Big data for predictive maintenance of industrial machinery

Abstract The operation of industrial manufacturing processes can suffer greatly when critical components fail suddenly. Large manufacturing processes can have plenty of critical components whose failure can interfere with the process operation. Typically these parts are changed periodically according to preventive maintenance strategy. Industry is eager to move towards predictive maintenance in order to make savings in spare parts and lower downtime. Predictive maintenance requires several measurement campaigns from a single part in order to make a working model or finding condition thresholds. A single measurement campaign from a certain part can take lots of time and give limited information about developing condition in certain environment. Multiplying the amount of this measured data leads to a more reliable estimate for the aspects affecting the condition and thresholds. The idea is to gather condition monitoring data from several similar machines or machine parts from a wide range of different environmental and stress conditions. This data can be used to generate models for several varying fault types. Data used for this system can include condition monitoring data from the target, automation system data describing operating conditions, metadata for describing environmental factors and maintenance reports in standardized form, including pictures of faults and events