Rory Telford

Hi, I'm Rory, and welcome to my personal web page. Here you will find information about how I keep myself busy in my professional life, including some of the projects and research I'm involved in. I am currently an Associate Researcher in Electronic and Electrical Engineering at Strathclyde University, Glasgow. My original background was in electrical engineering and, in particular, researching and developing methods that improve operation and reliability of power systems. I do also have particular experience and interest in using machine learning, artificial intelligence and probabilistic modelling to solve problems that would otherwise be difficult to overcome. My interest in these fields stem from my time completing a Ph.D. which focussed on fault diagnostics within aircraft electrical sub-systems. Since finishing my Ph.D., I have led and worked on a number of different projects with both industrial and government partners. Please do read on to find out a bit more about myself and these projects and feel free to get in touch using the above contact details if you want to discuss any of the information here.

Knowledge & Skills

• My honours degree was in Electronic and Electrical Engineering and I was particularly focussed on the "power" aspects of this (the electrical bit) as opposed to the electronics side. I was interested in modelling large interconnected systems, which tends to involve running power flow simulations i.e. develeoping a model of the electrical network and running hypothetical scenarios to see how power flows round the network under certain generation and load conditions. After finishing the honours, I went on to complete an MSc in the same subject and continued to focus on the power system aspect of things, including operation of power markets and trading arrangements.


• As part of my MSc project I was first introduced to aircraft and shipboard electrical sub-systems - I enjoyed this project so much I accepted an offer to undertake an Airbus sponsored Ph.D. investigating the move to "more-electric" aircraft. Throughout the Ph.D. I reserached methods for diagnosing series arc faults, conditions that are particulaly dangerous as they pose a significant risk of fire and they are also extremely difficult to accurately detect. It was throughout the Ph.D. that I really became interested in machine learning and mining datasets to extract useful information. I developed and implemented the IntelArc method for series arc fault diagnosis - this method is based on machine learning and depends on transformation of raw sensor data to extract distinct fault features.

• Since completing my Ph.D. I have worked as a postdoc on a number of different projects. Notably, I began life postdoc as the postdoctoral research fellow on a Scottish Governement funded project (via ClimateXChange in Edinburgh). As part of this I was using quantitative methods to understand the impact of renewable energy on Scotland's electrical distribution systems. Most receently I have been working on the AMIDiNe project which is investigating and developing data driven analytics for improving management and operation of electrical distribution networks.

• Electrical Engineering I have detailed knowledge of distribution networks and assets, including operation of medium and low voltage networks. I also have significant experince in developement of power system models and running power flow simulations. I have working knowledge of protection systems (including techgnologies and schemes) as well as power electronics.
• Data Science I am comfortable undertaking analysis of large datasets, including probabilistic modelling and extraction and evaluation of statistical features. I have particular knowlegde of analysing multivariate data as well as formal analysis of correlations between variables and application of parametric and non-parametric techniques. I also have experience setting up and evaluating SQL databases and in using and analysing Geographic Information System (GIS) data.
• Machine Learning I have significant experience in the development of machine learning based methods for tackling difficult (and non trivial) problems. I have developed using a number of different techniques and am accustomed to manipulating data features and rigorously testing, tuning and validating to optimise performance. I have developed applications for both classification and forecasting problems.
• Signal Processing My backgrounds in electrical engineering and machine learning means I have experinece in the transformation of signals/data into other domains to find and extract latent (hidden) information. I have particular practical experince using Fourier and Wavelet Transforms.
• Computer Programming This includes algorithm and procedural type coding while I have also developed object oriented based programs. I have experince across a number of languages, although I am probably most comfortable in Python. I have also previously developed using web framewroks (Django mainly) and front-end coding (CSS, html, bootstrap).

Experience

Education

Publications

Below are a list of articles and reports that have I have published as part of the various projects that I have worked on as well as my Ph.D.

