-
实验样品包含直径为3 mm的正方形铁氧体电感和直径为4 mm的不规则圆形铁氧体电感。在实际工业检测中开口宽度尺寸在1 μm级别的裂纹检测难度大,此类裂纹被称为隐裂。
文中通过光学显微镜下成像和热成像下人工观察识别样品,综合比对结果对待检测的样品进行预分类,共计610个样品:其中含完好样品400个,裂纹样品130个,隐裂样品80个。
样品实物图如图6所示,其中图6(a)为高倍光学显微镜下正方形铁氧体裂纹样品图,图6(b)、(c)分别为高倍显微镜下不规则圆形铁氧体电感裂纹样品图及其在电子显微镜下观察到的隐裂。
针对电感表面微裂纹,光学显微镜下可清晰成像,但是针对隐裂,需在高倍的电镜下才能观察到。
-
实验装置如图7所示。该系统采用试样静止、两个激光器A、B依次扫描样品的运动模式,通过热像仪记录试样表面温度变化并成像。
其中线激光波长为980 nm,功率为30 W,线光斑有效线宽为0.5 mm,线激光运动速度为10 mm/s,将电感试样表面的最高温升控制在80 ℃左右。采用帧频为50 Hz,结合镜头空间分辨力为50 μm的Flir A655SC热像仪对试样表面的温度变化情况进行记录。该装置可在18 mm×24 mm的视场范围下,在5 s内一次批量检测20个不规则圆形铁氧体电感样品或35个正方形铁氧体电感样品。
-
对比电感表面裂纹在光学成像和激光热成像的效果,200x光学显微镜成像和激光热成像的结果如图8所示。第一行为光学显微镜下样品的裂纹成像图,第二行为文中激光扫描后消除非均匀性后的样品灰度图。其中,图8(a)、(b)为裂纹样品,(c)、(d)为隐裂样品。
从裂纹成像结果可以看出,两种方法都能清晰裂纹样品表面的裂纹。但针对隐裂样品裂纹成像上,由于制作工艺上铁氧体电感是由铁氧体颗粒压制而成,在高倍显微镜下其表面不规则纹理同隐裂的纹理对比度不明显,常规的光学显微镜成像难有效区分出隐裂区域与非裂纹区域,而采用文中激光热成像的方法可以清晰地成像出隐裂,大大提高后期裂纹的识别成功率。
如图9所示,为文中实验试样中80个隐裂样品在消除非均匀性后的裂纹热成像图。裂纹信噪比较之光学显微镜得到大幅提高,通过热成像图,人工可以完全识别出试样表面的隐裂。
-
在裂纹自动识别算法方面,综合文中上述方法,对610个铁氧体电感样品进行裂纹自动识别,其中检测结果如表1所示。
Category Sample number Detection number Detection rate Intact sample 400 380 95% Crack sample 130 122 94% Microcrack sample 80 72 90% Table 1. Automatic detection results of ferrite inductor cracks
在裂纹自动识别结果中,针对完好样品和裂纹样品,自动识别准确率可达94%以上;针对隐裂样品,自动识别率可达90%。
由于隐裂尺寸过小,热成像中隐裂信噪比低、自动识别算法不够完善,在自动识别环节存在一定漏检率。其中,隐裂样品中漏检的8个样品为图9中红色虚线框内所标注的样品。但是通过人工观察电感样品的热成像图,隐裂样品表面的裂纹均能被识别出来,文中研究在电感裂纹检测上仍然具有意义。
Detection of microcrack in inductor based on orthogonal scanning line laser thermography
doi: 10.3788/IRLA20190522
- Received Date: 2019-12-05
- Rev Recd Date: 2020-02-20
- Available Online: 2020-04-29
- Publish Date: 2020-07-23
-
Key words:
- laser thermography /
- microcracks /
- automatic detection /
- ferrite inductor
Abstract: In the detection of microcrack on the surface of ferrite inductors, traditional machine vision detection has problems such as low signal-to-noise ratio and low detection accuracy. In order to solve these problems, a microcrack thermography detection system based on line laser orthogonal scanning was built. The surface temperature change of the sample was recorded by the thermal imager and imaged. Sub-maximum filtering was used to eliminate non-uniformity and edge contour interference of the thermographic image. And the multi-directional fan-shaped filtering was used to obtain the grayscale image of the sample in different directions. Finally, qualitative detection of microcracks on the surface of the inductor was realized by BP neural network and morphological processing. The results show that all cracks and microcracks are correctly imaged based on 610 samples in two specifications. The automatic identification algorithm has a false detection rate of 5%, a crack miss detection rate of 6%, and a microcrack detection rate of 10%. The system detects 20 to 35 inductors every 5 s and can be used for automated quality inspection in production.