The interest in research and development for additive manufacturing (AM) processes has grown significantly over the last years and attracts both industry and academia alike. Among the available AM technologies, stereolithography (SLA) is one of the most discussed, researched, and employed. On the other hand, being based on thermoset resins, all the limitations of this typology of materials still apply, limiting the range of applications of this highly versatile process. To overcome these limitations, especially brittleness, this research analyzes the effects of Tungsten (W) micro-size (average size 1 μm) particles reinforcement on a methacrylate base material. First, the manufacturing process for creating the W-reinforced methacrylate material is presented and investigated to define the effect of pre- and post-processing operations on the quality of the pre-cured solution considering 4% and 10% wt. W particles concentrations. Afterward, tensile, compressive, and impact specimens were manufactured with both concentrations and compared with the experimental results from clear (unfilled) resin-based specimens used as the benchmark. The addition of tungsten particles showed a strong improvement in the impact strength of the methacrylate base material, quantified in 28% for the 4% and 55% for the 10% wt., respectively, although at the expense of a slight reduction in elastic and yield properties on average −12%. Furthermore, using Scanning Electron Microscope (SEM) analyses, the particle–matrix interaction was investigated, showing the interaction between the polymer matrix and the reinforcement and the mechanism by which the impact resistance is enhanced.

Fiber reinforcement orientation in thermoplastic injection-molded components is both a strength as well as a weak point of this largely employed manufacturing process. Optimizing the fiber orientation distribution (FOD) considering the shape of the part and the applied loading conditions allows for enhancing the mechanical performances of the produced parts. Henceforth, this research proposes an algorithm to identify the best injection gate (IG) location/s starting from a 3D model and a user-defined load case. The procedure is composed of a first Visual Basic Architecture (VBA) code that automatically sets and runs Finite Volume Method (FVM) simulations to find the correlation between the fiber orientation tensor (FOT) and the IG locations considering single and multiple gates combinations up to three points. A second VBA code elaborates the results and builds a dataset considering the user-defined loading and constraint conditions, allowing the assignment of a score to each IG solution. Three geometrical components of increasing complexity were considered for a total of 1080 FVM simulations and a total computational time of ~390 h. The search for the best IG location has been further expanded by training a Machine Learning (ML) model based on the Gradient Boosting (GB) algorithm. The training database (DB) is based on FVM simulations and was expanded until a satisfactory prediction accuracy higher than 90% was achieved. The enhancement of the local FOD on the critical regions of three components was verified and showed an average improvement of 26.9% in the stiffness granted by a high directionality of the fibers along the load path. Finite element method (FEM) simulations and laboratory experiments on an industrial pump housing, injection-molded with a polyamide-66 reinforced with 30% of short glass fibers (PA66-30GF) material were also carried out to validate the FVM-FEM simulation frame and showed a 16.4% local stiffness improvement in comparison to the currently employed IG solution.

This research presents an automatic algorithm, implemented in a Visual Basic Architecture (VBA), for the optimization of the gate location in thermoplastic injection molding of short fibers reinforced composite materials. The algorithm receives, as input, the geometry of the component and, according to the user’s choice, defines the injection points grid, and relevant versors, on a pre-constructed mesh, automatically runs the finite volume method (FVM) simulation and exports the fiber orientation tensor (FOT) on each node of the mesh. The nodal coordinate of the part and the relevant FOT are then used as the training dataset for a Gradient Boosting (GB) algorithm for the full correlation between injection gate locations and the resulting fiber orientation distribution (FOD), allowing to define the injection gate configuration better suited to maximize the effectiveness of the reinforcement fibers. By coupling the trained GB algorithm with a finite element method (FEM) simulation it was confirmed that the developed algorithm can predict the influence of the gate location on the FOD and the resulting mechanical performances, improving the stiffness between 3.8% and 32.6%, on simple and complex geometries alike.

It is undeniable that artificial intelligence (AI) and machine learning (ML) have become rooted at every level of our society. It is also true that engineers have always strived for improvement in des...

Force estimation in metal forming processes is crucial for identifying the correct machine during production and determining the best process parameters to achieve good quality, reduce scrap and minimize energy consumption. In the hot Radial-Axial Ring Rolling (RARR) process force estimation is normally carried out by highly time-consuming finite element analysis (FEA) or empiric-analytical models. To achieve a reliable and yet real-time estimation of the process loads, this research presents a freeware software based on a Python script, to estimate mandrel and axial roll forces based on the initial/final geometry of the ring, its material and temperature, process time and tools motion laws’ initial and final values. The Mandrel Force (FM) and Axial roll Force (FA) forces are predicted from 10% to 100% of the process time, excluding the calibration phase. The prediction algorithm is based on the Extreme Gradient Boosting (XGB) ensemble algorithm, trained and validated with FEA and experimental data. A total of 50 cases belonging to rings with the final outer diameter ranging between 429 and 946 mm were gathered and standardized, and data augmentation was carried out to improve the robustness of the model. Training and validation sets are randomly constructed, employing a 5-fold cross-validation technique to ensure uniformity of the database and reliability of the results. During the validation phase, the prediction of the mandrel force and axial roll force showed an average error of 8.2% and 9.9% respectively. A simple Graphic User Interface (GUI) has been developed to easily utilize the algorithm to predict the mandrel and axial roll forces over the completion percentage. The proposed methodology aims to answer the call for agile, user-friendly and reliable analysis tools to be employed directly at manufacturing plants, limiting the utilization of FEA simulations to highly complex scenarios, and thus enhancing global efficiency.

This contribution presents a proof-of-concept and benchmark analysis between two different approaches for multiple domain modeling, employing numerical and experimental results relevant to the Radial-Axial Ring Rolling (RARR). First, a finite element method (FEM) simulation data-based Deep Neural Network (DNN) was employed as base model for a subsequent transfer learning (TL), carried out by employing 40 experimental (EXP) data belonging to a different RARR domain. The DNN-TL model is benchmarked against the proposed solution, in terms of a Proxy-Physics Informed Neural Network (P-PINN), defined using proxy equations for the outer diameter (OD) expansion, the radial force (RF), and the axial force (AF) and trained on 30 FEM and 30 EXP data. The proxy equations for the P-PINN are based on known knowledge and analytical formulations developed for the RARR. The results show that employing the whole 218 cases strong FEM database for the training of the base DNN model results in high prediction accuracy on the FEM cases and can be adapted well through TL. When employing only 30 FEM cases for the training of the base DNN model results in slight improvements for RF and in a significant drop of performance for AF, after TL. Instead, in a limited data scenario, the P-PINN model, powered equally by data and proxy equations steering the learning process, is capable of modeling well and simultaneously both FEM and EXP cases with reasonable accuracy, averaging at 3.7% for OD and 24.0% for RF/AF.