Monitoring the serial console logs of JINR MLIT MICC servers has always been a pressing issue. To reduce the response time to failure, the classification of a large number of lines, which is a routine and labor-intensive process, must be repeated several times a day. The solution to this problem was to create special regular expressions that allowed separating logs with errors. However, such an approach resulted in a multitude of false positives. The problem was settled by creating a neural network model, LOGmon. It enabled to increase the percentage of log recognition, while significantly reducing the number of false positives. The application of this model in the LITmon monitoring system will enhance the level of server reliability by providing timely warnings about possible emergency situations. The report presents a neural network algorithm and the implemented neural model that allows creating an alternative to the typical algorithm on top of regular expressions used in the JINR MLIT MICC.