Forum: EXAKT

Not irrelevant, but somewhat distracting from the issue of more immediate concern to the designer of a CBM program. Namely, the reliable detection of "P".

1. From NAVAIR 00-25-403 the CBM Interval I= PF/n where n would depend on:
   1. Probability, θ, of detecting a potential failure with one occurrence of the proposed CBM task, assuming the potential failure has occurred, and
   2. Acceptable probability of failure (Pacc).
2. We focus too much on the PF Interval. Neither Nowlan and Heap, nor Moubray intended that it be used for more than determining a rough first approximation of a CBM inspection interval in the context of a continuous improvement process.
3. Focusing on the PF Interval detracts from the bigger challenge, that of declaring P. (After which confirming tests can provide more information on the actual rate of degradation.)
4. If we want to discuss it, the maintenance "Lead Time" is the time required for maintenance to respond to a P alert. Then invoking classical CBM theory as depicted by the following graph:
   
5. In the article it says: "In the worst case, according to the graph, if an inspection predates the potential failure by only a small amount, the subsequent inspection will still catch it in time, provided that the maintenance organization is capable of acting within the net P-F interval."
6. From the graph, the Net P-F will always be greater than PF/2 (when I = PF/2). If n =3 then the Net P-F will be greater than 2PF/3, and so on. Thus we must choose an "n" such that the Net P-F is greater than the Lead Time. The bigger the value of n used, the smaller the inspection interval, and the more costly will be the CBM policy.

The P-F interval, while a simple model to explain CBM, is a sort of Catch 22. Discovering the P-F interval requires the repeated experience of having detected and reacted to P. However to detect P we need the experience of having applied and confirmed our P-F model. The EXAKT/LRCM process gets around this circularity with a model that derives from actual work order data representing instances of failure modes. The model, when applied day-to-day, not only provides practical CBM decisions but also tracks and improves their predictive confidence.
Murray

So how do you propose that the CBM inspection interval be determined?

I think Moubray had the right idea. Get the right people together and arrive at a consensus for the PF interval. Divide that number by two and you have your first stab at a CBM inspection interval. That's about it. I don't see much value, in most cases, added by applying further formula (for example, the NAVAIR n = ln(Pacc)/ln(1-θ))*

But the important thing is not to stop there. Thereafter apply a natural living RCM (LRCM) process to improve, if necessary, that first approximation. This is not as difficult a task as it might seem. In fact, it will usually reduce, rather than add to the overhead of the maintenance information process. More importantly LRCM will build the RCM knowledge base and link it intimately with the Work order system. This approach facilitates analysis and subsequent improvement. The analysis methodology is described in a previous post. The LRCM forum has information on, for example, continuous improvement metrics.

*Pacc and θ are defined in first reply above.