Načrtovanje in analiza poskusa za optimizacijo proizvodnega procesa

Založnik, Anamarija (author), Kastelec, Damijana (mentor)

V proizvodnem procesu podjetja so bile zaznane prenizke vrednosti fluksa, ene izmed glavnih lastnosti končnega izdelka, kar je povzročalo povečan izmet. Zaradi nepopolnih podatkov redne proizvodnje kakovostna analiza obstoječega stanja ni bila mogoča. Zato smo si v magistrskem delu zastavili cilj načrtovati in izvesti poskus z namenom ugotoviti vpliv posameznih spremenljivk na fluks na podlagi analize obstoječih proizvodnih podatkov. Na podlagi prvotne analize obstoječih proizvodnih podatkov in izkušenj zaposlenih v podjetju so bile izbrane spremenljivke, vključene v načrt poskusa z ustreznimi kombinacijami: magnetna lastnost materiala 1 in 2, jakost kletke, temperatura orodja 1 in 3 ter temperatura sušilca. Po izvedbi poskusa so bili izdelki pomerjeni, pridobljeni podatki pa analizirani z linearnimi mešanimi modeli, ki poleg fiksnih vplivov omogočajo tudi upoštevanje slučajnega vpliva orodja. Izbira modela je bila izvedena z uporabo elastic net regularizacije s poudarkom na LASSO metodi, s katero smo poiskali statistično pomembne interakcije. Zanesljivost izbire modela je bila dodatno preverjena s stabilnostjo modela. Z analizo načrtovanega poskusa je bilo ugotovljeno, da fluks poveča višja magnetna lastnost materiala 1, višja jakost kletke in višja temperatura orodja 1 ter nižja magnetna lastnost materiala 2 in nižja temperatura sušilca. Med spremenljivkama jakost kletke in magnetna lastnost materiala 2 obstaja interakcija. Spremenljivka temperatura orodja 3 se ni izkazala kot statistično značilna. Linearni mešani model je pokazal tudi pomemben slučajni vpliv orodja, kar nakazuje na obstoj dodatnih virov variabilnosti, ki niso bili zajeti z omenjenimi spremenljivkami. Na podlagi načrtovanega poskusa analiziranega z linearnimi mešanimi modeli so bili ugotovljeni vplivni procesni parametri za fluks. To podjetju omogoča ustrezno prilagoditev parametrov v primeru prenizkih vrednosti fluksa. V zaključku pa podamo tudi smernice za nadaljnjo raziskovanje vplivov.

• Language: Slovenian
• Keywords: načrtovan poskus, linearni mešani model, LASSO regularizacija, stabilnost modela
• Work type: Master's thesis/paper
• Typology: 2.09 - Master's Thesis
• Organization: FE - Faculty of Electrical Engineering
• Year: 2026
• PID: 20.500.12556/RUL-186746
• COBISS.SI-ID: 290514947
• Publication date in RUL: 04.09.2026

Secondary language

• Language: English
• Title: Experimental design and analysis for production process optimization

Abstract

Low flux values, one of the key properties of the final product, were detected in the company’s production process, resulting in an increased rejection rate. Due to incomplete routine production data, a comprehensive quality analysis of the existing situation was not possible. Therefore, the objective of this master’s thesis was to design and conduct an experiment to determine the effects of individual variables on flux, based on the analysis of available production data. Based on an initial analysis of existing production data and the experience of company employees, the variables included in the experimental design were selected with appropriate combinations: magnetic property of material 1 and 2, cage intensity, tool temperature 1 and 3, and dryer temperature. After conducting the experiment, the products were measured, and the collected data were analyzed using linear mixed models, which, in addition to fixed effects, also allow for the inclusion of the random effect of the tool. Model selection was performed using elastic net regularization, with an emphasis on the LASSO method, to identify statistically significant interactions. The reliability of the selected model was further assessed through model stability analysis. The analysis of the designed experiment showed that flux increases with higher magnetic property of material 1, higher cage intensity, and higher tool temperature 1, as well as with lower magnetic property of material 2 and lower dryer temperature. An interaction was identified between cage intensity and the magnetic property of material 2. The variable tool temperature 3 was not found to be statistically significant. The linear mixed model also revealed a significant random effect of the tool, indicating the presence of additional sources of variability not captured by the included variables. Based on the designed experiment analyzed using linear mixed models, the key process parameters affecting flux were identified. This enables the company to appropriately adjust process parameters in cases of insufficient flux values. Finally, guidelines for further research on influencing factors are also provided.

• Keywords: designed experiment, linear mixed models, LASSO regularization, model stability