Why Time Domain Analysis of Vibration is Critical in 24x7 Monitoring for Predictive Maintenance
Vibration analysis of machines, specially rotor, is nothing new. By 1995, as computers with 1GB+RAM became available in the market, FFT based vibration analysis became popular as harmonics of vibration signature could identify a particular bearing defect or a gearbox defect. Over time, this method became popular and experts who understood FFT signature of machine failure, started advising on the maintenance issues all over the manufacturing Industry.
With the advent of Industrial Internet of Things (IIoT) , couple of start-ups like Augery and Petasense, tried to automate this process via cloud so that machines could be monitored 24x7.
However, there is a catch to this new paradigm of predictive maintenance!
There are two important aspects of FFT based analysis.
First, experts will be running the machine in service mode or in a mode with known RPM so that harmonics of the rotors can be clearly understood. In 24x7, that is not guaranteed as rotor of the machine has to change RPM based on loads, process, etc. In general, that should not be an issue if the vibration sensor can identify the process cycles (all the fundamental frequencies) and baseline automatically according to multiple RPMs.
However, that is easier said than done. So, let's look at the second difficulty.
FFT is Nlog (N) algorithm. Even a good FFT for gearbox will need 2000+ samples. Such FFT would require at least 1GB RAM to execute within few tens of seconds. As a result, given the CPU and memory limitation, a system will be able to pick up a maximum of 2-4 FFT per 10 minutes of batch. Essentially, that is 2-4 seconds of 600 seconds of total operation. But to understand process pattern using vibration, samples have to be collected continuously. That is an issue, if vibration analysis has to understand the process pattern as well as isolate each class of data separately for reliable vibration analysis of failure.
This is where Time domain-based analysis of MachineSense (patent pending) scores over traditional FFT based analysis in predictive maintenance.
Set to understand vibration pattern in each second, it can quickly classify different modes of vibration that exists in a machine due to process variation and only then a meaningful failure algorithm can start. Besides, because it takes a sub-sampling approach, it is more sensitive to failure compared to frequency domain, which means even an early failure can be detected and can be trended properly.
In addition, since time domain does not require periodic pattern, this analysis is universal and has been applied to very specific cases like oil leak, oil change, press machine, etc. Which goes to show why the MachineSense method of vibration monitoring and predictive maintenance is superior to that of its competitors.
With the advent of Industrial Internet of Things (IIoT) , couple of start-ups like Augery and Petasense, tried to automate this process via cloud so that machines could be monitored 24x7.
However, there is a catch to this new paradigm of predictive maintenance!
There are two important aspects of FFT based analysis.
First, experts will be running the machine in service mode or in a mode with known RPM so that harmonics of the rotors can be clearly understood. In 24x7, that is not guaranteed as rotor of the machine has to change RPM based on loads, process, etc. In general, that should not be an issue if the vibration sensor can identify the process cycles (all the fundamental frequencies) and baseline automatically according to multiple RPMs.
However, that is easier said than done. So, let's look at the second difficulty.
FFT is Nlog (N) algorithm. Even a good FFT for gearbox will need 2000+ samples. Such FFT would require at least 1GB RAM to execute within few tens of seconds. As a result, given the CPU and memory limitation, a system will be able to pick up a maximum of 2-4 FFT per 10 minutes of batch. Essentially, that is 2-4 seconds of 600 seconds of total operation. But to understand process pattern using vibration, samples have to be collected continuously. That is an issue, if vibration analysis has to understand the process pattern as well as isolate each class of data separately for reliable vibration analysis of failure.
This is where Time domain-based analysis of MachineSense (patent pending) scores over traditional FFT based analysis in predictive maintenance.
Set to understand vibration pattern in each second, it can quickly classify different modes of vibration that exists in a machine due to process variation and only then a meaningful failure algorithm can start. Besides, because it takes a sub-sampling approach, it is more sensitive to failure compared to frequency domain, which means even an early failure can be detected and can be trended properly.
In addition, since time domain does not require periodic pattern, this analysis is universal and has been applied to very specific cases like oil leak, oil change, press machine, etc. Which goes to show why the MachineSense method of vibration monitoring and predictive maintenance is superior to that of its competitors.
