Networked, Intelligent I/O
The Truly Distributed Control Revolution
By : Jim Pinto,
San Diego, CA.
First published: ISA Proceedings '94 and Intech July '95
ABSTRACT : Control technology is being strongly impacted by the availability of low cost silicon chips - processing, memory and communications. Control architectures based on central-host processors (DCS and PLC centralized control architectures) with host-dependent I/O are steadily becoming obsolescent. Networked, intelligent I/O overcomes the complexity constraints of conventional deterministic control systems. Peer-to-peer networking of intelligent I/O allows autonomous agent type control, with performance far beyond conventional architectures.
IntroductionToday, a significant segment of industrial automation systems is I/O processing : measurement and signal-conditioning of a variety of sensor inputs; recording, displaying, trending, alarming and controlling based on the measurements. Final control outputs are typically solenoids, valves, heaters and various other actuators, which are manipulated based on control algorithms related to the inputs.
I/O typically refers to measurements that are "subservient" to the central computer or controller, with minimal "intelligence" at the I/O point. The term "smart" I/O was coined for I/O with at least minimal local memory and processing capabilities, for example, keeping the output at a fixed level when communications are lost. In computer-based systems with remote I/O, control typically cannot be done in the central computer since loss of communications would result in loss of control. Therefore full PID type local controllers are used for closed-loop control, supervised by the remote computer. The computer is relegated to supervisory tasks such as modification or adjustment of set-points and control parameters. This is an expensive alternative, but accepted as necessary.
The two largest segments of industrial measurement & control systems are PLC (programmable logic controllers) and DCS (Distributed Control Systems). Other market segments, such as SCADA (Supervisory Control and Data Acquisition) are hybrid approaches, using PLCs or computers. At the I/O level, both types of systems are similar - clumps of I/O connected via a data highway (bus) to the central system, which itself may be networked in a hierarchy. There may be some local processing at the I/O level, but this is usually focused on multiplexing, and faster/better communications with the central host.
During the last decade, front-end I/O architecture was typically clusters of micro-controlled I/O, also called "block I/O". Today, as much as 50-60% of PLC and DCS sales are I/O related - signifying that the complex and expensive clumps of processing are still at the front end - the I/O. This is precisely where the next major change will occur, with a price/performance revolution putting processing where it belongs - the front end. Concentrating the processing in a central, or even shared processor, will steadily become a thing of the past. Shared processing power was necessitated by considerations of cost/size/complexity. A significant change in those relative relationships heralds the next revolution in control.
Local I/O at or near the computer bus (STD, VME, PC-BUS or MULTIBUS) typically cannot really handle more than perhaps 50-100 channels, because of the sheer complexity of having the terminations all in one place. Practical convenience typically demands remote clusters of 16 or 32 channels connected to the host computer via a serial line. In many DCS systems, all I/O is clustered together in large cabinets, the termination cabinetry typically being physically more bulky and complex than the actual I/O electronics.
A multi-faceted revolution is now occurring in the industrial automation business, with single-point, intelligent, programmable I/O connected directly to the sensor and actuators, accomplishing all the measurement and control functions locally and independently, interconnected by peer-to-peer networking. Not only is dependency on a central processor (PLC or DCS) eliminated, but significant new features, advantages and benefits become available, through software and firmware architectures that were not previously possible.
Deterministic Systems vs Autonomous AgentsConventional control systems were designed to be deterministic, because non-deterministic systems were considered unacceptable. Over the past few years, Complexity (Chaos) Theory analysis is demonstrating that even so-called deterministic systems are in reality non-deterministic (brittle or prone to failure) when complexity increases. Conversely, excitement is being generated by demonstrations that interacting autonomous agent systems on the edge of chaos provide predictable performance - locally unstable, though globally stable and predictable. (Ref. 1 - Morley et al)
A hierarchical control system is subject to complex software constraints, network communication conflicts and network-speed limitations. The upper-limits of I/O count capability for conventional DCS and PLC systems appears to be limited only by data-rates; however, it is almost self-evident that the maximum I/O count is primarily limited by sheer complexity issues and secondarily by cost. The practical I/O count for programmable, intelligent, networked "autonomous agents" greatly exceeds that limit. Intelligent I/O, with algorithmic response mechanisms, has no theoretical complexity limit.
