Generative Adversarial Networks to Improve the Robustness of Visual Defect Segmentation by Semantic Networks in Manufacturing Components

Latent Diffusion Models to Enhance the Performance of Visual Defect Segmentation Networks in Steel Surface Inspection

This paper explores the use of state-of-the-art latent diffusion models, specifically stable diffusion, to generate synthetic images for improving the robustness of visual defect segmentation in manufacturing components. Given the scarcity and imbalance of real-world defect data, synthetic data generation offers a promising solution for training deep learning models. We fine-tuned stable diffusion using the LoRA technique on the NEU-seg dataset and evaluated the impact of different ratios of synthetic to real images on the training set of DeepLabV3+ and FPN segmentation models. Our results demonstrated a significant improvement in mean Intersection over Union (mIoU) when the training dataset was augmented with synthetic images. This study highlights the potential of diffusion models for enhancing the quality and diversity of training data in industrial defect detection, leading to more accurate and reliable segmentation results. The proposed approach achieved improvements of 5.95% and 6.85% in mIoU of defect segmentation on each model over the original dataset.

Image Generation and Recognition for Railway Surface Defect Detection

Railway defects can result in substantial economic and human losses. Among all defects, surface defects are the most common and prominent type, and various optical-based non-destructive testing (NDT) methods have been employed to detect them. In NDT, reliable and accurate interpretation of test data is vital for effective defect detection. Among the many sources of errors, human errors are the most unpredictable and frequent. Artificial intelligence (AI) has the potential to address this challenge; however, the lack of sufficient railway images with diverse types of defects is the major obstacle to training the AI models through supervised learning. To overcome this obstacle, this research proposes the RailGAN model, which enhances the basic CycleGAN model by introducing a pre-sampling stage for railway tracks. Two pre-sampling techniques are tested for the RailGAN model: image-filtration, and U-Net. By applying both techniques to 20 real-time railway images, it is demonstrated that U-Net produces more consistent results in image segmentation across all images and is less affected by the pixel intensity values of the railway track. Comparison of the RailGAN model with U-Net and the original CycleGAN model on real-time railway images reveals that the original CycleGAN model generates defects in the irrelevant background, while the RailGAN model produces synthetic defect patterns exclusively on the railway surface. The artificial images generated by the RailGAN model closely resemble real cracks on railway tracks and are suitable for training neural-network-based defect identification algorithms. The effectiveness of the RailGAN model can be evaluated by training a defect identification algorithm with the generated dataset and applying it to real defect images. The proposed RailGAN model has the potential to improve the accuracy of NDT for railway defects, which can ultimately lead to increased safety and reduced economic losses. The method is currently performed offline, but further study is planned to achieve real-time defect detection in the future.

Robustness optimization for rapid prototyping of functional artifacts based on visualized computing digital twins

This study presents a robustness optimization method for rapid prototyping (RP) of functional artifacts based on visualized computing digital twins (VCDT). A generalized multiobjective robustness optimization model for RP of scheme design prototype was first built, where thermal, structural, and multidisciplinary knowledge could be integrated for visualization. To implement visualized computing, the membership function of fuzzy decision-making was optimized using a genetic algorithm. Transient thermodynamic, structural statics, and flow field analyses were conducted, especially for glass fiber composite materials, which have the characteristics of high strength, corrosion resistance, temperature resistance, dimensional stability, and electrical insulation. An electrothermal experiment was performed by measuring the temperature and changes in temperature during RP. Infrared thermographs were obtained using thermal field measurements to determine the temperature distribution. A numerical analysis of a lightweight ribbed ergonomic artifact is presented to illustrate the VCDT. Moreover, manufacturability was verified based on a thermal-solid coupled finite element analysis. The physical experiment and practice proved that the proposed VCDT provided a robust design paradigm for a layered RP between the steady balance of electrothermal regulation and manufacturing efficacy under hybrid uncertainties.

Generation of Time-Series Working Patterns for Manufacturing High-Quality Products through Auxiliary Classifier Generative Adversarial Network

Product quality is a major concern in manufacturing. In the metal processing industry, low-quality products must be remanufactured, which requires additional labor, money, and time. Therefore, user-controllable variables for machines and raw material compositions are key factors for ensuring product quality. In this study, we propose a method for generating the time-series working patterns of the control variables for metal-melting induction furnaces and continuous casting machines, thus improving product quality by aiding machine operators. We used an auxiliary classifier generative adversarial network (AC-GAN) model to generate time-series working patterns of two processes depending on product type and additional material data. To check accuracy, the difference between the generated time-series data of the model and the ground truth data was calculated. Specifically, the proposed model results were compared with those of other deep learning models: multilayer perceptron (MLP), convolutional neural network (CNN), long short-term memory (LSTM), and gated recurrent unit (GRU). It was demonstrated that the proposed model outperformed the other deep learning models. Moreover, the proposed method generated different time-series data for different inputs, whereas the other deep learning models generated the same time-series data.