N A N O P R E C I S E

Prediction with Precision

Data Streams & Types During my engineering experience, I used to design the asset to certain parameters which we called, Design & Operating parameters (such as temperature, pressure, flow rate, density etc.). For example, a certain flange at the pump nozzle is not supposed to be operated beyond the +-10% of Po (operating pressure) and it is not supposed to be subjected to a pressure beyond a Pl (flange leakage pressure), which is basically its design pressure limit. Operating parameters are meant to be a certain percentage below the design parameters, just to ensure safe operation. Something to observe here is that the above parameters are internal to any system. Let us define what is a system in an industrial world. Without getting too technical, a system in an industrial setting from a structural point of view is defined as a point from the nozzle of one static equipment such a (column, tank, heat exchanger etc.) to another static equipment. In between, there are machines such as pumps, turbines compressor etc. that are needed to force or transform the fluid that travels through the pipe. As shown below:
All the sensor parameters in this system can be categorized as follows:
  • Input characteristics (such as internal temperature, flow rate, pressure, density etc.)
  • Output characteristics (such as vibration, RPM, current, sound, external temperature, pipe wall thickness)
The output characteristics here are the system’s response to internal characteristics. By measuring internal characteristics, we cannot really tell the performance of the rotating equipment (shown as pump above). The simple reason is those input characteristics are usually created from static equipment such as heat exchanger, reactor, agitator, fractionating column. Thus, the rotating equipment just reacts to the variations in the input characteristics. Input characteristics are usually defined during the engineering and design stage of the asset, which means the asset is supposed to operate within the operating temp, pressure or flow rate. Thus, input characteristics are basically range-bound:
Relevance to PdM of Rotating Equipment All the initiatives in industrial IoT serve two basic purposes:
  • Process & System Usage Optimization
  • Predictive Maintenance
Since this article is about Predictive & Prescriptive Maintenance (PdM), we will focus on what is useful to achieve that. After scouring through tonnes of resources, I feel following are the major tasks in Predictive Maintenance:
  • Anomaly Detection
  • Fault Characterization
  • Remaining Time to Failure Determination
An anomaly can either be due to a process or system upset or due to an actual fault in rotating machinery. To accurately characterize an anomaly, whether it is due to a process upset or due to an actual fault or a combination of both, is probably the biggest challenge for the AI algorithms behind the plethora of IIoT software (sold by “reliability philosophers”). A rise in vibration, sound, current or RPM (input characteristics) can be correlated to a rise in flow rate, pressure or temperature (output characteristics). If the correlation is really obvious, this is basically a process upset. However, just monitoring the output characteristics can also tell whether the anomaly is a process upset or a machine fault. This is a topic for my future articles. Thus, input characteristics are not really necessary for anomaly detection. The fault characterization & Remaining Time to Failure prediction can only be achieved through analytics on output characteristics. Input characteristics being range-bound will not really show any trend (like output characteristic) that signals an increasing severity like below:
Thus, even though it is tempting to gather as much data as possible from the SCADA, DCS & other such sources in cheapest possible way, because that really suits the (VC investment principle, invest in software development, not the hardware development), its only the output characteristics that are really relevant to PdM for rotating equipment. We at Nanoprecise understood this long back and focused simply on this approach. I will close by saying that the Quantity of Data might actually be a bad thing as it will simply cause you to spend your computational efforts somewhere else. When it comes to PdM in Industrial 4.0, its the Quality of Data that matters.

Leave a Comment

This site uses Akismet to reduce spam. Learn how your comment data is processed.