Using Spresense for predictive maintenance

Spresense PDMSpresense PDM

On March 25, we ran a hands-on webinar together with our partner company Senseye for how to use Spresense in Predictive Maintenance (PDM) solutions. Below is an intro to PDM followed by a link to the technical case study that we went through during the webinar.

What is PDM?

Predictive maintenance (PDM) is one of several schemes used to handle maintenance scheduling of an asset, such as a motor in a factory. Let’s compare PDM with some other maintenance schemes to see why it is so such an important concept:

  • Reactive maintenance is where maintenance only takes place after a failure has already taken place. For example, we don’t replace a light bulb until it has completely failed. For a more complex machinery this could prove costly since a broken part might also cause damage to other surrounding parts once it fails.
  • Preventive maintenance is when maintenance is done according to a set schedule based on estimations and previous experience. The problem is that we don’t know for sure when the equipment would actually fail, so maintenance might be done prematurely resulting in “wasted” lifetime of replaced parts and unnecessary downtime.
  • Predictive maintenance is where the current “health” of an asset is monitored by using a number of sensors to collect, for example, vibrations or sound. By comparing this data with previous measurements, we can detect anomalies and eventually predict when a failure is about to happen. This way, maintenance is only done when needed, which keeps cost and downtime to a minimum.

The basic idea behind predictive maintenance is that all parts in some machinery leave a footprint in the form of audible noise, vibrations and so on. When the machine is not performing optimally or is about to break, this footprint will most likely change slightly because of grinding or parts being under some kind of stress.

The footprint consists of a number of condition indicators which are extracted from the gathered data set using statistical operations and data transforms, such as the fast Fourier transform (FFT). This kind of analysis is often, but not exclusively, done with machines that have moving parts involving bearings. Bearing vibration analysis, for example, is a well-understood scientific field.

In a real-life production environment, an asset would probably have multiple types of sensors attached to it. These would then gather data about different condition indicators, such as temperature or pressure, possibly on multiple moving parts that need to be examined.

PDM with Spresense

The Spresense main and extension board include the following built-in features suitable for use with predictive maintenance:

  • 6 cores running at 156 MHz
  • 4 analog or 8 digital microphone inputs
  • Audio codec with support for recording at 192 kHz sample rate
  • SD card for temporary storage

The Spresense package, consisting of a main and an extension board, is an attractive alternative for collecting and processing data “on the edge” due to its small footprint and processing power. The six cores allows the application to run multiple tasks in parallel, which enables efficient and advanced data processing on the edge.

When complementing this with an add-on module that provides connectivity such as Wi-Fi or LTE, we can do most of the data pre-processing and analysis on the board itself. The aggregated result from this is can then reported to an internal historian/data store, or send it directly to a web service.

This makes Spresense a powerful and flexible tool with a small footprint, which is very suitable for situations where, for example, a PDM solution is retrofitted onto older machinery where space and availability is limited.

Predictive maintenance tutorial with Spresense

Step-by-step tutorial for using “Spresense for predictive maintenance”:

Predictive maintenance with Spresense tutorial

iot
predictive maintenance
Spresense
Webinar