Who Looks at a Billion Measurements?

We hear how the ‘Internet of things’ (IoT) will connect billions of devices using digital networking. Most of these ‘things’ will be sensors; many of these ‘things’ will have multiple sensors in them. So who will look at a billion of measurements? Who can digest all that input? To paraphrase an old saying; we don’t want to be drowning in PVs and starving for wisdom. The answer is that we need analytics software to distill multiple raw data points into information, knowledge, and ultimately leading to greater wisdom. Modern plants are already installing thousands of more sensors than plants in the past and are also deploying software to analyze the raw data helping plants detect abnormal condition of equipment and abnormal energy consumption etc. Here are my personal thoughts:

Modern plants are being built with digital networks for sensors and actuators. Existing plants are being modernized with wireless sensor and actuator level networks. This is partly because new processes are more complex to handle inferior feed and to meet more stringent product requirements therefore requiring more automation. More sensors are also required to automate data collection associated with equipment condition and performance monitoring. That is, digital networks go deeper than ever before. Digital networks enable a drastic increase of the number of sensors in plants because the wiring cost and I/O system cost for fieldbus and wireless is lower.

Thousands of additional sensors mean thousands of additional data points in a plant. Moreover, all control valves provide real-time position feedback and all electric actuators provide torque, temperature, and other information. Fortunately, people need not monitor these additional data points. It is mostly used by analytics software to extract useful information, knowledge, and ultimately wisdom as per the DIKW pyramid model in the illustration above. It is shaped as a pyramid to illustrate how lots of raw data is distilled into smaller amounts of information and so on. For instance, many process variables (PVs) which individually mean little or nothing might be used to extract a single piece of information which is easy to understand and act upon.

There are two kinds of data analytics; real-time data analytics and historical data analytics. Data from thousands of sensors collected and stored over a long period of time results in a lot of data; referred to as Big Data. Big Data analytics is thus after the fact, but can uncover insights from historical data pattern not visible from real-time data, particularly when correlated with other variables that might at first appear to not be related. Real-time analytics looks at the data immediately to detect issues. Examples of real-time analytics include detecting pump issues, steam trap failures, and energy overconsumption etc.

The sensors automatically collect the data and does so much faster than using traditional manual data collection methods such as reading gauges, or using portable IR temperature guns, acoustic tester, and vibration tester etc. Data is sampled once every second, minute, or hour instead of once a day, week, or month. Faster collection means developing issues are detected earlier, before outright failure. However, there may not be sufficient skilled personnel around to interpret all the collected data. Therefore the raw data must be distilled into information.

As raw data is distilled into information using analytics algorithms it percolates up the DIKW pyramid. For instance, inlet and outlet temperature signals for product and steam/coolant may be used to compute the duty of heat exchanger bundles to know how fast they are fouling. That is, six or so dynamic data points distilled into a single piece of information. Fourier transform on vibration signal dataset is another example. Moreover, since the monitoring is done automatically by the software, people never see most of this data. Only when there is a deviation from normal, such as high vibration, overconsumption, or low level will a human be prompted. In the information sciences this referred to as reducing the entropy of information, meaning reducing the disorder of the information. I wrote a paper for the ISA show in 1997 which touched on this relation between data and information: “Fieldbus - Key to Diagnostics and Efficient Instrument Management”

https://www.isa.org/store/fieldbus-key-to-diagnostics-and-efficient-instrument-management-isa-tech-1997/123885

Data analytics can be embedded in a device or it can run on a computer somewhere in the plant, or on a virtual machine in the cloud. Multi-variable analytics on data from multiple sensors may done on a computer and cloud analytics allows comparison against data from multiple similar plants.

The plant historian will play an ever increasing role in the operations & maintenance of plants going forward. Since many of the additional sensors now being deployed in plants are not related to process control, they will not connect to the control system. Most of these sensors are related to maintenance and energy efficiency etc. and will therefore connect directly to the equipment condition and performance monitoring applications and energy management applications used for these purposes. Many of these applications will take their data from the plant historian.

However the historian in most aging plants today does not have all the raw data input signals required to analyze equipment condition and performance. These are referred to as ‘missing measurements’. Therefore existing plants are now deploying hundreds and even thousands of additional wireless sensors to cover these missing measurements and get them into dedicated software applications and the analytics algorithms through the historian. That is, operators are not flooded with thousands of additional PVs. Instead raw data is distilled into information and knowledge, and most of it is used by reliability and maintenance engineers as well as HS&E officers.

FOUNDATION fieldbus and WirelessHART form the basis for the Industrial Internet of Things by digitally networking the transmitters, analyzers, and valves around the plant. Access to data from these sensors may one day be given across the Internet; so called the Industrial Internet of Things (IIoT) although most plants are not doing it yet. Software applications are key to converting the raw data into actionable information, knowledge, and wisdom. See further explanation of these applications in this article: Instrumental to Success

https://www.ceasiamag.com/2015/04/instrumental-to-success/11137/

Well, that’s my personal opinion. If you are interested in how the digital ecosystem is transforming process automation click “Follow” below my photo above. Let me know what you think by providing your feedback below.