Active Computing & AI Physics & Astronomy

Ultra-intense laser-plasma interactions: Application of machine learning

Funder EPSRC (part of UKRI)
Recipient organisation University of Strathclyde
Funding Not disclosed
Period Feb 2025 — Jul 2028

In plain English

AI plain-English summary

A laser-driven ion accelerator is being taught to stabilise itself using deep neural networks, replacing the current trial-and-error approach with machine-guided control. These accelerators produce unique ion beams with transformative potential for medicine, security, and manufacturing. But their output is notoriously unstable, and traditional research relies on single-shot experiments that explore only a narrow range of conditions. The emergence of high-repetition-rate lasers and high-performance computing now makes it possible to collect data at unprecedented scale, enabling real-time statistical analysis. This project will train a deep neural network on thousands of particle-in-cell simulations using the EPOCH code, then test the model in live experiments on the SCAPA 350TW laser. The goal is to predict optimal laser and target conditions, then feed those predictions back into the experiment to stabilise the beam. If successful, this approach could make laser-driven ion sources reliable enough for practical use—in proton therapy for cancer, in security scanning, or in manufacturing. It would also reduce the time wasted on suboptimal shots at expensive laser facilities, benefiting the entire user community. The work is applied fundamental science: it advances understanding of laser-plasma interactions while building a practical control system.

View original technical description
High-power laser-driven ion accelerators offer unique beam properties with transformative potential in science, medicine, security, and manufacturing. However, achieving stability and control of these sources remains a major challenge. Traditional research relies on single-shot experiments with limited parameter exploration. The emergence of high-repetition-rate lasers and advances in high-performance computing now enable data collection at an unprecedented scale, allowing for real-time statistical analysis and machine learning applications. This project aims to develop deep neural network (DNN) models to optimise, stabilise, and control laser-driven ion beams. Key objectives include: 1. Developing a DNN surrogate model trained on simulation data to predict optimal laser and target conditions. 2. Training the model using extensive particle-in-cell (PIC) simulations performed with the EPOCH code on HPC machines. 3. Demonstrating model-guided optimisation through live experiments at high-power laser facilities. 4. Investigating laser-plasma interactions to refine predictive capabilities and enhance beam control. The student will gain expertise in laser-plasma physics, numerical simulations, machine learning, and experimental techniques. The machine learning approach will be tested in live experiments using the SCAPA 350TW laser, with potential application to the EPAC laser. This project will drive more efficient use of laser facilities and benefit a broad user community.

View the original record at the funder ↗

Researchers

Sreejith Leena Sadasivan Nair (Student)

Related Research

Grants with similar aims, by meaning.

Optimisation and control of a laser-driven ion accelerator using machine learning
High repetition rate optimisation and control of laser-driven radiation sources
Automated multi-dimensional mapping of dynamic laser-liquid interactions
Machine learning applied to laser-plasma interactions.
Laser-plasma interaction physics for the development of plasma accelerators and radiation sources

Original classification

Studentship

Plain English summaries and category classifications on this site are generated by AI and may not perfectly reflect the original research.