Predictive Maintenance

High value capital equipment in industries such as utilities, oil & gas, manufacturing and mining are in general highly instrumented with sensors which measure at frequencies up to and beyond 1 Hz. This equipment is generally not deployed in isolation but is instead part of a network which works together to deliver a business outcome (generation and delivery of electricity, manufacture of X items per day etc). A defining characteristic of such networks is that non-optimal performance or even total failure of a single piece of equipment has a negative effect on the overall performance of the whole network.

Examples of such networks are the electrical grid (generating equipment, substations, transformers, lines all connected together in a redundant mesh), an oil field (downhole pumps, surface pumps, wells, gas-oil separation plant, injectors, valves and storage tanks which all work together to deliver oil production), a rail or road network etc. For example a wind turbine nacelle has approximately 2000 different sensors reading macro parameters (power output, rotor speed etc.), internal parameters (temperature of cooling fluids, speed of individual gears in the power train etc.) and environmental parameters (wind speed & direction, air pressure, temperature etc).

The data captured by these sensor networks provide insight into the historical and current performance of the equipment. As a result a specialised class of time-series databases have been developed to efficiently store and make query-able this sort of sensor data. These are generally referred to as historians (eg OSISoft’s PI Data Historian, Aspentech IP.21, GE Proficy and others) as they are mainly focused on dealing with long-term storage and analysis of sensor data and are usually used by domain specialists who are motivated to learn these specialised tools. Business Analytics tools such as TIBCO Spotfire® focus on the democratisation of data analytics making it easy for the more business-focused user to analyse data, finding trends and generating statistical models. They may also provide stronger data governance capabilities which is important in the enterprise. Streaming Analytics tools such as TIBCO StreamBase® can be used to apply the heuristics discovered by the citizen data scientist using the business analytics tools to the data being collected in real-time by the sensor network and thus identify optimisation opportunities or incipient failure scenarios which can be remediated immediately. This can increase the overall utilisation or performance over time of the network of equipment being monitored and thus improve business outcomes (more stuff produced for less cost).