What to Learn from a Plant’s Frequency Data

If you’re old enough to remember the Internet browser called Netscape, then you may remember the company’s CEO – Jim Barksdale – who once quipped: “If we have data, let’s look at data. If all we have is opinions, let’s go with mine.” Fortunately for those that come to this blog there’s typically sufficient amounts of data. That said, a question often raised relative to a plant’s process data is: What gems are hidden in it?

A key to the Internet of Things (IoT) is the value it creates by uncovering hidden gems – those otherwise ambiguous details that provide insight and enable progress. Indeed, it’s the advance warning of impending equipment failure often made possible thru machine learning and data clustering technologies that provides value, not an alert that the asset’s vibration levels have been exceeded. The value of such IoT innovations manifests itself in the production staff’s ability to shut down the process and to do so safely without catastrophic damage to the assets involved. IoT applies equally to optimization as it does to reliability. When applying advanced forensic tools like Power Spectrum to a plant’s process data other hidden gems can be uncovered that facilitate improvements in production efficiency and throughput.

Power Spectrum is an increasingly common albeit more complex method for evaluating the relationships that exist among and between a production facility’s many PID control loops. It involves a Fourier analysis of discrete frequencies which indicates the relative power of each frequency. When applied to process data Power Spectrum reveals both the frequency at which some dynamic event occurs and the associated magnitude. By plotting a control loop’s oscillatory behavior Power Spectrum can expose the following:

Sizing Matters

It’s not uncommon to see unambiguously sharp peaks on a Power Spectrum plot. These highlight clear oscillatory behavior and they often can be linked directly to the root-cause of a performance issue – whether as the source or situated in close proximity to the source. What’s more, it’s not uncommon to see echoes in the graphs pertaining to other control loops. The size of the peaks tend to shorten as the control loop’s relationship to the root-cause becomes increasingly distant.

Figure 1 – A Power Spectrum graph can reveal important insights into the relationship between PID control loops. What’s clear from the graph above is a particular frequency peak at 28 minutes that is shared by two controllers. The dynamic behavior of these loops suggests a close linkage.

Timing is Everything

Also relevant to root-cause analysis is the location of the peak along the X-axis of the Power Spectrum plot. While not a date/time stamp, per se, the X-axis identifies the period at which the dynamic event is occurring. Measured in terms of Time the period provides an indication of the regularity with which an event occurs. It’s not uncommon for the period to coincide with a mechanical condition such as a fan that cycles on and off every few minutes or the hourly change that occurs to a tank system’s feedstock. While Power Spectrum and its analysis of frequencies is not a run-of-the-mill diagnostic, it successfully reveals performance insights that would otherwise remain invisible to the naked eye. With the ability to compute frequency data in large quantities and to both compare and contrast the results, tools like Power Spectrum empower production staff to mine an existing resource – their plant’s readily available process data – and to leverage their findings in conjunction with routine maintenance and optimization initiatives.

Again, this is the power of advanced analytics and IoT.