Interpretable or Explainable Classification of 3D-Printer Failure Conditions


External Member

Luke Connal (Research School of Chemistry), Jekaterina Viktorova (Research School of Chemistry)


A new 3D printing method that allows for printing electronic devices within a single machine aims to improve reliability and reproducibility while reducing failure rates, down-time and waste. The technology has several advantages over existing 3D printing methods by allowing simultaneous monitoring of the printing process in real time, which thus allows for simultaneous quality control, a high degree of reproducibility and avoids defects in the printed structure by utilising a feedback algorithm.  It is envisaged that this method can add to the current tools in electronics manufacturing and even has the potential to replace many, if not all, of the current multistep and environmentally unfriendly procedures, but problems arise.  Due to the nature of the technology with each print job a specific signal can be generated and assessed, but there is currently no way to both identify and anomaly (and stop to correct production) and gain insights into what caused it. 

In this project machine learning will be used to develop a classifier to allow self-monitoring and in-situ quality control of the prints, as well as determination of the failure mode. A binary classification model for the system has been developed, but lacks interpretable or explainable outcomes that enables real-time data to identify conditions that lead to problems and rank the importance of experimental features that should be controlled in the lab. In collaboration with researchers in the Research School of Chemistry, you will develop an interpretable classification model (based on a tree-based classifier), or an explainable classification model (based on a neural network), which will contribute to a first-of-its-kind self-learning 3D printer that is capable of failure detection and, potentially, self-correction, with the potential to revolutionise printed electronics.


Develop an interpretable or explainable classificer to rank conditions of a prototype 3D printer with outomes (pass/fail).


python programming, COMP1730/6730, COMP3430/8430, COMP3670/6670


This is a 12cp project.


machine learning, classification, data science, 3D printing, additive manufacturing, python

Updated:  10 August 2021/Responsible Officer:  Dean, CECS/Page Contact:  CECS Marketing