Luke Frost explains how predictive analytics can optimise mining asset maintenance.
As the world collectively moves towards a high-tech future, mining in Australia stands to gain a lot from these advances. In particular, the Internet of Things (IoT) and the availability of big data can make a large impact in the way that Australasian mines are run, increasing productivity and improving the lifespan of equipment. Collecting and distilling data and extracting meaning from it holds great potential for reducing the total cost of ownership in Australasian mining companies, as well as optimising existing resources and improving compliance.
Companies are increasingly tuning in to the potential of using predictive analytics – a process of using statistical and data-mining techniques to analyse historic and current data sets, create rules and predictive models and predict future events. Big data is fast becoming a vital instrument to realise asset lifecycle cost reduction and improve the speed and accuracy of decision-making.
PREDICTIVE ANALYTICS FOR ASSETS
Mining assets are seldom standalone. They exist as part of an ecosystem whereby resources feed off each other. Between asset procurement/commissioning and decommissioning/salvage lies the productive life of an asset. Regular upkeep or maintenance is needed to maximise this life. There are two perspectives on how predictive analytics can help optimise asset maintenance:
- Individual equipment perspective: Typically, maintenance frequency is based on various parameters such as asset age, asset criticality, operating environments, and risk of failure. The risk of failure often forces over-maintenance of critical assets, leading to higher maintenance expenses. Over-maintaining can mean replacing lubricants or bearings that still have life left, and replacing tyres based on their age rather than the conditions under which they operate. Maintenance also means taking the machine off-line for a period, thereby reducing overall productivity of the mine’s assets. The ability to predict a failure can help by reducing unnecessary maintenance activities.
- Productivity perspective: As can be imagined, a system of assets, designed and scheduled for maximum productivity, can benefit immensely if the failure of an event requiring maintenance is known in advance. Alternative schedules or plans can be prepared to ensure maintaining productivity levels.
Ironically, in almost all common maintenance scenarios, the ability to predict failures and move maintenance procedures closer to failure already exists. It is just that the risk of failure keeps many organisations from taking those steps.
Big data can be used to predict when a machine is in need of maintenance attention before it becomes a problem, thereby reducing risk, while often increasing the time that the machine is operational and productive.
“Data is a powerful tool for any business, no matter what the industry might be,” says Daniel Ng, Senior Director, APAC, Cloudera.
“In the case of a mining, energy or resources enterprise, there is a host of useful information that can be collected every second of the day which can tell a company an infinite number of things about their operations. It makes perfect sense to collect and use as much of that information as possible to improve performance and maintain a competitive edge in these times.”
Let us consider a multi-stage centrifugal air compressor. Some of the typical maintenance procedures carried out today on centrifugal equipment include:
- Oil change: This involves oil replacement at a fixed schedule because real-time oil quality analysis is not possible.
- Intercooler cleaning: This involves intercooler cleaning at a fixed schedule because real-time analysis of coolant and intercooler efficiency is not available, even though inlet and outlet temperatures are measured.
- Bearing replacement: This involves bearing replacement at a fixed frequency because no scientific way to determine bearing life left is deployed even though lubrication oil temperature, viscosity, and bearing vibrations are measured regularly.
- Major overhaul: This involves a major overhaul of the equipment at a fixed frequency, usually more than that of the aforementioned procedures, because the manufacturer has prescribed it, even if all vital parameters look all right.
All this entails avoidable downtimes, unnecessary maintenance expenses, lots of collected yet unutilised data, additional capital expenditure for back-up equipment and complexity of operations, due to the need to accommodate multiple downtimes, coordinate with downstream and upstream business units, and so on.
That can cost a lot of money. However, the maintenance requirements are real. Operating equipment close to failure thresholds carries an inherent amount of risk as well. So, you might ask, how then does predictive analytics improve operational efficiency?
Primarily, big data and predictive analytics can help to determine failure threshold to a much more exact point, identify buffer zones and quantify risks in operating equipment within known boundaries. Data from the machines can be collected, collated into usable information, then built into a matrix to work out patterns of typical behaviour about that machine, defining boundaries and working out exactly how much use it can take before it needs to be maintained.
Consistency across all similar machines in a mine is very hard to pinpoint. Some will be operating in different environments to others, some may be exposed to dust and vibrations, others may be used in areas of high humidity. All of these factors can affect smooth operations, and reduce the lifespan of a machine or device.
