In manufacturing, new technologies are changing the way companies maintain their equipment to achieve better efficiencies and limit downtime. It’s called prescriptive analytics, and it uses IoT sensors, AI, and processing at the edge to help operators and engineers make better decisions about production equipment on the factory floor.
Maintenance of factory machinery is critical to keeping plant staff safe, maintaining output, reducing unplanned downtime and defects, and preserving the value of capital expenditures. Manufacturers have traditionally been reactive when it comes to maintaining machinery—waiting for something to break or show signs of wear—or performing maintenance on a fixed schedule. Both approaches are terribly inefficient and costly.
Prescriptive analytics looks deep into the data coming off individual assets or connected systems and tries to identify patterns or anomalies that indicate problems or looming outages. The data and analysis can point to the origin of the error and its potential impact. This informs human operators about whether they need to make a change right away, or if the issue can wait until the end of the shift to be remedied. Ultimately, the goal is to minimize the impact on the business, and hopefully make better use of available resources.
Data requirements for prescriptive analytics
Data lies at the heart of prescriptive analytics. Companies considering a pilot have to understand what data they need, what data the machinery or IoT systems can provide, and how the data can be transformed into a format that the analytics application can use.
Once the data is in the right format, then you need to be able to understand what it is telling you. This is where the AI at the heart of the analytics engine comes into play.
An AI needs lots of data to be trained. Tens or even hundreds of thousands of samples or data points might be required to teach the AI the difference between optimal and sub-optimal performance states for an individual CNC machine, robot, or pump. That data has to be available at the edge of the network, where the equipment is installed, to be processed and analyzed without having to incur bandwidth costs or potential outages. Once the system has modelled what falls within the range of normal behavior and what's considered abnormal, it can start to predict failures.
The prescriptive analytics AI can also be trained to inform operators what’s causing the problem. If sensors on a milling machine detect excessive vibration, is that caused by bearings, an imbalance, or something else? This information can also be tied into criticality within the production line. If the milling machine has a buffer and the cause is a minor criticality, it may be possible to sustain an outage for an hour or two. If it’s a major criticality, the production line may need to be immediately shut down to address the issue.
You can use the AI aspect of prescriptive analytics to understand what's happening to a piece of equipment, but you also need to consider the business data around it. Understanding where a particular piece of equipment sits within the production process and what ties into it is crucial. What are the maintenance schedules, and who is responsible? Where are parts sourced from? How will the shift pattern be impacted? These kinds of business data, as well as the data around the equipment you are monitoring on the edge, allow operators to proactively maintain plant equipment – and keep production lines running.