“Mr. Wallace sees, at a glance, that he may as well try to find a lost shell on a sea-shore, or a needle in a haystack, as attempt to discover what he is desirous of picking out of this documentary chaos.”
— Charles Dickens, Household Words (Vol. 2)
A typical industrial plant is made up of many different physical units such as furnaces, compressors, and pumps. These units can be divided into smaller components such as heat exchangers within furnaces, which can then be subdivided further and further until you have the base components of valves, tubes etc. It is at this low level that most sensors operate, reporting the pressures within each pipe, the temperature of each component and the status of every valve. It is usual to have dozens if not hundreds of sensors monitoring every aspect of a plant unit, which leads to thousands upon thousands of sensors monitoring the plant as a whole. This fine level of detail is the most accurate (and possibly only) way to measure a plant in the physical sense, but generates so much data that is presents challenges that are rarely addressed satisfactorily: How do I monitor my plant? What’s important? Where do I look?
It is clear that the number of sensors is so large there is no way to monitor each one, so often this is approached as a simple prioritisation problem. Find the most “important” sensors for each asset, decide that’s still too many to monitor, so pick your favourite and hope for the best.
The wealth of rich sensor data that is collected is then relegated to the role of disaster analysis. That is to say, something went wrong, why? After a failure process engineers go through the recorded sensor data for the failed asset to determine what went wrong, often finding that if they had been monitoring the right sensors the fault could have been identified in time to do something about it before it failed.
In an ideal world you would be able to monitor all of these sensors, identifying the early signs of failure without needing a team of people to stare at thousands of sensor graphs day after day. Sabisu has developed a way to do just that.
Metasensors® are a higher level sensor construct that combine a number of real sensor time series into one. This is done in such a way that mathematically guarantees the minimum loss of information, ensuring that the important information is encoded in the metasensor®. Rather than the base level covered by physical sensors, metasensors® can be built at the asset level, with a metasensor® for a whole heat exchanger made up of its constituent components.
Using this technique it is possible to effectively monitor a whole plant. With metasensors® operators can monitor the plant at the asset level, and be quickly directed to the important sensors when something goes wrong. Sabisu’s system goes one step further, and actually performs the monitoring for you! This is achieved through Anomaly Detection – which we’ll tell you all about in our next post.
We’re always interested in hearing from you with any comments or suggestions, feel free to get in touch.