Supervisory Control and Data Acquisition (SCADA) systems are nowadays widely used in modern industry. Their utility has been proven over the past decades to supervise any automated processes. The scope of applications of such a system has been extended from industrial to infrastructure and facility processes. Classic SCADA systems offer tools to help monitoring and analyzing data from this type of systems, but the existing approaches are often limited to Data Warehousing and Reporting. The main problem of these solutions is that they are not designed to work on online mode and can be seen as batch processing on archived data. In this thesis we study how to link machine learning solutions to SCADA systems on large infrastructures. The solution we propose in this work is a complete framework with its architecture providing a ’toolkit’ with different modules to build online solutions on top of a SCADA system.