As SOFC commercialization progresses and mass production is realized, quality control during production will become extremely important. Currently, quality control is based on visual inspection and testing of single cells and stacks under operating conditions. This method, although reliable, is time-consuming, costly, and unsustainable.Therefore, the authors have been developing evaluation methods with the aim of developing a system that enables rapid, safe, nondestructive, automatic, and accurate quality assurance assessment. One of them is developing evaluation methods for internal microdefects using THz waves. Previously, the authors have been conducting nondestructive evaluation during SOFC operation using the acoustic emission (AE) method [1-5]; the AE method detects the sound (elastic waves) generated when cells, stacks, etc. are deformed or mechanically damaged, and is suitable for in-situ observation during operation. In this situation, we have also succeeded in classifying deformation and various damages of cells and stacks by machine learning of the obtained signals [6,7,8]. However, there are two main disadvantages of using SOFC as a quality assurance inspection method: First, it is passive. In other words, it cannot detect deformation or damage that has already occurred. The second is that it is dependent on the measurement environment, since AE signals are not physical quantities and therefore change depending on the measurement environment. In other words, the waveform shape, amplitude, and frequency components of the signals may change significantly by attaching a sensor to the waveguide. Therefore, signals that have been successfully classified may need to be readjusted or re-verified if the measurement conditions change.On the other hand, signals obtained from THz waves are often highly reproducible physical quantities. In addition, it is rapid and nondestructive because of the contactless and pulsed wave. However, THz waves have a long wavelength, which results in low resolution in multilayer thin-film structures. If it becomes possible to detect small defects in the layers and visualize the internal structure by utilizing machine learning, it will surely be a very effective evaluation method for SOFCs.In this paper, we report on the clustering and recent results of machine learning using fluctuations of THz reflected waves.AcknowledgementsThis research was conducted as part of the New Energy and Industrial Technology Development Organization (NEDO) (JPNP20003). In addition, it was supported by Grant-in-Aid for Scientific Research (22K18865) and (22H01353).[1] K. Sato et al et al., Journal of Testing and Evaluation 34, JTE12707 (2006).DOI: 10.1520/JTE12707[2] K. Sato et al., Journal of Power Sources 195, 5481-5486 (2010).DOI : 10.1016/j.jpowsour.2010.03.077[3] K. Sato et al.,ECS Trans., 7 (1) 455 (2007).DOI: 10.1149/1.2729123[4] K. Kumada et al.,ECS Trans., 91 (1), 825 (2019).DOI: 10.1149/09101.0825ecst[5] T. Tsuchikura et al.,ECS Trans., 91 (1), 579 (2019).DOI: 10.1149/09101.0579ecst[6] D Inaba et al., Transactions of the Japanese Society for Artificial Intelligence 27 (3), 121-132 (2012).DOI: 10.1527/tjsai.27.121[7] D. Inaba et al., Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), 7301 LNAI(PART 2), 49–60(2012).DOI: 10.1007/978-3-642-30220-6_5[8] K. Fukui et al.,ECS Trans., 57(1), 571–580 (2013).DOI: 10.1149/05701.0571ecst
Read full abstract