By Rob Azevedo, product manager & strategic alliances with TrendMiner
Applying advanced analytics solutions to industrial processes usually involves attaining insights from the data and using that data to make decisions.
In my experience, obtaining the highest level of achievement—the augmented factory—comes with unique challenges. In an augmented factory, process experts use prescriptive analytics to change a predictive outcome in process behaviors.
Early production knowledge typically includes the status, performance and health of facilities, production lines, and machines. Data-driven insights address issues with production time, downtime and defects, while helping to predict when unwanted events may interfere with production.
Plants usually express these “contextual process events” in units such as batches, phases, cycles and campaigns. Downtime events also include planned or unplanned maintenance, emergency stops, changeovers, defects and quality inspections.
Process-data insights improve cycle times, overall equipment efficiency (OEE), product quality, and asset health. They correct anomalies quickly and identify when they might reoccur.
A review of the four types of analytics
It’s best to quantify options and test assumptions continuously using four data analytic types:
1. Descriptive analytics (identifying what happened)
2. Diagnostic analytics (identifying why it happened)
3. Predictive analytics (predicting when something will happen)
4. Prescriptive analytics (knowing what to do to change a predicted outcome)
All four types generate decisions and actions in what we refer to as the “augmented analytics cycle,” which demonstrates challenges in seeing end-to-end use cases using only descriptive, diagnostic, predictive or prescriptive analytics. For example, predictive maintenance is attained when descriptive and diagnostic data is used to make insights, identify what happened, and explain why. Predictive insights can help generate a model for building an end-to-end predictive solution.
When we worked with a company that was having problems reaching the correct temperature in a heat exchanger, engineers applied the first three analytics types to derive a prescriptive-maintenance schedule. The result? The heat exchanger temperature was consistent.
The company used all analytics types to accomplish its goal:
1. Descriptive analytics: The heat exchanger cannot reach the set output temperatures anymore.
2. Diagnostic analytics: Fouling causes a performance drop, but the data-analytics solution revealed that cleaning solves the problem.
3. Predictive analytics: The solution knows when fouling will occur again based on the discovered operating parameters.
4. Prescriptive analytics: When the heat exchanger does not meet setpoint temperatures, the solution notifies stakeholders that it’s time to clean.
To implement the right predictive-maintenance schedule, companies must apply all four analytics types at the augmented level. For example, the company would need to gather new insights if fouling no longer is the source of the problem with the heat exchanger. Engineers would need to start at the descriptive level and work through the steps to close the cycle again.
Formulas for prescriptive decision-making
“Prescriptive” can mean improving decision-making, forecasting production, or simply alerting someone to go check equipment. Success depends on knowing enough about the process anomaly to determine when it might reoccur.
Typically, the same formula is used for all production processes. Attaining the prescriptive level means gaining insights through data exploration and hypothesis testing. It also includes using machine learning (ML) or algorithms that predict outcomes from early indicators to make smart decisions.
In one recent use case, our data-analytics solution informed oil-and-gas maintenance teams about which tasks to prioritize. Engineers were able to avoid a predicted machine failure. In another, a food-and-beverage company used the solution to decide how to optimize raw materials by sending prescriptive-recommendation notifications to operators regarding extractors’ use and dosage.
Prescriptive analytics
During a visit to a German plant in 2021, we learned that its biggest challenge didn’t involve analytics, but rather knowledge retention. More than 50% of its workers were planning to retire within 10 years.
“Using prescriptive analytics, every person joining our company can now onboard to become, in two years, 100% knowledgeable about the production process and be fully independent from their experienced counterparts,” said a plant representative. “This used to take 10 years.”
Although interpreting the data itself wasn’t the biggest challenge, it provided the perfect opportunity to apply prescriptive analytics and create a knowledge base. Improving the ability to onboard new employees was as important as keeping processes functioning correctly. By applying prescriptive analytics, they accomplished both.
