The purpose of this research is to contribute to the understanding of the adhesion mechanism which takes effect at the interface of metallic coatings on ceramics substrates with cold spray. There are several reports that the adhesion mechanism of metallic coating on ceramics substrates is due to heteroepitaxial growth. On the other hand, there are reports that the oxide layer exists at the adhesion interface between the ceramic substrate and metal coatings, and which affects the adhesion. In this research, we prepared intentionally oxidized AlN substrates and Al₂O₃ substrates in order to investigate the effect of the oxide layer at the adhesion interface on the adhesion strength. Then, the pre-heating temperature of the ceramic substrates were adjusted and cold sprayed aluminum coatings. As a result of evaluating the adhesion and the interface, heteroepitaxial growth was not inhibited in the oxide layer having a thickness of about 1μm. However, as the oxidation progressed, the bonds between the ceramic particles weakened, which hindered the strong adhesion.

View full abstract

The electromagnetic characteristics of a thermal sprayed coatings were evaluated by the KEC method, which measure SE (dB) from 1 to 100kHz, and by the complex magnetic permeability. The coatings were formed on a SS400 or PEEK substrate by: (a) coating of Zn (80μm) deposited by arc spraying, (b) coating of PC-permalloy (500μm) deposited by plasma spraying on the coating of Zn (80μm) deposited by arc spraying, (c) coating of powdered Fe-Si-B amorphous sheet material deposited by plasma spraying (200μm) . The SE (dB) of the samples were (a) about 10 dB (b) about 50 dB (c) about 5 dB at 100 kHz, respectively. The results of this measurement are summarized and discussed in a simplified formula for the magnetic flux quantity Φ and the magnetoresistive loss quantity P. As a result, it is presumed that SE (dB) in (b) depends on the amount of magnetoresistive loss P and SE (dB) in (c) depends on the amount of magnetic flux Φ. (a) is a nonmagnetic highly conductive metallic material, and it was found that it is only effective for magnetic shielding above 100kHz.

View full abstract

This paper introduces machine learning approaches for process parameter optimizations of the laser powder bed fusion (L-PBF) process, which is one of the most representative additive manufacturing technologies. The first topic is an exploration of laser conditions for obtaining dense Al-12 mass%Si alloy parts using a neural network. A neural network architecture and training manner for precisely predicting optimized laser conditions using limited data are introduced. The second topic is laser parameter optimization for controlling the microstructure of L-PBF-fabricated WC/Co composites. Two characteristic microstructural regions (WC/Co two-phase region and WC decomposition region) are formed in the L-PBF-fabricated WC/Co composites. The WC decomposition region is formed by the melting of the WC particles and has a detrimental effect on the mechanical properties. Therefore, its formation should be suppressed. To quantify the amount of the WC decomposition region formed, a convolutional neural network model, in which micrographs of single-laser-scanned WC/Co composites were trained, was constructed. The support vector machine was also used for optimizing the laser parameters. These machine learning methods can make an efficient estimation of optimized laser parameters for the L-PBF process.

View full abstract

Materials data is generated every day. By analyzing these data using machine learning, we can expect to promote understanding of materials, accelerate materials development, and discover innovative materials. In this report, we introduce some examples of applications of black-box optimization methods in the field of metallic materials. First, an overview of Bayesian optimization methods for manufacturing process optimization and their application to gas atomization process is explained. The target is the yield of Ni-Co based superalloy powders. By using Bayesian optimization, better process parameters with higher yield can be found. Second, an overview of machine learning methods for phase diagrams and demonstrations for phase diagram constructions by machine learning are noted. We show that if machine learning is used, the number of experiments to create detail phase diagram is decreased as 20% against random sampling.

View full abstract

The application of artificial intelligence and machine learning technologies to engineering research and industrial technology development has been increasing in recent years. This paper gives an overview of AI and machine learning technologies and how they can be applied in terms of engineering research and industrial technology development, and introduces the authors' research results. Specifically, the paper describes the development of modelling techniques for complex input-output relationships in arc welding (welding condition-welding part property relationships) and the development of techniques for estimating boundary conditions (heat input conditions) in numerical simulations of welding heat transfer for the evaluation of welding thermal history.

View full abstract

Electromagnetic thickness gauges are commonly used for measurement of coating thicknesses in quality control. The gauges are convenient and relatively precise on smooth substrates. However, on rough substrates, measured values of coating thicknesses always have some positive errors and dispersions. We investigated the errors and dispersions of measurement on rough substrates with and without spacers, and showed histograms, scatter plots, a regression line and a prediction interval of thickness of spacers and measured values. The errors on rough substrates are correlated in some degree with their roughness, Ra. The measured values with spacers are distributed around the prediction interval of the regression line.

View full abstract