Jia Li Janessa Thong

Bio: Jia Li Janessa Thong is an academic researcher from Nanyang Technological University. The author has contributed to research in topics: Computer science & Imputation (statistics). The author has co-authored 1 publications.

Data imputation strategies for process optimization of laser powder bed fusion of Ti6Al4V using machine learning

Abstract: A database linking process parameters and material properties for additive manufacturing enables the performance of the material to be determined based on the process parameters, which are useful in the design and fabrication stage of a product. The data, however, are often incomplete as each individual research work focused on certain process parameters and material properties due to the wide range of variables available. Imputation of missing data is thus required to complete the material library. In this work, we attempt to collate the data of Ti6Al4V, a popular alloy used in aerospace and biomedical industries, fabricated using powder bed fusion, or commonly known as selective laser melting (SLM). Various imputation techniques of missing data of the SLM Ti6Al4V dataset, such as the k-nearest neighbor (kNN), multivariate imputation by chained equations, and graph imputation neural network (GINN) are investigated in this article. It was observed that kNN performed better in imputing variables related to process parameters, whereas GINN performed better in variables related to material properties. To further improve the quality of imputation, a strategy to use the median of the imputed values obtained from the three models has resulted in significant improvement in terms of the relative mean square error. Self-organizing map was used to visualize the relationship among the process parameters and the material properties.

Effects of carbon fibre reinforcement on the geometric properties of PETG-based filament using FFF additive manufacturing

Abstract: The increasing worldwide demand for high-quality on-demand products manufactured with flexible and efficient productions systems has led to the development additive manufacturing technologies (AM). One of the most popular AM technologies, is fused filament fabrication (FFF) due to its ability to manufacture complex parts using a broad range of thermoplastic polymers with low production costs. However, FFF still cannot compete with traditional manufacturing processes when it comes to producing high quality end-use products. To improve mechanical properties and geometric quality features of end-use products, researchers are developing new advanced filaments infused with nanoparticles, short and continuous fibres. In the search for enhanced materials, glycol-modified polyethylene terephthalate (PETG) filaments and PETG reinforced with carbon fibres (PETG-CF) have been developed for FFF, but the effects of the addition of these fibres on geometric properties have not been analysed. The main objective of this study is to evaluate the impact of fibre addition to PETG filaments on the geometric properties of FFF manufactured parts by analysing their dimensional accuracy, and surface roughness. The effect of the 3D printing parameter − speed, layer thickness, and build orientation – on the geometric behaviour were assessed. Artificial neural network based predictive models of geometric parameters of the two candidate PETG-based filaments were used to find optimal printing parameters. In general terms, the carbon fibre addition to PETG-based polymers negatively affected the dimensional accuracy, flatness, and surface roughness in most of the printing conditions, significantly reducing the printing parameter combinations where the optimal values were achieved.

Anomaly Detection in Fused Filament Fabrication Using Machine Learning.

Abstract: Fused filament fabrication (FFF) has been widely used in various industries, and the adoption of technology is growing significantly. However, the FFF process has several disadvantages like inconsistent part quality and print repeatability. The occurrence of manufacturing-induced defects often leads to these shortcomings. This study aims to develop and implement an on-site monitoring system, which consists of a camera attached to the print head and the laptop that processes the video feed, for the extrusion-based 3D printers incorporating computer vision and object detection models to detect defects and make corrections in real-time. Image data from two classes of defects were collected to train the model. Various YOLO architectures were evaluated to study the ability to detect and classify printing anomalies such as under-extrusion and over-extrusion. Four of the trained models, YOLOv3 and YOLOv4 with "Tiny" variation, achieved a mean average precision score of >80% using the AP50 metric. Subsequently, two of the models (YOLOv3-Tiny 100 and 300 epochs) were optimized using Open Neural Network Exchange (ONNX) model conversion and ONNX Runtime to improve the inference speed. A classification accuracy rate of 89.8% and an inference speed of 70 frames per second were obtained. Before implementing the on-site monitoring system, a correction algorithm was developed to perform simple corrective actions based on defect classification. The G-codes of the corrective actions were sent to the printers during the printing process. This implementation successfully demonstrated real-time monitoring and autonomous correction during the FFF 3D printing process. This implementation will pave the way for an on-site monitoring and correction system through closed-loop feedback from other additive manufacturing (AM) processes.

A data-driven machine learning approach for the 3D printing process optimisation

Abstract: ABSTRACT 3D printing has become highly applicable in modern life recently. The industry has brought a facelift to most others. However, this technology still exists some shortcomings, and it therefore has not been generalised to bring the best benefits to users. In this paper, based on multilayer perceptron and convolution neural network models, we propose a new data-driven machine learning platform for predicting optimised parameters of the 3D printing process from a model design to a complete product. This finding can open up great advances in the current 3D printing technology. Accordingly, the results obtained allow us to predict quickly and accurately some decisive parameters of the traditional 3D printing process such as time, weight and length while the input was fuzzy with a part of the initial information missing. The proposed approach does not need to account for the shape, size and material of the printed object, but it can perform the process automatically without other extra factors. After completing the model, a configurator is proposed to set the parameters for the respective printer types, which makes the 3D printing process simple and fast.

Deep-Learning-Based Microfluidic Droplet Classification for Multijet Monitoring.

Abstract: Inkjet printing, the deposition of microfluidic droplets on a specified area, has gained increasing attention from both academia and industry for its versatility and scalability for mass production. Inkjet printing productivity depends on the number of nozzles used in a multijet process. However, droplet jetting conditions can vary for each nozzle due to multiple factors, such as the surface wetting condition of the nozzle, properties of the ink, and variances in the manufacturing of the nozzle head. For these reasons, droplet jetting conditions must be continuously monitored and evaluated by skillful engineers. The present study presents a deep-learning-based method to identify the droplet jetting status of a single-jet printing process. A convolutional neural network (CNN)-based on the MobileNetV2 model was employed with optimized hyperparameters to classify the inkjet frames containing images captured with a CCD camera. By accumulating the classified class data in order by frame time, the jetting conditions could be evaluated with high accuracy. The method was also successfully demonstrated with a multijet process, with a test time of less than a second per image.