Remaining useful life estimation of bearings using trend–detail decomposition and extended 1D-LBP features with a hybrid deep learning model

Purpose This study aims to develop a robust and accurate remaining useful life (RUL) prediction framework for rolling bearings by overcoming the limitations of conventional health indicator construction under noisy and nonlinear degradation conditions. To achieve this, a unified end-to-end framework integrating trend–detail decomposition, cumulative damage modeling, extended one-dimensional local binary pattern (1D-LBP) encoding and a hybrid Convolutional Neural Network–Bidirectional Long Short-Term Memory–Gated Recurrent Unit (CNN–BiLSTM–GRU) architecture is proposed. The objective is to improve degradation representation, prediction accuracy, robustness and interpretability for intelligent predictive maintenance under varying operating conditions. Design/methodology/approach Raw vibration signals are first standardized on a bearing-wise basis and decomposed into trend and detail components. An extended 1D-LBP operator and a cumulative damage indicator are then employed to extract complementary degradation information and construct a four-channel health indicator. These features are processed by a hybrid CNN–BiLSTM–GRU network within a multi-task learning framework that jointly predicts health indicators, RUL percentage, degradation rate and degradation phase. The proposed framework is validated on the PRONOSTIA benchmark using 11 bearing prediction scenarios, supplemented by ablation studies and multi-seed experiments to evaluate robustness and model stability. Findings The proposed framework consistently achieved accurate and robust RUL prediction across 11 PRONOSTIA bearing prediction scenarios, yielding average values of MAE = 0.0884, RMSE = 0.1225, RA = 0.855, and Score = 0.537. Ablation experiments and three independent random-seed evaluations demonstrated that each architectural component contributes to the overall performance, with the extended 1D-LBP encoding providing the largest improvement. The results also confirmed the framework's stability, robustness against noisy degradation signals and strong generalization capability under different operating conditions, highlighting its effectiveness for intelligent predictive maintenance. Research limitations/implications The proposed framework was validated using the PRONOSTIA benchmark under controlled laboratory conditions. Although the results demonstrate strong predictive performance and robustness, further validation on additional bearing datasets and real industrial environments is required to confirm its generalization capability. Future research will focus on domain adaptation, self-supervised learning, uncertainty-aware prognostics and multi-sensor fusion to improve adaptability under varying operating conditions and reduce dependence on labeled run-to-failure data. Practical implications The proposed framework provides a practical solution for intelligent predictive maintenance by enabling accurate and stable RUL estimation under varying operating conditions. Its robust degradation representation supports timely maintenance scheduling, reduces unexpected machine failures, minimizes maintenance costs and improves equipment availability. The framework can be readily integrated into condition monitoring systems for rotating machinery and has the potential to support Industry 4.0 and smart manufacturing applications by enhancing maintenance decision-making and operational reliability. Originality/value This study presents a unified prognostic framework that integrates trend–detail decomposition, cumulative damage modeling, 1D-LBP encoding and hybrid CNN–BiLSTM–GRU learning within a multi-task architecture. Unlike existing approaches that employ these techniques separately, the proposed framework jointly exploits complementary degradation representations to improve RUL prediction. Comprehensive ablation studies, multi-seed stability analyses and a monotonicity constraint further validate the effectiveness and robustness of the proposed methodology, providing an accurate, interpretable and reproducible solution for intelligent bearing prognostics.