Posters

Abstract

Asset inspections are a key element of predictive maintenance. Visual condition monitoring of equipment already contributes a lot to revelation of not only cosmetic, but also structural or mechanical findings and defects. This, of course, provides valuable input for analysis or planning of reactive maintenance. Here, the main benefit is to repair defects that already impact safe and normal operation of assets and handle, in the most efficient way, those that might in a foreseeable future evolve into such. Procedures like this help lower OPEX and keep the cost of repairs lower than they would be if the defect is detected later in asset operation lifecycle or, in worse cases, prevent hazardous situations or disasters. There is, however, more to the data that inspections generate. Large and variable enough data set is necessary for generalizable patterns to be detected and verified. This detection is itself not a trivial nor cheap achievement, which is why most of our initial Machine Learning efforts have been within this process. Machine Learning techniques such as classification and clustering help to separate visually severe findings from those on cosmetic end of spectra. This have supported a great decrease in data generation/classification costs. When this data foundation is available, observed behavior can be used in simulations and predictions. This provides insight not only on the component's future through predictions, but also insights on the past. Further analysis of severe findings can help reveal failure mechanisms and, in some cases, root causes. This information can then be correlated with other operational data such as sensor readings, weather conditions, vibrations but also equipment age, lifetime, loads, etc., to increase our Machine Learning performance and optimize several related business processes. In Siemens Gamesa we lead the way to a sustained blade healthful future for the fleets in operation, therefore, while before we have relied on deterministic simulation models, now, by using blade inspection findings in connection with operational and climatology data, we create models that can predict the risks of component failures and leading edge erosion impact. With these better tools we can more accurately calculate costs associated with warranty or service programs tailored to our customers' specific needs. In future, we target to prevent certain failures, and those that we can not fully prevent, ensure the right stocking of spare parts, materials and human resources and their availability in the regions where action will be necessary. In addition to this, it helps us recommend the best time for inspections and also offer other relevant products to our customers, such as leading edge protection or LPS solutions.