It’s a good time for some future casting to break the focus on the dismal present.
The history of automation, in ICS and elsewhere, has been about reducing the number of people it takes to accomplish a task. It is amazing to see how a handful of people can operate a plant, a pipeline, and other industrial processes. In the next one to three years that number is going to get smaller as Operators are replaced by computers.
Much of the AI / Machine Learning predictions are at best hopes for the future and in many cases, especially cybersecurity, marketing hype. Mark me down as an optimist for these fields as I believe that once you have a Digital Twin there are wins for machine learning based predictive maintenance and efficiency improvements. In the security and reliability area, unexpected correlations will drive process variable anomaly detection that will detect even sophisiticated state-actor attacks and bad data from non-malicious causes, such as a faulty sensor.
Creating and analyzing the Digital Twin is the sticking point now, especially when you have plants that are all either a little or a lot different. It is one thing for GE to have a Digital Twin for a GE turbine run by a GE Mark VIe control system with a standard deployment. It is much harder to create a Digital Twin when you have fertilizer plants cobbled together with different products over decades. Even if they are all using the same DCS, each is different enough that it makes creating and analyzing each Digital Twin cost effectively a unique challenge.
However unique plants / processes isn’t always a problem in applying machine learning and getting the benefits. Replacing the Operator with a computer is an example. There are two main reasons why this can be successful:
- Consistent Operational Tasks – Consider how Operators are trained. Sometimes it is structured on simulators with various events occurring or by studying training material and taking tests. Other times it is sitting next to a seasoned operator. It both cases, and everything in between, it is “when that happens, you do this”. Sometimes it is a single Operator step, that may include a number of actions that have already been automated. Other times it is trigger to look at other factors and then take steps based on what is found. 99+% of Operator activities are repetitive actions that follow rules / standard procedures. The Operator is not developing a new set of actions or principles based on the situation displayed on the screens.
- Machine Learning Training Data Exists – Historians have collected years and even decades of process state information and corresponding operator actions to feed the machine learning algorithms. This data is plant / location specific.
Over the past two decades I’ve heard that in most control rooms, an Operator’s time is filled with hours and days of boredom, monotony, and then minutes of excitement / terror when an unusual state or situation occurs. Those hours and days go to the computer, and the human becomes even more important during the minutes of excitement.
The Engineer and Automation Professional have always been the keys to designing, deploying, maintaining/fixing and improving operations. They are often called during the 1% of time when something unexpected happens (or there is an Operator who knows the process at or near the level of the Engineer). This problem solver becomes more important, even more valuable. There always needs to be one of these people available to deal with the new situations when the machine learning flags a new situation or parameters exceeded a threshold.
It is a smaller number of higher skilled and higher paid individuals. The Operator goes away and the Engineer becomes even more important.
Ideally some of this reduced Operator headcount can also be transferred to maintenance, including cyber maintenance which has been under-resourced since Ethernet and Windows computers found their way into ICS.
Now where those ‘computers’ physically sit and who owns and maintains them are the next questions.