Inside of Machine

Predictive Maintenance with Machine Learning and IoT

Maintenance of machinery is vital in all industries.  Not only does maintenance cost time and money, but improper maintenance could lead to inefficient performance or potentially be a safety hazard depending upon the purpose of the machinery.

For example, if you’re operating a transportation company, your vehicles will eventually need maintenance to keep running in top condition as well to prevent accidents from occurring from a catastrophic malfunction.  Here it would be much better, and potentially cheaper, to know about these situations prior to them occurring, without the need for a mechanic to check out each individual vehicle around the clock.

General Overview

Sensors gather our data, which will then be sent to the cloud for storage.  This data can then be visualized for easier digestion by humans, utilized for machine learning to gain valuable new insights into when maintenance is needed, or provide notifications to necessary parties.

Phase 1: Sensors

There are numerous types of sensors that can be included with machines to provide data about that machine. Some examples of sensors are: temperature readers, cameras, microphones, motor statistics.

The location of these sensors varies based upon what data we are trying to record.  For example, a camera sensor checking on the condition of the outside of the machine could either be an always-recording camera attached near the machine or it could be an autonomous robot that patrols from machine to machine periodically checking on the condition of the machine.  A sensor getting details about the motor, for example, rotation speed, would need to be connected to the motor.

Phase 2: Transmission to the Cloud/Private Server

Now that streams of data about our devices has been captured, how are we going to keep that data so we can eventually use it to gain valuable insights?  The first thing we should do is store it somewhere secure for continued accessibility. All three major cloud vendors (AWS, Azure, and Google Cloud) provide large database solutions for big data.  

Phase 3: Data Visualization

Having all of this data present does nothing unless we can work towards actionable conclusions.  A simple step towards this goal would be to visualize the data so that trends could be seen by key decision-makers.

The simplest form of data visualization is just showing some charts, graphs, and tables.  However, sometimes animations and time lapses will show some of the bigger picture, especially in the case of any audio/visual data collection.  Seeing the changes in the images helps to extrapolate trends that could show machines are failing faster than normal.

Phase 4: Machine learning

Machine learning allows us to teach computers to tell us if a machine is failing.  This can be done using a combination of photos, audio, and other sensor readings. Having historic data is necessary, so this phase may not be able to be implemented immediately

For example, if you have a large enough sample size of images of machines in proper working order, and those in a failing state, we can use those to create a program that will tell us when the images from the camera feed indicate that a component is failing.  There is a potential that we can also use sample images to create a program that will tell us approximately how long until a machine would fail if no maintenance was done to it.

Machine learning isn’t just capable of telling us that a machine is currently failing, but it can learn to tell the signs that machines will need maintenance and when to keep them in full working order.  Once it has learned how to predict the window when maintenance will need to occur, maintenance can be performed when it is most economical, potentially lowering maintenance costs by having to service machines when they fail or having to maintain them too often and when it is not necessary. 

Phase 5: Real-time Alerts

Finally, if anything is discovered that requires urgent attention during the automated processes, alerts can be sent to any necessary parties (via email, push notifications, text messages, etc.).  These alerts could then be acted upon as necessary.

Date posted: November 22, 2019

Add new comment

Restricted HTML

  • Allowed HTML tags: <a href hreflang> <em> <strong> <cite> <blockquote cite> <code> <ul type> <ol start type> <li> <dl> <dt> <dd> <h2 id> <h3 id> <h4 id> <h5 id> <h6 id>
  • You can enable syntax highlighting of source code with the following tags: <code>, <blockcode>, <cpp>, <java>, <php>. The supported tag styles are: <foo>, [foo].
  • Web page addresses and email addresses turn into links automatically.
  • Lines and paragraphs break automatically.

Metal Toad is an Advanced AWS Consulting Partner. Learn more about our AWS Managed Services

Have questions?