The author theorizes that DCS systems are indeed theoretically limited because of complexity. Several major DCS and PLC manufacturers have been asked directly about the maximum number of I/O controlled by current systems - with a number in the region of 30,000, typically at a cost of several million $. Honeywell has achieved a point-count of 80,000 in the SASOL project in South Africa - but this, in reality, is several inter-connected systems, rather than one truly networked DCS. The author postulates that the practical I/O count for programmable, intelligent, networked virtual host I/O greatly exceeds that limit. Intelligent I/O, with algorithmic response mechanisms, has no theoretical complexity limit.
A host-independent (peer-to-peer communication) intelligent networked I/O system can provide vastly improved performance at a fraction of the cost of conventional DCS. Software in a deterministic DCS or PLC system is complex, and typically the most expensive part of the system, controlled and modified by only an engineering elite, inaccessible to the end-User. Software in a virtual host I/O system is rule-based, consisting of minimal instructions in each node, coordinated and supervised by several user-orientated supervisors.
Autonomous Agent architectures - intelligent, networked, programmable I/O processing - is where the next major change will occur, with significant price/performance improvements putting the processing right at the sensor and the actuator. New plant-floor networks are providing the ability to directly connect all the measurement and control devices, initially with supervisory control (through PLCs and PCs), but eventually allowing totally autonomous control systems. When individual devices become programmable, and re-programmable over the network, the ability to self-organize becomes inherent. The software architectures to achieve "self-organizing systems" is still somewhat theoretical and perhaps esoteric, but practical applications are indeed emerging. The change in the relative relationships between cost/size/complexity heralds the next revolution in control.
Silicon chips, firmware, and miniature packaging developments have yielded a completely new type of I/O architectures which provides significant cost/performance advantages over conventional block I/O. Single-point intelligent, networked I/O modules provide performance equivalent to single-point PLCs (Ref 2 - Pinto). This is exactly the equivalent of small, local, autonomous agents as defined in chaos and complexity-theory (Ref. 3 - Kompass) An integrated system based on this approach yields many advantages.
The ASIC RevolutionOnly recently, software and hardware technologies have combined to make available new, intelligent, programmable I/O products which provide vastly different distributed control architectures, with potential performance that could obsolete conventional host-based systems.
Three silicon-chip developments have contributed to the new I/O architecture, to achieve practical cost/performance/complexity which compares with conventional block I/O based system.
b/ Processing & Communications : LON (Local Area Network) and CAN (Control Area Network) chips are now available from Motorola, Intel and a variety of other sources at a cost of approximately $3, allowing individual devices to be connected to a fairly sophisticated and reliable industrial network. Local circuitry (within the device) can include microprocessor-based intelligence, non-volatile memory (for operating-parameter and program storage) and complete networking protocols and services.
c/ Memory : Cheap silicon memory is now almost a commodity, allowing local parameters (functions such as calibration adjustments and control algorithms) to be retained in non-volatile storage. Processing and communications chips typically include at least minimal memory.
Physical PackagingI/O for conventional DCS systems is typically clustered in large I/O cabinets, with multiplexed signal-conditioning cards connected via a tremendous number of cables to termination panels within the same cabinets. Imagine a system with 1000 sensors - each sensor (2, 3 or 4 wires) would need to be wired to the termination panel at the cabinet, which in turn is connected (within the cabinet) to the multiplexed I/O signal-conditioning system, which in turn may be connected to the central host computer via a I/O data highway or network.
Figure 1 : Intelligent I/O modules, built within a standard terminal-block package.
By contrast, a truly distributed I/O system would include ASIC-chip based electronics directly within a terminal-block housing (Figure 1). The single-point input or output point would be connected directly to the nearest available network access via 2-wires. The vast conglomeration and complexity of wiring is eliminated, providing vastly improved reliability - aside from other less obvious advantages.
