Condition Based Maintenance

Condition-based maintenance (CBM), and later Reliability-Centered Maintenance (RCM), are ideas that sound irresistibly rational. Measure the condition, detect degradation, intervene only when needed, increase uptime, extend asset life. Conceptually I like the CBM principle: it is clean, elegant, and in some environments it genuinely works. Where I struggle is with the way it is sold to power plants.

CBM was developed by the US Air Force in the ‘50s, and RCM followed in the ‘70s, again for aviation. Over the years, CBM has been packaged as a quasi-scientific certainty, supported by glossy business cases and increasingly baroque algorithms, all promising fewer outages, lower O&M costs, and a future where failures politely announce themselves well in advance.

Although the Space Shuttle tragedies are widely quoted as a profound failure of how CBM assumptions were applied, these approaches work well in some sectors: nuclear, oil  gas rotating equipment, semiconductor manufacturing, and aviation itself.

(Thermal) power plants are neither as well-funded as the Air Force nor they operate in laboratory-like environments such as nuclear reactors. They are, without disrespect, complex, constrained systems operating under commercial pressure, with imperfect instrumentation, limited data quantity, intensive human intervention, and maintenance windows dictated more by dispatch requirements than by sensor confidence intervals.

In this environment, the signal-to-noise ratio is brutal. Most condition data is ambiguous, and interpretation still depends heavily on experienced engineers making judgment calls that no algorithm can fully replace where the business cases assumed that better data automatically leads to better decisions.

CBM works best where failure modes are slow, observable, and decoupled from operations, where there is a fleet of similar machines operating in comparable environments. Solar PV plants, work in this way,  with rows of panels and inverters generating comparable signals and large datasets. Yet most CBM investment cases are still written for thermal assets (a 2 GW CCGT plant is definitely sexier than a 100MW solar).

There are many presentations built around biased business cases, proudly showing how, thanks to CBM, a pump was changed just in time and millions of dollars were saved. What I question is the way these cases are constructed, and I am even more skeptical of the algorithmic complexity layered on top to give them authority. Many critical power plant failures are operationally induced, accelerated by cycling, load changes, fuel quality, ambient conditions, and human behavior.

So yes, condition-based maintenance can be useful, as a tool within a broader decision framework grounded in operational reality. What it is not is a universal upgrade that magically turns O&M into a science experiment with guaranteed returns.