Recognize the value of both physical and data models
The physical nature of industrial devices and systems implies a knowable, objective truth: there is real meaning to a temperature or pressure measurement on a physical asset. Failure modes of a device have a physical origin and explanation. For this reason, companies like GE build sophisticated models of the assets they support. Human expertise can be encoded in decision rules and analytics that help inform the actions required given a set of observed inputs.
In contrast, data-driven models, especially those used in machine learning, typically incorporate little physical understanding as part of their creation—inferring their outputs solely as a function of the training data input into the system. As a result, data-driven models may not incorporate all the information from the “real world” that we already know about the physical system being modeled.
Even in a world of perfect data-driven models then, physical models can play an important role. First, they are essential for applying machine intelligence to the over-engineered environments that make up many industrial systems, where failures are very rare, so there is little data to learn from. Physical models can also work alongside and inform more complex data-driven models, as well as serve as a kind of “reality check” against the behavior of pure data-driven models as they evolve.
There is ample opportunity to marry physical and data-driven models provide better outcomes. Companies bringing machine learning to industrial systems successfully should recognize the importance and distinct roles of both of these types of analytics, and the different ways (and by whom) they are created and managed.