Condition based maintenance (CBM) is one of the latest buzz words in the control industry. CBM is being touted as a significant time and money saver across a broad spectrum of industrial applications. I was recently at a bearing production facility where spindle speed can reach 120,000 rpm or higher. Given that bearing life spans are usually rated in the millions of revolutions, it should not be surprising that the bearings on these spindles are changed out about every 90 days or so, if all goes well. I was told that there are two key elements to successfully maintaining these machines. First is to have a number of replacement spindles, and second is to have the ability to predict when a particular spindle is going to fail. Addressing the first issue is a matter of capital investment and inventory control. However, the second item, failure prediction, is more interesting.

First of all, if the spindle actually fails and the grinder must be taken off-line, it is a disaster as the entire production line is down. Second, such a failure typically happens after a significant amount of scrap has been produced. It is critical to detect the onset of failure as soon as possible so that system maintenance can be scheduled before product quality or production operations are significantly affected.

Now that we know to “detect an incipient failure as early as possible,” how do we do that? It turns out that the bulk of CBM really targets understanding a baseline of operations and then following a trend. When the process does something unexpected, or its parameters begin changing rapidly, this is usually a good indication that maintenance will be required soon. For a bearing spindle, it may be a simple vibration analysis, or even a rapid rise in the spindle lubrication temperature. Of course, there are numerous other metrics that can be used. However, the point is that setting a baseline when the system is in good health and watching it change over time is an approach that has been proven successful in CBM not only in spindles, but in many other operations and processes.

The beauty of this concept is that with modern technology, it is quite easy to implement. For example, if a specific parameter in a process is known to be the key to process health, it can easily be monitored for a baseline and subsequently for a significant change in a trend. Off-the-shelf technology such as wireless networking in conjunction with local area networks (LANs) and wide area networks (WANs) can let production control personnel know of pending issues for equipment in the building or on a different continent, and corrective action can be scheduled. For processes that are new or so complex that simple metrics are not known or available, models can be built using a variety of approaches such as basic or enhanced statistics, or even neural networks. Many of these capabilities are available for well-known packages.

It does not matter if you are running a well-documented and understood process such as a spindle, or if it is a new one such as a batch bio-manufacturing process. As long as you can establish a baseline for the process when it is healthy, you can easily and inexpensively monitor for a change that indicates a pending problem. This can be done even if good process models are not available, and it can be done remotely with existing technology.

As time goes on, the modeling aspects of these systems will become more accurate, enabling increased process control, quality, up-time, and, ultimately, profit. This should make managers and investors happy. It is even possible for the monitoring system to send an email or make a telephone call if a potential problem is detected. Whether or not that call makes someone happy probably depends on the time of day that the CBM system detects the problem.