Abstract: There are many challenges in the detection of appearance defects of plastic components. Due to the variety of plastic components and various types of defects, and the instability of the detection environment of plastic components such as changes in light, angle and material reflection, it is extremely difficult to establish a comprehensive and effective defect datasets. Aiming at the industrial requirements of plastic component appearance defect detection and identification, high-quality appearance defect datasets are constructed to provide powerful data support for component appearance defect detection. The PCDD datasets is constructed using a Leica S9i stereo microscope to capture surface defect images of plastic components, including stains, scratches, pin anomalies, and other defect types, and labeled using AnyLabeling tools. DRAEM, GANomaly, RIAD, f-AnoGAN and PaDiM models are used to conduct experiments, and AUROC/mAP is selected as the model evaluation index, and the differences of AUROC/mAP of each model are compared to verify the effectiveness of PCDD datasets in abnormal defect detection tasks. The construction of PCDD datasets helps to optimize algorithm performance and promote the development of defect detection technology to improve detection accuracy and component quality.