The production process of industrial manufacturing is a complex process with multiple factors coupling together, any abnormality during the production process can lead to the production of defective products. In order to detect surface defect in industrial production accurately, an Attention-Enhanced Diffusion Model, AEDM is proposed in this paper. The AEDM is built upon the DiffusionDet, and two innovative modules are designed to improve the model. Firstly, a Multi-scale Attention Enhanced (MAE) block is designed in the shallow layer of the encoder to capture local and global features simultaneously. Secondly, an Attention-enhanced Parallel Bidirection Mamba (APBM) block-and-layer is designed in the deep layer of the encoder to focus on key information related to defects, and strengthen the feature representation ability of the model. Due to the lightweight design scheme of these two blocks, little extra computation is increased to the model. Extensive experiments on a public dataset (NEU-DET) and our self-collected dataset (LDD) demonstrate the superiority of our proposed AEDM over the state-of-the-art models, the mean Average Precision (mAP50) achieves 82.54% and 99.48% respectively. Specially, the performance on small defect, such as the oil category is improved by 5.36%.