By Pat Dixon, PE, PMP
President of DPAS, (DPAS-INC.com)
Is it time to perform maintenance on that critical piece of equipment? You know, the one that if it suddenly fails would shut you down, and possibly create a safety concern?
One approach is to use your calendar. When was the last time maintenance was performed on this unit? An Asset Management System can help you with that. The system can tell you when any asset you have is due for service.
The problem with that approach is that the unit may have been idle that whole time, and not require maintenance. Your technicians likely are stretched thin and it is important to have them effectively utilized, so having them service equipment that doesn't need it is costly in several ways. On the other extreme, it may have been in use more than you expect and is overdue. It can fail before you expect it to.
A better approach is for your control system to calculate runtimes for your equipment. When the service exceeds a certain number of runtime hours, service should be scheduled. This is a simple form of Predictive Maintenance. Predictive Maintenance is the use of real-time process data to predict the health of an asset.
The problem with this approach is that a failure can occur before the predicted service life.
The third approach is a more comprehensive use of data for Predictive Maintenance. You have process data that can indicate the health of your asset. Predictive Maintenance uses that data and compares it to patterns to identify whether there is a problem.
In this third approach, there are two techniques being used in industry:
- High speed sampling of vibration can yield a failure signature. As the equipment is running and in production use, the system can be continuously sampling and comparing the spectrum to known failure patterns. An advantage of this approach is that when there is a match it can tell you exactly what is wrong and what to fix. The disadvantage is that if there is a problem that vibration does not identify, you won't know about it.
- There is instrumentation other than vibration that pertain to the asset. Amperage, inlet and outlet pressure, flow, winding temperatures, and other measurements can be combined into a Machine Learning model. This model can identify what normal patterns are, and therefore notify when the asset is behaving abnormally. This is a more comprehensive use of data to catch failures that vibration alone might not be able to predict. A disadvantage is that when there is a prediction of abnormal behavior, you may not be able to tell the technician what to fix. Also, the caveats of Machine Learning that I covered in my prior article still apply.
You are dependent on your assets to operate. They have finite service lifespans. Your maintenance staff is also finite, and in many cases have more work on their plate than time in the day. The data in your process can help keep your assets healthy and efficiently deploy your maintenance staff. Predictive Maintenance, properly applied, can help you stay running efficiently and safely.
