Most of industrial factories as well as textile factories interest to increase quality of production, profit and promote along decreasing their expenses. One of the most important effective factors to achieve above goals is favorable function of the producing plants, distinguish the faults and remove them rapidly.Because having identified the faults specially the mechanical ones it would be possible to correct the deficits, find the shafts roller eccentricity and rollers creating periodic faults and decreasing the quality of the threads highly. The study of the problems connected with textile machines and their rollers has formed the basis of previous research work. These problems have attracted the attention of researchers for decades. This investigation has been revealed so much of principle and theory of problems assessment from voice analyzing.The problems are found in quality control unite by Ouster plant so a defined length of the thread passes through the plant and thecurve plant draws the mass changes by which they find the faults that is by virtue of the tests conducted on the products and computations they find the origin of the fault in the plant , but this is not a favorable method because the faults are not often found because of test and computation errors or perhaps it lasts a long time the control section finds the fault so many deficit products are produced because of high speed of the plants led to high expense and low output . So if it is possible to find the find online, the output increases considerably. In this study we try to execute the primary step of finding the faults rapidly through the sounds of the rollers. Five types of rollers were selected in order to distinguish the faults of the rollers out of the center through the sound. They should not select the eccentricity rollers primarily because we are to distinguish the faults out of the center so first they should become sure that the rollers work well by indicator watch. Then the rollers are coated , lathed and four of them of 0.5 , 1 , 1.25 and 1.5 mm are led out of center ; the rollers are put on the machine, their sounds are saved on the disc; the saved data (Time and frequencies ) are given to LVQ network. Some data not present in the network previously are given to it to distinguish the fault(s). The results showed that the time data were not effective to separate the faults, the frequencies were very sufficient in a manner that the categorization by the network was right in 99 percent of thecases