So let’s say we have a big datalake of high quality oil & gas or petrochemicals data. What represents ‘good’ and ‘bad’ performance?
If we were a marketing company, we’d invest in tag management software, with the aim of improving the quality of our analytics, but also improving other areas; key operational tools such as web site performance, taking the IT department out of the iterative marketing loop, standardisation, cost reduction.
Marketing teams have plenty of metadata but it relates primarily to a one-dimensional system. They’re interested in segmenting markets and tuning communications to achieve a larger degree of engagement. The industrial challenge is in applying algorithms to a massive multi-variate system where tag management isn’t going to help much beyond simple categorisation. Often you’ll see solutions using tags to identifying groups of like parameters; tagging all the data from a particular location as, say, electricity in kWh, or feeds to furnace D1092A.
This approach misses two key points when applied to industrial data: the datalake contains a lot of structured data, and there’s lots of metadata available already.
At Sabisu we specialise in the process industries like oil & gas, petrochemicals, or manufacturing. Within those industries the basic tagging job is already done; there are usually well configured, well trusted DCS and historians which describe pretty well what parameters relate to what parts of the system.
So if a group of Naphtha feed tags are all related to furnace D1092A, they’ll be called something like D1902A_N_FEED.PV. There’ll be plenty of other instrument reading tags relating to that furnace; feed preheat, dilution steam flows, alarm suppression indicators, coil output temperatures, coil output pressures and so on. There’ll be plenty of tags that provide context too; feedstock type or selection, trip/alarm overrides & disarms or plant running modes.
This is valuable data. It’s not just process data but descriptive, guide or technical metadata because of the context it provides.
So we have a head start compared to other sectors. We’re in the most heavily instrumented sector of all, which means our algorithms can take this metadata into account.
What do we do with this metadata? Well, things like this.
We hope that’s useful. As always if you have any questions or suggestions, head over to our LinkedIn group and share your thoughts.
