Overcoming complex background noise and poor small-target detection in industrial settings, this paper introduces YOLOv8-BBP2, an enhanced YOLOv8 model. To better extract dynamic features, the backbone integrates a BiFormer dual-level routing attention mechanism. Moreover, a learnable Bi-directional Feature Pyramid Network (BiFPN) replaces the standard module, optimizing multi-scale feature integration. A P2 detection head is also added to accurately identify tiny objects, such as early flames and thin smoke. Tested on a custom factory fire dataset, YOLOv8-BBP2 yields 95.231% precision, 94.612% recall, and 89.677% mean average precision (mAP@0.5). These metrics represent respective gains of 3.31%, 4.934%, and 7.451% over the baseline YOLOv8s. Ultimately, with an inference speed of 20 ms per frame, the proposed network ensures highly robust, real-time performance.

Aiming at the difficulties in extracting gear fault feature information and the unsatisfactory performance of commonly used machine learning classification models in terms of feature separability and stable within-distribution recognition, this paper proposes a fault diagnosis method combining variational mode decomposition (VMD) optimized by the coati optimization algorithm (COA) and a BP neural network optimized by the sparrow search algorithm (SSA). Firstly, a composite index based on permutation entropy and mutual information entropy is adopted as the fitness function, and the COA is used to optimize the key parameters of VMD to improve the decomposition accuracy of the VMD algorithm. Secondly, the intrinsic mode function (IMF) components sensitive to fault information are selected for signal reconstruction according to the time–frequency-domain comprehensive evaluation factor algorithm, and the refined composite multiscale fuzzy entropy (RCMFE) is used to extract the features of the reconstructed signal to form a state feature set. Finally, the SSA is employed to optimize the parameters of the BP model to achieve accurate diagnosis of gear working conditions and fault types. The proposed method is verified using measured gear single and compound fault data and compared with commonly used signal decomposition methods and machine learning classification models. The results show that the proposed method can effectively extract gear fault feature information and has the advantages of high accuracy, good feature separability and stable within-distribution recognition. It can achieve ideal diagnosis accuracy for both single and compound gear faults, providing a certain reference for solving such problems.