited for industrial real-time applications. 

Although rooted in quantum theory, the algorithms are entirely implemented on classical hardware. This means advantages of quantum technology can be leveraged today—without waiting for future quantum computers. The models integrate with nebumind’s data platform and run directly on production data from MT Aerospace, enabling fast, scalable, and sustainable quality control. 

The Potential

Harnessing quantum-inspired tensor networks, we can construct novel anomaly detection models at reduced costs in terms of energy, compute power and data, by removing unused information in the data and the models themselves. Additionally, these models are portable to computing platforms operating on GPUs or TPUs, where they can greatly benefit from a high degree of parallelization. This opens up new opportunities not only to improve and accelerate manufacturing processes in the industrial sector, but also in the direction of developing sustainable machine learning approaches requiring less resources for their training and deployment.