A multi-objective prediction and optimisation method for additive manufacturing technology

Yang, Jimeng (2021) A multi-objective prediction and optimisation method for additive manufacturing technology. PhD thesis, University of Glasgow.

Full text available as:
Download (13MB) | Preview


Cleaner production in a sustainable and customised industrial environment has gradually become the focus of attention in industrial manufacturing. Additive manufacturing (AM) proposes a revolutionary paradigm for customised engineering design and manufacturing, attributed to its design freedom with limitless structural constraints. As an emerging manufacturing technology, most manufacturers and researchers are dedicated to the innovation of AM’s manufacturing mechanism and the improvement of part quality. However, the understanding of this emerging manufacturing technology is not yet sufficient in the resource efficiency perspective, which includes three environmental dimensions, namely production time, electrical energy consumption and material usage.

In order to improve AM’s resource efficiency, this thesis aims to provide a general modelling scheme to predict time, energy and material consumptions of the AM process, utilise meta-heuristic algorithms to optimise the process parameters of AM, and minimise the three consumptions (i.e. time, energy and material).

A hybrid data-driven and physics-based modelling method is proposed to build up the predictive models of AM’s time, energy and material consumptions. To start with, all consumption-related components of the existing AM technologies are classified into five types of module: axis movement, material processing, material feeding, component heating and auxiliary components. Then, hybrid modelling is performed on each module to obtain the relationships between the consumptions and process parameters. In physics-based modelling, the time, distance of axis movement and amount of material usage are calculated from the computer numerical control (CNC) programming language (also named G-code). In data-driven modelling, the remaining parameters are measured through experiments. A power meter is used to measure the apparent power and time of each module under different process parameters. The relationships between the measured parameters and process parameters are derived through regression analysis methods. In addition, some parameters in the predictive models are affected by the characteristics of machine and material in a practical manufacturing context. For example, the actual speed of axis movement is affected by the high loads of stepper motors during high-speed printing. To further improve the prediction accuracy, additional experiments are conducted to test the actual values of affected parameters. The nature of additional experiments is determined by the machine characteristics.

Meta-heuristics are developed to approximate the Pareto front of process parameters that consume the least time, energy and material. The predictive models are used as three objective functions to evaluate the performance of each solution of process parameters. Since the non-dominated sorting genetic algorithm (NSGA-II) has been widely used to solve optimisation problems with two or three objectives in industrial manufacturing, this study improves and applies NSGA-II to this optimisation problem. Experiments are designed to perform the optimisation under different combinations of optimisation parameters. A set of Pareto fronts is obtained. Hypervolume (HV) indicator is used to compare all obtained Pareto fronts before finally selecting the optimum solution sets of process parameters. In a practical manufacturing context, the optimisation result can provide guidance and a trend for selecting a feasible solution of process parameters.

To validate the effectiveness of the prediction and optimisation method, two case studies are conducted on two different types of fused deposition modelling (FDM) 3D printers. The predictive models of time, energy and material consumptions for each printer have been built by following the proposed prediction method. To improve the prediction accuracy, additional experiments are performed on both FDM 3D printers, including testing the actual speed of axis movement and the actual density of thermoplastic material. According to the prediction results and experimental results, the feasibility of prediction models has been proved, which achieves an acceptable prediction accuracy. The consideration of machine characteristic has also been proved to further improve the prediction accuracies.

The effectiveness of the optimisation method using NSGA-Ⅱ are also verified in two case studies. To evaluate and compare the qualities of obtained Pareto fronts, the hypervolume (HV) indicator has been used as the response of each optimisation test. The non-dominated iv solutions of the Pareto front that has the maximum HV indicator are the optimum solutions for the AM task. This result can provide guidance for setting a feasible combination of process parameters in the prefabrication stage. The optimal solutions of process parameters are compared with the default setting of process parameters. The comparison results prove that the consumptions of optimal solutions are significantly reduced. Furthermore, the significances of optimisation parameters (i.e. population size, number of generations, crossover probability and mutation probability) for the response are analysed by using the range analysis and analysis of variance (ANOVA) methods. According to the analysis results, the significances of optimisation parameters for the HV indicator are not found to be consistent in these two cases. Since the predictive models are customised, there is no general rule to recommend the setting of process parameters and optimisation parameters for general AM technologies.

The proposed prediction and optimisation methods provide a modular, customisable and flexible interface to personalise the predictive models, the optimisation objectives and the process parameter to be optimised. The method fully considers the characteristics of AM machine and material, process parameters, production environment, and customer demands. The use of manufacturing information provided by G-code significantly reduces the workload of the modelling process and achieves an acceptable prediction accuracy. Furthermore, the proposed method is unrestricted to any AM machine, task or complex structure of CAD design, and is also applicable to any other manufacturing technologies that fabricate through numerical control (NC) programming.

Item Type: Thesis (PhD)
Qualification Level: Doctoral
Subjects: T Technology > T Technology (General)
T Technology > TS Manufactures
Colleges/Schools: College of Science and Engineering > School of Engineering
Supervisor's Name: Liu, Dr. Ying and Li, Dr. Peifeng
Date of Award: 2021
Depositing User: Theses Team
Unique ID: glathesis:2021-82355
Copyright: Copyright of this thesis is held by the author.
Date Deposited: 30 Jul 2021 08:39
Last Modified: 30 Jul 2021 08:39
Thesis DOI: 10.5525/gla.thesis.82355
URI: http://theses.gla.ac.uk/id/eprint/82355
Related URLs:

Actions (login required)

View Item View Item


Downloads per month over past year