Predictive Analytics for Automated Systems
What Can We Do with A.I. and Machine Learning?
I’m a data guy. There – I admitted it. I hear that is step one. But seriously, it is a bit of a running joke that I will likely produce a spreadsheet or a pivot table when faced with some type of data challenge. Perhaps it is my analytical side that likes to take data and try to turn it into information that can help me make decisions for future actions. I am not alone here. Many of our customers are trying to do the same – use historical and live data to make intelligent business decisions to predict a possible problem before it happens, or to avoid that problematic situation altogether.
How We Use Data
A little history. In the 30 years or so that I have been involved with automation, there has been a desire to collect data. Often data was collected only to be used after an undesirable event to determine what happened. For example, if a bad batch of product was made, you could look back at the data and see that a temperature varied from setpoint during the process. You now know WHAT happened, but you still made scrap. Today we would call that “descriptive historical data”. Is my drive faulted? Is my temperature too far from the setpoint? Which production line made the most good parts last week? These are all questions that collecting descriptive historical data can answer.
Let’s take a quick look at four different types of data and what they mean:
- Descriptive — What happened?
- Diagnostic — Why did it happen?
- Predictive — What will happen?
- Prescriptive — What should I do?
Manufacturing and process data is typically collected and evaluated via an industrial computer such as this VersaView® 5400
Descriptive Historical Data – Is it good enough?
Is that good enough? Maybe – but along with WHAT, people typically want to know WHY. Why did my drive fault? Why didn’t I reach my temperature setpoint? Why did Line 3 make 20% scrap last week? As devices get smarter, they produce more information that we can collect. This data is often called “Diagnostic Data”. So, to answer “Why did my drive fault?”, I might use the diagnostic information collected to determine that the motor current was excessively high for several minutes before the drive fault. That information can now be given to a technician to determine why the motor current was high, and hopefully, prevent that problem in the future. So now we know WHAT and WHY, but we still wasted production, and the human interaction required is high.
This graphic correlates types of data to the amount of human input. This graphic is courtesy of Rockwell Automation®.
A Predictive Analytics and Data Analogy: Your Car
Today, in addition to the basic analytics of WHAT and WHY above, customers would like to use all the data being generated by their production assets to learn what should be happening and predict imminent anomalies. For example, wouldn’t it be better to know that your motor is drawing abnormally high current before the drive faults and shuts down production?
- This is part of the advanced analytics that we see our customers start to implement today known as “Predictive Analytics”
- So now we’ve gone from Descriptive “My car ran out of gas” to…
- Diagnostic “My gas light just came on – better get gas” to…
- Predictive “I have 120 miles until empty – I can get gas at my next rest stop”
We experience Predictive Analytics in our daily life – we should expect the same in manufacturing.
A.I. and Machine Learning for Process Automation
So, all this data that is available today allows technology like Machine Learning (a form of Artificial Intelligence) not only to make predictions about our process but also tell us how to avoid problems. Instead of “your drive will fault in 5 minutes” we can now get “decrease load on motor”. This is often called “Prescriptive Analytics”. If predictive analytics tells us what is about to happen, prescriptive analytics tells us what we can do to avoid it. In my low gas example above, this would be the equivalent of showing you all gas stations along your route in your mileage range and routing your GPS navigation system to them.
An example screenshot of a dashboard from FactoryTalk Analytics for Devices
A.I. and Machine Learning are here today. Rockwell Automation has introduced several offerings in the FactoryTalk Analytics platform and is investing heavily to continue to provide solutions for manufacturing production analytics. What solution is right for you?
Not sure if you or your manufacturing facility are ready for analytics? Contact Horizon Solutions for more information.