In the 2000s, EDF's hydropower sector launched a project to modernise its Information System, including national harmonisation of nomenclatures. This new indexing, ranging from the regional scale to the details of the sensors installed on our equipment, now makes it possible to create a tool for handling all raw data previously unimaginable on the scale of the national hydraulic fleet.

The various databases supplied (production time data, current and exceptional maintenance reports, operational contingencies, etc.) remain however always stored on different media, not designed to interact together.

EDF R&D has therefore produced a tool for the joint cross-referencing and display of heterogeneous data in order to demonstrate the additional added value that can be derived from the concomitant use of data of different kinds. This new tool primarily targets 2 types of data:

• Hydropower plant process data, i.e. time series from the sensor fleet present at each production site.
• Maintenance data, essentially containing text data from structured forms containing irregularly filled in free fields.

In practice, the tool allows, on the one hand, to display several time series of sensors between two dates, and on the other hand, to superimpose on the same graph the relevant maintenance acts, linked to the displayed sensors. We then use the efficient structuring of sensor nomenclatures and maintenance data to sort by relevance the maintenance events to be displayed first, based on expert knowledge

The main actor benefiting from this tool is the e-monitoring of the hydraulic fleet. The engineers of the regional e-monitoring centres remotely supervise the monitoring of a whole fleet of machines in order to detect slow drifts or the crossing of alert thresholds. On the contrary, the operational units manage the maintenance operations directly from site. This decoupling leads to a phase shift between the e-monitoring cells, which do not have a perfect knowledge in real time of the work performed, whereas these works, by affecting the behaviour of the machines, generate many alerts or events wrongly interpreted as suspicious. This situation generates numerous telephone exchanges between maintenance and e-monitoring teams in order to explain the changes in trend induced by maintenance operations.

The tool developed aims to reduce the number of exchanges by making e-monitoring cells more autonomous in interpreting abnormal behaviour detected by direct access to contextual information of primary importance: maintenance operations carried out on sites. E-monitoring teams will also be able to build statistical indicators related to maintenance (time before new failure following repair, warning signals before failure, etc.) using historical data.