成果報告書詳細
管理番号20160000000484
タイトル*平成27年度中間年報 SIP(戦略的イノベーション創造プログラム)/インフラ維持管理・更新・マネジメント技術 点検・診断技術の実用化に向けた研究開発 学習型打音解析技術の研究開発
公開日2016/9/10
報告書年度2015 - 2015
委託先名国立研究開発法人産業技術総合研究所 首都高技術株式会社 東日本高速道路株式会社 株式会社テクニー
プロジェクト番号P14031
部署名ロボット・AI部
和文要約
英文要約Title: Development of an analysis technology for hammering-echo based on machine-learning techniques (FY2014-FY2016) FY2015 Annual Report

The objective of this project is to develop hammering test technology that will enable detection of flaws inside the structure depending little on skills of inspectors. To this goal, this project develop (1) analysis system that can estimate the structure condition under the surface by collecting pairs of hammering sound data and its existence of flaw or sort of flaw to learn their correlations (2) dedicated hammering apparatuses operating together with the analysis system (3) on-site display devices such as tablet computers which can visualize flaw maps for inspection results. These developments will realize not only improvement of inspection accuracy, but also reduction of labor costs and total man-hours including report writing works.
In this fiscal year, we developed a two-stage learning-type flaw detection algorithm. In the first stage, it learns statistically a “normal” pattern in non-supervised manner corresponding to every target structure under the assumption that most of the collected hammering sounds are normal. In this stage, it can judge relatively the existence of flaws even if sufficient hammering sound data for various kinds of flaws are not collected. In the second stage, it carries out discriminant analysis in supervised manner utilizing the labels given by skilled engineers for detected results in the first stage. This improves the detection precision for the flaws.
We also developed two types of hammering apparatus, one is a hand-truck type, and the other is an apparatus dedicated for high place hammering, by reflecting the evaluation results. We have improved the hammering mechanism and sound sensing part for the second and third prototypes of hand-truck-type machines. We have also adopted two kinds of actuators in the high place type for comparison.
As for the analysis system, we have developed an analysis server on the cloud, which cooperates with the data collection server developed in the last fiscal year. We started development of the image processing program for tracking of the hammering apparatus.
For evaluations of the developed system, we have conducted (1) collection of hammering echo data with the developed high-place-type apparatus in 12 different places, such as highway floorboard, bridge pier, and bridge abutment in Kanagawa, Chiba, Ibaraki prefectures. (2) collection of hammering echo data with the developed hand-truck-type apparatus in 6 highway bridges including a newly constructed bridge. Results of these experiments were effectively fed back to the development of the hammering apparatus. A testing concrete block was constructed, which simulates the bridge pier with frequently observed surface flaws to investigate their influences on the detection of inside flaws.
ダウンロード成果報告書データベース(ユーザ登録必須)から、ダウンロードしてください。

▲トップに戻る