Research On Machine Vision-based Lithium Battery Swollen Detection

Lithium batteries as a new generation of environmentally friendly energy,its increasing application demand puts forward higher requirement for automatic production and safety performance.In the process of production,battery packs may occur swollen phenomenon because of bad encapsulation,formation abnormalities,etc,which will cause the performance serious failure and be a potential safety hazard.These swollen batteries must be picked out to reprocess before leaving factory.At present,most factories separate the swollen batteries using artificial detection method which greatly rely on workers' experience and feeling,that is inefficiency and susceptible to the subjective factors,and is also not conducive to the intelligent management of lithium production.To solve this situation,this paper put forward a new method based on machine vision to detect and separate the swollen battery by taking surface defect detection technology for reference and using square batteries as the research object.Firstly,under light irradiation with certain angle and intensity,the distribution of light-spot reflection area on qualified lithium battery surface is different from swollen ones.Grasping the differences,this paper designed a set of complete battery feature extraction scheme.Set ROI region on batteries images first,then morphological processing method is adopted for smoothing and denoising,extracting battery images' geometrical characteristics as the input of classifier.Secondly,to solve the problem that reflection spots are easily expanded even submerged in silver background,an improved lateral bimodal threshold segmentation method is proposed to segment target area.Filtering the histogram peak generated by shaded area,relocate segmentation threshold between the reflected light spot and silver surface.The experimental results show that,the improved lateral bimodal threshold segmentation method is more suitable for this detection system,and segmentation result is closer to real images.Lastly,considering the limitation of battery samples,this article gives full play to the support vector machine classification in solving small sample of good generalization ability,building the swollen detecting classification model based on C-SVM theory,and presents a simplified discernibility matrix covering rough set attribute reduction method optimizing sample characteristics.Experiment shows that,the proposed method can achieve a recognition rate of 90.1%.On the basis of the above algorithm,we describe the prototype of lithium battery swollen detection system and implementation method,then develop the software system with aid of OpenCV visual library in VS2010 platform.Lithium battery swollen detection based on machine vision is a way with high practicability,which provides an effective method for realizing nondestructive detection.