Improved Quality‐Loss for Degradation‐State Assessment of Planetary Sun Gear via Fault Recognition

Condition‐monitoring methods commonly provide discrete fault labels, whereas maintenance assessment requires a quantitative representation of damage severity. This paper proposes a quality‐loss‐based degradation‐state assessment framework integrating cross‐speed fault recognition with severity‐weighted loss mapping for planetary sun gears to improve cross‐speed fault identification and provide severity‐aware quantitative support for degradation‐state evaluation and maintenance decision‐making. Firstly, multi‐domain features are extracted from the frequency, energy, and order domains to characterize vibration signals from the public WT‐Planetary gearbox dataset, and four machine‐learning classifiers are evaluated using a leave‐one‐speed‐out protocol to assess generalization under unseen rotational speeds. Secondly, the recognized health state is mapped to an ordinal damage index and then converted into a dimensionless relative quality‐loss index using a quadratic loss function. The proposed loss index penalizes misclassification according to the severity difference between the true and predicted states. Next, the best‐performing configuration, Boosted Trees with enhanced features, provides the most accurate and balanced cross‐speed fault recognition and severity‐weighted quality‐loss estimation among the evaluated models. Finally, increased assessment errors are observed at boundary speed conditions, particularly for Healthy and Wear states. The framework converts discrete fault‐recognition outputs into a severity‐aware quantitative indicator for degradation‐state assessment and maintenance‐oriented condition evaluation.