The concept of digital twins was born from the marriage known as cyber-physical systems. The cyber representation of a product or process was often held digitally within CAD/CAM or PLM systems. These became linked to the physical object through a feedback loop that kept the two in sync.
Digital Twin has moved from the esoteric to mainstream within industrial culture. And digital no longer is consigned to drawing databases, as my recent conversation with Michael Kanellos and Perry Zalvesey of OSIsoft reveals.
They described the process this way, “From devices all the way to buildings and factories, we’re now living in a world where everything is connected. And as these operations become more connected, it’s increasingly important to identify the strongest solution to monitor them. With the introduction of IoT, sensor and even AI technology to industrial operators, there’s been a surge of unfamiliar digital strategies – the latest being digital twins.”
OSIsoft prefers to consider digital twin as a loose term, as it can be either a complete network doppelganger or just a copy of key data streams to narrow in on specific issues. Everyone has their own preference and iteration.
OSIsoft named its digital twin technology the Asset Framework, which allows companies to take a project-by-project approach, creating solutions for each need on a rolling basis.
When one of its customers, DCP Midstream, began deploying OSIsoft’s AF tool it rolled out 12 AF based applications in two months, experiencing a $20-$25 million one-year return.
Application of OSIsoft’s Asset Framework has been strong in the water industry. Zalvesey says that his first work in the area was with modeling processes that were only static models. Today’s digital twins are dynamic. Designers can model the facility and objects within it. Each object has attributes that data are then associated with. Where originally there was a pump object—say we define “Pump 12” and associate data such as temperature and pressure and more. Now with Asset Framework, designers can create a template class “pump” and be able to replicate for as many pumps as a facility contains.
1. Asset Framework is the core digital twin offering. It’s as a relational layer on top of PI that combines all the data streams (temp, pressure, vibration) of an asset into one screen. A lot of people get fancy with the digital twin term but to us it’s a simulation combined with live data.
2. A simple AF template for a pump probably takes a half an hour to build. It can then be replicated ad inifinitum. It’s a drag and drop process. AF is part of PI Server (it was a separate product years ago but combined into it.) Complex ones can take months. Element, a company that OSIsoft helped incubate (and has since culled investment from Kleiner Perkins, GE and others) has built a service called AF accelerator. Basically, they parachute a team of data scientists to study your large assets and then develop automated ways to build AF templates for complete mines or offshore oil platforms. It still takes two months or so but they can streamline a lot of the coding tasks. BP used them.
- DCP. In 2017, the company launched an effort to digitize operations. One of the first steps was using PI to collect the data and use AF to create simple and complex digital twins. DCP has 61 gas plants for instance. Each one has been modeled with AF. Plant managers are show a live feed of current production, idealized production, and the differential in terms of gas produced and revenue. DCP discovered that it could increase production per plant on average $2000-$5000 per day, or millions a year, by giving the plant managers better visibility into current production and market pricing. In year one, it saved $20=$25 million, paying off the entire project (including the cost of building a centralized control center in Colorado and staffing it.) The next year (2018) it saved another $20 million.
- MOL. One of the largest uses of AF. MOL tracks 400,000 data streams and has 21,000+ AF instances based on 300 templates (a single template can be replicated several times.) MOL says that it has added $1 billion EBITDA since 2010 by using its data better. With AF, for instance, they figured out why hydrogen corrosion was exceeding the norm. In some instances, they’ve used advanced analytics—an experiment to see if it could use high sulfur crudes required deep analytics—but most of the time MOL has made its improvements by creating AF templates, studying the phenomena and taking action.
- Colorado Springs. Complete opposite end of big. It’s a small, regional utility.
- Heineken uses AF to model its plants to reduce energy. Aurelian Metals used it to boost gold extraction from ore from 75% to 89%. Michelin saved $4 million because AF let them recover more quickly from a previous outage. Deschutes Brewery meanwhile boosted production by $450K and delayed a plant (per our 2018 meeting.