Elsevier

Microelectronics Reliability

Volumes 88–90, September 2018, Pages 1151-1156
Microelectronics Reliability

A novel method of reliability-centered process optimization for additive manufacturing

https://doi.org/10.1016/j.microrel.2018.07.149Get rights and content

Abstract

Process optimization problem of additive manufacturing nowadays is a research hotspot in the field of manufacturing industry. However, parameter uncertainty has not been considered in the past. In this paper, the process optimization methods of additive manufacturing are reviewed, and a novel reliability-centered optimization combining stochastic finite element analysis (SFEA) with particle swarm optimization (PSO) method is proposed and have been explained in details. Finally, the Direct Metal Deposition (DMD) process is used as an example in this paper, and parameters including the layer thickness, heat generation of the melt, scanning speed, as well as the hot bed temperature flux are taking into account. Deformation after the part cools down is the single optimization objective, and the reliability performance is treated as an uncertain constraint in the optimization problem. As a result, the best process parameters of DMD are obtained, and the case study verified the superiority of the proposed method.

Introduction

During the last decade, additive manufacturing (AM) has been studied very well for its great application potential in both industrial and commercial products making. Yet the uniformity of the parts and the reliability of the process are still an unsolved problem since the complex and intricate relation between the process parameter and the part performance [1].

There are several kinds of AM process, and the formation mechanism differs from each other. For instance, selective laser melting (SLM) is a process that powder layer is melt by laser at the proper place to realize connection and formation. While direct metal deposition (DMD) is another kind of AM, which fabricate parts by delivering metal on the melt pool where needs to be printed. For DMD, performance of the part can be determined by many process parameters, such as bead width, scanning strategy, layer thickness, melt temperature, scanning speed, air gap, and so on. Additionally, performance of the part can be also described in various aspects, for example, the mechanical property, geometric accuracy, time, cost, et al.

Process optimization problem of AM is regarded as a deterministic problem by most of the scholars. A typical solution is to design an experiment, and then construct the limit state function, finally predicting or optimizing leveraging the above function [2,3]. Several design of experiment (DOE) methods [4,5] have been employed, including grey Taguchi [6], central composite design (CCD) and face centred central composite design (FCCD) [6,7]. Then the experiment results (a bunch of point sets) are to be fitted using response surface method (RSM) [8], artificial neural network (ANN) [6], etc. Once the limit state function (also call the response model) is obtained, one can predict the part performance easily instead of experimenting. Furthermore, the response model can be treated as the objective function of process optimization. For this purpose, different evolution (DE) [4], quantum-behaved particle swarm optimization (QPSO) [6], analysis of variance (ANOVA) [7] and genetic algorithm (GA) [8] have been implemented and are proven to be proper method to solve the optimization problem. However, in literature the optimization has been treated as a single objective problem. Hence, the part performances such as dimension accuracy, residual stress cannot be optimized synchronously. In addition, a deterministic optimization process easily to converges on the boundary of the feasibility region. This is dangerous as the printing part or printing process may tend to fail because of parameter uncertainty.

In this paper, a novel method of reliability-centered process optimization for AM is proposed to solve the above problems. Besides, the stochastic FEA method is introduced to avoid the time-consuming and material-wasting of experimental based reliability optimization method.

Section snippets

A reliability-centered process optimization method

In order to solve the AM process reliability optimization problem, a novel method combining the SFEA with design optimization method is proposed. The SFEA is used to substitute the complicated experiment, and its goal is to calculate the part accuracy or other performance under various parameters. After that, a response model will be constructed. Finally, the design optimization model will be established, taking the parameter uncertain into account.

The calculation flow of the proposed method is

AM FEA modelling

In this paper, direct metal deposition (DMD) is selected to be a sample AM process since DMD is a common used AM method in the industry. In the present study, a thin wall is considered to be the analysis part, and the hot bed is included as it has a great impact on the heat transfer of the built part. Further, the CAD model is meshed using thermal-structural brick type element. In our proposed FEA model, temperature field is calculated first in the thermal model, and then the temperature is

Conclusion

A Reliability-Centered Process Optimization method is introduced in this paper to solve the reliability problem of Additive Manufacturing. Furthermore, the calculating flow is illustrated in details with a case study called DMD process. Several results and advices about the proposed method are summarized below:

  • i)

    The proposed reliability-centered process optimization method is proved to be effective and case study shows that a less better process parameter set will be obtained when reliability is

Acknowledgment

This work is supported by the Ministry of Science and Technology of P. R. China (National Key R&D Program of China, 2018YFB1105200) and Guangdong science and Technology Department of P. R. China (Natural Science Foundation of Guangdong Province (2018A030310019)).

References (10)

There are more references available in the full text version of this article.

Cited by (0)

View full text