I/O NetworksToday, networking of I/O usually means remote I/O connected via serial communications - RS 232, 422 or 485. Clumps of I/O (typically 8, 16 or 32, seldom mixed-signal - analog and digital signals are in separate clusters) are connected to a central computer, which itself may be on a network (such as Ethernet, or an equivalent). The I/O may be connected to a PLC on an independent network or I/O bus. Even the most advanced I/O of this type, operating on its own bus, is still in clumps of like-signals and still relatively dependent on central host-processors (PLC or computer) for operation in a system.
In March '94, at the ICEE in Chicago, two major new networked industrial I/O systems were introduced - DeviceNet by Allen-Bradley, and SDS (Smart Distributed System) by Honeywell Micro Switch. Both were beyond just proprietary product launches and were startlingly similar, in that both were based on CAN - the Control Area Network chip - and both claimed to provide a platform for open and interoperable industrial networking, strongly soliciting third-party vendor involvement and support. (Ref. 4).
Networking on the factory floor is advancing inexorably, because of the many benefits it brings. Networked device installation is much more cost-effective than traditional I/O wiring in most applications, and allows much improved maintenance - a device may be removed or replaced without shutting down other devices in the same system. An open device network will quickly provide common solutions, reducing the need to support a wide variety of devices and networks currently proliferating in the market. Openness provides interchangeability between multiple vendors, something for which users have always been clamoring.
The latest network introductions, specifically DeviceNet, still connects relatively dumb devices to PLC processors, typically in master/slave scanning configurations. However, SDS allows peer-to-peer I/O networking, which eventually allows hostless operation. This will lead eventually, and inevitably, to the development of self-organizing capabilities. When a networked device has intelligence (memory and processing power), the capability to network and be programmed over the network, and perhaps the ability to re-program itself based on the environment, that provides the elements of self-organizing systems.
Fault Tolerant & Fail-safeWith central processing - processing separate from the I/O - the physical connections between the I/O and the processor become important, since the I/O processing is dependent on the separate and remote processor. Several I/O become dependent on a host and the host reliability is critical - if the host fails, everything fails. Conventional PLC-system techniques include redundant-systems and the use of hot backups : manually or semi-automatically switched alternative PLCs which can take over when the primary PLC fails. In a deterministic DCS system, fault tolerance follows arduous and expensive design criteria - with each important component (CPU, power supply, etc.) being backed-up. This is typical of old, brittle system design. Failure is dependent on any number of critical items.
A significant problem occurs when dependent I/O is separated from the central processor. If the connections are cut, the I/O becomes relatively incompetent. Smart I/O supposedly overcomes this problem, but the intelligence is usually limited to holding an output to a predetermined value while waiting for the connection to be re-established.
With the type of intelligent, programmable I/O systems being described, fault-tolerance is achieved by simple duplication (or even triplication) of the measurement and control where it occurs. Loss of communications with the host, or even between I/O points, simply transfers control to a local tolerance algorithm processed independently within the I/O modules. The concept of a fault-tolerant computer becomes irrelevant, when the I/O itself is fault-tolerant. This is a significant advantage. With autonomous agent I/O systems, the system design is robust as opposed to brittle.
Fault tolerant or Fail-safe operation in a system of this type has several connotations - failure of the network connections, or failure of one or more of the network nodes. Network path redundancy is accomplished by connecting a ring topology, providing alternate cable path among the modules, should a physical break in the ring occur. I/O point redundancy can be achieved by adding modules to the network, with an operating algorithm to assure correct operation - for example, three input modules may measure the same temperature, and the input temperature is measured as the average of the three, provided all three are within acceptable limits; when any one is outside a predetermined difference from the other two, the two similar readings are believed and a diagnostic is flagged in the system.
Intelligent I/O Control SystemsIntelligent I/O networks can be designed to be cost effective and high-performance control systems, from a few points up to fairly complex DCS equivalents. The autonomous agent system is most effective when a large number of small groups of I/O clusters are needed to gather data and/or control a process or system, which in turn is part of a much larger enterprise. The techniques of control, safety, redundancy, diagnostics, emergency backup, are much more effective and user-orientated than conventional DCS.
Figure 2 : Control system based on networked single- point, autonomous-I/O modules. The input modules are near the sensors, and the outputs near the actuators.
