Most maintenance methods or policies like RCM, CBM and TPM (or 26^3 other possible abbreviations) are:
- limited in scope and based, explicitly or implicitly, upon incomplete or even false models of reality
- simple, prescriptive methods to solve complex problems that are not amenable to such solutions
- promoted and oversold by consultants who themselves do not fully understand the problems, and/or –
- containing false measures of success that ‘demonstrate’ that they have been effective
The same is true for Predictive Maintenance (PdM) there is the promise that by recording and graphing, items can be taken out close to failure before being maintained. Failure defined as an change in the level or gradient of the graph. Assumed that costs for sensors, monitoring and data storage are not too high and that indication of imminent failure can be accurate predicted, inspection models should improve availability and productiveness.
There are in literature basically two types of practices connected with PdM: Stochastic maintenance practices in which the interval between interventions are based on statistical information. In a deterministic practice, interventions are determined by a pre-determine physical change or limit. Predictive Maintenance is founded upon statistics relating to the variability of failure and repair times or on the basis of a physical measurement.
Downside of PdM is that for each fallible part in a system the behavior, modelling techniques are needed. Unfortunately, few engineers have the mathematical knowledge to analyze already available data, infer a mathematical or statistical model and calculate the relevant optima to plan the tight maintenance practices. System models should be built-up from the behavior of all of these parts within the system.
That System model must be connected with the purpose of the system (i.e. delivery of a product, service) and the more economical aspects (profit, customers, stakeholders). The maintenance interventions of a system must therefore become more depending on the purpose or goals of the organization than on the physical aspects of parts. Other aspects like availability, reliability and energy consumption must be treated even important as i.e. technical capability (accuracy and precision).
Another important aspect to include are the cost of downtime. These are seldom calculated accurately, which results in underperformance of systems. This aspect has also a relation with buffers between production stages. Just-in-Time (JIT) production forces organizations to avoid breakdowns. The need to fulfil orders can lead to acceptance of sub-quality product and spares, botched repairs and neglect of maintenance.
An ideal maintenance strategy minimizes the combined downtime or the total cost, or maximizes the long-term expected profit. Only with proper data (cost, failures, time) it can be possible to find an optimum. Within existing concepts of PdM part are considered in isolation from the rest of the system, i.e. the models do not consider any relationships.
Without this relationship all claims towards PrM are unsubstantiated. Everything depend upon full system models. Most maintenance policies are still being applied without making proper predictions of the cost and savings or full assessment of the accuracy.