31 Jul Q&A with Perth Forum Speaker Melinda Hodkiewicz
Interviewed by Chantal Hortop
Melinda Hodkiewicz is BHP Fellow for Engineering for Remote Operations at the University of Western Australia, a 2018 Visiting Fellow at the UK’s Alan Turing Institute for Data Science and AI and a Theme Leader in the new Australian $8.8M Centre for Transforming Maintenance through Data Science. She will present on “Semantics – a Necessary Tool for Maintenance 4.0” at the GMG Perth Forum, which will be held August 14-15.
The views expressed below are those of the Melinda Hodkiewicz and do not reflect the views of any of the organizations she is associated with.
Why do maintenance management practices need to change?
Maintenance can account for more than 50 percent of a site’s operating costs. Maintenance practices in the mining industry to date have been slow to evolve and asset downtime in the mining industry remains stubbornly high.
Mining operations are 28 percent less productive today than a decade ago [1]. Mining continues to rely on human problem-solving skills and today’s maintenance management work is still a mainly manual process without widespread adoption of automation. Historically mining senior leadership has focused on silver bullet initiatives for improving maintenance practice.
What benefits might the resource sector see from improving their maintenance management using AI?
The maintenance work that directly impacts asset availability includes tasks such as inspecting, cleaning, repairing, and replacing assets. Every task actioned by a maintainer needs to be identified, planned, scheduled, documented and analyzed. Collectively, this is called work management. To understand the opportunity for AI it is important to understand the socio-technical nature of the work management process.
Data generated through work orders and parts data, together with all the equipment performance and condition data, is a massive potential information resource. Mining companies rarely use this information effectively to generate insight. The use of AI to automate and augment selected parts of the work management process is a significant opportunity to improve productivity.
What role does natural language processing play in the interpretation of maintenance documents?
We see many uses for the information and focus on three use cases. The first is for validation of predictive maintenance recommendations. To know if a prediction is correct, we need asset condition information from the maintainer when the unit is pre-emptively removed. For example, was the bearing damaged or still as good as new? This information is contained in the short text of the work order or in rebuild reports. Closing this feedback loop is vitally important for gaining trust in failure predictions. Trust is an issue for those on site as these predictions are increasingly being made by data analysts based in capital cities with limited maintenance experience.
The second use case is for automatic failure mode identification and the third is for automatic estimation of mean time to repair and mean time to failure statistics [2]. The latter two are about transaction automation, removing humans from the tedious loop of manually tagging failure modes and trying to work out failures and suspension events. There is also a strong case for use of natural language processing for safety incident data analysis.
What are some of the major changes you hope to see in the next few years in terms of the way operators are using automation in their maintenance management?
The challenge for maintenance in future is that we need to integrate human and digital capabilities in the same system. Too often effort is expended on trying to automate tasks which require high levels of human knowledge and expertise.
I am of the view that no company operating on its own will be able to make significant strides in natural language processing or the ability to run reasoning queries over their maintenance data sets. The time and culture are right for effort on development to be a shared endeavour for the benefit of all. The prize will not be in owning data but having the capability to use the tools that unlock the information it contains.
[1] Chakravarthy, C. et.al., Optimizing asset maintenance in the mining industry, Final study executive summary Deloitte Consulting LLP, May 2014.
[2] Sexton, T., Hodkiewicz, M., Brundage, M.P. and Smoker, T., 2018, September. Benchmarking for keyword extraction methodologies in maintenance work orders. In Proceedings of the Annual Conference of the PHM Society (Vol. 10, No. 1).