Return to Blog

Blog

Data Aquisition Word Cloud

DATA ACQUISITION SERIES PART III: CLEANING AND INSPECTING YOUR DATA

DECEMBER 19, 2018

In the prior installments of this series, we established our data acquisition system by determining our project objectives, selecting our hardware, and choosing a software solution.

Once you have selected your data acquisition hardware and software, the next step is to determine where to place your sensors. Sensor placement is critical to collecting high-quality data.

For example, if you were examining collected data from an accelerometer mounted to a specific bearing within an engine, an accelerometer mounted to the engine cover, and an accelerometer on the ground near the engine, you can expect the engine bearing vibration to be significantly more apparent in the accelerometer near the bearing as opposed to the one on the cover. Not only this, but the bearing vibration would be least apparent from the ground accelerometer and that data would likely not be usable. This happens because the frequencies are amplified as they pass through the various structural components.

Another example would be observing lubricant temperature in a tank. A thermometer placed directly into a tank will gather a more accurate data set than if the thermometer were placed partially down the outlet pipe.

Occasionally, sensors cannot collect the exact data of interest because they cannot be physically placed in the proper location. While data quality will be degraded, these measurements could help you better understand the effects applying to the data.

In the engine example discussed previously, you might see a circumstance where the bearing is unreachable or perhaps the interior of the engine is too hot for the sensor required. The engine bearing vibration can be computed through first measuring the transfer function of the bearing from the engine housing to the engine cover and then factoring the amplification and attenuation out from the vibration measured on the cover. The result will be a fairly close estimate of the engine bearing vibration.

In our earlier thermometer example, a situation could arise where the tank is sealed and insulated, preventing the introduction of the thermometer. The tank temperature can be computed by taking the rates of conductive and convective heat transfer from the lubricant through the pipe wall and the atmospheric temperature.

These types of calculations factor out the effects of sensor placement while giving you the values you need for your data collection.

Another issue with data collection is random noise, which can be any disturbance in the surrounding environment beyond a user’s control, from the vibration of a person walking past a machine to a loud factory environment creating disruptive sound waves.

There are many different methods of denoising data. One common way is simple averaging. For example, vibration data can be easily averaged in the frequency domain, reducing the effect of white noise from any one data block.

Once your data is appropriately processed, you can invest in data analysis software to compute features and values from your data and to better understand the health of your machine. As mentioned in Part II, edge analytics can also prove helpful in this regard.

It also helps to have someone with data science domain knowledge when interpreting your data. A person with expertise can detect anomalies and false alarms, improving the quality of your data.

Once you have confidence in your data acquisition solution and you have analyzed enough data offline, you can deploy it to a cloud server or host it on-premise using a software application. This type of solution can apply your developed model to new data, as it is acquired by your data collection software, and can scale it to multiple machines or factories being monitored, if needed. This type of software organizes your data and processed results in an intuitive manner, enabling convenient visualization and the ability to set up scheduled reports and alarms for a variety of user roles.

Collecting data is a critical step towards creating data-driven solutions that meet your company’s needs and solve machine failure problems. This will ensure your business achieves its full potential, saving you time, money, and resources.

For more information and weekly updates, follow Predictronics:

Know What Happens Next with PDX!

TRY IT TODAY