Intelligent Systems for Monitoring Energy Usage Patterns & Maintenance Scheduling
The aim of this project is to investigate the role of artificial intelligence techniques in the management of energy consumption and maintenance by building on previous projects which have looked at energy (e.g. Defra project), condition monitoring and maintenance (Defra, Posseidon, Dynamite) and the application of AI techniques (Posseidon) to such problems.
This project will extend several previous projects, most notably the Defra Energy project by measuring energy usage. A lack of suitable data has been established as being the key barrier to developing a research direction looking at the application of artificial intelligence techniques, principally reinforcement learning, to automated scheduling of maintenance activities based on energy consumption.
Dr. David Baglee
Dr. Mike Knowles
Dr. Alan Yau
A study is undertaken to ascertain the barriers which exist for local firms in terms of improving their maintenance requirements. This is based around a survey available online. The survey looked at three main areas, current maintenance strategy, current energy management strategy and perceived barriers to developing new strategies.
Automatic Scheduling of Maintenance
The key to successful condition based maintenance (CBM) is determining the optimal time to perform maintenance in terms of the condition measurement. Optimality may be defined in terms of mean time between failures (MTBF) or cost. Reinforcement learning is a machine learning paradigm where the likelihood of a particular behaviour is increased by offering some reward when the behaviour occurs. In computational terms RL is concerned with maximising long term reward following a sequence of actions.
A tool has been developed to allow us to investigate the application of RL to the problem of maintenance scheduling. Ongoing experiments with this tool will support the development of new, energy-aware condition monitoring applications.
Measuring Energy Consumption
In order to further investigate the link between maintenance and energy consumption we will be using the latest in affordable non-contact energy measurement equipment. Surveys will be carried out in industrial and domestic settings to provide data for ongoing research and development.
Detecting Unusual Power Consumption Patterns
The link betweem maintenance and energy consumption is widely known but has received little coverage in terms of quantification or as an indicator of equipment condition. Our premise is that energy consumption can be used as both a condition measurement and as an indicator of maintenance efficitiveness. Unlike other measures, however, energy usage cannot be assessed using simple static thresholds, averages or other statistical calculations since energy consumption is in most instances a function of variables such as time, equipment use, temperature conditions and others. We are currently developing a tool to monitor energy usage using Gaussian Mixture Models to detect abnormal usage patterns.
Knowles M.J., Baglee D. and Wermter S. Reinforcement Learning for Scheduling of Maintenance. Thirtieth SGAI International Conference on Artificial Intelligence (AI-2010).December 2010.
Knowles, Michael and Baglee, David (2011) Energy usage modelling as a condition monitoring tool. In: MIMAR 2011 - 7th International Conference on Modeling in Industrial Maintenance and Reliability, 18th - 19th April 2011, Cambridge UK.
This page was published on 8 April 2019