基于数学形态学的航空发动机涡轮叶片故障自动检测方法

摘要:

涡轮叶片的故障类型多样,且有些故障特征可能比较隐蔽,不易察觉,在检测时无法充分挖掘每个分量中蕴含的故障信息,导致难以精准提取感兴趣区域,使得故障检测准确性较低。因此,提出基于数学形态学的航空发动机涡轮叶片故障自动检测方法。设计图像采集装置和光源设备,完成航空发动机涡轮叶片图像采集。采用几何推导法实现叶片图像颜色空间变换,将RGB图像变换为更适合观察的HIS图像。利用数学形态学算法,对HSI彩色图像中H分量、S分量、I分量图像分别进行滤波和边缘检测,从多个角度和层面提取故障特征以充分挖掘每个分量中蕴含的故障信息,提取出感兴趣区域(Region of Interest,ROI),提高故障检测的准确性和可靠性。将提取的ROI输入改进YOLOv8网络模型中,通过自动学习完成特征提取、特征融合和分类检测,输出叶片故障自动检测结果。测试结果表明:所提方法进行检测后的平均精度(Average Precision,AP)结果为98.33%,能够实现对涡轮叶片故障问题的准确描述。

Abstract:

The types of faults in turbine blades are diverse, and some fault characteristics may be hidden and difficult to detect. During detection, it is difficult to fully explore the fault information contained in each component, resulting in difficulty in accurately extracting the region of interest and lower accuracy of fault detection. Therefore, a mathematical morphology based automatic detection method for turbine blade faults in aircraft engines is proposed. Design image acquisition devices and light source equipment to complete the image acquisition of aircraft engine turbine blades. Using geometric deduction method to achieve color space transformation of blade images, transforming RGB images into HIS images that are more suitable for observation. Using mathematical morphology algorithms, filtering and edge detection are performed on the H, S, and I components of HSI color images, respectively. Fault features are extracted from multiple perspectives and levels to fully explore the fault information contained in each component, and the Region of Interest (ROI) is extracted to improve the accuracy and reliability of fault detection. The extracted ROI is input into the improved YOLOv8 network model, which completes feature extraction, feature fusion, and classification detection through automatic learning, and outputs the automatic detection results of blade faults. The test results show that the average precision (AP) of the proposed method after detection is 98.33%, which can accurately describe the problem of turbine blade faults.