Most wind turbine and solar plants use sophisticated SCADA systems that generate Alarms and Events based on pre-set conditions. While these Alarms are essential for on-site operations and maintenance teams, their application can often be compared to a post-facto (or in our case post-mortem) analysis. We have seen a number of instances when the Alarm system has been turned off, due to large number of alarms that the system generates. In many cases O&M teams use the Alarm only to identify failure of a key component. Other information which can be indicative of impending failure is often ignored due to the design of the alarm system. Let us understand the alarm system of a modern turbine and understand the drivers and use case for predictive analytics in such systems.
The alarm & event system of a turbine logs the following types of information
a. General information – Wind speed has crossed max speed limit or GHI (Global Horizontal Irradiation) level has crossed minimum threshold.
b. State change updates – Turbine or inverter state changes like start, stop, reset, emergency and so on.
c. Warning information – Component wise information change to a warning level like worn out brake pad, low oil pressure, panel temperature in warning zone and so on.
d. Critical information – Turbine shut down, gearbox temperature limit reached, icing of blades, grid failure and so on.
One common point to note with regards to these limits is their static nature. A warning level is pre fixed based on parameters for example, tower acceleration above 400 units is considered warning while above 1000 units is emergency shutdown. Take an example where wind speed is as low as 5 m/s and the vibration level is 350 units, an experienced operator will tell you that something is wrong, but the alarm system will not even indicate a warning. So the alarm and event system is driven by extreme events and does not correlate with the operating environment. In summary, the alarms and events module of a SCADA system is not intelligent.
|Failure symptoms sequence curve|
The design of SCADA systems provides a lot of room for multi-parameter analysis which can prevent failure and predict impending problems. This kind of multi-parameter analysis is exactly what Algo Engines is built for. We have a range of predictive data mining models that can help you identify anomaly, classify abnormal behaviour and assess the deviation. Here are a few examples
a. Predictive model based validation – Compare actual value of vibration or temperature with a model that provides standard vibration or temperature based on historic operation under similar conditions (wind speed or GHI under similar ambient temperature)
b. Cluster model – We identify movements out of the standard cluster for any parameter like the vibration or temperature
Implementation of appropriate models, back testing and tuning are needed before we can use the model to make intelligent decisions. At Algo Engines, we leverage our knowledge of wind turbines and solar plants to ensure our models deliver measurable results. With expertise across linear & logistic regression, clustering, artificial neural networks, classification and regression trees, support vector machines and an array of such machine learning models, we make predictive models to help increase share of renewable energy in the grid.