I'm a software engineer and data scientist with experience of translating engineering problems into machine learning solutions. Here is a sample of the interesting projects I've worked on.
There's some related code on my Codeberg page. If you have any questions, send me an email at vic@cowlet.org.
Engineers need to periodically access offshore wind turbines to perform maintenance. This work developed a suite of wave height forecasters to predict whether waves will stay below the access limit, and an improved metric for assessing the strength of a forecaster.
Transformers are the most expensive components of power networks. This research resulted in a probabilistic model for predicting remaining transformer life, based on the ambient conditions it operates in. This can be used to speculate about the effects of emergency overloading, as well as predicting normal behaviour.
Insulation is used in all electrical equipment to prevent short-circuit faults. This work improved the accuracy of diagnosis of insulation defects by using a deep approach. It also led to insights about important parts of the data, through a comparison of features learned by the deep network and typical engineered features.
The data used to diagnose insulation defects can be distorted by harmonics on the power network. This work proposed an alternative to costly harmonic monitoring, by diagnosing the presence of harmonics from the same condition data. This gives a clearer picture of equipment health, separating harmonic effects from defect effects.
Older transformers develop a signature pattern of behaviour that is specific to each one. Anomaly detection is more useful than diagnostics in this case, since it's more important to spot changes in behaviour than to repeatedly classify minor problems. This work applied Conditional Anomaly Detection to keep two transformers in service for 18 months, while minimising false positives.
A growing number of people are interested in data science, but academic papers and formal training can be difficult ways to break in. My personal site shows how to get started with data science in R, through a series of articles covering feature extraction and selection, clustering, and classification using multiple techniques. These are demonstrated on an open dataset of faults within bearings.
I was asked to contribute to the Smart Grid Handbook by writing a chapter on "Data Analytics for Transmission and Distribution". I drew on my experience of developing and deploying data analytics in this field to highlight the increasing role that online analytics can play, and some lessons learned from each case study.