High-Performance Defect Detection Methods for Real-Time Monitoring of Ceramic Additive Manufacturing Process Based on Small-Scale Datasets

Xinjian Jia,Shan Li,Tongcai Wang,Bingshan Liu,Congcong Cui,Wei Li,Gong Wang,Jia,Li,Wang,Liu,Cui,Li,Wang

DOI: https://doi.org/10.3390/pr12040633

IF: 3.5

2024-03-23

Processes

Abstract:Vat photopolymerization is renowned for its high flexibility, efficiency, and precision in ceramic additive manufacturing. However, due to the impact of random defects during the recoating process, ensuring the yield of finished products is challenging. At present, the industry mainly relies on manual visual inspection to detect defects; this is an inefficient method. To address this limitation, this paper presents a method for ceramic vat photopolymerization defect detection based on a deep learning framework. The framework innovatively adopts a dual-branch object detection approach, where one branch utilizes a fully convolution network to extract the features from fused images and the other branch employs a differential Siamese network to extract the differential information between two consecutive layer images. Through the design of the dual branches, the decoupling of image feature layers and image spatial attention weights is achieved, thereby alleviating the impact of a few abnormal points on training results and playing a crucial role in stabilizing the training process, which is suitable for training on small-scale datasets. Comparative experiments are implemented and the results show that using a Resnet50 backbone for feature extraction and a HED network for the differential Siamese network module yields the best detection performance, with an obtained F1 score of 0.89. Additionally, as a single-stage defect object detector, the model achieves a detection frame rate of 54.01 frames per second, which meets the real-time detection requirements. By monitoring the recoating process in real-time, the manufacturing fluency of industrial equipment can be effectively enhanced, contributing to the improvement of the yield of ceramic additive manufacturing products.

engineering, chemical

What problem does this paper attempt to address?

The problem that this paper attempts to solve is that in the ceramic additive manufacturing process, due to the influence of random defects during the coating process, it is difficult to ensure the yield rate. Currently, the industry mainly relies on manual visual inspection to detect defects, which is inefficient and highly dependent on personal experience. To address this limitation, this paper proposes a defect detection method for ceramic vat photopolymerization based on the deep - learning framework. This method adopts a two - branch object detection approach. One branch uses a fully convolutional network to extract features from the fused image, and the other branch uses a differential Siamese network to extract difference information from two consecutive - layer images. By designing a two - branch structure, the decoupling of the image feature layer and the image spatial attention weight is achieved, thereby reducing the influence of a few abnormal points on the training results and playing a crucial role in training on a small - scale data set and stabilizing the training process.

Specifically, the main contributions of the paper include:
1. Proposing a two - branch object detection model based on multi - image fusion characteristics. In the first branch, two - layer coating images are input into the differential Siamese network to extract image differences; two - layer coating images and one - slice image are fused into a three - channel image and input into another branch, thereby achieving multi - feature extraction.
2. By adopting the method of multi - prediction - head parallel prediction, each prediction head is responsible for predicting defect targets of different scales, and then the duplicate results are filtered through the non - maximum suppression principle, which significantly improves the efficiency of the framework.
3. Based on the detection results and 3D reconstruction technology, the close - range quality control of the DLP manufacturing process is achieved, which can be used to guide the subsequent processes.

Through these methods, the paper aims to improve the defect detection accuracy and real - time monitoring ability in the ceramic additive manufacturing process, thereby increasing the product yield rate and the intelligence level of manufacturing equipment.