Load-dependent fault signature evolution in three-level NPC inverter-fed induction motor drives via physics-based characterisation

Reliable detection of faults in inverter-fed induction motor drives depends on a precise understanding of how diagnostic signatures evolve across the full operating range. Most published fault characterisation studies, however, are conducted at a single operating point and do not formally distinguish between features whose diagnostic value holds across the load range and those confined to specific load conditions. This paper addresses that gap through a systematic, physics-based characterisation of fault signatures in a three-level neutral-point-clamped (NPC) inverter-fed squirrel-cage induction motor drive under closed-loop voltage-to-frequency (V/f) control. Seven fault categories—broken rotor bar (BRB), air gap eccentricity (AGE), bearing degradation (BD), DC voltage imbalance (DCI), DC undervoltage (DCU), DC overvoltage (DCO), and mechanical overload (MO)—are injected at three severity levels through physics-motivated parameter perturbation in MATLAB/Simulink across thirteen discrete load levels spanning 10%–110% of rated torque, yielding 288 observations (46 healthy-baseline runs and 242 faulty-condition runs, with 31 cases excluded due to mechanical instability at high load under severe fault conditions). Nine signal-derived mechanical, electrical, and statistical features are extracted and their evolution with severity and load is characterised and physically explained within the studied simulation framework. A load-invariance classification reveals that mean rotor speed satisfies the load-invariance criterion (coefficient of variation <2%) across all fault types within the studied conditions, while RMS stator current, total harmonic distortion, and skewness are strongly load-dependent. The DC voltage imbalance model produces the largest feature deviations, with skewness reaching $$-0.53$$ (a sign reversal from the healthy value of $$+0.03$$ ) and THD increasing by 92.4% of the healthy mean under severe conditions. The mechanical overload model exhibits near-zero feature sensitivity across all severity levels within this framework, reflecting that it displaces the operating point along the healthy characteristic curve rather than modifying any circuit or mechanical parameter. Reference signature tables and a feature sensitivity analysis are provided as quantitative benchmarks for diagnostic threshold design and feature selection in three-level NPC motor drives operating under scalar V/f control.