Reliability is an essential criterion for gas turbine and generator equipment for power generation systems. With a variety of power generation assets operating under differing market strategies, a unit’s reliability is characterized by successful starts, staying online for the course of dispatch, and delivering the performance that was anticipated when bidding into the electricity market. Regardless of the market strategy, maintenance, repair, and overhaul (MRO) constitutes a critical cycle in the lifetime of any power generation gas turbine.

MRO activities have historically been conducted according to a routine schedule of outages. These outages are typically scheduled based on generic measures of the unit’s remaining useful life, typically estimated from runtime and number of starts. The costly downtime of these outages is minimized through careful planning of contractors, repair activities, and availability of a prescribed set of replacement parts. Unlike planned outages, a forced outage occurs due to an unanticipated failure of reliability of one or more parts. This leaves the plant with an ad-hoc exercise in contractor sourcing, parts procurement, repair procedures, unexpected costs, and missed market opportunities from unplanned downtime. Planned outages are without a doubt favorable to unplanned outages. However, both types of outages have disadvantages: planned outages replace parts that may have remaining useful life, and forced outages have obvious real costs including opportunity cost implications.

Condition-based monitoring offers a promising opportunity that avoids many of the downsides of traditional planned outages and possibility for forced outages. The basic idea is to monitor the health of each component real-time, and to strategically plan outages to repair or replace only parts that are approaching end of useful life. These predictive technologies can also forecast performance to help time outages appropriately. This strategy has recently seen rapid development enabled by advances in computing. This is evidenced by a growing literature on digital twins, industrial internet of things, and edge analytics, all of which are technologies that promote condition-based monitoring.

Despite its growing popularity, condition-based monitoring for MRO faces challenges for certain critical components. Successful implementation of this strategy requires sufficient gas turbine design knowledge and instrumentation to infer the condition of a large number of parts: rotor blades, turbine nozzle guide vanes, turbine buckets, combustor and transition piece components, etc. This can be especially challenging in the harsh environment of the combustion system, which can introduce rapid wear to compromised parts and which precludes many reliable means of instrumentation.

Due to operational complexities, the combustion system has had decades of monitoring focus despite challenges in data interpretation. This is due to the fact that modern, low emissions gas turbines for power generation are instrumented with combustion dynamics monitoring systems, which simply monitor the acoustics of each combustor. These monitoring systems exist as a protective measure against thermoacoustic oscillations, or combustion dynamics, which are a common challenge in modern low NOx combustor strategies. The intended use of these systems is to provide real-time feedback of acoustic amplitudes in the combustion system to avoid damage from acoustic oscillations. This information is fed to a human tuner or an autotuning system, which can tune the combustion system to simultaneously maintain acoustic amplitudes at acceptable levels while maintaining compliant pollutant emissions. However, the availability of acoustic data sampled at a high rate from each combustor provides a valuable data stream that can be re-purposed for health monitoring. While most plants monitor these data as a means to prevent damage and tune the unit, some savvy plants have utilized these data to predictively identify combustor component issues.

The MRO opportunity provided by combustion dynamics data stems from the high sensitivity of combustor acoustic signatures to combustion system component health. The degradation of combustion system components produces measurable changes to combustor acoustics. These changes might not be the high amplitudes that combustion dynamics monitoring systems traditionally monitor for and protect against. Instead, these changes may be subtle drifts in frequency or amplitude that signify a change in the underlying hardware. For example, a combustor whose dynamics amplitudes trend lower might be ignored due to reduced threat of damage; however, the low amplitude trend may be symptomatic of a developing hardware fault.

To change combustion dynamics perspective from protective to predictive can be a challenging adoption. Nevertheless, several plants have successfully demonstrated the sensitivity that lies within combustion dynamics data. Success stories include the early detection of resonator cracks, fuel-air premixing system erosion, and even inlet bleed heat hose failures  [1]. In all cases, early detection provided an opportunity for a brief borescope inspection to confirm the failure and planning of repairs, rather than discovering the failure resulting in a forced outage.

Predictive monitoring from combustion dynamics data requires advanced anomaly detection. This anomaly detection benefits from subject area expertise. For example, a well-informed anomaly detection code knows how to rule out changes due to ambient conditions and operational changes. A well-informed anomaly detection code can also benefit from combustor-to-combustor comparisons on a given engine, and can benefit from knowledge of common instrumentation fault signatures. An advanced code may also include a “fault matrix,” consisting of a database of experience relating anomaly signatures to specific hardware faults. All together, such a code can find anomalous dynamics signatures, rule out instrumentation issues or operational anomalies, and point the team to the most probable components to borescope and repair.

The re-purposing of combustion dynamics monitoring data as an automated fault detection tool has enjoyed several success stories [2], avoiding forced outages and informing planned outages.  These limited success stories embody the concept of condition-based monitoring, and they are born from necessity within the engine’s most difficult to monitor components. Today, the industry wrestles with its strategies around whole-engine condition-based maintenance and predictive health monitoring. Meanwhile, the silent digital health monitoring success stories from the acoustics of the combustion system serve to validate the condition-based monitoring paradigm and raise confidence from a nearly impossible to instrument environment.

References