著者

竹田 憲生 成瀬 友博 河野 賢哉 服部 敏雄

出版者

公益社団法人 日本材料学会

雑誌

材料 (ISSN:05145163)

巻号頁・発行日

vol.52, no.2, pp.204-209, 2003-02-15 (Released:2009-06-03)

参考文献数

9

被引用文献数

4 4 2

---

Silicone gel is usually applied to electrical automotive devices to protect them from corrosion. However, under a vibration environment, the silicone gel vibrates bonding wires in the devices, thus, to evaluate the reliability of the devices, the vibration analysis of the gel/wire structure is indispensable. In this study, we clarify the relation between the fatigue life of gel-protected bonding wires and the geometry of the gel and bonding wires experimentally. It was founded that the diameter of wires and the thickness of the gel have a significant influence on fatigue life. Then, we developed a method, based on a vibration analysis model that takes into account the visco-elasticity of a gel, for predicting the fatigue life of the wires. It was confirmed that the predicted fatigue life showed good agreement with the measured fatigue life. Finally, we developed a design tool for easily calculating the fatigue life of the wires. This tool estimates the strain range by using a response surface, i. e., a neural network. As Bayesian regularization was executed in learning of unknown parameters in the neural network, we could make the response surface and ensure good generalization ability.

著者

竹田 憲生 亀山 達也

出版者

一般社団法人 日本機械学会

雑誌

日本機械学会論文集 (ISSN:21879761)

巻号頁・発行日

vol.88, no.910, pp.22-00095, 2022 (Released:2022-06-25)

参考文献数

14

---

A practical structural health monitoring has been proposed for evaluating the structural health of a whole mechanical asset by using digital twin with data collected during the operation of the asset. Digital twin can be utilized to predict the remaining useful life by estimating the variation of the physical quantity that dominates the life, even though any records of failure do no exist. However, a mechanical asset includes huge number of local hot spots where structural health should be evaluated, and accordingly, huge man-hours are required to integrate a monitoring system that evaluates the health at all the hot spots by using digital twin. A hierarchical structural health monitoring has been therefore developed to overcome this drawback. In the first stage of the health monitoring, the overview of the mechanical damage of the components included in a asset is evaluated according to a metric, D factor, that defines the cumulative damage of the components, and the assets having relatively large damage are extracted. The extracted assets are then evaluated in detail in the second stage; that is, structural health is checked at the local hot spots. The monitoring system that employs digital twin and the hierarchical health monitoring was applied to the health management of wind turbines. As the result of evaluating the structural health of the main components of wind turbines, about a hundred wind turbines can be prioritized according to the D factor. In this first stage, a surrogate model based on a machine learning was utilized for evaluating the overview of the damage with low computational cost; the approximation error of the D factor was less than 3 % by using the surrogate model. It can be therefore concluded that this practical structural health monitoring is useful for the decision making of fleet health management of mechanical assets.

著者

竹田 憲生 亀山 達也

出版者

一般社団法人 日本機械学会

雑誌

日本機械学会論文集 (ISSN:21879761)

巻号頁・発行日

pp.22-00095, (Released:2022-06-07)

参考文献数

14

---

A practical structural health monitoring has been proposed for evaluating the structural health of a whole mechanical asset by using digital twin with data collected during the operation of the asset. Digital twin can be utilized to predict the remaining useful life by estimating the variation of the physical quantity that dominates the life, even though any records of failure do no exist. However, a mechanical asset includes huge number of local hot spots where structural health should be evaluated, and accordingly, huge man-hours are required to integrate a monitoring system that evaluates the health at all the hot spots by using digital twin. A hierarchical structural health monitoring has been therefore developed to overcome this drawback. In the first stage of the health monitoring, the overview of the mechanical damage of the components included in a asset is evaluated according to a metric, D factor, that defines the cumulative damage of the components, and the assets having relatively large damage are extracted. The extracted assets are then evaluated in detail in the second stage; that is, structural health is checked at the local hot spots. The monitoring system that employs digital twin and the hierarchical health monitoring was applied to the health management of wind turbines. As the result of evaluating the structural health of the main components of wind turbines, about a hundred wind turbines can be prioritized according to the D factor. In this first stage, a surrogate model based on a machine learning was utilized for evaluating the overview of the damage with low computational cost; the approximation error of the D factor was less than 3 % by using the surrogate model. It can be therefore concluded that this practical structural health monitoring is useful for the decision making of fleet health management of mechanical assets.

著者

竹田 憲生

出版者

一般社団法人日本機械学会

雑誌

日本機械学会論文集. A編 = Transactions of the Japan Society of Mechanical Engineers. A (ISSN:03875008)

巻号頁・発行日

vol.73, no.733, pp.105-112, 2007-09-25

参考文献数

12

被引用文献数

1 3

---

This paper verifies the response surfaces of artificial neural networks (NN) learned by using a method based on Bayesian inference. Mackay showed that the Bayesian method due to Gull and Skilling can be applied to regularization for NN. However, generalization ability has not been verified sufficiently for the NN response surface regularized by using the Bayesian method. If the NN response surface has good generalization ability, it can be used in the optimization process of response surface methodology (RSM) NN therefore was learned by using the Bayesian method to investigate generalization ability. We tried three rules for updating the regularizing constants in an objective function minimized during NN learning. All of the update rules were derived from the Bayesian method. As a result, the response surface of NN had good generalization ability, with the exception of one update rule. The poor update rule failed to determine the regularizing constants. This tendency for the update rules was recognized regardless of response surface geometry. After we selected a superior update rule, the NN response surface by using the Bayesian method was applied to an optimization problem. The response surface didn't fit noises included in teacher data, and consequently, it was effectively used to reach a solution. Finally, we concluded that the NN learned by using the Bayesian method can be used as the response surface in the process of RSM.