Research On Vision-based Ampoule Liquid Particle Inspection Machine

Ampoule injection is the first choice of clinical salvage medicine and plays agreat important role in clinical medicine because of its rapid drug effect, dependableaction and intravenous injectability. During the process of manufacturing, some tinyinsoluble foreign particles such as glass chips, aluminum chips, fibers or hairs appearin the liquid medicine due to incomplete cleaning, filter material’s breakage,disqualification of the workshop’s purification level or carbonization phenomena.They can cause thrombus, phlebitis, tumor, anaphylactic reaction or even death whenthese kinds of particles injected into the vein. Nevertheless, more than99%pharmaceutical corporations in China adopt light inspection method by workers indark working rooms. Frankly, this method is simple but the inspection efficiency andrepeatability are poor, omission rate increasing synchronously influenced by workers’attitudes or feelings. Studies on intelligent foreign particle inspection machine forinjection based on machine vision are rare. Only several European corporationsdeveloped this kind of machine. Similarly, quite a small number of pharmaceuticalindustries in China utilized these machines well because the liquid filteringprocedures and packaging materials are quite different in China from those used inwestern countries, which results in high error-inspection rates. Hence, developing anautomatic foreign particle inspection machine for ampoule injection, suitable withdrug-making standards and having independent intellectual properties, has importanttheoretical and practical significance. For this purpose, the dissertation made a deepand systematic study on foreign particle inspection machine for ampoule injection.Based on the combination of technologies such as machine vision inspection,lighting, foreign particle inspection in liquid, the dissertation firstly designed thismachine’s mechanical structure, electrical control system, optical path and imagingmethod. Then a single-working-station inspecting platform was manufactured. Afterlots of experiments’ proof, foreign particle inspection machine based on machinevision was finally proposed. Main results and contribution can be concluded asfollows:1. Ampoule injection’s manufacturing, usage and research status of foreignparticle in liquid medicine are demonstrated. Study significances and necessities offoreign particle inspection machine and feasibility of the machine vision inspection methodology are pointed out. Important technologies related with machine areemphasized and some application cases with machine vision inspection technologyare proposed. Furthermore, common equipments’ research status at home and abroadare analyzed.2. Foreign particle’s resource and classification are introduced before system’soverall design. After high speed revolving and abruptly stopping, from the point ofview of mechanics and kinematics, moving objects’ vertical and horizontal force areanalyzed, which offers an important reference distinguishing the foreign particlesfrom random noises.3. Inspection machine’s overall design based on machine vision was proposedcombining with the light scattering technique and manual inspection method. Detaileddevelopment of ampoule conveying method, mechanical structure, key parts(such aspressure level, bottle twist, products’ separation mechanisms), imaging system andillumination plans are demonstrated. To meet the necessity of long-time, real-time andstable running, distributed control scheme combing with industrial PC and PLC wasproposed, software architectures was designed.4. Mechanical parts’ manufacturing and assembling accuracy, offset load forceand et al. will bring great challenges to the recognition of the visible particles in theliquid. To simplify the algorithms and reduce the particle detecting time, ampoule’ssequential images must be registered in advance. Regions of interest are rearranged toreduce computational complexity and improve the running efficiency, correctdetection rate. To avoid disturbances of the background noise, pixel probabilitystatistics based background modeling is proposed according to their characteristics.Visible foreign particles can be extracted effectively and quickly through backgroundsubtraction algorithm. Some remaining noises can be found in the difference images.An adaptive de-noising model based on total variation and difference curvature whichcan preserve the image’s details effectively and avoid TV model’s staircase effect.5. Better image segmentation can be observed to noisy and low-contrast imageswith pulse-coupled neural networks model. This model can segment the overlappingregions effectively even their gray values are close. Hence, it was selected as themodel to segment the visible foreign particles. Nevertheless, lots of parameters haveto be set manually in advance and the segmentation results judged subjectively by thehuman. Two modified PCNN segmentation models based on two-dimensional Tsallisentropy or water region area have been proposed and segment the foreign particlesautomatically. 6. Comparing with computational intelligence method like traditional artificialneural network, support vector machine, extreme learning machine has thesecharacteristics: input weights can be selected randomly; the hidden layer need not beiteratively tuned; fast learning speed; rare manual intervene; good generalizationcapacity. It has distinct advantages to classify the foreign particles. However, thetraining samples usually came one-by-one or batch-by-batch, online sequentialextreme learning machine (OS-ELM) was proposed to satisfy the high-speed learning.The false alarm rate will be large because of the small SNR and low contrast of theampoule images, also only depending on foreign particles’ size, shape or texture.Hence, foreign particle classification and recognition algorithms based on OS-ELMhas been discussed. All the objects’ eigenvectors of the sequential images were inputinto the OS-ELM model. Those eigenvectors are classified into one group which canbe thought as one object. Connect the object’s position into a trace according to theirimage sequence number and distinguish the foreign particles from remaining regionsusing the traces’ smoothness and noises’ randomness.7. Single working-position-platform was developed firstly to improve the successrate, reduce the research risks and save the costs. To perfect the system design scheme,ampoules’ mechanical conveying, clamping, illumination were simulated and theclassification algorithm was demonstrated on the platform. The foreign particleinspection machine was developed at the last. It was proved that the machine waseffective, dependable and completely satisfied with production line’s inspectionaccuracy and speed through Knapp-Kushner test, human-machine comparison test androbustness judgments.During the ampoule inspection machine’s manufacture, amount of practicalproblems were found and solved effectively, which accumulated priceless experienceson further upgrading and regeneration. The developed software lays a solidfoundation for the forthcoming theoretical research and potential exploration.