The Under-Emphasized Role of Ascertaining an Actionable P-F Interval

Most of the conventional resources that explain the implementation of Reliability Centered Maintenance (RCM), Predictive Maintenance (PdM) or Condition-based Maintenance start from a point of discussion in which the Potential Failure to Functional Failure (P-F) Interval is taken for granted and the discussion is primarily based on how to design a maintenance program around that.

In my opinion, the point that needs to be emphasized and expanded the most, especially in introductory phases, is how the P-F Curve and the P-F interval will be ascertained? and how implementation of Conditioned Based Maintenance or Predictive Maintenance is almost predicated on knowing the P-F intervals of different failure modes in different equipment.

After all, if you don’t know how much time different defects (potential failures) found in your equipment take in converting into functional failures, how are you going to decide about the maintenance intervention?

To give an example from my experience with airport equipment, a critical failure mode of Preconditioned Air (PCA) Units is fatigue failure of any of the blades of its high speed rotary fan of its condenser heat exchanger. It can be caused by imbalances in the fan, abnormal vibration or material defects.

Unless you know how much time it takes for any anomaly in fan balancing, vibration or material to convert into a fatigue failure, you cannot decide how long can you allow the blade to run before taking the unit out of operation for blade replacement.

Application of any Condition-based Maintenance or Predictive Maintenance (PdM) program on the PCA Units is predicated on having reliable knowledge of P-F intervals i.e., how much time will it take for an observed defect or anomaly in condenser fan to convert into a failure?

The above is the most challenging question to answer. It is harder than realized to know the P-F interval.

What is the P-F interval of a 1mm x 2mm surface defect converting into failure of the fan blade? Or, what is the P-F interval of an abnormal vibration (off by say 1 mm/s) converting into blade failure??

The true P-F interval will be found when you leave a small surface defect observed on the condenser fan blade to grow until it results in blade failure. However, it is obvious that you can’t find P-F intervals by allowing failures, unless you are in a laboratory environment specifically working to find the P-F interval.

Hypothetically speaking, even if you allow the above to happen, a single observation for just one of the fan blade failures in one of the many PCA units will not be representative of that particular failure mode. You can’t say that it will take same time for a surface defect to convert into blade failure on all PCA units.

In view of above, it becomes a critical question that how an actionable P-F interval is ascertained?

One of the practical ways an actionable P-F interval is ascertained is opinion-based. The opinion of experienced operation and maintenance personnel is taken regarding what do they think, based on their experience of maintaining a piece of equipment for so long, would be the time taken by a certain defect to convert into a failure.

The above method gives an actionable P-F interval only if a consensus is developed over a P-F interval [1]. If the experienced professionals agree closely over a certain time period, it can be reliably taken as an actionable P-F interval. If the professionals have widely different opinions about it, you cannot rely on the resulting P-F interval.

It is important to note that in the above scenario, the P-F Interval will not be the time taken by a potential failure to convert into a functional failure but the time interval that reliable opinions predict would be.

The other way of predicting the same is through predictive data models. These are mathematical models that predict when a failure would take place based on extrapolating existing data on degradation.

For instance, how the crack has propagated in a mechanical structure in the past 2 weeks? You could then fit the curve and predict a future point of failure.

Such data models can either have linear degradation paths as their challenge for prediction or non-linear paths. A linear degradation, such as the gradual degradation of a rubber tire tread, is relatively easy to extrapolate. What’s hard is to predict a non-linear degradation path.

What makes the matter more challenging is that majority of industrial equipment and its components follow a non-linear degradation path [2] which makes ascertaining an actionable P-F interval one of the most difficult tasks in the whole implementation process [3] [4].

The P-F interval in non-linear degradations is therefore predicted by different probabilistic models developed through research [5]. The model is trained by providing it available data accompanied with suitable assumptions based on equipment understanding.

The balance between data and assumptions while training the model depends on the extent and depth of available data and is critical to the prediction accuracy of the model.

Consequently, such models are data intensive and may require application of Condition Monitoring on a piece of equipment to collect the data in the first place.

Paradoxically, Condition Monitoring application through installation of sensors and measurement systems requires investment and is a decision that needs data from the P-F Curve to know if the nature of failure mode even makes it cost effective and practical to predict failures and decide maintenance interventions accordingly.

In short, if P-F interval cannot be ascertained with confidence, the whole idea of implementing predictive maintenance and on-condition monitoring of equipment becomes impractical.

The above mentioned challenges are the reason why on-condition monitoring and predictive maintenance remains viable for only 25-35% of all the failure modes of an equipment [1].

The purpose of this article is to emphasize the importance of developing understanding about the P-F Curve and P-F Interval from a practical standpoint. it may not be confused as a case against practicality of predictive and condition based maintenance.

?

References & Further Reading

[1] J. Moubray, Reliability-centered Maintenance. Industrial Press, 1997.

[2] S. Ochella, M. Shafiee, and C. Sansom, “Adopting machine learning and condition monitoring P-F curves in determining and prioritizing high-value assets for life extension,” Expert Systems with Applications, vol. 176, p. 114897, Aug. 2021.

[3] F. Besnard, K. Fischer, and L. Bertling, “Reliability-Centred Asset Maintenance — A step towards enhanced reliability, availability, and profitability of wind power plants,” in 2010 IEEE PES Innovative Smart Grid Technologies Conference Europe (ISGT Europe), Oct. 2010, pp. 1–8.

[4] A. K. S. Jardine and A. H. C. Tsang, Maintenance, Replacement, and Reliability: Theory and Applications, Second Edition, 2nd ed. Boca Raton: CRC Press, 2014.

[5] J. A. Nachlas, Reliability Engineering: Probabilistic Models and Maintenance Methods, Second Edition, 2nd ed. Boca Raton: CRC Press, 2017.