The value of predictive maintenance
As data has become easier to collect, store and analyze, many maintenance professionals are turning to prediction-based maintenance strategies. Let’s see why.
This article is a chapter taken from “The business case for predictive maintenance“.
The value of time
Predictive maintenance technologies enable an organization to take proactive actions, such as performing targeted maintenance, clustering maintenance activities, and adjusting asset usage. Few of these actions can be performed instantaneously, however (preparing a maintenance activity, for example, takes time), nor can they be initiated at every moment in time. In reality, most organizations have a response time—the time required to respond to a request for action—which depends on the action to be taken, the organizational context, and the timing of the request. The consequence is logical, yet important: the earlier you know an asset is going to fail, the bigger the range of proactive actions you can take.
In fact, for many actions, organizations have both a minimum response time and an optimal response time. Let’s take the overhaul of an electric motor as an example. The minimum response time is based on an emergency scenario. If you find out right now that the motor is about to break down, how long will it take you to start the maintenance activity? That might require that you stop the production process (wasting product), hire a skilled maintenance technician from a contractor (at a premium), or obtain a spare electric motor (with emergency shipping). If the consequences of breakdown are great enough, it’s possible to save money with such an emergency approach. But the motor’s overhaul would be much less costly if the organization had more time to react.
The optimal response time denotes how long an organization needs to optimally perform an action. In this example, the optimal response time depends on the time between planned production stops, the scheduling horizon for maintenance technicians, and the standard delivery time for electric motors. If production is stopped once every month, for example, and the scheduling horizon and delivery time are three weeks, the optimal value of preventive overhaul can be derived if the organization knows more than a month in advance that the motor is about to break down. The technology’s prediction horizon— how far in advance a prediction system produces a correct prediction—therefore determines the value that can be derived. This idea is visualized in figure 1.
Figure 1. The value of time: an example of a relationship between a condition monitoring technology’s prediction horizon and the potential value you can derive.
The value of accuracy
Even if you have all the time in the world, few predictive maintenance technologies are capable of perfectly predicting failures— neither from the start nor over time. Most new applications require learning—by machines, by humans, or both—while over time the predictive performance is subject to changes in the asset itself (e.g., modifications) and its operational context (e.g., process or product changes).
Two important performance indicators for a predictive maintenance technology are its sensitivity and its specificity. The sensitivity, also known as the true positive rate, indicates the percentage of failures that are identified beforehand (providing the organization sufficient response time).
Specificity, or the true negative rate, indicates how well the technology is able to identify that an asset is not about to fail. The higher the specificity, the lower the number of false alarms. The higher the sensitivity, the lower the number of unexpected breakdowns. Together, the sensitivity and specificity determine the technology’s accuracy: the percentage of failures and non-failures that are correctly identified as such.
Figure 2. The value of accuracy: classifying predictive sensitivity and specificity.
To calculate the business case for a new predictive maintenance technology, we have to take into account that its accuracy is not perfect. Especially for complex assets with multiple failure modes and degradation mechanisms, business case analyses should incorporate the probability of missing an upcoming failure and the probability of raising a false alarm. In addition, it should be noted that for many assets, the current accuracy— before implementing the new predictive maintenance technology—is rarely zero. Anomalies and upcoming failures can, for example, be detected during visual inspections, functional tests, and via production interference, although the prediction horizon of these methods is typically lower than with predictive maintenance technologies. Sound business cases therefore focus on the difference in accuracy between the old and new situations.
The value of decisions
Almost by definition, predictive maintenance is intended to reduce the cost of maintenance, by enabling you to skip scheduled maintenance activities, prevent unexpected breakdowns, reduce the frequency of inspections, cluster maintenance activities, or perform focused maintenance. This value is generated by making decisions that are better informed. But insight into the current and future state of assets can also benefit other stakeholders in the organization, such as the production department—by reducing energy and materials usage, increasing availability, reducing slowdowns and reducing quality losses—and the project department— by extending assets’ useful life. Table 1 summarizes common value drivers for predictive maintenance, including the range of realized benefits I’ve observed in practice (in percentages).
Table 1. Common value drivers for predictive maintenance, including benefits observed in the field.
Two things should be noted here. First, while each predictive maintenance use case can have multiple value drivers, only one or two are generally dominant. For example, if predictive maintenance is used to extend an asset’s useful life, the cost of maintaining the asset (and the associated risks) tend to increase. If predictive maintenance is used primarily to maximize the asset’s uptime, maintenance costs tend to remain stable.
Of course, there are examples in which predictive maintenance results in less unnecessary and time-consuming maintenance, thereby automatically increasing the asset’s overall equipment effectiveness (OEE) and limiting the number of risky maintenance activities. In these situations, the asset’s context mainly determines which benefit is dominant: for some contexts, uptime is much more valuable; for others, capital expenditure or the cost of maintenance.
Second, in my research I’ve observed that it can take up to several years before the information provided by the predictive maintenance technology is used in decision making, especially if the technology is not yet perceived as “proven” and the decisions carry risk. This reduces short-term benefits, as the technology only provides value if the organization’s decision-making is improved.
During that time, it’s possible for the predictive maintenance technology to generate “negative benefits.” If the costs of maintenance and capital expenditure haven’t yet declined, the initial investment in purchasing and installing the technology and the operational cost of using it to perform measurements and analyses can actually increase overall capital expenditure and maintenance cost. Moreover, if the predictive maintenance technology generates many false alarms, the number of maintenance activities might actually increase, further raising the cost of maintenance.
This was a chapter taken from “The business case for predictive maintenance” white paper by Roland van de Kerkhof.
To learn more about how to calculate the business case of predictive maintenance, click here to download the full white paper
Resources67 See all resources
Learn about three digital tools and methodologies, powered by artificial intelligence (AI) and machine learning, to help you make your maintenance and reliability work better, smarter and more efficient.