Just like with preventative maintenance, predictive maintenance (as the name implies) is a strategy that schedules maintenance visits in anticipation of equipment issues or failures (i.e. predicts when issues will occur and schedules a visit in advance).
The objective of predictive maintenance is to provide cost savings to its users by scheduling service visits only when required, as opposed to reactive maintenance (which only schedules a visit when an issue occurs) or preventative maintenance (which uses either time intervals or usage, along with equipment life expectancies as the basis for maintenance).
This in turn reduces the costs of unscheduled downtime (associated with reactive maintenance) or ‘over maintenance’ of the equipment (a risk associated with preventative maintenance).
For more information on the difference between preventative and predictive maintenance, check out this article.
How does predictive maintenance work?
Predictive maintenance can be characterised as ‘condition based maintenance’ as the driver of timing of maintenance visits is the condition of the asset or equipment itself. Decisions on the remaining life of an asset or or equipment are driven by a combination of the current condition of an asset and an expectancy of the remaining useful life or time to failure.
In order to utilise a predictive maintenance strategy there are a number of precedent requirements including:
- A data set that contains life expectancy statistics for the asset or equipment
- Data that can be filtered according to measurable conditions
- Thresholds regarding prediction of failure
- Condition measurement (for example sensors or inspections)
Once the above have been established, the condition of the equipment as a result of periodic inspection (offline) or continuous transmission (online) will trigger the raising of a work order and the scheduling of a maintenance visit.
This data is then used to optimise a cost function so that the expense of performing maintenance (or replacing equipment) can be minimised. Because the data set that is used to inform these decisions is one that will grow, predictive maintenance is an area of facilities management that may benefit from the use of artificial intelligence (or machine learning).
In order to minimise the expected cost function, the model will require inputs such as the business costs of downtime; replacement costs; residual values; useful life remaining and an expectation of future maintenance costs.
Assets and equipment will typically have a known (or estimable) cost function associated with their ownership.
This can be thought of through the example of car ownership. If you replace the car too frequently, you will keep suffering the sharp initial depreciation even though the car has a lot of life left in it. If on the other hand you hang onto your car year after year, there will be a point where the maintenance costs are higher than the costs of getting a new car.
Cost functions can be simple or complex, but must be known in order to decide the point at which to take action, as per the below.
An example of a replacement cost function
An example of a predictive maintenance workflow
How is asset or equipment condition measured?
As can be seen above, the key variable in assessing the maintenance schedule is the ‘condition’ of the asset or the equipment.But how does one measure condition?
Condition (or a proxy for condition) is measured differently from one piece of equipment to the next, but in essence the method of collection is the same. Condition is assessed via the capturing of quantitative information following non-destructive tests of the asset or equipment in question.
These tests or monitoring of the equipment can take the following forms: oil levels, vibration analysis, infrared scans, sound level assessments, discharge levels (such as gas or pathogen emissions) etc.. As you can imagine, these functions can get very complicated as additional metrics or data are added to the decision criteria.
Just like preventative maintenance, predictive maintenance is used as a strategy in order to try and minimise the costs of asset and equipment maintenance and/or ownership. Decisions on maintenance are taken on the basis of information endogenous to the equipment itself rather than based on independent assumptions.
In order to make these decisions there are a number of requirements that must be satisfied such as forecasts of the remaining life and/or the future cost of maintenance.
The decision criteria that can be used in a predictive maintenance strategy can get very complicated and an investment is required upfront in order to collect the necessary data. As such, it is often used where equipment costs and/or the costs of downtime are high.