1. Predictive Thermal Relation Model for Synthesizing Low Carbon Heating Load Profiles on Distribution Networks, Anderson, A, Stephen, B, Telford, R and McArthur, S, 2020 IEEE PES Innovative Smart Grid Technologies, Nov. 2020,
2. Design of a DSO-TSO balancing market coordination scheme for decentralised energy, Edmunds, C, Galloway, S, Elders, I, Bukhsh, W and Telford, R, IET Generation, Transmission and Distribution, 14, 5, p. 707–718
3. Dirichlet sampled capacity and loss estimation for LV distribution networks with partial observability, Telford, R, Stephen, B, Browell, J and Haben, S, IEEE Transactions on Power Delivery, Sep. 2020
4. Non-Gaussian residual based short term load forecast adjustment for distribution feeders, Stephen, B, Telford, R and Galloway, S, IEEE Access, 8, p. 10731-10741, Jan. 2020
5. Predictive thermal relation model for synthesizing low carbon heating load profiles profiles on distribution networks, Anderson, A, Stephen, B, Telford, R and McArthur, S, IEEE Access, 8, p. 195290-195304, Oct. 2020
6. The Impact of Energy System Decarbonisation and Decentralisation on Scotland's Electrical System , Telford, R, Galloway, S, Robertson, E, Frame, D, ClimateXChange Commissioned report, Nov. 2019
7. DC arc fault detection methods in MEA distribution systems, Thomas, J, Telford, R, Rakhra, P, Norman, P, and Burt, G, Aerospace Systems and Technology Conference, London, UK, Nov. 2018
8. Diagnosis of series DC arc faults - a machine learning approach, Telford, R, Galloway, S, Stephen, and Elders, I, IEEE Transactions on Industrial Informatics, 13, 4, p. 1598-1609, Aug. 2017
9. Analysis tool for initial high level assessment of candidate MEA architectures,Telford, R, Jones, C, Norman, P, and Burt, G, Aerospace Systems and Technology Conference, Hartford, Connecticut, Sep. 2016
10. Fault classification and diagnostic system for UAV electrical networks based on hidden Markov models, Telford, R and Galloway, S, IET Electrical Systems in Transportation, 5, 3, p. 103–111, Sep. 2015
11. Evaluating the reliability & availability of more-electric aircraft power systems, Telford, R, Galloway, S, and Burt, G, 47th International Universities Power Engineering Conference (UPEC), London, UK, Sep. 2012

1. Classification and characterization of intra-day load curves of PV and non-PV households using interpretable feature extraction and feature-based clustering, Ge, J, Hu, M, Telford, R, Stephen, B, Wallom, D, submitted to Applied Energy
2. Automated Quantification of Photovaltaic Penetration on Unmonitored Distribution Network Feeders, Ge, J, Hu, M, Telford, R, Stephen, B, Wallom, D
3. Intermittent Series DC Arc Fault Detection in DC More-Electric Engine Power Systems Based on Wavelet Energy Spectra and Artificial Neural Networks, Thomas, J, Telford, R, Rakhra, P, Norman, P, and Burt, G

• Novel Methods for Improving Fault Protection & Health Management within Advanced Aircraft Electrical Power Systems, Telford, R, Apr. 2017

All accepted articles are indexed on Scopus.

Case Studies

A brief outline of some of the projects I have worked on over the past few years are described here. I don't want to go into too much detail, so here I have provided informal, level summaries of some of the ideas that I have developed and implemented. Most of the case studies described here have been descriebd in more detail in published articles, so please refer to my publications for more details if you wish. Although, if you prefer, you can always contact me to discuss anything in greater detail!

• Back in the 1880’s there was a battle between two great minds which shaped the way in which electricity is today transmitted. On one side, Thomas Edison was fighting the corner for direct current (DC), while Nicola Tesla was on the side of alternating current (AC). To cut a long story short, Tesla and AC were the victors, and the majority of nations designed their electricity systems based on AC transmission. There were a though a few exceptions, such as Finland, while some stand-alone systems found in aircraft and ships have DC sub-systems.

• More recently, DC has made a comeback of sorts with the integration of AC-DC transmission links as well as increasing uptake of photovoltaics and large-scale battery energy storage. There are a number of challenges that come with relying on DC based systems – the series arc fault is a particularly hazardous condition which is difficult to detect in an AC setting, while, in DC, the condition poses even more difficulty. Series arcing defies the logic of traditional electrical faults. Under typical fault conditions, current flowing around circuits increases; however, in series arc faults, the opposite happens, and current actually decreases. This means that more advanced techniques are required to detect them.



