Brierley N, Tippetts T, Cawley P
Department of Mechanical Engineering , Imperial College London , London SW7 2AZ, UK.
Proc Math Phys Eng Sci. 2014 Jul 8;470(2167):20140167. doi: 10.1098/rspa.2014.0167.
In industrial non-destructive evaluation (NDE), it is increasingly common for data acquisition to be automated, driving a recent substantial increase in the availability of data. The collected data need to be analysed, typically necessitating the painstaking manual labour of a skilled operator. Moreover, in automated NDE a region of an inspected component is typically interrogated several times, be it within a single data channel due to multiple probe passes, across several channels acquired simultaneously or over the course of repeated inspections. The systematic combination of these diverse readings is recognized to offer an opportunity to improve the reliability of the inspection, but is not achievable in a manual analysis. This paper describes a data-fusion-based software framework providing a partial automation capability, allowing component regions to be declared defect-free to a very high probability while readily identifying defect indications, thereby optimizing the use of the operator's time. The system is designed to applicable to a wide range of automated NDE scenarios, but the processing is exemplified using the industrial ultrasonic immersion inspection of aerospace turbine discs. Results obtained for industrial datasets demonstrate an orders-of-magnitude reduction in false-call rates, for a given probability of detection, achievable using the developed software system.
在工业无损检测（NDE）中，数据采集自动化越来越普遍，这使得近期数据的可获取性大幅增加。所收集的数据需要进行分析，通常这需要熟练操作人员的艰苦体力劳动。此外，在自动化无损检测中，被检测部件的一个区域通常会被多次询问，这可能是由于多次探头扫描在单个数据通道内进行，也可能是同时采集多个通道的数据，或者是在重复检测过程中。人们认识到，将这些不同读数进行系统组合，有机会提高检测的可靠性，但在手动分析中无法实现。本文描述了一种基于数据融合的软件框架，它提供了部分自动化功能，能够以很高的概率判定部件区域无缺陷，同时能轻松识别缺陷迹象，从而优化操作人员的时间利用。该系统设计适用于广泛的自动化无损检测场景，但以航空航天涡轮盘的工业超声浸没检测为例进行处理。对于工业数据集获得的结果表明，使用所开发的软件系统，在给定的检测概率下，误报率可降低几个数量级。
