Staying One Step Ahead: Transforming Machinery Maintenance with Machine Learning
Written by Dr Nick Mayhew
Metals & Mining
Machine Learning, Failure Prevention Of Machinery
Mining is an incredibly machinery-intensive industry. At every step of the extraction process, heavy machinery is involved, and it is machinery that is really put through its paces.
Drilling, cutting, crushing – all of these activities are high impact, and machinery failure is an inevitable outcome during the process. When equipment does break down, it is then put out of commission until repairs can take place. This results in an increase in costs, both for the repair itself and for the loss of productivity during the down time. And the costs of downtime are considerable – a single incident can cost up to $180,000. Unexpected machinery failure is also a safety issue that can put machine operators in danger. 80% of fatal accidents in the coal mining sector of China are a result of equipment failures. To avoid and mitigate the problems associated with machine failure, different maintenance strategies are used, and the use of machine learning presents a real opportunity as the strategy of the future.
Traditionally, there have been three main types of maintenance. The first type is reactive maintenance. This approach involves running machines until they fail, and only then repairing them or performing maintenance. This strategy is highly disruptive for the production process, as when a machine fails unexpectedly, the whole production process will grind to a halt, resulting in a loss of earnings. On the one hand, it is a low cost approach to maintenance as technical staff are only required in the event of a break down, but on the other hand the costs of unplanned disruptions to production are high, and equipment failure can be a dangerous outcome.
The second type of maintenance is planned maintenance. This involves periodically checking and servicing machinery and equipment. As an approach, it helps to avoid the machine failures associated with reactive maintenance, and also allows companies to allocate for the downtime that comes with planned maintenance. Periodic maintenance can help to extend the service life of machines and avoid catastrophic failures before they happen, but the repeated periods of downtime are costly.
Proactive maintenance is the third type of maintenance. Here, analytics are used to work out what the capacity of a machine is and to understand what the critical moments may be. In this way, it can be determined when a machine is in need of repair, allowing for a better allocation of maintenance resources. At the moment, most proactive maintenance is undertaken manually by people, and it is this area that presents the biggest opportunity for machine learning to transform the machinery maintenance process.
A New Frontier For Machine Maintenance
Through the use of AI and machine learning, maintenance has the opportunity to move beyond being merely reactive or proactive, and instead become predictive. In predictive maintenance, repairs only happen when they are necessary, and arise through the analysis of data from various sources. Modern mining machinery is equipped with a wide array of sensors that collect huge amounts of data. This data can then be analysed in real time in order to pre-empt the onset of faults and failures. Machine learning can analyse and understand data in ways that are far beyond the abilities of people, allowing for more accurate and timely notification of impending maintenance needs.
Predictive maintenance will reduce the costs of maintenance by enabling repairs to take place before a failure shuts down production. Over time, AI and machine learning will advance to the point where it will understand an impending problem and know what the solution is, even going as far as to be able to automatically schedule repairs, order parts, notify technicians and so on. Machine learning always improves over time as more data becomes available, meaning that the understanding of when a machine will break down and how it needs to be addressed is always getting better, ultimately leading to minimal downtime and decreasing costs.
The success of predictive maintenance depends on a number of factors: availability of the right data, appropriate framing of the problem, and proper evaluation of the predictions. As mining equipment already gathers a huge amount of data and the problems of failure are well documented, it makes the mining industry a prime candidate for successful application of this emerging technology.
Some analysts are even predicting that the future will bring machines that are capable of self-maintenance. 3D printing can be combined with AI to enable machines to pinpoint problems and order replacement parts without any need for human intervention. Until such a time, predictive maintenance is still reliant on human intervention.
At the moment, predictive maintenance is still in its early stages, but the mining industry has been an enthusiastic early adopter of the technology. Given that 40% of costs of mining are derived from asset management, it’s easy to see why it has such an inherent appeal. More effectively managing the costs associated with these assets really does hold the power to transform the industry.