It is still hard to diagnose faults when the sampling conditions are very low because the fault signatures are partly overlapped with normal operating behaviour. This is particularly true in current-based monitoring systems
with limited sensing capability. This research suggests a simple way to identify problems with bearings using three-phase stator current signals, which are measured at 10 Hz. The research is about coming up with a new way of using low-frequency features, together with a way of combining
information and making decisions based on confidence, to improve weak fault detection when there is not much data available. The trial configuration made use of a three-phase induction motor functioning under Healthy, Ball Bearing Fault, and Inner Race Fault circumstances with 21
grouped physical trials. The transformation of current signals was achieved through the implementation of Clarke representation, where data was segmented into overlapping windows. These windows were then processed using a combination of low-frequency statistical and nonlinear descriptors. Fisher score-based feature selection and Linear Discriminant Analysis (LDA) were implemented before Support Vector Machine (SVM) classification using grouped validation to minimise data leakage risk between windows from the same trial. Experimental evaluation was 100% accurate at the trial level and 97.22% accurate at the window level when
tested in groups. The results showed that direct LDA projection without PCA improved the separability of weak ball bearing faults under ultra-low sampling conditions. The proposed framework exhibited reduced computational intricacy in comparison to deep learning methodologies whilst concurrently sustaining efficacious classification outcomes for constrained datasets. These findings substantiate the notion that dependable bearing fault diagnosis can nevertheless be accomplished under 10 Hz sampling through the utilisation of discriminative low-frequency feature learning and lightweight classification strategies. However, it is still necessary to test it more widely under different conditions, such as different loads, speeds, and levels of industrial noise, before it can be used on a large scale.