Embedded analytics do the job of a data scientist to help reliability professionals proactively address production issues.
EVER WISH YOU could anticipate production downtime and scrap problems before they happened?
Advanced manufacturing analytics can help you predict problems such as these, but, until now, the technology wasn’t feasible for many companies. Implementing advanced analytics required the specialized skill set of a data scientist and a lot of time and effort spent cleaning and organizing data. Today’s advances make it possible for the technology to do more of the analytical heavy lifting without a data scientist. This puts advanced analytics in reach for more companies.
Predictive-analytics technology is the latest example of this. It uses artificial intelligence (AI) to learn your process, build models, and deliver actionable, predictive analytics to workers. This can help you proactively identify and resolve production issues before they become problems.
Predict, Prevent Problems
A key use of analytics technology with AI is to learn, then continuously monitor your operations and detect anomalies. If anomalies are detected, the technology can alert workers to investigate or intervene before they become issues.
Consider, as an example, a cutting machine in a wood-products factory. When the blade on the machine dulls, it can cause splintering. Workers then either need to rework the product or scrap it entirely.
AI-based predictive-analytics technology can learn the process and the variables that lead to dull blades, such as the lumber type, feed rate, and humidity level. Then, by continuously watching the process, the technology can help workers know when to change a blade before splintering occurs.
Even if splintering isn’t a common problem, the predictive analytics can still provide value. For example, workers may be changing the blade every three days, but analytics could inform them that the blades could, in fact, last five days. This would help workers keep production running longer and reduce maintenance costs.
Consider another asset such as a mixer in a rubber plant. It must be kept at a specific temperature set point to maintain a rubber product’s quality and to avoid the downtime and clean-up that’s required when a product’s viscosity reaches a point where it can’t be extruded.
An engineer can even take anomaly detection a step further by linking specific changes detected in the analytics technology with automatic adjustments in their controller or human-machine interface (HMI). In the mixer application, for example, the engineer could program specific corrective steps in the controller when an anomaly is detected.
A virtual or “soft” sensor is another valuable way to put embedded analytics to work. Sometimes a physical sensor just isn’t practical. It could be that the sensor is too expensive, making manual readings with a sensor too time and labor intensive. Physical sensors also can be inadequate when the situation involves a physical measurement that loses integrity, often due to environmental conditions. In these instances, predictive analytics can use existing data to estimate the value you’d get from a physical sensor.
A good use of a virtual sensor would be at the end of a production process. It’s almost impossible to use an actual sensor on some finished products, such as packaged cookies. As a result, workers need to physically pull products off a line to read key variables, such as humidity. But this wastes product and labor.
A virtual sensor can use variables from the production process, such as sprayer flow rate and dryer burner rate, to estimate the finished product’s humidity. Adding a virtual sensor can reduce the need to make manual readings or specific process adjustments.
How do you implement and use advanced analytics with AI? The technology itself is a module that plugs into your control chassis. Once the module is installed, a controls engineer can begin configuring it. This involves selecting the output that the technology will model and monitor, such as temperature, pressure, or other variables that obey the laws of physics. The variable models should have a direct correlation to product quality, asset reliability, scrap, or rework to drive a financial impact.
The engineer also identifies the input variables that will be needed to build the predictive model. If the predictive-analytics solution being used can do its own data cleansing, the engineer can cast a wider net when identifying data.
After the model is built, a training mode is used to establish a confidence level of the model. How long this process takes depends on the complexity of your application. In a simple process unit, it can take less than 30 min.
Finally, the technology will have learned normal operations and can start predicting problems in your process. Workers can be notified of issues however you choose, such as through alarming on an HMI or dashboard.
Even as digital technologies change industrial operations every day, the use of AI in the control system is a profound development. It breaks down barriers to advanced analytics and can help improve operations in ways that will make your enterprise more competitive.
Jennifer Mansfield is Business Development Manager at Rockwell Automation, Milwaukee (rockwellautomation. com). As an IIoT strategist, with 30+ years of experience, she helps customers integrate all aspects of their operations to achieve a strong, correlated foundation for their automation systems.