Making maintenance audits digital for railway stations

Context and challenges

Our client is a major player in France's railway industry. The network, and more specifically the stations, have to face challenges concerning maintenance activities, especially in terms of the collection and export of data. Documents produced by technicians are heterogeneous in their format and highly diverse in terms of classification.

The client needed a solution to aggregate and export station maintenance data to provide greater visibility on the state of infrastructures and to plan future maintenance operations ahead of time.

As part of an innovation project, Assystem proposed a solution for this client to optimise its audits: a text mining and machine learning solution that could explore maintenance documents and automatically extract information from these documents.

Project scope

Preparation and deployment of the Assystem Smart Editor application - a web-based application that handles all document formats (.doc, .pdf, mail, etc.) and delivers the following functions:

Prediction of field of document application by a machine learning algorithm (IFTE - Fixed Electric Traction Equipment, signalling, buildings, cross-functional) prediction of the sub-domain / assets (overhead power line, switching points, etc.). This function enables the user to view the number of documents for each location by type of domain, to deliver a report on asset management for each station.
Automatic extraction of main document theme: maintenance, scheduling, organisational structure etc. The documents injected are automatically indexed using several models of machine learning.
Recognition of entities named in the documents (locations, dates, people, domains etc.)
Identification of sentences / paragraphs containing incident descriptions.

Client benefits

Secure platform enabling multiple users to log on simultaneously.
Audit processing lead time shortened to one week for all stations (previously the extraction of data required for the audit needed one month of document analysis for each station).
Reduced loss of data as each document is collected, processed and stored in the application. This enabled the introduction of a maintenance document governance policy.
Anticipation of work to be done in each station through the automatic retrieval of events recorded in documents and incident reports.