<p>Accurate Remaining Useful Life (RUL) estimation is essential for industrial operational safety and cost-efficient maintenance. Traditional approaches, such as physics-based, statistical, and machine learning methods, often rely on strong prior assumptions and extensive feature engineering, which significantly increase modeling and computational costs and hinder large-scale online deployment.&#xa0;Meanwhile, the growing availability of industrial sensor data and high-performance computing resources has substantially facilitated the training of scalable deep learning models. However, under evolving environments and unstable operational loads, existing methods inadequately model complex intra-series variations and inter-series dependencies throughout the degradation process. To overcome this, we propose DS-IID, a novel Dual-Stream Framework to model Intra-series and Inter-series Dynamics through two complementary branches. Technically, we split the raw data into odd and even series, forming “two streams” to reduce redundancy before feature extraction and thus lower computational cost. One stream is designed to capture the degradation trend and the other aims to capture unstable fluctuations. Additionally, the model generates variable-wise weights that evolve over time at each time step to model inter-series dependencies and adapt to collaborative degradation across components through selective fusion, avoiding dense cross-variable interactions. Empirical evidence on a benchmark of standard evaluation settings shows that DS-IID is superior to state-of-the-art methods, indicating its effectiveness in preventing unexpected failures and enhancing system reliability. Moreover, further efficiency analysis suggests that the proposed framework has favorable computational feasibility, highlighting its practical value for industrial deployment and applications.</p>