Predicting Quality Control Test Outcomes using Machine Learning
Location: Clyde, NSW
Duration: 3 months
Proposed start date: 1 September 2018
Keywords: Machine Learning, Programming, Proteomics, Predictive modelling, Data processing, Optimisation, Mass Spectrometry
Please note: Due to funding requirements, students must have Australian Citizenship or Permanent Residency to apply. Any applicants not meeting this requirement will be ineligible for this project.
ETP Ion Detect builds various products for ion detection. Our major product is discrete dynode type electron multipliers. Every multiplier that we ship to a customer has passed a set of quality control tests. Each of these tests examines a different aspect of the multiplier’s operation. One of these tests is a `noise test’. We check that the electron multiplier won’t produce random output beyond an acceptable level under certain operating conditions. This is currently our most time-consuming test. This single test takes up as much as 80% of the testing time for some multipliers. We want to reduce the time it takes to carry out noise testing. This will allow us to improve our manufacturing output per hour per test system. This in turn will make our testing more scalable.
We think it might be possible to reduce the time it takes to carry out a noise test, by using machine learning to predict the outcome. The volume of noise testing that we carry out should provide an adequate training set. Additionally, it might be possible to train a neural net to predict the appropriate corrective action when a multiplier fails a noise test. This would further reduce the time spent noise testing. We are open to exploring all forms of machine learning. This includes adapting machine vision techniques to recognise `good’ and `bad’ noise histograms and chronograms.
Either a computer science Ph.D. or an equivalent with knowledge of machine learning is required. The more techniques they are familiar with the better. There’s no guarantee that any particular machine learning technique will be most applicable.
Research to be Conducted
The principal goal of this research is to determine which machine learning techniques can be used to predict the outcomes of our noise testing. The next goal is to measure the `efficiency’ of these applicable techniques. The efficiency here refers to the amount of test data required to make an accurate prediction. Lastly, can we develop an implementation that runs on a desktop in an appropriate amount of time.
ETP will support this research in the following ways. First, we can provide a substantial training dataset with labelled outcomes (pass/fail/corrective action taken). Second, we have substantial in-house computational resources. This includes a 7-node Beowulf cluster and 2 Xeon Phi equipped super-computer desktops capable of 228 concurrent threads. We would consider purchasing GPGPU cards. Lastly, programming and algorithmic support will be provided by the scientists.
We are looking for a PhD student with the following skills:
• Knowledge of machine learning. (The more techniques the better)
• Use of the Xeon Phi desktops is limited to Fortran code
• Ability to get on with others and work as a member of a team.
• Strong research and problem-solving skills
We expect this project to have two deliverables. At a minimum, we expect a feasibility report that presents the applicability of various machine learning techniques to our noise testing. If applicable techniques were found, we expect the feasibility report to include measurements of their efficiency. If possible, a version of at least one of these applicable techniques should have been developed for a desktop computer. This software and accompanying documentation is the second deliverable.
If a suitable technique can be found and implemented, then our test engineer will implement it on our test systems and log the long-term success of the predictions. After adequate validation in production we would then replace our current noise testing with these predictions, and periodically batch test to re-validate the predictions.
The intern will receive $3,000 per month of the internship, usually in the form of stipend payments.
It is expected that the intern will primarily undertake this research project during regular business hours, spending at least 80% of their time on-site with the industry partner. The intern will be expected to maintain contact with their academic mentor throughout the internship either through face-to-face or phone meetings as appropriate.
The intern and their academic mentor will have the opportunity to negotiate the project’s scope, milestones and timeline during the project planning stage.
To participate in the APR.Intern program, all applicants must satisfy the following criteria:
• Be a PhD student currently enrolled at an Australian university
• PhD candidature must be confirmed
• Applicants must have the written approval of their Principal Supervisor to undertake the internship. This approval must be submitted at the time of application.
• Internships are also subject to any requirements stipulated by the student’s and the academic mentor’s university
29 July 2018
INT – 0462
FOR ANY ENQUIRIES ABOUT THIS INTERNSHIP03 8344 1785