Take the example of a PID controller - conventionally a single, panel-mounted instrument, with adjustable set-point and control parameters (P-I-D), and displays of input, output and set-point. Typically, the input and output would need to be wired to this instrument. With the networked I/O architecture being described, the input module would be located near the sensor, the set-point module (potentiometer adjustment) would be at a location convenient for the operator, and the output module (perhaps 4-20mA) would be near the actuator (pneumatic valve). There would be no mechanical P-I-D adjustment potentiometers, since these parameters are simply non-volatile memory-adjustments within the output module, adjustable from an operators computer console. The three networked modules (input, set-point and output) would be bound together during configuration of the system, at the operators console. The output module monitors the input and set-point (over the network) and provides the PID control output directly.
When the operator-console computer is dis-connected, the 3 modules which comprise the PID control would continue to work normally, provided they are connected to the network. This provides the complete equivalent of an independent PID controller. To allow for disconnection of the network between the modules the intelligent output module can be programmed to have several backup or fault-tolerant modes of operation :
b/ If the local-set-point module is inoperative, then the output is forced to a pre-determined safe value.
c/ If the input module measurement is disconnected, an alternative input-module can be recognized as a backup.
d/ If there is no input backup, then the output is forced to a pre-determined safe value.
In a networked autonomous I/O system, several physical advantages are available, almost as an unexpected bonus. The input modules can be physically located near the sensors, the display devices near the operators and the output module near the actuators. The only connection is the network. The network monitor is typically a standard PC. A laptop, or PDA (Personal Digital Assistant, such as Apple Newton, or Tandy Zoomer) serves as a portable monitor or calibrator. Input devices can be calibrated/adjusted at a convenient location near the sensor, with programmable displays located anywhere along the network. For a completely networked installation, Honeywell Micro Switch estimates that wire-savings account for just 18% of the total. The other savings include documentation, maintenance, trouble-shooting, elimination of system-downtime (Ref. 4).
Distributed Autonomous IntelligenceDick Morley (Ref. 5) has compared intelligent I/O to chicken brains. It is well known that the brains of a chicken are distributed down its spinal column, and not totally in its head, so that when its head is cut off, the rest of the body still runs around, controlled by the remaining intelligence. Intelligent I/O , with local intelligence and control functions - almost like a single-point PLC - continues to work when the host computer (or network) gets disconnected - just like chicken brains. And, a significant benefit is that Chicken brain I/O has no conventional complexity limitations.
A network of taxis is an excellent metaphorical illustration of how a chicken brain control system would work. Each taxi (node) is completely independent, with at least some, minimal intelligence. Network communication is via a radio link. Programming is rule-based and relatively simple, so that a novice is easily instructed on the protocol to follow. Each taxi reports within predetermined rules, time-based if there is no incident to report, or event-based in a changing situation. Typical response time (to order, and have a taxi arrive) may be a few minutes - one cannot guarantee which taxi responds, but speedy response is guaranteed. If the radio-link is interrupted for any reason, an individual taxi follows pre-agreed backup rules to continue operation. When communication is restored, improved network operation continues. Fault-tolerance and redundancy rules are simple extensions. If a specific taxi breaks down during the assignment, another taxi can immediately be called up, within parameters of convenience and availability.
ConclusionSilicon chip developments have yielded a completely new type of I/O architecture which provides revolutionary price/performance improvements in industrial applications. Intelligent, networked I/O modules provide performance equivalent to the autonomous agents of Chaos and Complexity theory. An integrated system based on this approach yields many advantages : host-independence, fault-tolerance, fail-safe features, vastly simplified wiring installation.
2/ Pinto James - 'Chicken Brain' I/O - Will This Replace DCS? - Controls & Inst. (UK) May '94
3/ Kompass Edward J. - Can Control, Run Better at the Edge of Chaos? - Control Eng., January '93.
4/ Pinto James - Two New Networks Target Factory-floor Devices - I&CS, June '94
Return to Index of all JimPinto Writings
Return to JimPinto.com HomePage
If you have ideas or suggestions to improve this site, contact: firstname.lastname@example.org