Artificial intelligence, along with big data and the industrial internet of things (IIoT), is receiving its fair share of attention. Often times it is presented as the downfall of some dystopian future in movies like The Terminator or The Matrix. But the vast majority of AI applications have led to tremendous improvements (think Google maps). Similar to conversations around IIoT, it is important to understand what AI is.
So let’s start with an understanding of intelligence. I like the definition that was put forth by Joanna Bryson in a recent Exponential View podcast. Simply put, intelligence is behaving appropriately for the context and transforming perception into action. Organisms have evolved as environmental changes have forced them to adapt. This adaptation can be thought of as a change in behavior with a changing context. More importantly, the actions of organisms change with this new perception. This same idea must be applied in manufacturing. As we understand processes better, we must behave appropriately for the context and take proper action (I will discuss context in a forthcoming post).
Now that I have defined intelligence, we can move to artificial intelligence. The purpose of an IIoT effort is to use an increasingly larger set of data to improve manufacturing and operational processes. As previously discussed, there are additional data sources, like sensors and instruments that enlarge the data set. But once the analysis is done, people still make the decision on how to proceed (decision making is a topic for another forthcoming post). Artificial intelligence can be thought of as extension to this process. Rather than people analyzing the data, the machines themselves determine the proper course of action and make necessary adjustments.
Machine learning of one of the most common AI tools currently in use. It starts with the analysis of a large data set of data. The machine learning algorithm analyzes the data and determines correlations. When process conditions suggest a potential problem, the system either corrects itself or provide an alerts of a potential issue. There is not an explicit set of instructions for the alert. It should be noted there are roughly dozen models for machine learning including regression, decision tree and artificial neural networks to name a few. A third forthcoming topic is proper model selection – you want to make sure your model accurately represents your actual process.
One of the most common applications of machine learning is for predictive maintenance as part of a reliability centered maintenance (RCM) strategy. A typical piece of turbo machinery will have several process instruments, including pressure, temperature and vibration. Using a machine learning algorithm, the interaction of these variables is processed and correlations are made. When these conditions appear to be outside normal operating conditions, an alert is given. After an evaluation, the operator will define the condition as either good or bad. Over time the algorithm improves its predictive capabilities and will automatically create a work order in your maintenance system. This helps ensure there is not any unplanned downtime.
Artificial intelligence is but one tool in a vast array of tools. As with every IIoT effort it should be part of a journey. Because of the complexity (and investment) involved, AI should be one of the last pieces in your overall strategy. But in order to resolve some of your manufacturing inefficiencies, AI may be necessary.
Tagsacoustic advanced diagnostics analytical artificial intelligence asset management business development condition based maintenance control valve differential pressure downtime ers filtration flow genealogy general historian IIoT impulse piping level machine learning manufacturing intelligence measurement MES mixer orifice plate positioner pressure project management real time visualization rotating equipment safety SCADA temperature vibration wireless WirelessHART