Project Overview
Why ReneWind needs failure prediction, what each prediction outcome costs, and how the solution is sequenced.
ReneWind builds and maintains wind turbine generators. A generator that fails in the field forces an unplanned replacement — the most expensive outcome in the entire maintenance chain.
Sensors across the gearbox, tower, blades and brake systems already stream data, plus environmental factors such as temperature, humidity and wind speed. That data is ciphered for confidentiality and arrives as 40 anonymous predictors, V1–V40, with a binary failure Target.
Because replacement >> repair > inspection, the model must minimise false negatives first.
Solution approach
Shape, dtypes, duplicates, missing values, target balance
Univariate, bivariate and correlation analysis of all 40 predictors
Median imputation and scaling fit on training data only
Single hidden layer with SGD optimizer as reference point
10 configurations: depth, optimizer, dropout, class weights, oversampling
Best model retrained on full train set, scored once on Test.csv
Click Generate to produce an evidence-based observation, insight, recommendation, and business benefit for this section.