Monocular Vision Detection And Control Technology Of Micro-Droplet Dispensing Volume

The precise distribution technology of adhesives,fillers and other materials has played an essential role in ensuring the quality of microelectronic chip packaging.In the process of surface mounting,wire bonding,field soldering,etc.,the droplet dispensing technology must be able to realize the rapid distribution of the different desired volume to different types of droplet,and achieve a placement task for a large number of chips of different sizes and different packaging types in a short time.However,at present,in the domestic industrial market,the process of droplet's distribution is still mostly open-loop control to achieve the distribution of droplet's volume of different sizes.There is low distribution accuracy,poor anti-interference ability,and manual calibration is required when dispensing droplet's condition changes.Therefore,it is significance for the adaptive closed-loop control of droplets is carried out to realize the real-time adaptive control of the desired droplet volume size,and to meet the requirements of micro-precision,controllable volume and rapid automated dispensing operations to various droplets in fields of microelectronic packaging.To achieve closed-loop control,it first need to extrac-the volume of droplets online.In view of the high cost and low real-time performance of the traditional 3D measurement method,it chose the method based on the Shape From Shading(SFS).In order to further improve the accuracy and speed of the SFS algorithm,and solve the problem that the traditional Lambert model's low accuracy for reconstruction of highlight region and the Phong hybrid model's low running speed,this paper combines the advantages of each model algorithm and proposes a Lambert-Phong hybrid model algorithm which uses the Lambert model and the Phong model to reconstruct the droplet's 3D shape in the diffuse reflection region and the highlight region respectively.For this method,it is necessary to segment the highlight region of the droplet firstly.In order to ensure the accuracy of segment highlight region in different environments,Mask RCNN deep learning semantic segmentation network in feature detection RCNN series is adopted.This article introduces the basics of Mask RCNN and the overall network architecture.In the experiment,the theoretical graph dataset and the actual droplet dataset are put into the Mask RCNN network for training.And the results show that the Mask RCNN semantic segmentation network can achieve highlight region's egmentation.Based on results of the segmentation of highlight regions,the Lambert-Phong combination model proposed in this paper is solved.For the highlight region,firstly the image gradient weighting coefficient is introduced into the model equation to correct the error;then the target equation is discretized by the central difference.Next,the Newton expansion is performed on the equation,and the high?order is discarded because of only taking up a small proportion.The original equation is converted into a linear equation.Finally,the whole equation is solved by Newton iteration.For non-highlight areas,firstly the Newton expansion is performed on the equation,and the high-order terms are discarded and converted into linear equations.Then,the Newton iteration method is used to solve the whole equation.Finally,we use the Lambert-Phong combination model for theoretical image and actual droplet image experiment with accuracy of 7.8%and speed of 0.015s,which shows the effectiveness of our proposed Lambert-Phong model.It not only ensures the accuracy of the algorithm,also greatly increases the speed of the algorithm.Based on the previous volume measurement of droplets,an experimental platform for the droplet micro-spray monitoring system is developed,which includes the dispenser,driven module,motion control module,volume detection feedback module and human-computer interface.The dispenser utilizes a piezostack to vibrate the needle and the air-pressure to move the liquids.The driven module can generate signals for piezostack and air-pressure controller.The motion control module can drive the micro-spray valve to perform accurate dispensing operation at a specified position;.the volume detection feedback module can perform three-dimensional reconstruction ofthe collected droplet's image by using the Lambert-Phong model equation,and the volume information is extracted;The human-computer interface lets the operator set the system parameters.By analyzing the influence of PID control parameters on the system,a PID controller with closed loop control of droplet micro-spray's volume is designed.Based on the above,the whole algorithm is finally applied to the final droplet volume closed-loop control experiment.Given the desired droplet volume,the droplet volume is precisely controlled by the PID closed-loop control algorithm.Finally,the control system overshoot is 14.9%;the rise time is about the sixth time of the micro-spray valve dispensing;the volume fluctuation does not exceed 0.4ul;volume deviation is about 8.14%.The experimental results show that when the dispensing environment changes,the droplet micro-injection control system can adaptively guide the micro-spray valve to perform the droplet dispensing operation..