Smart Workload Balancing

Smart Workload Balancing - Yokogawa

The cognitive workload of operators working with automated systems should neither be too high nor too low. A static level of automation is unable to cope with systems that produce large fluctuations in cognitive workload, therefore a method for adaptive automation is discussed that could balance workload by intelligently choosing what to automate and when. The following text is a short extraction from a paper by Ferdinand Coster introducing the concept of the Cognitive Workload Value factor, which takes into account both workload and situation awareness. The full paper will be presented at the H-Workload conference in June where a framework will be discussed for categorizing and using different workload and situation awareness measures effectively.


Today’s industries are highly automated, sometimes to a point where the human operator is supposed to just sit back and monitor. There are many good reasons to use automation, and it is fair to say that many things we take for granted would not be possible without automation. For example, automation is executing tasks that require faster responses than humans possess (e.g. a process safety system) or accurately track many data points 24/7 without interruption.

Another common reason to use automation is to lower the workload for users by taking over the more tedious tasks so the human can focus on their main task. However, care must be taken to choose an appropriate level of automation, for taking too many tasks away from the human operator can introduce problems of its own. An operator that becomes detached from the actual processes in the plant because automation is doing practically everything will have a very hard time understanding what is going on when that automation fails (“Out-of-the-Loop syndrome”). It could be argued that the primary role of the human operator has actually become to take over when the automation systems can’t cope with whatever abnormal situation is happening. An operator that is Out-of-the-Loop can’t perform his primary role effectively.

The balance that must therefore be considered is a trade-off between workload and Situation Awareness. It must be noted that in this paper when we talk about workload, it means cognitive workload; physical workload is not considered and assumed appropriately designed for the operator working in a control room. For systems that have a fairly constant workload it is possible to design and choose an appropriate level of automation that will impose a manageable workload on the operator while still maintaining a good level of Situation Awareness. However, depending on the type of process, there can be large fluctuations in workload and in these cases a static level of automation will either impose a workload that is too high during peaks or too low during normal operation.

Workload Balancing Mechanisms

A fluctuating workload might be balanced by modifying the distribution in multiple ways, such as:

  • Distribution in time
  • Distribution in executing entity
  • Distribution in available processing power
  • Distribution in priority

Distribution in time is basically a task scheduling activity. A priori knowledge of workload associated with certain tasks can be used to plan for a certain workload over time. This should be the basis of any workload balancing strategy, but does not account for unforeseen situations (like process upsets). Ad-hoc changes to the schedule might be difficult because many tasks and procedures, once started, do not allow for pausing and picking up at some point later in time.

Distribution in executing entity means choosing who will do a specific task. This can be a choice between human or automation, but also a choice between different humans. Operators often work in teams, and an operator more experienced with the task might experience a lower workload than an inexperienced operator.

Distribution in available processing power means splitting and dividing a task between multiple executing entities. Typically this means getting more operators involved (e.g. during a plant startup).

Distribution in priority is a mechanism to help decide which tasks are most important at any given moment. This could mean that tasks that would be seen as important during normal operation change to a lower priority during critical situations and postponed to a later time or even get dropped completely.

Intelligent Adaptive Automation

Fig 1: Evolution of automation technologies and their relationship to different design approaches. (From: Hou, M., Banbury, S., Burns, C.: Intelligent Adaptive Systems: An Interaction-Centered Design Perspective (2015))

As static automation is not capable to regulate large fluctuations in workload, a different type of automation is needed. Alternatively, adaptive automation techniques that modify their level of automation based on models of operator behavior and workload, and more recently based on psycho-physiological measures.

Hou, Banbury and Burns introduce the idea of Intelligent Adaptive Automation (IAA) that goes one step beyond Adaptive Automation as illustrated in Figure 1. While flexible automation aims to reduce the negative effects of static automation by dynamically shifting tasks between operator and automation, it is based on task and user models only and does not take external effects into account. Intelligent Adaptive Automation explicitly adds world models so the external effects are incorporated.

The Cognitive Workload Value Factor

When trying to determine what to adapt it is important to keep in mind the trade-off between workload and situation awareness. For a specific task one can look at the workload imposed and the situation awareness provided by manually executing that task. Tasks that impose a high workload but provide little situation awareness should preferably always be automated, whereas tasks that impose little workload but provide high situation awareness should preferably always be done manually. The space of flexibility where (intelligent) adaptive automation can exist is somewhere between these extremes. In order to compare tasks to select which ones to automate the Cognitive Workload Value factor is introduced, which takes into account both workload and Situation Awareness.

If we would implement the intelligent adaptive automation by using an adaptation threshold based on the calculated Cognitive Workload Value factor resulting in switching sub-tasks between completely automated and completely manual, these changes might be too drastic for the operator to keep understanding what the automation is doing. Therefore it is suggested to use a simple Automation Level Taxonomy as a transient automation spectrum. In this way the operator can always be in control of the level of automation.

For more detailed information and to read the full article on Smart Workload Balancing, please consider joining the H-Workload 2017 conference to be held in Dublin Ireland June 28-30.