| During the production process of pharmaceutical injection,it is inevitable to produce undesirable products mixed with insoluble foreign particles such as tiny glass chips,hair and fiber.It will seriously harm h?man health if the pharmaceutical injection containing foreign was injected into h?man body.The national pharmacopoeia clearly stipulates that the diameter of insoluble foreign particles shall not exceed 50?m.Limited by the backward domestic detection technology and the lack of high-end insoluble foreign particles detection equipment in China,more than 99% of enterprises still use manual detection methods to control the quality of drugs,which is likely to cause serious drug safety problems.Manual detection has many shortcomings such as subjective judgment,serious omission and lax standards.With the rapid development of machine vision technology,optical sensor technology,industrial control technology and high-performance processor technology,the performance of Automated optical inspection(AOI)equipment gradually improves to achieve or exceed manual detection capability,and the development of new optical detection technology also opens up a new way for automatic detection of insoluble foreign particles in pharmaceutical injection.This article was support by the national 863 project and the key project of the National Natural Science Foundation of China,and cooperates with Hunan CHINASUN pharmaceutical machinery Co,Ltd.Multiple bottleneck technical problems are gradually solved from processing technology,sample analysis,mechanical design,optical solutions,image algorithm,software development and electrical control.After the trial production test of the three generations of prototypes,the small-capacity,large-capacity weak insoluble foreign liquid particle visual perception and detection robot equipment was successfully developed.The main work and innovative achievements of this paper are as follows:(1)This paper analyzes the harm,generation and detection means of insoluble foreign particles in medical injection in China,compares the existing detection technology level in foreign countries,and points out the necessity,importance and market promotion value of developing automatic visual detection robot for insoluble foreign particles in medical injection.Many technical problems in the development of detection robot,such as various kinds of foreign particles,complex optical imaging and high detection efficiency,are introduced.(2)The requirements and key indicators of the development of detection robot are extracted.Through the research of a large n?mber of bad sample,classification,analysis,the physical properties of main kinds of foreign particles are determined,and based on this,The detection principle of "rubbing flask-rotation-emergency stop-dynamic imaging-sequence image detection" of detection robot was proposed,and the kinematics and mechanical models of foreign bodies were established,and the bright field optical scheme and dark field optical scheme of moving foreign bodies were designed.Finally,the vision detection robot's optical,mechanical,electrical,software,control the whole machine program and implementation strategy.(3)Aiming at the problem that mechanical vibration seriously restricts the image detection effect,the registration algorithm based on phase correlation and SIFT feature is researched,which is suitable for bright field sequence image registration and dark field sequence image registration,respectively.Firstly,the low-contrast images under complex background are preprocessed by dynamic dual-threshold salt-pepper noise filtering algorithm,histogram equalization algorithm and TOPHAT contrast enhancement algorithm,which improves the discrimination scale between weak foreign object and dynamic background.Then,bicubic b-spline interpolation method is adopted to scale the image to improve the registration accuracy and stability through sub-pixel image operation.Based on the above preprocessing and aiming at the problem of bright field image jitter,this paper proposes an improved Fourier registration method based on Canny edge feature,which takes the obvious edge feature as the feature input to eliminate the influence of dynamic liquid level,bottle feature changes and other factors on the registration.In order to eliminate the vibration of sequence image in dark field,an algorithm of bottle random SIFT feature point registration is proposed,which solves the problem of registration in dark field image with less information and random features.A large n?mber of experiments have proved that the algorithm is suitable for different bottle bodies,different brightness,noise interference environment,good robustness,high registration accuracy.(4)The segmentation effect and precision of foreign object in image sequence are the core indexes of the detection robot.The detection robot needs to efficiently segment and extract foreign object under the condition of detecting tiny target,random change of liquid level reflection brightness and poor consistency of foreign object features.Traditional direct difference,c?mulative difference,energy entropy and other segmentation effect is not good.In order to segment images effectively,a BPNN segmentation algorithm for sequential images based on time spectr?m is proposed.First,the same pixel of the image is sorted according to the size of the grayscale value,the disordered image is changed into an ordered image,and the sequence image is transformed into an ordered sequence of extracted features.Then,the normalized operator is used to eliminate the inconsistency of eigenvectors under the inhomogeneous background.Finally,BP neural network is trained and sequence images are segmented.The segmentation results show that the interference and anti-vibration ability of the proposed segmentation algorithm is significantly improved,and the foreign body segmentation rate reaches more than 98%.To further improve the segmentation accuracy,make up for the division of BPNN small target decision marginal problem,put forward in backward PM segmentation algorithm,through partial differential gradient feature to subpixel non-uniform background image segmentation,and then by foreign body BPNN segmentation for PM morphological hit the segmentation results for operation,to hit the results as the final characteristics of high precision foreign body.BPNN and PM can be used together to ensure the segmentation accuracy and the segmentation performance,providing technical support for the accuracy and precision performance indicators of the detection robot.(5)Classifying unqualified products according to the characteristics of foreign particles is an important index to evaluate AOI detection equipment,which can not only be used to characterize the discrimination accuracy of detection equipment,but also provide reference for manufacturing enterprises to improve the process.This paper proposes a PSO-RBF-SVM multi-level classification method based on geometric features of foreign particles,gray features and texture features as feature vectors.In order to solve the problems of large dimension and uncertain correlation of the feature set of foreign body,the method of reducing dimension of PCA is proposed.The multidimensional vector is compressed into a few orthogonal feature vector sets,which improves the distinguishable scale of the feature set of foreign body and reduces the interference of invalid features.Then the parameters of SVM classifiers were optimized by PSO method to obtain the high-dimensional optimal classification surface,and the multi-level classification of foreign particles was realized.The classification accuracy and classification speed could satisfy the online classification of foreign particles,and the classification results could be used for the fault prediction in the production process.(6)This paper deeply analyzes the opto-electromechanical technical problems faced by the development of medical foreign body detection robot.Starting from theoretical modeling,mechanical simulation,motion analysis,prototype test and other aspects,it gradually analyzes and penetrates layer by layer.Through a large n?mber of test verification and technical upgrading,it successfully overcomes hundreds of core technical problems.After the trial-manufacture of three generations of prototypes,the project team developed the optimal feeding mechanism,rubbing mechanism,testing mechanism,feeding and removing mechanism of the robot,and developed the control system and real-time software system to ensure the high-speed and high-precision operation of the robot.The detection robot has been used by many customers and tested by the Knapp-Kushner standard,and the detection effect of the equipment has been unanimously recognized.The detection accuracy can reach less than 40?m.After upgrading and iteration,the software system of the detection robot has been able to directly connect the high-speed production line for injection filling.The successful development of the detection robot is of great significance for ensuring the medical quality of pharmaceutical enterprises and reducing drug safety. |
PDF Full Text Request