Rather than using a blanket formula for maintenance, as has been the case in the past, predictive analysis allows a much deeper insight into the workings of an individual machine.
Predictability does not mean the maintenance software will tell you to replace a certain bearing on a certain equipment in a certain process on a certain date at a certain time. It would, however, indicate the probability (for example, there is 89 per cent chance) of a bearing failing at a given time. Business rules can be built to instruct a bearing replacement above a certain threshold (for example, when there is a more than 70 per cent probability of failure). This therefore allows the risks an organisation is willing to take to be quantified (for example, 70 per cent is the risk threshold).
Oil change: Oil quality over a period of time can be plotted to predict with a degree of confidence what the oil quality (oil viscosity or decomposition) will be at a given time. This can then help to answer the question: Is there a significant difference in quality of oil supplied by different vendors?
Intercooler cleaning: Temperature delta between inlet and outlet for coolant water and air is a good way to measure efficiency.
A lower delta would mean less heat taken away and lower efficiency, assuming everything else is intact. This delta over a period of time can be plotted to determine statistically what the value will be at a given point of time and whether or not there is a need for maintenance. This can help figure a trend.
Bearing replacement: Predicting bearing life can be tricky as identical bearings tend to display different endurance lives. Reliability of bearings has to be predicted based on installed population of bearings and analysing failure data to identify patterns.
Major overhaul: Something like this will be difficult to predict statistically (or there may not be a need). But if individual components—such as intercoolers and bearings—are working fine, making a judgment or qualitative decision on the need for a major overhaul should be straightforward.
All this can help widen maintenance intervals, push maintenance activities closer to when needed and consequently, optimise asset operations.
IT’S NOT DIFFICULT OR EXPENSIVE
Does building predictive analytics capability cost lots of time, money and effort? The answer to this big question is an emphatic “no”. There is always a sweet spot where savings from predictive capabilities (reduction in maintenance expenses, capex, spares utilisation, and so on) outweigh the cost of building those capabilities (software, human resources, etc). Once done, the benefits through direct asset-related saving (fewer maintenance dollars, fewer spares, longer life span), and organisational saving (optimal teams, increased operational efficiencies) are immense.
“Using an open-source platform allows an enterprise to tailor their analytics solution to their specific needs, and take advantage of many specialised tools that can drive their data further. There really are an abundant amount of possibilities for optimising machinery performance by using data and information better than ever before. Predicting when a machine will break down and be offline is one thing, but knowing exactly how that machine is operating within a huge ecosystem of other devices, and how its individual performance is impacting other machines, workflows and business units, gives resources companies a huge and powerful advantage,” says Mr Ng.
KPIS TO MONITOR ASSETS KPIs
(Key Performance Indicators) are different for different asset types, operating conditions, criticality, and so forth. Due consideration needs to be given to the role of statistics and the related process changes and change management, among others, in order to build a predictive maintenance culture. Data and statistics will provide quantification, but will need to be applied with business insights to derive benefits (for example, the 70 per cent risk quantification mentioned earlier).
Here are a few commonly used KPIs:
Reliability given time: This indicates the probability that a unit will operate successfully at a particular point in time. For example, there is an 88 per cent chance that the product will operate successfully after three years of operation.
Probability of failure given time: This indicates the probability that a unit will fail at a particular point in time. This is also known as “unreliability” and is the reciprocal of reliability. For example, a 12 per cent chance that the unit could fail after three years of operation (probability of failure or unreliability) is the same as an 88 per cent chance that it will operate successfully (reliability).
Mean life: This indicates the average time that the units in the population are expected to operate before failure. This metric is often referred to as “mean time to failure” (MTTF) or “mean time before failure” (MTBF).
B(X) life: This indicates the estimated time when the probability of failure will reach a specified point (X per cent). For example, if 10 per cent of the products are expected to fail by four years of operation, then the B(10) life is four years.
As mining and the resources industries move into a modern, high-tech world, the use of smarter systems and information gathering will be of paramount importance for keeping us competitive on the global stage. Just as we find better ways to mine precious resources from the ground, we should also strive to mine as much information as we can from the tools at our disposal, and learn as much as we can about improving performance.