Harnessing deep learning for physics-informed prediction of composite strength with microstructural uncertainties

Zhou, Kai; Sun, Haotian; Enos, Ryan; Zhang, Dianyun; Tang, Jiong

doi:10.1016/j.commatsci.2021.110663

Cited by 15 publications

(5 citation statements)

References 41 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…-Implementing deep learning for material research in the context of mechanical components involves several key steps, ranging from data preparation and model development to validation and deployment. [7] Below is an outline of the implementation process:…”

Section: Implementation Of Deep Learning For Materials Research For M...mentioning

confidence: 99%

Investigating the use of Deep Learning, in Materials Research for Predicting Material Properties, Identifying new Materials, and Optimizing Material Selection for Mechanical Components

Mohan Raparthi

2024

dxjb

View full text Add to dashboard Cite

The rapid advancements in deep learning techniques have spurred a paradigm shift in materials research, revolutionizing the way we predict material properties, identify novel materials, and optimize material selection for mechanical components. This paper explores the integration of deep learning methodologies into materials science, presenting a comprehensive investigation into their efficacy and potential applications. The paper explores the development of deep learning models for predicting material properties.[1] Leveraging vast datasets containing information on diverse materials and their corresponding properties, we delve into the application of neural networks to establish robust predictive models. By extracting complex relationships within the data, deep learning facilitates the accurate estimation of material characteristics, enabling researchers and engineers to streamline the materials discovery process. In addition to property prediction, the study explores the role of deep learning in the identification of new materials with superior or tailored attributes. By training models on extensive databases encompassing known materials and their functionalities, we investigate the ability of deep learning algorithms to suggest novel materials with specific desired properties. This capability holds immense promise for accelerating the discovery of innovative materials, especially in fields where tailored material performance is critical. Furthermore, the paper examines the utilization of deep learning in optimizing material selection for mechanical components. By considering a holistic approach that factors in mechanical, thermal, and other relevant properties, we explore how neural networks can assist in selecting the most suitable materials for specific applications. This not only enhances the efficiency of the design process but also contributes to the development of more durable, efficient, and sustainable mechanical components. Through a systematic exploration of the integration of deep learning in materials research, this paper provides valuable insights into the transformative potential of these techniques. The findings contribute to the ongoing discourse on the intersection of artificial intelligence and materials science, paving the way for accelerated advancements in materials discovery, design, and application.

show abstract

Section: Implementation Of Deep Learning For Materials Research For M...mentioning

confidence: 99%

Investigating the use of Deep Learning, in Materials Research for Predicting Material Properties, Identifying new Materials, and Optimizing Material Selection for Mechanical Components

Mohan Raparthi

2024

dxjb

View full text Add to dashboard Cite

show abstract

“…38 , Simonyan and Zisserman have shown increased efficiency with deeper networks where a small kernel size ( ) coupled with delayed pooling operation is used. The CNN architectures with this idea, known as VGG CNN, have been extensively used in different domains, including some micro-structural applications 18 , 19 , 24 . The advantage of using a smaller kernel size with increased depth (or more layers) over a big one is to reduce the number of training parameters and probably enhance learning capability as the non-linear activation function is applied more times through the depth.…”

Section: Model Developmentmentioning

confidence: 99%

“…Innovative architectures of the first stage have led to efficient CNN models like AlexNet, VGG, and ResNet. Among these, the VGG model has been adopted widely in many micro-mechanical models 18 , 19 , 24 , either directly by transfer learning or using its principle of stacking convolutional layers with delayed pooling operations. For example, Li et al 19 used pruned VGG-16 model for learning and reconstructing micro-structure features wherein high-level layers or those away from the input layer are removed to reduce the computational cost.…”

Section: Introductionmentioning

confidence: 99%

A generalised deep learning-based surrogate model for homogenisation utilising material property encoding and physics-based bounds

Nakka

Harursampath

Ponnusami

2023

Sci Rep

View full text Add to dashboard Cite

The use of surrogate models based on Convolutional Neural Networks (CNN) is increasing significantly in microstructure analysis and property predictions. One of the shortcomings of the existing models is their limitation in feeding the material information. In this context, a simple method is developed for encoding material properties into the microstructure image so that the model learns material information in addition to the structure-property relationship. These ideas are demonstrated by developing a CNN model that can be used for fibre-reinforced composite materials with a ratio of elastic moduli of the fibre to the matrix between 5 and 250 and fibre volume fractions between 25 and 75%, which span end-to-end practical range. The learning convergence curves, with mean absolute percentage error as the metric of interest, are used to find the optimal number of training samples and demonstrate the model performance. The generality of the trained model is showcased through its predictions on completely unseen microstructures whose samples are drawn from the extrapolated domain of the fibre volume fractions and elastic moduli contrasts. Also, in order to make the predictions physically admissible, models are trained by enforcing Hashin–Shtrikman bounds which led to enhanced model performance in the extrapolated domain.

show abstract

“…Unless otherwise specified, the results throughout the manuscript are analyzed through the AlexNet network with transfer learning hyperparameters shown in option 1. The Adam optimizer built upon the stochastic gradient descent (SGD) algorithm is adopted to direct the backpropagation training (Zhou et al, 2021). The epoch and batch sizes are set as 10 and 5, respectively, which are proven to enable the adequate training through observing the training and validation loss trends with respect to epoch.…”

Section: Fault Diagnosis Implementation On Crwu Bearing Fault Datamentioning

confidence: 99%

A hierarchical deep learning framework for combined rolling bearing fault localization and identification with data fusion

Liang

Zhou

2022

Journal of Vibration and Control

Self Cite

View full text Add to dashboard Cite

Fault diagnosis of rolling bearings becomes an important research subject, where the data-driven deep learning-based techniques have been extensively exploited. While the state-of-the-art research has shown the substantial progresses in bearing fault diagnosis, they mostly were implemented upon the hypothesis that the location of bearing prone to failure already is known. Nevertheless, in actual practice many rolling bearings are installed in a complex machinery system, any of which is likely subject to fault. As such, fault diagnosis essentially is a process to achieve both fault localization and identification, which results in many fault scenarios to be handled. This will significantly degrade the fault diagnosis performance using conventional deep learning analysis. In this research, we aim to develop a new deep learning framework to address abovementioned challenge. We particularly design a hierarchical deep learning framework consisting of multiple sequentially deployed deep learning models built upon the transfer learning. This can improve the learning adequacy for a high-dimensional problem with many fault scenarios involved even under limited dataset, thereby enhancing the fault diagnosis performance. Without the prior knowledge regarding the fault location, this methodology is greatly favored by the sensor/data fusion which takes full advantage of the enriched pivot fault-related features in the measurements acquired from different accelerometers. Systematic case studies using the publicly accessible experimental rolling bearing dataset are carried out to validate this new methodology.

show abstract

Harnessing deep learning for physics-informed prediction of composite strength with microstructural uncertainties

Cited by 15 publications

References 41 publications

Investigating the use of Deep Learning, in Materials Research for Predicting Material Properties, Identifying new Materials, and Optimizing Material Selection for Mechanical Components

Investigating the use of Deep Learning, in Materials Research for Predicting Material Properties, Identifying new Materials, and Optimizing Material Selection for Mechanical Components

A generalised deep learning-based surrogate model for homogenisation utilising material property encoding and physics-based bounds

A hierarchical deep learning framework for combined rolling bearing fault localization and identification with data fusion

Contact Info

Product

Resources

About