本文へジャンプ

成果報告書詳細
管理番号20190000000251
タイトル*2018年度中間年報 次世代人工知能・ロボットの中核となるインテグレート技術開発 人工知能技術の社会実装に向けた研究開発・実証 AI活用によるプラント保全におけるガス漏洩の発見と特定の迅速化、並びに検出可能ガスの対象拡大
公開日2019/6/19
報告書年度2018 - 2018
委託先名コニカミノルタ株式会社 国立大学法人神戸大学
プロジェクト番号P18002
部署名ロボット・AI部
和文要約
英文要約Title: Accelerate detection and identification of gas leaks at plant maintenances by using AI and expand the scope of detectable gases. (FY2018-FY2019) FY2018 Annual Report

Konica Minolta, Inc.
Kobe University

As plants in Japan has been aging, it is concerned that explosions, fires and destructions may occur due to damage caused by the aged deterioration. Explosions, fires and destructions at plants have a major impact on the decline in Japan's international competitiveness. Furthermore, current maintenance work at plants relies too much on experienced worker’s skill and their skills is not properly being handed down to younger generations. So, the more experienced worker retires the more maintenance work will be done by untrained young workers, which results in the inefficiency work flow at plant maintenance. To improve these problems, KONICA MINOLTA incorporates AI technology into the gas monitoring system that visualizes gas leaks with the infrared camera and the visible camera. KONICA MINOLTA’s new camera system is able to identify the location of the leak source, the amount of leaking gas, and the flow direction. This enables plant to be highly maintained out of limited human and financial resources without stopping operation.
In 2018, KONICA MINOLTA developed a basic AI model to convert two-dimension into three-dimension. The conditions were simple (straight piping, no background).
1. Collect an experimental data
We set up multiple cameras to monitor gas quantitively released under noise free environment while controlling the background temperature. This is how data was collected at the outdoor test site and the actual plant
2. Develop of flow simulations, and match of the flow simulation data and the experimental data
Next, we simulated gas distribution by three-dimensional fluid simulation, and created two-dimension flow simulation data. Furthermore, we matched flow simulation data and experimental data to see if it could be used for AI model development. As a result, there was no difference between flow simulation data and experimental data and no obvious abnormality was found from flow simulation data. We concluded that it can be used for AI model development.
3. Develop a basic AI model
Finally, we developed a basic AI model. When we tested using the flow simulation data, we were able to construct a three-dimensional model that can determine whether the flow of gas is near or far with 90% probability.
Additionally, Kobe University worked on the following two themes in 2018.
4. Basic research for accelerating detection and identification of gas leak by utilizing AI
4-1. Quantitatively estimate leaked gas concentration distribution and leak amount based on infrared measurement in multiple light paths
We verified the method of reconstructing three-dimensional concentration distribution of gas leakage by experiments using a simple solid model (gas phantom). From this experiment, it was possible to reconstruct two-dimensional images and found it useful. Next, we analyzed the distribution error factors during reconstruction.
4-2. Estimate leak source using inverse analysis
We calculated the residual sum of squares using the distributions obtained from flow simulation data and experimental data. We confirmed whether the residual sum of squares is smaller as the distance between the leak sources is closer, and we obtained a prospect after examining the possibility of estimating the location of the leak source when in a calm and windy conditions.
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