Multi-modal learning for real-time diagnostic and prognostic control in additive manufacturing

Collaborator on a real-world project developing real-time control systems using multi-modal deep learning

Picture of amanda-barnard.md Amanda Barnard AM

1 May 2025

Real world science and engineering often involves data from multiple sources and different scientific instruments. Each of these data sets has unique information that is necessary to enable advanced applications and real time control of scientific and industrial systems. Combining structured data from instrument settings (inputs) with sensor data from observation (outputs) involves complex multimodal relationships and will require advanced learning methods. In this project you will explore the relationships between three different complementary datasets recording different experimental scenarios in metal additive manufacturing (AM). In collaboration with CSIRO, you will develop advanced machine learning and deep learning models capable of recognising the current state of the AM system in real time, and potentially forecasting future states to allow prognostic control (the “holy grail” of AM). While the research problem has been clearly established, this project offers great freedom to determine the best AI approach and platforms. Pre-processed data will be provided.

Research Questions and Tasks

This project will focus on the following tasks:

  1. Relating the layer-by-layer thermal profiles of the part (images) with the instrument input settings (tabular), and vice versa.
  2. Relating the layer-by-layer thermal profiles of the part (images) with the final product characterisation (tabular), and vice versa.
  3. Combine data to predict the final product state and forecast future stages.

Supervision

Hybrid, including periodic project meetings with CSIRO collaborators.

References

Requirements

Background and experience in machine learning and deep learning (i.e. COMP3670/4670/4660/4650). Experience with Python is strongly desirable..

arrow-left bars search times