Search

Machine learning is a promising method to model production processes and predict product quality. It is challenging to accurately model complex systems due to data scarcity, as mass customisation leads to various high-variety low-volume products. This study conceptualised knowledge accumulation, extraction, and transfer (KAET) to exploit the knowledge embedded in similar entities to address data scarcity. A sequential cross-product KAET (SeqTrans) is proposed to conduct KAET, integrating data preparation and preprocessing, feature selection (FS), feature learning (FL), and transfer learning (TL). The FS and FL modules conduct knowledge extraction and help address various practical challenges such as changing operating conditions and unbalanced datasets. In this paper, sequential TL is introduced to production modelling to conduct knowledge transfer among multiple entities. The first case study of auxetic material performance prediction demonstrates the effectiveness of sequential TL. Compared with conventional TL, sequential TL can achieve the same test mean square errors with 300 fewer training examples when facing data scarcity. In the second case study, balancing anomaly detection models were constructed for two gas turbines in the same series using real-world production data. With SeqTrans, the F1-score of the anomaly detection model of the data-poor engine was improved from 0.769 to 0.909. [ABSTRACT FROM AUTHOR]

Additive manufacturing (AM) techniques are maturing and penetrating every aspect of the industry. With more and more design, process, structure, and property data collected, machine learning (ML) models are found to be useful to analyze the patterns in the data. The quality of datasets and the handling methods are important to the performance of these ML models. This work reviews recent publications on the topic, focusing on the data types along with the data handling methods and the implemented ML algorithms. The examples of ML applications in AM are then categorized based on the lifecycle stages, and research focuses. In terms of data management, the existing public database and data management methods are introduced. Finally, the limitations of the current data processing methods are discussed and suggestions on perspectives are given. [ABSTRACT FROM AUTHOR]

The progress of machine learning (ML) in the past years has opened up new opportunities to the field of gas turbine (GT) modelling. However, successful implementation of ML algorithms remains challenging, particularly for complex problems such as multi-mode faults. An important tool for enabling applications are the feature selection and feature learning (FSFL) techniques. In particular, FL techniques have recently facilitated and improved the applicability of ML to GT modelling. This review paper conducts a review on 46 studies that utilised FSFL for GT modelling with ML. The purpose of this review is to investigate how FSFL techniques can help address GT modelling challenges and when researchers should deploy them. Therefore, the theories behind the techniques are illustrated in depth along with practical application examples from the analysed literature. The advantages and limitations of FSFL are discussed, the computational costs of different techniques are compared, and trends in the field are highlighted. Consequently, a novel categorisation framework for FSFL techniques and recommendations regarding when and how to implement them are provided. A new knowledge accumulation, extraction, and transfer concept is proposed to address GT modelling challenges. [ABSTRACT FROM AUTHOR]

Auxetic metamaterials have been applied in many domains due to their unique auxetic behavior, tunable local kinematics, and morphological intelligence. However, the classic auxetic designs normally cannot be directly applied due to their defects. How to modify the existing classic design to achieve diverse but also valid variant designs remains a challenging problem. One common approach is to use predefined parameters and fine-tune these parameters. Yet, this approach limits the design space and highly relies on expert knowledge. In this paper, we proposed a variant design generation method based on a path-finding algorithm to achieve pixel-level design freedom. Meanwhile, to support the evaluation of designs in such a large design space, we developed a multi-scale geometry-informed Graph U-net. More specifically, we proposed MeshPool—a predefined clustering-based pooling technique, and distortion loss—a geometry-informed loss function inspired by finite element analysis (FEA). A variant design family for re-entrant structures is generated as a case study to validate the variant design generation method. And the Graph U-net is developed to predict complex nonlinear deformation with simple linear FEA results. The results from the case study demonstrated that the proposed variant design generation method covered the properties space of re-entrant structures and the developed Graph U-net achieved accurate prediction with a nodal displacement mean absolute error (MAE) of around 0.0169 mm under an average nodal displacement of 0.5058 mm. More importantly, we demonstrated the graph neural network's unique capability of learning gradient information compared with convolutional neural network in a benchmark on strain and stress. • Path-finding algorithm generates pixel-level auxetic designs for three classic types. • Algorithm enhances diversity, reduces duplicates, improves training data fairness. • Multi-scale GNN predicts auxetic deformation using geometry-informed pooling. • Geometry-informed loss term enforces physical constraints efficiently. [ABSTRACT FROM AUTHOR]

The existing mechanical metamaterials are often designed with periodic inter-connected structs with simple cylindrical or uniform hierarchical geometries, which relies on their parent materials to either have a good mechanical performance with low recoverability, or significantly sacrifices their mechanical performances to be highly recoverable. Biological fibrous structures are often evolved with a composition of different fibrous morphologies to possess a desired balance of mechanical performances and recovery. In this study, we developed digital design algorithms to generate the next-generation metamaterials with composite bio-inspired twisting fibrotic structs that are rubber-like recoverable without significant scarification of their mechanical performances. A machine learning predictive model is trained based on experimental data to reveal the resulted specific energy absorption (SEA) and SEA recoveries for such metamaterials with complicated fiber-composition mechanisms. To further understand the fundamental structural recovery mechanisms of the natural fibers, we derived the elastoplastic theories of the twisting fibrotic structs, and revealed that such structs possesses a rubber-like fracture strain with significantly improved specific energy absorption. Our studies combined the structural recovery mechanisms of the composite natural fibrous structures and mechanical metamaterials, liberates the design potential of materials with engineerable optimal balances of their mechanical performances and recoverability. [Display omitted] • Designed the metamaterials with natural-fiber-inspired (NFI) structs for simultaneous high specific energy absorption (SEA) and SEA recoveries. • Designed, and successfully 3D printed metamaterials with composite NFI morphologies with digital design algorithms. • Trained machine learning aided prediction model of the SEA and SEA recoveries for NFI metamaterials with variations of fibrous morphologies. • Developed NFI metamaterials with programmable rubber-like recoverability without reducing their SEAs through a data-driven approach. [ABSTRACT FROM AUTHOR]