Text mining of the DU-46 inspection log using frequency-based text processing methods

For making well‑founded decisions on the elimination of failures and malfunctions occurring in railway infrastructure facilities, prompt access to information on previously identified faults and the dynamics of the irresolution is essential. Inspection logs such as DU‑46 contain valuable data on the condition of these facilities (tracks, turnouts, signals, power supply, contact lines, etc.); however, they are hardly used in practice when analyzing the causes of newly emerging failures. Aim. To develop an algorithm for processing DU‑46 log records that allows operators, up on request, to obtain information on previous malfunctions or maintenance activities on specific infrastructure objects. Methods. Text preprocessing, lemmatization using M. Korobov’s morphological analyzer, frequency‑based text analysis, TF–IDF, L2 normalization, cosine similarity calculation, and result sorting. Result. A prototype application has been developed that enables search for relevant records and displays a similarity metric between the query and the retrieved fragments, which, among other things, may serve as a recommendatory function for determining the causes of failures. Conclusion. The use of operation al inspection logs in combination with text mining methods can form the basis for building recommendation systems and decision support systems in the maintenance of railway infrastructure facilities.