Search

Physics-Informed Neural Networks (PINNs) have gained popularity in solving nonlinear partial differential equations (PDEs) via integrating physical laws into the training of neural networks, making them superior in many scientific and engineering applications. However, conventional PINNs still fall short in accurately approximating the solution of complex systems with strong nonlinearity, especially in long temporal domains. Besides, since PINNs are designed to approximate a specific realization of a given PDE system, they lack the necessary generalizability to efficiently adapt to new system configurations. This entails computationally expensive re-training from scratch for any new change in the system. To address these shortfalls, in this work a sequential meta-transfer (SMT) learning framework is proposed, offering a unified solution for both fast training and efficient adaptation of PINNs in highly nonlinear systems with long temporal domains. Specifically, the framework decomposes PDE's time domain into smaller time segments to create "easier" PDE problems for PINNs training. Then for each time interval, a meta-learner is assigned and trained to achieve an optimal initial state for rapid adaptation to a range of related tasks. Transfer learning principles are then leveraged across time intervals to further reduce the computational cost. Through a composites autoclave processing case study, it is shown that SMT is clearly able to enhance the adaptability of PINNs while significantly reducing computational cost, by a factor of 100. [ABSTRACT FROM AUTHOR]

This work presents a prototype of an intelligent quality inspection tool for application to fibers distortion monitoring in the fabric composites forming processes. To this end, a series of hemisphere draping tests on a typical commingled fiberglass/polypropylene twill weave were conducted at the dry form, and the defects (in-plane and out-of-plane) for each formed part were inspected via camera vision. By gathering data from around 30 samples and over 1200 images from different forming regions, different machine-learning algorithms were trained and validated. As an application scenario in a smart factory, the developed simple AI tool would be used by an operator, or a robot, to scan different areas of the formed parts and identify 'defected' (fail) versus 'non-defected' (pass) scenarios. It was found that the K-nearest neighbors and Support Vector Machine models detect the defects with an error rate of less than 5% in the present case study, regardless of the background noise in the images such as external objects, marks on samples, or blurriness. [ABSTRACT FROM AUTHOR]

Composite materials have been widely used in the aerospace industry and are critical for safe operations. However, the delamination, caused by cyclic loads and corrosive service environment, poses a serious threat to the structural integrity of composite laminates. The acoustic emission technique has been adopted to assess the structural integrity by characterizing damage location, type, and size. This article proposes and compares data-driven prognostic methods to quantify the delamination area efficiently and accurately. To address the problem of insufficient inspection data, the prediction model adopts the path length across the delamination area as the target value. The delamination area can then be estimated with the predicted path length based on formulated geometric relationships. This solution will augment the model training datasets, and consequently, avoid the overfitting problem during the training process. Experimental results on composite coupons demonstrate that the proposed ensemble learning-based model outperforms other state-of-the-art methods in terms of prediction accuracy and efficiency. [ABSTRACT FROM AUTHOR]

There remain major concerns over the increasing use and waste of materials and energy resources in multiple manufacturing sectors. To address these concerns, some manufacturers have begun to align their R&D efforts with the circular economy principles: Reduce, Reuse, Recycle and Replace (RRRR). Focusing on advanced composites manufacturing sector, this paper presents an innovative approach for process design and analysis of a new waste heat recovery system for carbon fiber manufacturing. Namely, the stabilization process is known to be one of the most critical steps in the production of carbon fibers, as it consumes the most energy, has the largest factory footprint, is a complex system composed of many components, and is the largest capital investment within the factory line. The heat recovery system in this step of the manufacturing can notably reduce energy consumption, emission, cost, and conversion time, while aiming to maintain the mechanical properties of the final product. Here, via an actual industry-scale fibre production setting, the energy consumption factors were obtained and used to model the total energy and its balance in the thermal stabilization step. Two machine learning approaches with limited data, Artificial Neural Network and Non-Linear Regression were then constructed to predict the energy consumption. Results suggested that using the recovery system by means of a heat exchanger, can yield over 62.7 kW recovery, corresponding to 64% of total exhausted energy from the entire process. The electric energy consumption was reduced from 73.3 kW to 10.2 kW, which corresponded to an 86% improvement in the total energy efficiency. The model also confirmed that, by preheating the make-up air with the recovered energy, the energy performance index of the thermal stabilization can be increased from 0.08 to 0.44, along with a reduction in the process carbon footprint by 28.5 t/y. This is especially crucial as we are turning on smart digitalisation in manufacturing inspired by industry 4.0 concept with limited data. Waste heat recovery system for stabilization process of carbon fiber manufacturing. [Display omitted] • Measurement of key factors in energy and resource flows in the carbon fiber stabilization process. • Development of two machine learning models for prediction of the stabilization energy consumption. • Reducing the energy consumption of stabilization process through a heat recovery system. • The electric energy consumption can be improved up to 86% in the total energy efficiency. • The ANN model confirmed that the energy performance index can be increased from 0.08 to 0.44. [ABSTRACT FROM AUTHOR]

In industrial composites applications, drilling is one of the most common operations and complex processes during the final assembly, which can generate undesirable damage to the manufactured part. Data collection from a given composite's fatigue life is often costly and time-consuming. To address this challenge, the current case study aims to adapt a hybrid machine learning framework to predict the fatigue life of the drilled Glass Fiber Reinforced Polymer composite laminates (with both unidirectional and woven lay-ups) under a limited and noisy data assumption. Composite specimens were drilled at various cutting speeds and feed rates. The size of the delamination around the hole was scanned by a microscopic camera. Cyclic three-point bending tests were conducted, and results indicated that the drilling-induced delamination size and the composite lay-up affect the specimens' fatigue lives. The latter were then modeled in two steps. In the first step, an offline deterministic model was established using the group method of data handling along with a singular value decomposition. Pareto multi-objective optimization was applied to prevent overfitting. In the second step, the Kalman filter was employed to update the polynomial of the deterministic model based on minimizing the mean and variance of error between the actual and modeled data. Results showed an excellent learning reliability, with a correlation coefficient of 97.6% and 96.5% in predicting the fatigue life of unidirectional and woven composite laminates, respectively. A sensitivity analysis was performed and indicated that the fatigue life of the samples has been more affected by the drilling feed rate, compared to the cutting speed. [ABSTRACT FROM AUTHOR]