doktorska disertacija
Franci Koplan (Avtor), Zvone Balantič (Mentor)

Povzetek

Medicinski pripomočki za selektivno lasersko trabekulopastiko, v katerih so vgrajeni kratki Nd:YAG-sunkovni izvori s spreminjajočim se profilom laserskega žarka in nestabilno izhodno energijo med delovnim ciklom, so nezvezno delujoči kompleksni, dinamični, mnogoodzivni, nelinearni in temperaturno odvisni visokotehnološki sistemi, za katere še nimamo ustreznih modelov. Za izpolnitev visokih zahtev po dolgoročni kakovosti in zanesljivosti je v vseh fazah proizvodnje nujno meriti ključne procesne parametre in karakteristike proizvoda ter zagotavljati hitre in učinkovite povratne krmilne zanke. Pri montaži mehanskih, optičnih in elektronskih komponent ter sklopov in njihovem justiranju v pogojih maloserijske izdelave igra človek v delovnem procesu odločilno vlogo. Za povečanje njegove uspešnosti in učinkovitosti smo razvili ekspertni sistem, primeren za uporabo v specifičnem delovnem okolju laserske svetlobe različnih valovnih dolžin. Na ta način smo delovni proces prilagodili človekovim kognitivnim zmožnostim in omejitvam, ki nastanejo zaradi uporabe zaščitnih očal z omejeno prepustnostjo vidne svetlobe, ter zmanjšali številna tveganja. Osrednji del raziskave je predstavljal razvoj ekspertnega sistema za obdelavo in merjenje različnosti posnetkov profilov laserskega žarka za posamezne preskusne cikle, overjanje ter validacijo algoritmov. Na osnovi računalniškega vida temelječi sistem za podporo odločanju je učinkovito prepoznaval dejanske in tudi potencialne zgodnje odpovedi sunkovnih Nd:YAG-laserskih izvorov in tako povezal rudarjenje podatkov z napovedno analizo. S pomočjo statistične analize smo najprej overili sposobnost merilnega sistema za snemanje profilov laserskega žarka s CCD-kamero ter določili relevantni model in parametre za opazovanje v okviru preskusnih ciklov. Nato smo vzpostavili podatkovno bazo potekov parametrov laserskih profilov in jo dopolnili z ekspertnim znanjem ter informacijami, pridobljenimi v okviru že pretečenega dela življenjske dobe. S pomočjo multivariatnih časovnih vrst in z metodo dinamičnega ukrivljanja časa smo izračunali indeks različnosti med izbranimi elementi v podatkovni bazi. Pri določitvi parametrov algoritma smo upoštevali specifične lastnosti sunkovnih laserskih izvorov. Ugotovili smo, da je zaradi sprememb karakteristik, ki so posledica tehničnih in tehnoloških sprememb, potrebno hkrati opazovati vsaj 8 parametrov profila laserskega žarka. Z uteženim glasovanjem najbližjih sosedov smo laserske izvore razvrstili po podobnosti. S pomočjo izračunanega indikatorja za verjetnost zgodnje odpovedi sistema smo olajšali odločanje o sprejemljivosti proizvoda in razbremenili delavca ter ga osredotočili samo na proizvode s povečanim tveganjem. Optimum smo dosegli pri 10 glasujočih sosedih. S statistično analizo merilnega sistema za atributne karakteristike smo na osnovi 1.700 posnetih merilnih ciklov na več kot 1.500 različnih laserskih izvorih, izdelanih po 5 indeksih tehnološke dokumentacije, potrdili, da je verjetnost pravilne razvrstitve vsaj 80 %, napačne 5 %, ostalo pa so bili lažni alarmi. Z večanjem števila posnetih profilov, podatkov o že pretečeni življenjski dobi in, na osnovi preteklih izkušenj temelječem posodabljanju razvrstitev v podatkovni bazi, se zanesljivost informacij in uspešnost kibernetskega sistema s ključno vlogo človeka v krmiljenju delovnega procesa pomembno povečuje. Spoznanja, dobljena z uporabo predstavljenega sistema, odpirajo možnosti razvoja novih in izboljšanih proizvodov, predvidljivo vzdrževanje in diagnostiko. Pričakujemo, da bo s podobnim pristopom možno razviti sisteme za podporo odločanju tudi v drugih okoljih, kjer imamo opraviti z velikimi množicami podatkov, ki se sočasno spreminjajo in pri katerih so informacije zakrite v šumu.

