We discussed Industrial AI at the recent OPTIMIZE 21 conference and people have asked what does it mean, what’s the definition? So, here’s a treatise...
The term Industrial AI has roots in Aspen Mtell, where “industrial machine learning” characterized the product and the team’s passion for democratizing AI/machine learning so that everyday engineers in manufacturing industries could achieve great analytical accomplishments without being hard core data scientists and engineering specialists. The change to Industrial AI assures it is more inclusive but is still about making the hard stuff easier. It has spread like kudzu all over AspenTech and its products. But what does it mean, and what must it provide to be truly Industrial AI?
First, it begins with the idea that the Industrial AI process must hide the hard stuff and just do it so you don’t have to. Just like the way Steve Jobs designed the iPhone – the expert is in the box, so you do not need to be the brain trust. The AI is built into AspenTech products to help and assist without needing deep end user analysis and data science skills. It works to complement what you know already.
Second, Industrial AI recognizes that AI alone cannot solve industrial problems. A great deal of specific domain knowledge and experience are essential. Otherwise, as an extreme example, you might get a European company that paid a lot of money for data science analysis being told “we found a correlation between temperature and viscosity” – Chemistry 101! Your knowledge and experience are the foundation. So, the essence is to supply guiderails around AI to make sure it hunts for causation and not simple correlation.
An associate gave me these insightful examples of the guiderails provided by Industrial AI that elude pure data science:
- Clustering highly multidimensional data/events to uncover similarities between current situations and previous experiences to learn potential issue-resolution approaches.
Using an experiential industrial data-driven root cause analysis to reduce and include only relevant data is not a simple algorithm selection activity; it facilitates such methods and can eliminate a problem from recurring.
- An AI-driven data validation and cleansing routine based on deep multi-dimensional / temporal analysis of data patterns that exposes far more than boxing out all data around simple outliers.
Detecting anomalies from normal behavior can be very difficult in the face of messy industrial data across all the variables’ dimensions. Knowing what data to pick and what to throw away is critical. Pick too much or too little and false alerts will dominate. The anomalies are sparse, with often imbalanced distributions from normal, where a smaller of amount of known relevant data sets detect anomalies better. Industrial AI knows the industrial problem space and provides tools and methodologies to assist the user, where the AI workbenches missing the deep connection to the process industry, do not.
An Industrial AI solution assures:
The embedded technology is sound, proven to do the job and not just a science project.
The methodology to construct a working 24/7 solution is quick, easy and sustains well.
The work process for responding to anomaly and specific failure alerts is thorough and immediate and fits the way you work – not another deep analysis routine to determine what’s real and what’s not.
The solution adapts and adjusts, scaling rapidly and easily.
In the end, it’s not about the fact that AspenTech implements AI. Industrial AI is about how we did it and what we did to make the end users work less complex, easier, and faster to accomplish great outcomes.