@conference{
author = "Miletić, Milan and Radonjić, Aleksandar and Sibinović, Predrag",
year = "2023",
abstract = "The work that will be presented in the rest of the document deals with the application of artificial intelligence AI, neural networks, machine learning on machine maintenance, which is a key resource for production in industry. It is a specific machine that must not have an interruption longer than 30 minutes during one shift. Due to the specific nature of the job of inserting fresh air into the blast furnace, the machine must work continuously during the entire furnace operation campaign. This campaign can last up to 12 months. By looking at the situation before the introduction of AI into the system, it was established that the stoppage is mainly caused by damage to the rolling bearings, which are the basis for starting the fan turbines. Further research led to the startling conclusion that bearings ran shorter when they were more lubricated than when they were not lubricated at all. Based on these observations, it was decided that it is necessary to create a program that will collect data on the sensors and based on this data, create an AI that will decide when and how much it is necessary to lubricate the bearings. The advantages of the system are related to the application of algorithms that significantly improve the efficiency of the software in the maintenance application, which significantly reduces the downtime of the machine, and increases its timeliness, availability and efficiency. The method of learning with incentives was applied. The program receives data from the sensors (pressure, temperature, vibrations and ultra sound), then performs an action on the machine via the actuator. The machine returns feedback via sensors to the program, which corrects the settings depending on the results (good or bad). The goal is for the program to learn during operation to have as high a percentage of good results as possible. Due to the complexity of the machine, there are limited limit values in the program, so that the program cannot cause damage to the machine during learning. The research results are presented using statistical methods in the paper.
Specifically, the paper deals with the application of the Convolutional neural network CNN. The data measured on the sensors are sent to the database located on the server. The program groups this data and selects them based on the results - good and bad. The data is then used to train the network and create an optimal algorithm that, with its timely actions, should extend the service life of the rolling bearings on the machine, which is a key resource for the complete production of the factory. Based on the learning, the AI can generate reports based on which the procurement and replacement plan of critical components can be planned. By using the mentioned solution, the service life of the rolling bearings was increased by 20%, while the emergency outages of the plant were reduced to 0. The advantage of the used solution is reflected in high timeliness, availability, reliability, since there were no emergency outages since the implementation of the mentioned solution., Rad koji ce biti predstavljen u nastavku dokumenta bavi se prime-nom veštačke inteligencije AI, neuronskih mreža, mašinskog učenja na odrzavanju mašine koja je kjucni resurs za proizvodnju u industriji.Radi se o specificnoj masini koja u toku jedne smene ne sme imati prekid veci od 30min. Zbog specificnosti posla ubacivanje svezeg vazduha u visoku pec masina mora da radi kontinulno tokom cele kampanje rada peci. Ova kampanja moze da traje i do 12 meseci. Sagledavanjem stanja pre uvodjenja AI u sistem ustanovili smo da do zastoja uglavnom dolazi zbog ostecenja kotrljajucih lezajeva koje su osnov za pokretanje turbina ventilatora. Daljim istrazivanjima dosli smo do zapanjujucih zakljucaka da su lezajevi krace radili kada su bili vise podmazani nego kada uopste nije ni bilo podmazivanja. Na osnovu ovih zapazanja odlucili smo da je potrebno napraviti program koji ce vrsiti prikupljanje podataka na senzorima i na osnovu ovih podataka uraditi AI koja ce odlucivati kada i koliko je potrebno podmazati lezajeve. Prednosti sistema se odnose na primenu algoritama koji znatno poboljšavaju efikasnost softwera u aplikaciji održavanja čime se znatno smanjuje vreme otkaza mašine, a povećava njena ažurnost, dostupnost i efikasnost. Primenjena je metoda učenja uz podsticaje. Program prima podatke sa senzora (pritisak, temperatura, vibracije i ultra zvuk), zatim preko aktuatora vrši akciju na 97mašini. Mašina vraća povratnu informaciju preko senzora programu, koji koriguje podešavanja u zavisnosti od rezultata (dobri ili loši). Cilj je da program tokom rada nauči da ima što veći procenat dobrih rezultata.
Zbog složenosti mašine u programu su ograničene granične vrednosti tako da program ne može da prouzrokuje oštećenje mašine prilikom učenja. Rezultati istraživanja prikazani su statističkim metodama u radu.
Konkretno rad se bavi primenom neuronske mreze Convolutional neural network CNN. Podatke izmerenih na senzorima salju se u bazu podatka koja se nalazi na serveru. Program grupise ove podatke i selektuje ih na osnovu rezultata dobri I losi. Podaci se zatim koriste da se izvrsi ucenje mreze i napravi optimalan algoritam koji ce svojim pravovremenim akcijama treba da produzi radni vek kotrljajucih lezajeva na masini koja predstavlja kljucni resurs za kompletnu proizvodnju fabrike. Na osnovu ucenja AI moze da vrsi generisanje izvestaja na osnovu kojih se moze planirati nabavka I plan zamene kriticnih komponenata. Upotrebom pomenutog resenja radni vek kotrljajucih lezajeva je povecan za 20%, dok su havarijski ispadi postrojenja svedeni na 0. Prednost upotrebljenog resenja ogleda se u velikoj azurnosti, dostupnosti pouzadnosti posto nije bilo havarijskih ispada od implementacije pomenutog resenja.",
publisher = "Beograd : Fakultet za poslovne studije i pravo Univerziteta Union - Nikola Tesla : Fakultet za informacione tehnologije i inženjerstvo",
journal = "Zbornik apstrakata - Osma međunarodna naučna konferencija Pravo, ekonomija i menadžment u savremenim uslovima - veštačka inteligencija (AI) - LEMiMA 2023.",
title = "Application of artificial intelligence in factory maintenance",
pages = "169-172",
url = "https://hdl.handle.net/21.15107/rcub_dais_14757"
}
