Artificial Intelligence–Based Load Classification and Imbalance Detection Using Vibration Signals in Drum-Type Washing Machines

Vibration and noise in drum-type washing machines are primarily driven by load variability and mass imbalance, which can amplify resonance response, reduce user comfort, and accelerate component wear. Reliable state recognition from vibration signals is therefore essential to enable adaptive operational strategies and safer spin-up behavior. Objective: This study aims to develop a physically grounded AI-ready framework for load classification (empty/dry/wet) and imbalance-risk detection using vibration measurements, so that operational states can be inferred and mapped into vibration-mitigation decisions. Methodology: The research used a quantitative experimental design with controlled operating conditions (empty, 2 kg dry, 4 kg wet) and two damper configurations (OEM and high-damper). Vibration responses were characterized using free-decay and FRF-based identification, producing parameters such as effective mass, natural frequency, damping ratio, stiffness, damping coefficient, and peak transmissibility. These parameters were then organized into an AI-ready label structure to support supervised and semi-supervised learning pipelines.
Findings: The results show a clear mechanical signature for load separability, with natural frequency decreasing monotonically as load increases (2.95 Hz → 2.77 Hz → 2.63 Hz). Under the same wet load, the high-damper configuration substantially increased the damping coefficient (190 → 235 N·s/m) and reduced peak transmissibility (2.00 → 1.45), indicating a strong reduction in resonance amplification and transmitted vibration. Implications: The findings support the use of vibration-based state recognition as an input to adaptive spin control, enabling conservative decision rules to minimize resonance dwell and reduce vibration transmission without requiring major suspension redesign. The framework also facilitates scalable model development when labeled data are limited by leveraging physically interpretable anchors for validation. Originality: This study contributes a novel integration of repeatable vibration identification (free-decay/FRF/spin-up) with an AI-ready state and labeling framework for load classification and imbalance-risk inference, providing an interpretable bridge between vibration physics and supervised/semi-supervised learning for engineering deployment.