The Industrial Internet Consortium (IIC) has mapped out a “continuum” of edge computing in industrial settings, with the ‘edge’ variously located at certain points of remove from the data source depending on the technical requirements of the use case. For simplicity, it summarises four common placements for edge compute functions.
Its white paper, IIC Introduction to Edge Computing in IIoT, defines the ‘edge’ as a logical layer, rather than a specific physical divide. Its authors describe a “continuum of fundamental capabilities” for industrial internet-of-things (IIoT) solutions, which determine the ultimate placement of edge processing within the system.
“The definition of edge moves depending on your business problem,” says Todd Edmunds, one of the report’s co-authors, and senior solution architect at Cisco.
“There is no hard-and-fast saying that, ‘this is the edge’ – because you might be gathering and controlling data with an edge-compute temperature controller, and you might aggregate data in a single cloud instance from 35 factories around the globe, and then the edge becomes those factories themselves.”
Indeed, the IIC’s four IIoT edge scenarios – at device level to measure pump temperatures, at plant level to monitor machine performance, at the factory perimeter to optimise supply chain processes, and at enterprise level to predict equipment failure (as described below) – are only a snapshot of a layered decentralised compute architecture.
“The edge could be multi-layered. Some of those layers at the edge may look like many data centres at a big factory; it could start with something substantial like an oil rig, or something of that nature, and filter down to other smaller edges at a facility,” says Lalit Canaran, co-author of the report, and vice president at software firm SAP.
The white paper will be followed by a technical guide to offer practical advice on IIoT edge deployments. Edmunds says commonalities will be brought to bear on industrial use cases, to simplify the variety of edge placements.
“There will be similarities between use cases, no matter where you are on the continuum, and we will start to identify those, and build them into different models. No architecture will be exactly the same, but these similarities will mean [edge architectures] can be deployed with only minor changes,” he says.
1 | EQUIPMENT PROTECTION
“In this scenario, a “dumb” thermocouple measures temperature on a pump. A pump with
edge computing capability can perform basic analytics to determine if a defined threshold is exceeded and shut the pump down in milliseconds. There is no decision latency and no need for connectivity to perform this function.
“Connectivity is not necessary, but it may be used for notification. The time value of the temperature information decays rapidly as delayed response can result in equipment damage. In this case the edge is at the device level as it can achieve the key objective, even if connectivity to higher-level systems and networks are interrupted.”
2 | PERFORMANCE MONITORING
“The performance of equipment and production lines are often expressed through performance indicators like Overall Equipment Effectiveness(OEE). Near real-time analytics on multiple data points from sensors in the plant area can be processed on a local gateway and provide OEE trends and alerts to operational systems or personnel.
“In this case, the fundamental capability requires information from multiple equipment sources to perform simple analytics. The time value of information is high as response delays waiting for decisions from the cloud can cause significant losses. This business problem suggests that the edge is at the plant area level.”
3 | SUPPLY-CHAIN OPTIMISATION
“Optimising supply chain processes for a local facility, factory or an oil field requires data from multiple sources at short intervals to apply optimisation algorithms and analytics that will adapt supply-chain plans in business systems such as SCM or ERP. The fundamental capability requires local or factory-level connectivity with decisions made in hours.
“Additional information outside the perimeter of the factory may be useful, but not mandatory for effective optimisation. In this instance, the edge is at the perimeter of the factory, plant or local facility.”
4 | PREDICTIVE MAINTENANCE
“Machine learning models to predict Electric Submersible Pump (ESP) failures require data from multiple offshore platforms. The analytics models are complex and a large amount of data is needed to train and re-train the models. It also requires regular data feeds from operating ESPs to determine each unit’s remaining useful life. The data from individual ESPs need to be analysed regularly but information decay is much slower than in the other scenarios and decisions can be taken daily or weekly.
“Computation is typically performed at the enterprise level using a public or private cloud and is at the top end of the edge continuum. The edge can be anywhere along the time-value graph as these examples illustrate. It is ‘where’ data for sensors is used to achieve a specific key objective or address a specific business problem.”
