Kyoji Homma

Estimation of acoustic emission source waveform of fracture using a neural network

The applicability of a neural network to acoustic emission (AE) is presented. It is shown that the shape of the simulated source waveform using piezoelectric ceramics is steplike, similar to that of mode I crack extension, and its rise-time can be varied by the resonance frequency in the thickness direction. The results imply that the simulated source can provide learning waveforms for the network. Actual AE waveforms were also acquired by conducting a tensile test of a chevron-notched graphite specimen. It was demonstrated that the appropriate source waveform associated with mode I crack extension was successfully determined by the network taught with simulated waveforms.

Development of an implanted bone-conduction hearing aid using giant magnetostrictive material

To circumvent some of the disadvantages of conventional hearing aids such as sound distortion, feedback, and cosmetic factors, implantable hearing devices have been developed. However, these hearing devices also have problems such as insufficient output at high frequencies and inflammation. In this study, a new subcutaneously implanted bone-conduction hearing aid with an external unit and an internal unit is proposed. The external unit consists of a microphone, a speech processor, and a transmitting coil, which send the sound signals and energy to the internal unit by generating a magnetic field. The internal unit consists of a receiving coil, a driving coil, and a vibrator made of giant magnetostrictive material (GMM), which deforms its body by changing the magnetic field. The internal unit is surgically embedded in the temporal bone under the skin and vibrates the skull when the magnetic flux is applied by the external unit. For the first stage in the development of the new bone-conduction hearing aid, a prototype was made and its fundamental properties were examined. The GMM vibrator has a good linear response and high output especially at high frequencies. In contrast, the output at low frequencies is relatively lower than that at high frequencies. To enhance the output of the implanted bone-conduction hearing aid at low frequencies is an issue in the future.

Acoustic Analysis of Adhesive Surface in Granular Inhomogeneous Structural Material using Neural Network

Difficulty in quality assessment of adhesive interface between granular inhomogeneous materials and a disk plate has been pointed out in nondestructive ultrasonic inspection because of the effect of multiple-reflection at the contact point of each granular particle. A simulation technique is presented in this paper for estimating the adhesion quality of a CBN grinding wheel at the interface between the granular material and the disk plate. The grinding wheel was transposed to one dimensional serial model consisted of each segment of particle with random acoustic impedances at the interface. The propagation of incident ultrasonic waves in the grinding wheel was calculated and echoes from adhesive interface were analyzed. The domestic echo waveform from the adhesive interface was specified and learned by a neural network corresponding adhesion quality and the characteristic of the echo waveform. The effect of the grinding wheel structure to the accuracy of presumption was investigated by using the network. It was revealed that the neural network is effective to assess the adhesive quality of such inhomogeneous structural materials.

Application of Neural Network to Ultrasonic Flaw Detection in Layered Media.

Classification of the ultrasonic wave signals emitted from defects such as delamination and inclusions in layered media has been attempted using the neural network technique. In providing both reflected waveform from defects to the input and information of defect type, as well as location to the output, the relationships between the input and the output were learned beforehand by the neural network. In the system, the network learned in terms of detected waveform, however, could not infer correct information of defects for a waveform of a different phase. Therefore, appropriate parameters sampled from the detected waveform were added to the network. Locally connected networks restricting the connection between input layer and output one were proven in a highly precise estimation. It was revealed that the locally connected network is an effective technique to estimate the location of each defect, types of defect, and the amount of defects, within the range of the application of the model.