Influential factors modelling and injury prediction in mining

[J. Ivaz]

VII, 132 str.

Tehničko-tehnološke nauke -: Rudarstvo i geologija / Technical-technological sciences - Mining and geology

Rudarska industrija predstavlja pokretačku snagu razvoja privrede, ali ova industrijska grana poznata je kao jedna od onih sa najnepovoljnijim uslovima za rad, velikim brojem povreda na radu i mogućom pojavom kolektivnih nesreća. Povrede na radu predstavljaju nepoželjne događaje na svim radnim mestima, pa se teži da se njihov broj smanji i postigne „nula“ povreda na radu. Događaj povređivanja radnika je vrlo kompleksna problematika i ne postoji jedinstveni model koji objašnjava ovaj fenomen. Predmet ovog istraživanja su povrede na radu, dok je cilj istraživanja definisanje prediktivnog modela za povrede na radu kojim bi se predvidelo i preventivno delovalo na sprečavanje ovakvih događaja. Sprovedenim istraživanjem teorijski je definisan postupak analize uticajnih faktora koji dovode do pojave povređivanja radnika u rudarstvu. Ideja je da se, primenom i kombinacijom postojećih statističkih i analitičkih metoda i postupaka za analizu uzroka i verovatnoće pojavljivanja, napravi model koji bi ukazao na radna mesta sa najučestalijim povredama. Za modeliranje su korišćene tehnike zasnovane na veštačkoj inteligenciji, odnosno veštačkim neuronskim mrežama i na teoriji fazi logike. Primenom neuronskih mreža i teorije fazi logike moguća je obrada podataka o povredama na radu, predvidljivost ovakvih događaja i analiziranje uticajnih faktora na pojavu povreda. Veštačke neuronske mreže predstavljaju moćan alat pri odlučivanju, posebno u rešavanju kompleksnih problema sa velikm brojem ulaznih podataka. U ovom radu analizirana je mogućnost predikcije povreda na radu primenom neuronskih mreža. Istraživanje je sprovedeno u rudnicima podzemne eksploatacije uglja u Srbiji. Ulazni podaci mreže dobijeni su na osnovu upitnika koji je obradilo preko 1150 ispitanika. Nakon analize uticaja ulaznih podataka na izlaz mreže, izvršena je selekcija mreža koje su dale najbolje rezultate. Pored toga, analizom osetljivosti, izdvojeni su najuticajniji parametri. Izdvojeni parametri su direktan indikator problema koji mogu uzrokovati povrede. Značaj predloženog modela ogleda se u jasnim indikatorima za sprovođenje pojačanih bezbednosnih i organizacionih mera u cilju smanjenja broja povreda u rudnicima.

The mining industry represents the incentive for the development of economy. However, the mining industry is known as one with the most adverse working conditions, a large number of injuries at work, and collective accidents as well. Work-related injuries are seen as unwanted events in all workplaces, therefore, the aim is to reduce their number and to achieve “zero” injuries at work. The worker’s injury phenomenon presents a highly complex issue and there is no single model that can explain these events. Work-related injuries were the subject of research for this doctoral thesis and the aim was to define a predictive model for this type of injuries that could anticipate these adverse events and offer a preventive action. The conducted research provided a theoretical frame which would perform the analysis of influential factors that could cause the occurrence of injuries among workers in mining industry. The idea was to use a combination of the existing statistical and analytical methods and procedures to analyse the probabilities and causes of injury occurrence in order to create a model which could identify jobs with the highest number of injuries. Methods used for the modelling were based on artificial intelligence (AI), i.e. artificial neural networks (ANN) and fuzzy logic theory. The application of neural networks and fuzzy logic theory provided a possibility to process data on injuries, predict these incidents, and investigate the influential factors for the occurrence of such injuries. Artificial neural networks (ANN) represent a valuable decision-making tool, especially in addressing complex problems with considerable amount of input data. The possibility of predicting injuries at work using ANN was analysed in this research. The study was conducted in underground coal mines in Serbia. The input data for ANN were obtained based on a survey which included over 1,150 respondents. When the analysis of the influence of input data on output data was performed, the selection of the networks that provided the best results was completed. Moreover, the most prominent parameters were identified by the sensitivity analysis. The highlighted parameters were seen as direct indicators of problems that could induce injuries. The significance of the proposed model is reflected in the fact that it offers useful indicators for the implementation of improved safety and organizational measures that could reduce the number of injuries in mines

povrede na radu, predikcija, uticajni faktori, veštačke neuronske mreže, teorija fazi logike

622:614.82(043.3)

Tekst