Machine Health Optimization

The biggest challenge to overcome as you move toward machine health optimization is rejecting the way that things have always been done. If you perform an honest assessment, those traditional approaches to machine health management have likely resulted in:

• “Saving the day” but not addressing the root cause of equipment failure
• Maintenance being performed incorrectly
• Working reactively after a problem has occurred 
• Data analysis that is out of touch with the day-to-day realities in your plant
• Operators not knowing how their set points impact equipment life
• Institutional knowledge driving maintenance, which becomes more and more dangerous as maintenance gurus move on or retire
• Inconsistent practices from plant to plant, or even between production divisions in the same plant

Failure Is Almost Always Random

Any number of factors can lead to equipment failure — the products on your lines change, scheduled maintenance is performed incorrectly, or someone drops a tool on a machine. In this case, machine health degradation doesn’t follow a straight slope down to failure. It could even look like a straight line, with failure being equally likely at all times. 

Decades of research show that random failure models are more accurate than predictable wear-out models more than 80% of the time. Only a small minority of equipment exhibits predictable wear-out behavior.

Focusing solely on OEE also ignores other critical operational factors.

Vibration monitoring is often where PdM data collection starts and ends, but it’s just the beginning of an optimization-focused approach to machine health. Somehow, you have to measure the health of your machines, and sensors are the ideal way to do that.

When people talk about condition monitoring, they’re often still thinking about route-based data collection. We used to pay people to walk around collecting data from vibration sensors on a monthly basis. Thanks to advances in technology, it now makes much more sense to instrument machines with low-cost wireless sensors that can stream data continuously to a centralized location.

A distributed continuous asset health monitoring system that provides hundreds of data points per hour is critical to being able to spot systemic damage-causing conditions. The type of data points it provides is also a key consideration. It’s an excellent idea to monitor single values like peak acceleration, overall vibration, and temperature, but single-value trending will only show that there is a problem without providing insight into what that problem might be.

There is more data out there. SMARTdiagnostics sensors send a time waveform, not a peak or an average. Then, the software shows that data in a frequency spectrum. Different kinds of mechanical faults look completely different in the frequency spectrum, but would look almost identical in the overall vibration chart.

As the field has matured, many other tools are now being used to comprehensively characterize machine health. This extends to not just simple rotating machines but many other types of equipment — ranging from intermittent running equipment like converting lines in packaging plants, to robots and lifts in automotive assembly.

Some manufacturers have cobbled together various sensor systems to try to gain a better understanding of overall machine health, but the best approach is one that leverages a common network infrastructure to connect all sensors under one umbrella; dramatically simplifying sensor and network maintenance, overhead, and management of the data.

Data collection using the right technology is an important first step, but sensors and data don’t push an actionable plan to the surface by themselves.

In order to create an action plan to optimize machine health, the data collected from your smart sensor network needs to be correlated with data about plant conditions and the people who interact with your equipment. By doing this, we learn which conditions lead to damage and which machine components are contributing. Then, we can engineer those damaging conditions out of the process. This leads to extended machine life, greater uptime, lower costs, and better energy efficiency.

This is the point where other machine health solutions providers might start talking about applying artificial intelligence to asset problems. AI is exciting and full of possibilities, but in the early days, it’s much more practical to capture knowledge from plant employees and adapt it into logic and digital workflows.

Many manufacturers think it could take weeks, months, or even years of monitoring to establish a useful baseline. Using SMARTdiagnostics sensor technology, we can collect thousands of data points within hours and get to work. Identifying failures gets easy very quickly, and catching failures equals quick ROI and a success story that everyone can rally around.

When you start putting sensors on every piece of important equipment, the resulting data avalanche can get overwhelming very quickly. It is not unusual to have tens of thousands of indicators in a single plant. For this kind of system to scale, the first step is auto-alarming, so the correct people can be notified when something is out of spec.

When you reach this point, congratulations — you’ve successfully started using an IIoT system to facilitate the goals of a predictive maintenance program. The journey to machine health optimization does not stop there. We can do much more to get even smarter.

Implementing predictive analysis tools is where forward-thinking manufacturers begin to move past traditional PdM and toward true machine health optimization.

Machine health data has been collected and studied for decades, so many phenomena are already well-understood. We also have trillions of our own data points on tens of thousands of assets to supplement this global knowledge. With all of this information, we’ve built algorithms that can look at data coming in from a machine, compare it to thousands of known fault signatures, and quickly predict which fault conditions might be present in that machine.

At this stage in the journey, one of the primary goals is to deepen knowledge of your equipment’s health by understanding the physics of each machine and providing root cause analysis. For example, you may learn how changing pressure at a pump inlet influences other variables like vibration and temperature. Modeling and optimization at this stage are focused on finding the operating conditions that consume the least energy and minimize machine and component wear while still meeting production targets.

The final stages of the machine health optimization journey are what will drive the most value from your investments and solidly position you ahead of your competitors. Once damaging conditions have been identified and ideal operating conditions have been found, the next step is putting SOPs in place to ensure machines are consistently running in the best possible manner. This involves a number of factors that aren’t just OEE: machine health, energy costs, downtime costs, quality, labor, and so on. The definition of success includes both mechanical and business factors.

Once plant conditions are fully understood, we can use machine learning to apply advanced computational methods to determine the best way to run a machine. This looks like ongoing optimization on the fly — based on current conditions — in order to maximize value.

The final step is automation; removing as much of the human element from operational adjustments as possible. When this is fully and accurately implemented, the smart algorithms produced by the machine learning process can directly control your equipment, e.g. adjusting the speed of multiple parallel pumps in a system on-the-fly to minimize overall energy consumption, maximize output, or minimize machine damage.