If you are interested in analysing temperature measurements in SCADA for early component fault detection, there are several possible approaches. For example, you can regularly check the signals on the screen in your control centre. You can create some dashboards showing average temperatures over time, and compare each turbine with its neighbours. You can write some Python code and create a solution for comparing simulated temperature signals with the measured signals, perhaps using neural networks. Or you can buy an off-the-shelf product that solves the problem for you.
So there are plenty of options. Before making your decision, we recommend that you consider the following:
- Temperature signals obtained via 2nd-level SCADA systems are sometimes incorrectly mapped. Naming conventions used by the OEMs can be ambiguous and it is sometimes very hard to tell which signal has been measured at which position in the turbine. For example, we often see that temperatures from main bearings, gearbox bearings and generator bearings are mixed-up. If not corrected, this can lead to wrong conclusions and in the worst case, additional work for service technicians sent to inspect the wrong components. Before starting temperature monitoring, be aware of this and check signal mapping carefully.
- Wear and damage to major components such as main bearings or generator bearings often cause an increase in the temperature of the structure, which in turn can be measured through monitoring of the temperature signal. However, the sensors themselves are sometimes subject to faults which are likely to trigger alarms in the monitoring signal. Care must be taken to distinguish between component damage and sensor damage, to avoid faulty diagnosis.
- Similarly, if the nacelle cooling system in a turbine is not functioning correctly, the temperature of several components may raise due to the increased environmental temperature. This may cause multiple alarms in a monitoring system which again, need to be attributed to the correct root cause.
- Some parts of the wind turbine are cooled passively, with components such as main bearings or generator bearings typically releasing heat to their environment through conduction and convection. However, other parts are actively cooled, such as the gearbox, with the lubricant continuously pumped through a heat exchanger. This adds complexity to the accurate simulation of temperature signals and must be properly considered in any monitoring solution.
- Many monitoring systems simulate the expected temperature of components based on input variables (such as power, air temperature, rotational shaft speed) and then detect deviations of the measured temperature. For high accuracy, such models must be calibrated for each individual turbine and seasonal effects (i.e. the influence of environmental temperature on the component temperature) must be properly considered. Black-box modelling techniques typically, therefore, require datasets that cover long time periods to properly “learn” the sensitivity to such seasonal effects. However, this can be problematic since the monitoring system is then effectively “offline” following any fault, repair, change of sensor or anything influencing the validity of the model for a specific turbine. The manual effort for training models for multiple signals on large numbers of turbines can be huge.
So although there is high value in monitoring SCADA temperature signals and detecting component faults early, there are several pitfalls to watch out for. Why not give our application “i4SEE Heat” a try? We charge no setup costs, running costs are very low, onboarding is fast and painless and we can start delivering insights and recommendations to you almost immediately.