This study proposes a deep learning–based predictive maintenance model for condition monitoring and remaining useful life (RUL) estimation of a 1 kW brushless DC (BLDC) motor. Multi-sensor signals, including vibration (10 kHz), current (20 kHz), and surface temperature (10 Hz), were acquired under six health conditions: normal, bearing outer race fault (BPFO), bearing inner race fault (BPFI), unbalance, misalignment, and stator insulation degradation. To jointly exploit spatial patterns and temporal degradation behaviors, a hybrid CNN–LSTM model with a multi-task learning framework was developed to perform 6-class fault classification and RUL regression simultaneously. Experimental results on the constructed BLDC motor dataset show that the proposed model achieves a classification accuracy of 95.8%, outperforming conventional SVM and 1D-CNN baselines (85.2% and 90.7%, respectively). In addition, the proposed method significantly reduces RUL prediction error, yielding an RMSE of 9.6 and an MAE of 6.8, which corresponds to approximately 39% improvement over a single LSTM-based regression model. These results demonstrate that the proposed CNN– LSTM multi-sensor fusion framework is effective for intelligent condition monitoring and predictive maintenance of BLDC motor systems, and it can be extended to a wide range of rotating machinery applications.

Abstract
1. 서 론
2. 관련 연구 및 이론적 배경
    2.1 진동 기반 고전적 상태 진단 기법
    2.2 머신러닝 기반 상태 진단 기법
    2.3 딥러닝 기반 예지정비(PHM) 기법
    2.4 BLDC모터의 예지정비 필요성
3. 실험구성
    3.1 시스템 및 센서구성
    3.2 고장 시나리오 구성
    3.3 데이터 전처리
4. 예지진단 모델 구조
    4.1 CNN-LSTM 하이브리드 구조
    4.2 다중 태스크 학습(MTL) 구조
5. 실험결과 및 분석
    5.1 상태 분류 성능
    5.2 RUL 예측 성능
    5.3 Ablation Study(제거 연구)
    5.4 실험결과 분석 및 고찰
6. 결 론
References

• 박창규(Dept. of Mechanical & Automotive Engineering, Gwangju University) | Chang-Kyu Park Corresponding author