Ključne besede

profil laserskega žarka;časovne vrste;ekspertni sistemi;selektivna laserska trabekuloplastika;zanesljivost;disertacije;

Podatki

Jezik: Slovenski jezik
Leto izida:
Tipologija: 2.08 - Doktorska disertacija
Organizacija: UM FOV - Fakulteta za organizacijske vede
Založnik: [F. Koplan]
UDK: 617.7:66.08(043.3)
COBISS: 286841088 Povezava se bo odprla v novem oknu
Št. ogledov: 2151
Št. prenosov: 100
Ocena: 0 (0 glasov)
Metapodatki: JSON JSON-RDF JSON-LD TURTLE N-TRIPLES XML RDFA MICRODATA DC-XML DC-RDF RDF

Ostali podatki

Sekundarni jezik: Angleški jezik
Sekundarni naslov: A Model of a Robust Production Process of Ophthalmological Therapeutic Laser Systems
Sekundarni povzetek: Medical devices for selective laser trabeculoscopy with installed short Q-switched Nd:YAG laser sources with changing laser-beam profile and unstable output energy during operation cycles are noncontinuous, complex, dynamic, multiresponse, nonlinear and temperature dependent high-tech systems, with no suitable models. To fulfil high expectations for long-term quality and reliability it is necessary to follow key process parameters, product characteristics and to ensure fast feedback control loops. Human in the process still plays an important role during assembly of mechanical, optical and electronic parts and their adjustment in small-scale production. Expert system suitable for use in specific working environment of laser light with variety of wavelengths was developed to improve the process efficiency and efficacy. This way working procedure was adjusted to human cognitive capabilities and limitations arising from use of safety goggles with decreased visual light transmission, and several risks were decreased. Main research was focused on development of an expert system for processing and measurement of the dissimilarity of the laser beam profile images for selected test cycle sequences, verification and validation of algorithms. The presented computer vision based decision support system successfully recognised real and potential early failures of pulsed Nd:YAG laser sources. Statistical analysis was used to verify a CCD-camera based laser beam profile measurement system capability and observed beam parameters were defined. Database for sequences of laser beam profile parameters and images was established and later updated with expert knowledge as well as with information gained during product life cycle. Profiles were presented as multivariate time series and parameters were pre-processed to ensure efficient similarity comparison of sequences during test cycles. Index of dissimilarity between selected elements from the database was calculated by a method of dynamic time warping. Algorithm parameter selection considered specific properties of the observed pulsed laser sources. It was observed that characteristics changes as a consequence of technical changes led to the fact that at least 8 laser-beam-profile parameters required simultaneous monitoring. Distance weighted nearest neighbour algorithm was used to sort laser sources by similarity and a threshold value for classification was based on correlation of the algorithm result with time to early failure. Calculated probability-indicator for early failure of the system was used to make decision about acceptability of a product easier, relieved and enabled the worker to focus on products with higher failure risks. Optimum was achieved with 10 voting neighbours. Statistical analysis of the measuring system for attribute characteristics, based on 1700 recorded test cycles and more than 1500 different laser sources made in accordance with 5 various indexes of technological changes, confirmed that proper classification effectiveness was at least 80% and miss rate below 5%. All other cases were false alarms. With increasing number of profiles in the database, information gained during life cycle and database updates based on past experience the information quality showed an important increase. An insight into development and changes of selected parameters of laser beam profiles during observed sequences was enabled by analysis of dynamic time warping in multidimensional space. Related graphical result presentations, stored images and voting results were provided by means of dedicated user interface for workers and subject matter experts. Cybernetic system with human-in-the-loop was improved by adding new knowledge and regular updates of the database. Experience gained with usage of presented system opens possibilities for development of new and improved products, predictive maintenance and diagnostics.
Sekundarne ključne besede: laser beam profile;time series;expert system;selective laser trabeculoplasty;reliability;
URN: URN:SI:UM:
Vrsta dela (COBISS): Doktorska disertacija
Komentar na gradivo: Univ. v Mariboru, Fak. za organizacijske vede
Strani: 137 f.
ID: 9160640