• As part of my Ph.D., I developed the IntelArc method to accurately diagnose these conditions. Detailed testing has proven that IntelArc can correctly diagnose fault conditions and also eliminate false positives during network reconfiguration conditions (e.g. when loads and equipment are switched on or off).

• In real time, IntelArc is essentially a two-step process: the first step involves feature extraction from 50ms windowed raw current data. Raw current data is not particularly useful for detection due to the presence of noise disturbances. The Wavelet transform (or the discrete wavelet transform (DWT) specifically) is useful in that it converts raw data into bands of different frequencies, and also tells you at what point in time those particular bands are present. This was proven to be very powerful for detection.



• In the second stage of IntelArc, the extracted wavelet features are applied to trained Hidden Markov Models (HMM). HMMs are a type of model that relate observed data to hidden states, and training involves using data to tune model parameters, including transition probabilities and observation model means and (co)variances. I deemed HMMs ability to handle non-stationary data as being potentially useful for diagnosis of series arcing.

• Training was undertaken in a supervised fashion, and individual HMMs were trained that related to different nominal and fault conditions, with training data gathered from both experimental and computer simulations. In application, DWT features extracted from windowed current data is applied to each HMM, and a log-likelihood measure is used to determine what model (and thus condition) most closely matches the data.

• In experimental validation and testing, IntelArc showed its capability for accurate series arc fault diagnosis and isolation. For much more information, please refer to this paper.

• Electricity distribution network operators (DNOs) have traditionally managed passive networks to ensure power generated from large, transmission connected, generators is securely distributed to end consumers. However, the move towards a low carbon economy means that DNOs are increasingly required to consider, and perhaps revise, the ways in which distribution networks are operated. In particular, the rise in low carbon technologies such as electric vehicles (EV), wind turbines and photovoltaics (PV) integrated within distribution networks is likely to change power flows within them.

• Seemingly gone are the days when electrical load growth was relatively predictable, power flowed unidirectionally from large generators to many consumers, and DNOs could conservatively manage the networks through upgrading the network infrastructure. There is now greater impetus on actively managing distribution networks through matching supply and demand at these lower levels. Regulatory controls are now motivating DNOs to consider active management to overcome potential challenges as opposed to ever increasing network investment strategies. Within this shift from passive to active management lies the roots of DNOs transitioning to Distribution System Operators (DSOs).

• There is, though, significant uncertainty surrounding the evolution from DNO to DSO and how future changes will impact distribution networks. This is where a distribution network Digital Twin is likely to help out.

Digital Twin

• A Digital Twin is the accurate and detailed mapping of physical assets and their connections within a distribution network into a virtual domain to create a replica software model. This mapping allows DNOs to analyse:

  1. how assets perform across their networks;
  2. how future changes within networks may impact asset performance, and;
  3. how, and where, active management would be both feasible and strategically beneficial.

  A key aspect of a Digital Twin is how they can be manipulated to assess how different scenarios will alter the behaviour of assets, or even estimate when it is likely to require upgrade. The development of a Digital Twin relies on various data sources and significant domain knowledge. A main source of asset related data for electricity distribution networks is geographic information system (GIS) data, which DNOs typically store as shapefiles. These shapefiles are vector data formats that describe each asset within the network in terms of their geographic locations, attributes and connectivity.

• As part of the AMIDiNe project, a prototype distribution network Digital Twin methodology has been developed using GIS shapefiles. The methodology autonomously translates shapefiles into accurate power systems model, from which advanced power flow studies can be undertaken. There is the potential here to link network models with data from metered assets and even consumption data with smart meters installed.



• While there has been advances in Digital Twinning at higher voltage levels of the grid (e.g. Transmission Networks), the roll-out to medium and low voltage levels has been slower – electrical cables and lines are shorter, there are more of them, and data stored in the shapefiles is less likely to be accurate. Having a means of automatically generating Digital Twins from which scenarios can be tested and analysed in detail would be beneficial for the DNOs of the future. Distribution networks are key facilitators in the move to low carbon based energy systems - embracing digital assets and improving the integration of data and IT systems through development of Digital Twins is pivotal in enabling this transition.