This paper aims to diagnose an industrial-scale and slow-speed wind turbine blade bearing. It has been in service for 15 years in the wind farm, so the faults inside the bearing are naturally formed. Unlike most of the existing works with seeded defect bearings, this bearing provides the opportunities to demonstrate real-world blade bearing fault characteristics. To diagnose this bearing, the bearing acoustic emission (AE) signals are collected under very slow rotation speed and heavy load conditions. Then, a novel sparse representation technique, Sparse Augmented Lagrangian (SAL) algorithm, is applied to filter the raw AE signal and extract weak fault signals. The characteristic of SAL is that it can transform the original filtering problem into several sub-optimization problems, so SAL is easy to implement to produce a sparse filter for AE signal denoising. The diagnostic results indicate that our proposed framework is effective and outperforms the conventional discrete/random separation (DRS) technique.