The era of the Internet of Things (IoT) will transform many aspects of industry and affect all our lives. Interconnected devices of all sorts share rich and voluminous data, providing the opportunity for analytics to extract insight, distill intelligence, and make a multitude of “things” – and the way organisations manage them and users interact with them – smarter.
Today, smart refrigerators know when to order groceries. Home automation systems can learn to manage heating and lighting to suit the occupants’ preferences while minimising energy usage. But what happens when “things” fail? If your car breaks down because a mechanic missed a potential problem at a recent service, you could have the stress and frustration of being stranded far from home or missing an important appointment. If an equipment failure stops an oil extraction operation, lost production can cost millions of dollars per day. An undetected crack in a train wheel can lead to a catastrophic derailment.
Advanced analytics can predict when, how and why equipment will break down – helping avoid all these adverse outcomes, while at the same time reducing the associated costs of maintenance.
In maintenance, all companies have started with the “break-fix” approach: something goes wrong, and resources are assigned to fix it. This is impossible to plan around, and hence costly to resource, and the unforeseen failures often cause significant operational disruption. Many companies and equipment vendors try to address this by moving to scheduled maintenance. While this is to some extent based on statistics – the mean time to failure of types of devices or components – it’s a blunt instrument: a pump operating non-stop at full capacity in extreme desert conditions won’t have the same failure characteristics as an identical one operated infrequently, on demand, in Northern Europe.
To get down to the level of the individual device, organisations will often deploy condition monitoring: taking readings from equipment that may give early indicators of failure, for example vibration and temperature of rotating equipment. These signals, though, often only become clearly apparent very close to failure – so problems are often not detected before the equipment is significantly damaged.
In true predictive maintenance, analytic algorithms are applied to all available and relevant data. That can include descriptive and historic failure/repair information for this specific device and all devices of that type; operational and production data; streaming data from sensors; environmental information; and unstructured data, such as inspection and maintenance reports. From these, the algorithms derive the key capability of predictive maintenance: to predict, accurately and robustly, when and how a device or component will fail.
With this predictive foresight, maintenance operations can be transformed. Pre-emptive maintenance can be planned to prevent the predicted failures from happening, avoiding losses and disruption from unscheduled downtime. The excessive costs of emergency maintenance can be reduced or even eliminated. Supply chains can be optimised, ensuring parts are in the right location at the right time. And insights gained into the causes of failure can improve product quality and operating practices, leading to improved reliability and increased uptime.
Today, predictive maintenance is giving benefits in many industries.
- Manufacturers of complex, capital-intensive mining equipment monitor their machines in operation around the world in real time, using predictive algorithms to schedule pre-emptive maintenance with minimal production disruption.
- Electricity companies minimise supply outages by predictively prioritising their field maintenance activities.
- By dynamically predicting usage of toner and other consumables, huge fleets of printers can be kept 100% available with minimal wastage.
- In most countries auto manufacturers don’t yet stream data from vehicles due to privacy concerns, but downloading data when the car comes in for service enables analytics-based diagnostics that have massively reduced the need for of repeat repairs.
- Aeroengines, with terabytes of data generated for each intercontinental flight, can have that data leveraged to give very early warning of future problems. Addressing these at an early stage reduces maintenance costs by tens of percent and, because actions are taken before the engine suffers excessive wear or damage, can extend the engine’s life significantly, delaying the multi-million dollar cost of replacement or rebuild by years.
As with any application of predictive analytics, value depends not just on being able to predict failure, but to turn that foresight into the most effective actionable decision. A steel mill, for example, may roll a slab of steel to a total sheet length of 2km or more. While a predictive model could give a real-time alert that the current sheet is going to kink or jam in the rollers, it’s too late to do anything about it while that sheet is actually in the mill. The right analytics-based decision in this case is, before each new slab enters the mill, to predict the risk that the mill’s state will cause a jam on this run – and therefore delay the run until after a pre-emptive maintenance intervention.
And beyond this prescriptive approach, there is the possibility to optimise maintenance operations more holistically. For example, predictive models may identify 100+ possible faults on a production plant in the next week. By taking into account a range of factors and constraints - the probability, severity and impact of each fault; the cost of pre-emptively fixing it; and the availability of the necessary parts, and of resources with the required skills – an optimisation algorithm can work out the best and most effective set of maintenance actions that can be taken, to be automatically scheduled in the work planning system.
The Internet of things gives us wonderful opportunities to make our world smarter. And predictive maintenance makes the “things” we depend on more available, more dependable and more productive!