I have long been a supporter of Weibull engineering, RAM modelling and Crow-Amsaa as reliability methods. None of these are methods on their own and are generally useful as part of a larger exercise. For example, Weibull engineering is a fantastic supplement to RCA when you have the information to do so.
And there in lies the rub...
Just about every single time I get into methods like this it ends up being an exercise dedicated to finding the data rather than performing the analysis.
The end result is often probabilistic methods based on subjective engineering judgements, or on substitute data sets.
And both fo these are fraught with potential problems... but the killer issue is that once the results are being produced people tend to overlook the flimsy nature of the underlying information.
Meanwhile the end result ends up being a ball of assumptions overlying other assumptions.
Not a good way to manage a modern asst management department...
But when you DO have the data, there is absolutely no doubt that methods such as those above are far and away the best means of generating high confidence decisions over the physical assets. (In particular decisions of renewals and replacements)
So how can you get the data? Resnikov clearly pointed out that having a lot of failure data is a fundamental problem, because it means the maintenance department has already failed. before you can have failure data you need failures.
And today we are more dependent on assets than at any other time in our history. So failures can mean the Gulf of Mexico, Buncefield, The Houston Refinery Explosion and so on. IN short, failure cots production, money and potentially lives and environmental damage as well.
A strategy built upon crashing a few more assets is no longer possible either ethically, or legally in may countries.
Several years ago, almost a decade actually, I was at the forefront of some work in this area within the utility sectors of the United Kingdom.
The nature of their regulated market drove them to frequently make high confidence decisions related to their physical asset base. Decisions they were in no way able to make due to the continual issue of the lack of underlying data.
They were also very acutely aware of the dangerous path of obtaining data. Some companies continued to throw themselves into the deep end, in lieu of other methods, while others were quitely trying to build the failure data, and relying heavily on engineering judgement.
That was when we started to help clients using the Whole-of-Life thinking. The whole concept of Lifecycle costing had been with us for many years. I recall when we used to call it terotechnology.
But I had never been a fan of it. For me it represented the deterministic whole of life models. Or the place where most people started from. Unsure about their asset tactics, no clue where their initial CAPEX maintenance forecasts had come from, and no real understanding of their corrective maintenance forecasts.
At the other end of the scale was the probabilistic whole of life model. The one we were all aiming for, where you could regularly and easily produce high confidence CAPEX maintenance forecasts, determine Net present Costs, and forecast performance for many years ahead.
And between the two was nothing... which was when I started to promote the Proactive Whole of Life Model.
The core concept was around data management, understanding that it was only possible through application of two methods simultaneously. That was, a) codifying knowledge into data via a structured approach such a RCM, and b) proactive capture of failure data.
The results have been fantastic. Companies get their baseline strategies reviewed and revitalised, using the RCM approaches that I have also been at the forefront of for the past ten years. And int he process, particularly for companies with a large asset base, they captured the failure and performance data required in the course of every day activities.
Within very short period of time these companies, who already had pretty strong OPEX and CAPEX tactics from the implementation of RCM, now had the failure data at hand to delve even further into the root causes of certain failure modes via Weibull, or forecast CAPEX spending for five years via RAM modelling.