Wind farms generate thousands of heterogeneous alarms daily, whereas maintenance resources are limited. This paper converts multi-vendor fault records into plan-date–turbine–system candidates and ranks them using event intensity, persistence, stop indicators, within-turbine competition and time-decayed expert history. A reviewed dictionary maps 645 vendor–name–code records to 13 standard systems. In a chronological July replay, 278,812 events produced 9,079 candidates, including 84 positives, 8,953 negatives and 42 uncertain records. Eight expert tasks lacked matching candidate keys; excluding them increased the candidate-pool ceiling from 91.30% to 100.00% without changing the ranking data. With July 1–23 for training and July 24–31 for testing, the Top-15 list hit 15 of 24 positive tasks, yielding Recall@15=62.50%, Overlap@15=12.50% and NDCG@15=0.3788. Candidate observability and ranking quality must therefore be evaluated separately.

wind farm,fault alarms,maintenance planning,expert experience,candidate tasks