Research On Defect Detection Method Of Electrospun Membrane Based On Computer Vision

Electrospun membrane is produced by spinning natural or artificial polymer solution under the action of high-voltage static electricity.The striking characteristics of the membrane are its high porosity,specific surface area and good air permeability,which is very popular in the fields of battery diaphragm,sensor and medical protection.However,the electrospinning process is more complicated,and product defects are difficult to avoid.The defects damage the filtration and physical properties of the electrospun membrane,resulting in a decrease in production efficiency and waste of raw materials.At present,there are few studies on the quality inspection of electrospun membranes.However,most of them focus on the measurement of orientation,diameter and porosity,and there are few studies on defect inspection at the microscopic level.On the other hand,as labor costs increase year by year,and the concept of production cycle and quantitative analysis is gaining ground,the demand for intelligent defect detection solutions is becoming more and more urgent.In response to the above-mentioned problems,this paper studies the defect detection methods of electrospun membranes based on computer vision technology.The paper mainly does the following work.(1)The background and significance of defect detection on electrospun membrane were introduced first.Then,we proposed two research points,which were the development of the automatic control system of the stage and the design of semisupervised defect detection algorithms.(2)Hardware selection for image acquisition: The advantages and disadvantages of self-made and commercial electrospinning equipment were analyzed first.Then,we designed an automatic control system for microscope stage,which contained the drive system,main control chip and auxiliary peripherals.At last,The forms of defects in electrospun membrane were analyzed,and FE-SEM was determined as the imaging tool.(3)Production of the image set: Sample preparation and SEM image collection were firstly carried out.In the process of making the data set,the methods of data cleaning and image augmentation were studied.(4)The automatic control system of the stage based on STM32: We firstly designed the hardware circuit,drawn the PCB,and then developed the driver of the functional modules,including the temperature and humidity module DHT11,wireless chip n RF905,LCD1602 and an OLED of 0.96 inch,etc.,involving the single wire protocol,I2 C,SPI,and serial port protocols.At last,we studied the open and closed loop control of stepper motors,which produced higher displacement accuracy and faster speed.(5)Semi-supervised defect detection algorithm based on VAE: On the one hand,GMM was used to model VAE coding features,and anomalies were identified through density estimation;on the other hand,VAE was used to output repaired images,which were compared with original images to detect defects based on SSIM.The two ways could be merged to produce a robust and precise result.The dataset was firstly prepared,and then we built a deep learning framework,which was used to train models and conducted experiments.The results showed that the maximum AUC is 0.950,which was in line with expectations.