Research On Defect Detection System Of Emulsion Pump Based On Machine Vision

In order to realize the high-precision and high-efficiency automatic detection operation of the emulsion pump and reduce the production cost,this paper uses machine vision technology to develop a new emulsion pump defect detection system instead of manual sorting of qualified products and defective products.The main tasks completed in this paper include:First of all,starting from the system's functional requirements and design indicators,according to the design ideas,the overall scheme of the system is given at the macro level,and then the detailed design of each part of the system is made.The overall scheme of the system adopts multi-station layout,including emulsion pump feeding station,visual inspection station,qualified product sorting station and defective product sorting station.The delivery scheme of the emulsion pump is realized by intermittent rotation of the splitter..Secondly,the visual hardware design part,this article discusses in detail the selection of industrial cameras,optical lenses,light sources,network cards,computer configuration,the application of polarizers in image glare,emulsion pump lighting solutions.The structure of the system body is introduced,including the distribution of each station,the detailed design process of each device and the overall working mechanism of the structure.In the System Software section,the software development environment,workflow,and modular design are described.Image acquisition,processing,analysis,decision making,and control of the lower computer are realized.Then,the paper focuses on the design of the emulsion pump defect detection algorithm.According to different methods of adoption,the detection algorithms are divided into two categories.In the first category,traditional image processing methods are used for detection problems that are less interfered by the structure or pose of the emulsion pump,such as head lock detection,nozzle lack of glue detection,bottle cover lack of glue detection,and cap oil stain detection.,the positive and negative insertion detection of the suction pipe,the oil contamination detection at the upper end of the pump body.Before designing various detection algorithms,the key image processingmethods were first studied.In the second category,a model classification scheme based on convolutional neural network is adopted for the detection of white pump body which is largely interfered by the structure or pose of the emulsion pump.The paper introduces the structure,training and promotion methods of convolutional neural networks,sample image sets and marking cases.The design of the sorting model firstly adopts a shallow convolutional neural network with simple structure,and then proposes a model improvement method based on migration learning according to the training effect of the model.The final debugging effect of the detection system shows that the system has achieved the established design function as a whole,and the detection algorithm can detect the surface defects of the lotion pump more accurately.The developed detection system initially solves the non-contact detection problem of the emulsion pump,and has a good reference value for the application of the detection system on the production line.