Research On Surface Defect Detection Algorithm Of Elevator Compensation Chain Based On Machine Vision

Elevator compensation chain is used to balance the weight of traction wire rope and traveling cable in the process of elevator car running.It is one of the key components to ensure the stability and safety of elevator operation,and the market demand is huge.At present,manual detection methods are used in the production process,which are inefficient and labor intensive.And because of the strong subjectivity of detection and visual fatigue,it is easy to lead to missed detection and false detection,so it is difficult to meet the production requirements of modern enterprises.Therefore,enterprises have an urgent demand for automatic detection system based on machine vision technology.Elevator compensation chain is curve and intersects 90 degree between two adjacent chain rings,which makes the chain twist and sway in the process of traction.Moreover,because of the strong reflection on the surface of the chain ring and the different distribution positions of different types of defects,it is difficult to obtain the image with high contrast by a single lighting method.At the same time,there are many kinds of defects in the chain,and each kind of defect has different features and details.Not only the surface of the chain is rough,uneven,but also there are many interference factors such as stain and rust spots,which makes the defect segmentation and identification difficult.Aiming at the key technical problems mentioned above,we designed an image acquisition unit to acquire images with high contrast,and developed the detection algorithms for typical surface defects(stamping defects,weld and cold bend cracks).In the design process of image acquisition unit,to solve the problem of distortion and vibration during chain traction,cross grooved sleeve and sprocket matching chain structure are designed,and reasonably arranged on the inspection line to ensure the stability of chain imaging;For the problem of strong reflection of chain and various types of defects,the corresponding combined light source is designed to obtain the image with high contrast.On the research of defect detection algorithm,firstly,according to the location distribution of different types of defects,the ROI segmentation algorithms at the middle and location R of the chain body are proposed to narrow the detection range and reduce the computational burden of the subsequent algorithm.Aiming at the problem of large noise and relatively blurred edge around the stamping defect,anadaptive denoising and enhancement algorithm based on NSCT is proposed to improve image quality.Then the segmentation method based on texture analysis and watershed transform is used to segment and determine the stamping defects.For the weld defects,the linear contrast stretching and adaptive threshold segmentation are used to identify them.The proposed methods are verified by experiments.In view of the various manifestations of cracks in the chain at location R,it is difficult to universalize a single detection method,so the deep learning method is adopted.Firstly,data sets are produced by collecting ROI images at location R and processing them with normalization and data enhancement.Aiming at the problem that the training effect is not ideal due to the small data set,the network model of migration learning based on ResNet-50 is reconstructed.At the same time,the learning rate is changed to self-adaptive adjustment during the training process.By comparing with traditional CNN and multi-channel CNN model,it is verified that the model can not only accelerate the training speed,but also effectively solve the problem of defect detection of the chain at location R under insufficient data samples.