Combined Cycles Prognosis Strategies, a Reflection

My Experience

This paper explores the practical implementation of prognostic strategies for combined cycle power generation systems [1]. I have experience performing development testing on fully instrumented diesel engines on a test stand. Although this is not necessarily a combined cycle power generation system, the number of instruments and level of system monitoring was very precise. Thousands of parameters made up the exported data sets that we used for our development work and I wrote scripts that would help engineers monitor measurement equipment and facilities equipment health. This paper would have provided me with better strategies to monitor our systems performance in a way that was normalized. I was surprised by the lack of prognostic strategies in place and because of this errors and even complete failure of equipment could have been avoided and were seen in the data after the fact. Power generation plants that cannot afford these kinds of failures would be wise to implement some of the strategies discussed in this paper.

Purpose and Expectations

The purpose of this paper is clear in that it not only provides a method for monitoring a combined cycle power generator but also provides real world examples of this method succeeding. Many of these existing power systems already are required to monitor their systems performance for regulatory reasons, so it is a win-win for everyone to implement a way to monitor the health and performance of the system overtime. I would expect that the ideas presented in this article are embraced by plants that are operating with any kind of power system. Its amazing to me that there are not more people concerned about the safety and health of their power systems, as most of the problems that are detrimental to the system can be detected well ahead of time, or even forecasted using the methods discussed in this paper.

Reflections

The paper cites many other works done by other researchers in the field of prognostics that are complex and utilize strategies that involve fuzzy logic and neural networks to detect anomalies in the systems behavior. I can appreciate the simplicity of the approaches presented in this paper, because they provide straight forward and intuitive solutions that can be readily employed by engineers monitoring these systems. The state space approach is an interesting concept that I would have never considered. It leverages the historical system data to create a normalized view of the systems parameters without the need for modeling any of the systems physical dynamics. Outliers are easily identified and can be inspected upon discovery, potentially avoiding catastrophic failure. The paper takes this further and looks at the individual signals measured and the expected signals from the plant. These signals are then represented as a regression curve that is surrounded by tuples that contain within them 95% of the expected measurements from that signal. Once the data has been transformed into this form, it is necessary to filter out operating points that are not of interest since the system would measure different values there. Once this is done it can be seen that outliers are easily identifiable outside of the tuples, and it is even possible to forecast the anticipated future states of the system using statistical forecasts. All of this is achieved with only a few simple mathematical transformations, providing a relatively computationally efficient approach over other methods presented in the beginning of the paper. The case study for the HRSG provides the motivation for why it is important to monitor system critical parameters with a prognostics strategy such as the one presented in order to avoid potential failures or unintended damages to the system. The graphs shown in figure 5 show the success and drawbacks of this approach. The drawbacks being that outliers make the phase planes regression curve inaccurate and could lead to less accurate forecasts about the systems future errors. These errors are important to catch in order to fix broken sensors and maintain monitoring of the systems performance. In order to obtain a perfect trend of the anticipated future performance of the system the correct physical signals have to be accounted for in the creation of the phase space graph. Otherwise unmonitored issues could arise and perhaps impact your system in a way that you may not have expected.

Next Steps Moving Forward

This article builds a strong case for using simple algorithms to perform system prognostics. It is proven to be successful in a case study and should be embraced by other power generation systems such as diesel engines. It may be important to know how the major parameters are performing over time, but why not extend this to the minor components. For example, a fuel pump or major facilities system that must be in operation in order for your system to operate. More often than not from my experience it has been unanticipated facilities failures that have led to down time and ultimately loss of revenue because the monitoring of these systems simply was not adequate. This approach requires no complex or fancy method for implementing a straightforward way of ensuring a healthy system. Moving forward this research could be combined with some of the more complex neural network methods as a way of providing training data for the network to learn the system within the tuples. The trends developed here are the best case ground truth data for a neural network to develop its weights and offer better determination of sensor failures or shutdown anomalies from the measured data. Currently it appears that the only way to monitor the trajectory of the system is by visually inspecting the graph, which I agree is important however it could be improved by having a system that alerts when violations occur in the space.

Figure 1. Elements involved in performing the prognosis of the system.

Conclusion

I never knew how useful the phase plane and regression forecasting methods were until they were applied in this article. I have gained a new perspective with respect to prognostics and how easily they can be implemented using this straightforward approach. Table 2 does a great job of explaining the advantages and disadvantages of the commonly used approaches for prognostic strategies and to me the clear winner is the phase space trajectory approach. Maybe a combination of the neural network strategy and the phase space trajectory could offer a even better solution that may provide even more insight into the systems and facilities performance. This research is beneficial to everyone who enjoys their reliable source of electricity and should be embraced by more power generation systems in the future.

References

[1] Joshua Finn, John Wagner, Hany Bassily, Monitoring strategies for a combined cycle electric power generator, Applied Energy, Volume 87, Issue 8, 2010, Pages 2621-2627, ISSN 0306-2619, https://doi.org/10.1016/j.apenergy.2010.02.017.