APM, AI, Prescriptive maintenance, prescriptive analytics

Rethinking Traditional Condition Monitoring

July 15, 2021

Wikipedia tells us “Condition monitoring is the process of monitoring a parameter of condition in machinery (vibration, temperature etc.), to identify a significant change indicative of a developing fault. It is a major component of predictive maintenance.” Time-based monitored parameters include vibrations, current, voltages, temperatures, pressures, flows, lubricant properties and sound. It is generally accepted that condition monitoring provides clear value; early warnings of issues result in less downtime, more product throughput and even execution of smarter, safer, environmentally friendly maintenance. 

But is all condition monitoring created equally? How are all those diverse signals combined to provide detection of issues that humans can interpret and act upon properly? 

Traditional condition monitoring grew out of vibration analysis where Fast Fourier Transforms (FFT) converted time-series data into frequencies so that humans, with very limited vision of dimensionality, can detect frequency changes in two-dimensional charts. Further detection techniques involved multiple manually set alarm thresholds at different levels on single measurements, or logical rule sets combining variables in an anticipation that the copious alarms will be timely, dependable and not false. Furthermore, visualization of single point variables and alert thresholds in two or even three dimensions cannot tell the whole story and it’s all too hard and too complicated to implement and sustain. It is not enough to manually set multiple simplistic alarms and robust, dependable analytics needs much more. 

New technology has enabled superior solutions that are easier to implement and use, that are much more scalable, besides providing more accurate and earlier detection. With decades of maintenance experience, about twelve years ago the Aspen Mtell® team pioneered changes in the status quo of condition monitoring to bring about greatly improved equipment availability. The breakthrough came from a novel implementation of machine learning technology – as a key component of what we now call Industrial AI – to recognize the telltale signs of equipment degradation in complex patterns across simultaneous combinations of multiple variables and time. Two key factors assure success. Vibrations do not need to be converted to frequencies; machine learning can detect minuscule changes in the time domain. Second, as an example, a complex pump may include say, 25 measured variables on the machine and on the process around it. In this case machine learning ‘sees’ patterns in 26 dimensions – all the sensors plus time. Humans see well in three dimensions but barely across time; if something happens and we do not see a consequence within seconds, we’ve lost it. Machine learning patterns are subtle combinations of relationships between variables and they start earlier than any single variable disturbance. They are precise and each pattern matches a unique root cause. But unfortunately, humans cannot see them. 

A most important factor is that our technique is not modeling or estimating behavior. We are measuring the actual machine behavior with great precision and accuracy. So now we look for all the pattern combinations that represent normal behavior so we can detect abnormal conditions. We also can recognize the very earliest patterns that indicate explicit degradation that if unattended will lead to failure. Best of all, rather than looking to detect basic late-stage symptoms of damage, we foresee the earlier root cause conditions giving warnings in weeks and months not just hours or days along with accurate, measured time-to-failure. Detection is not correction and knowing you have a problem on an industrial machine is one thing. But that extra time is your friend! Embedded prescriptive advice can suggest process course corrections to avoid degradation and damage or well-planned service interventions that are much shorter, safer and avoid environmental mishaps that can happen with unexpected breakdowns. 


Rethink your condition monitoring! To stop your machines breaking down visit our solutions page for more information. 

There was a problem storing your subscription

Leave A Comment