A broad learning model with dual-path feature encoding for robust and expedient incremental fault diagnosis

Intelligent fault diagnosis (IFD) models that rely on one-shot learning often struggle in dynamic mechanical systems because they exhibit limited knowledge stability and high computational cost under incremental learning (IL). To address the trade-off between robustness and efficiency, this paper proposes a Multi-Scale High-Resolution and Source-Informed Broad Learning Model (MHRSI-BLM). By leveraging the computational efficiency of broad learning for rapid weight correction and pseudo-inverse updating, the proposed framework achieves efficient incremental updates. In addition, the MHRSI-BLM framework adopts a dual-path feature-encoding strategy to improve robustness: a multi-scale high-resolution pathway enhances feature representation, while a complementary source pathway provides a relatively stable feature anchor for incremental updates. Experiments on two rolling-bearing datasets demonstrate that the proposed method achieves strong diagnostic accuracy with competitive computational efficiency.