“Big Information” instead of “Big Data”

This short article will not repeat what you can read about Big Data. Almost all media channels cover this topic although the majority of these features scratch on the surface when outlining prospective industrial applications based on the new technology. “New” is not quite correct because “Big Data” was known in the 1990´s as “Data Warehousing” to be analyzed with “Data Mining” approaches and technologies.

The author of this article will show you in detail how Big Data can be used today in form of intelligent Online Diagnostics of Rotating Equipment. We start our journey with a well-known example of data collection and aggregation: a 6 hours flight of a twin engine civil aircraft produces 240 terabyte (TB) of data. These are analyzed onboard and feed the EICAS (Engine Indicating and Crew Altering System).

How compare the 240TB to the data analyzed of an Online Condition Monitoring system that runs 24/7 to safeguard a piece of critical Rotating Equipment?

As an example, we take a 4-cylinder reciprocating compressor with standard sensor instrumentation:

  • One vibration sensor on each crosshead slide
  • One vibration sensor on each cylinder
  • One dynamic pressure probe in each compression chamber (double acting)
  • One proximity probe on each piston rod
  • One trigger for speed and phase reference

Total: 21 sensors

The math lesson starts here:

To get a really good picture of what is going on inside the machine, a sampling rate of 25 kHz is highly recommended. Sampling rate x sensors = 525,000 samples per second. Each sample is 2 byte x 525,000 = 1,050 kb per second. Take this x 60 (seconds) and you receive 63 MB of data acquired and analyzed per minute. 63 MB per minute add up to 91 GB per day. Let this sink: 91,000 Megabyte of sensor signals acquired and analyzed in real time to ensure reliable machine protection and proper early failure detection. “Big Data” is defined as “Data sets with sizes beyond the ability of commonly used software tools to acquire, manage, and process data within a tolerable elapsed time”. Having in mind that our example is about machine protection of a critical production asset, it is obvious that a lot of specialized technologies are required to continuously monitor and safeguard the 91 GB reciprocating compressor.

What makes a Monitoring system capable to handle this task?

Several technologies are required to efficiently analyze the gapless stream of sensor signals and produce meaningful results for machine operators. “Machine learning technologies” is the umbrella term for what is needed. At PROGNOST Systems, we use Artificial Neural Networks; i.e. Learning Vector Quantization (LVQ) for machine diagnostics. And also, in parallel, Fuzzy Logic and Rule-based diagnosis.

Relax – I will waste your time with explanations how these Deep Learning disciplines work. Your take away are the results: Machine failure pattern recognition with clear text messages for the operator. PROGNOST® systems come with an integrated database of machine failure patterns that we identified in the monitoring data of millions of machinery operating hours. The failure patterns we are using to provide meaningful messages are real-life and not artificially produced in a R&D lab. Users rely on them and, better still, add individual patterns the monitoring system identified during operation of their machinery.

 A look into future

How will Online Condition Monitoring look like in e.g. 5 years from today? How much “Prediction” is in “Predictive Maintenance”? Will Cloud Services overcome the strict IT-regulations many petrochemical plants established? Stay tuned – I will share my thoughts in this column.