Industrial control and management at the edge
04 December 2018
Cloud applications and industrial processes used to be distinct spaces, but the IIoT is bringing data from the plant floor to the cloud and merging the distinctions between the two. And as this piece explains: at the boundary line is the ‘edge’. Edge computing – defined by devices like gateways or directly network-connected industrial devices – is where the cloud meets the concrete: where the action happens.
For the digital issue of this piece, please visit this link – or click here to register for EPDT's magazine.
The unique position that the edge is in, moreover, means that system architects need to understand both cloud software and networking, as well as industrial control and management.
Control versus management
In the context of edge devices, there is a need for a distinction between control and management: management tasks, such as firmware updates or diagnostics, are on a relatively relaxed timeframe and can be carried out from the cloud; whereas control must be done at the edge, since it needs to be performed in real time.
The latency between cloud and edge is simply too great for real-time feedback and control. This article provides an overview of control systems typically used in an industrial context, covering the actuators used and related AFEs (analogue front ends) (see Figure 1 – links to ActiveMag), in addition to some examples of complete control loops.
Control directs the physical interactions happening on the plant floor. These interactions are performed via actuators: electromechanical devices which convert electrical signals into physical interactions, such as to position a part or assemble a component.
Industrial actuators can be thought of as handling linear or rotary control. Electrical actuators include motors, piezoelectrics and solenoids. Non-electric actuators include jacks, brakes and pumps, which operate through hydraulic action, or other devices which use pneumatic control.
Solenoids work similarly to motors, but generate linear motion. As an electric current is passed through a coil, a magnetic field is induced that exerts a force on the rod inside to move it linearly. They can be voltage or current-driven, and are used for a range of applications from relays to valve control to circuit breakers.
For piezoelectric devices, the application of voltage causes mechanical displacement, as the material (usually quartz or ceramic) strains. They provide a rapid response, with precise control; and in addition, they are energy efficient as no energy is needed to hold them in place like a solenoid. On the downside, they require a high voltage, and they exhibit hysteresis. They are used for haptics in consumer devices, valve positioning and precision mechanics.
To drive the actuator, an analogue front end (AFE) is used. This consists of a digital-to-analogue convertor (DAC), then a high voltage buffer to generate the voltages needed to drive the actuator.
A resistance ladder-type DAC is usually sufficient because of the low conversion rate. A high voltage buffer is typically used between the DAC and actuator, depending on the load characteristics.
Motors work by creating a magnetic field when current is passed through them, creating a rotational mechanical force. They can be classified into AC, DC or stepper motors, which convert electrical pulses into discrete mechanical movements.
While control loops vary, they tend to share a few common elements: these include sensors, measurement, analysis and response.
To understand the principles of control, we can look at a relatively simple system at first. Figure 2 shows an environmental chamber with a set of lasers with controllers. The lasers need to be kept within a strict temperature range of just tenths of a Kelvin. To adjust the temperature of the chamber, we can inject warm or cool air.
Each laser and controller have thermistors nearby to measure temperature. To make sure we set the correct temperature, we have to take into account that there may be differences between actual temperature and temperature measured at the thermistors.
This difference is the temperature error. By combining the errors from each thermistor, and then doing the root mean square, we can minimise the worst-case error. This gives us a single number we can use to estimate temperature error, and with which we can determine how to control the air going into the chamber. The simplest method to do this is to multiply the error by a number, ‘P’. This gives us a difference we can apply to the air temperature setting.
The ‘P’ setting will determine how quickly the error is reduced. This control scheme, however, will not be able to completely eliminate the error. To accomplish that feat, we would need to account for memory. By adding in a term that sums previous errors and multiplies by ‘I’, we end up with a ‘PI’ controller. If we needed a faster response to rapidly changing temperatures, we could also add a term for the derivative of the error, or a ‘D’ term. Thus, we end up with a classic PID controller.
The controller scheme can get more complicated when there are multiple sensor inputs to consider. An example is a valve control system that uses compressed air set by piezo control. Pressure sensors check the compressed air pressure at various points; the rotational position of the valve is also monitored. All of these sensor inputs are used in the control loop.
Such systems, which have multiple sensor inputs, can be challenging: multiple sources of information have to be reconciled. It may be that the position sensor has shown the valve is in the desired position, yet the pressure sensors indicate the compressed air has not moved it correctly yet. The control system needs to be able to reconcile various sensors, which may provide conflicting information, and use a decision tree to perform the right action.
Other factors to consider are loop bandwidth (how fast the sensors will respond to changes in air temperature, as well as lag), and how long it takes for a change in an actuator to produce a change in the thing being measured.
Rather than building control systems separately, a popular and cost-effective solution these days is to integrate functionality into a single chip. Integrating a processor like a Cortex M3/M4, which can perform floating point calculations at high speed, can provide high loop resolution for rapid response control loops.
The said processors can be integrated on the same die with other digital and analogue components, such as AFEs to control actuators, or sensors to provide input, and even op-amps, power switches and other devices that would otherwise need to be separate board components. This type of integration can both increase reliability and reduce energy consumption, BOM costs and the risk of end-of-life sourcing issues for components.
For a ready-made solution, the S3 Semiconductors SmartEdge platform integrates sensor AFE, control loop, calibration, security and communication elements – all onto a single cost-effective ASIC suitable for smart edge devices.
Well-designed control loops can greatly simplify system design. The principles we’ve described are the fundamental building blocks of control mechanisms and can be used to understand systems of all sizes and complexity. Once tested at the board level, these control loops can be integrated into a single chip – helping to save BOM costs, improve energy consumption, and increase long-term reliability.
Contact Details and Archive...