Return to Blog





In the emerging world of predictive analytics, terms like big data, internet of things, and smart factories have left many people wondering where to begin the digital transformation of their factories and companies.

While many machines, including most older or legacy machines, do not intrinsically support data acquisition, this does not mean data collection is impossible. Even on modern machines, the controller may not take the data needed. If this is the case, various types of sensors (accelerometers, current meters, etc.) can be applied to the machines to measure and collect the necessary data.

Commonly-used sensors include devices like non-invasive (inductive) current sensors clamped around a cable, accelerometers glued to a machine chassis, or even reflective tachometer tape applied to a rotating shaft. Each of these types of sensors is capable of endlessly and non-obstructively recording data. Eventually, someone may find that a sensor isn’t recording the desired data, a sensor is in the way, or the data from the sensor simply isn’t needed to assess the machine’s health. When this occurs, these non-invasive sensors can be easily removed.

While it may sound simple, more work is needed than simply hooking up a couple sensors to a machine, connecting them to an ADC (analog-to-digital converter), and endlessly dumping the output to an industrial PC.

One of the first issues most people will run into is data storage. The most blatant example is vibration data from an accelerometer. When sampled at 20 kHz and recorded in double precision (8 bytes of data per number), the storage will be filled at an alarming rate of 156 kilobytes per second per accelerometer. In one single minute, 9.2 megabytes of data will have been generated. That corresponds to 13 gigabytes of data per day per sensor.

Clearly, storing 100 percent of the data is out of the question.

This brings us to data triggering, or the recording of data when a certain trigger occurs. These recordings can be triggered through various events, ranging from a specific time of day to a specific voltage level on a sensor.

Data acquisition software packages will generally provide several different options for triggering data collection and for how long to collect data after a trigger occurs. Once good data has been collected, predictive analysis can begin.

All things considered, measuring additional data than what a machine controller can record is commonly beneficial when trying to determine machine health. Non-invasive sensors like accelerometers can measure vibrations on the machine. Current sensors clamp around cables to measure load current heading to parts like motors. Finally, data acquisition software packages are used to measure only as much data as is needed to avoid the unnecessary waste of storage space and analysis time that occurs when too much data is recorded.

Brian Phillips

About the Author

Brian Phillips is a Mechanical Engineer for Predictronics Corp. with 4+ years of experience in software design and development, vibrating and rotating systems, fault detection algorithms and frequency domain analysis. Brian is the sole developer and project lead for Predictronics’ PDX DAQ, an application that allows users to synchronize data collection from multiple sources for any given period of time. Brian has a Bachelor of Science in Mechanical Engineering from the University of Cincinnati, where he studied intelligence maintenance systems at the NSF I/UCRC IMS Center founded by Professor Jay Lee. Connect with him on LinkedIn.

For more information and weekly updates, follow Predictronics:

Know What Happens Next with PDX!