When implementing a predictive maintenance strategy to improve your maintenance processes there are many things to take into consideration. Naturally the assets and the sensors are the core of industrial asset monitoring. The next step is to collect, process and analyze the data being measured by the sensors. This is where things like networks, databases and the cloud enter the conversation.
To understand edge computing in AI based predictive maintenance it is important to get the basics down first. With predictive maintenance solutions industrial assets are being monitored by smart sensors. These sensors measure things like humidity, temperature, motion, vibrations, gasses and other factors that affect the operation and wear of a machine. This sensor-data allows us to closely monitor the status of assets, so we can take action before they break down. But to actually be able to perform time series analysis to understand the collected data, it needs to be processed first.
This is where processors and a network come in. All the asset’s sensors are connected to a communication network, an Internet of Things network (IoT network), over which the data is transmitted to a place where it can be processed and analyzed. The location of this data processing is exactly what edge computing is about.
Not only in industrial settings, everywhere in your daily life too you’ll find devices that collect real time data: smartwatches, smart kitchen appliances, devices like Google Home and Alexa. Those devices collect historical data and often send this raw bulk of data through the network to a cloud service to get processed and analyzed.
Edge computing means that instead of sending all this raw sensor-data to a central spot, the cloud, it stays closer to the device to get processed locally. A smartwatch for example measures your physical activity and heart rate amongst other things, and gives you reminders when you haven’t moved enough. This smartwatch doesn’t send your data to a cloud, but processes it itself.
Edge computing incorporates artificial AI and machine learning models to create intelligent edge devices and networks, which can be integrated with the cloud for further analysing and processing.
In earlier years the cloud was the central place where all data was being sent to and processed. These days however, there is an exponential growth in smart devices which causes the amount of data being transmitted and processed to increase massively. And this will only increase more in the next couple of years up to a point where it’s not manageable anymore. This is one of the reasons why it’s recommended to compute data on the edge.
The exact definition of the edge computing can differ per situation. It could be that the data is processed within the device, in which case it’s called ‘device edge’. But in some situations a ‘network edge’ (from the device to the gateway) or ‘telco edge’ (from the device to the radio tower) might be preferred, depending on the type and amount of data.
The fact that we’re moving our data processing more and more to the edge is being made possible by the development of better components that require very little power. In the past data processing required a very heavy processor that would need 230V or a huge battery pack to process the data. That’s not ideal and very expensive so to avoid that, all the data was just being sent directly to the cloud. Nowadays however, processors require less power, can process more data and are becoming smarter, allowing us to operate more on the edge of the network.
5 benefits of edge computing
- Enhanced speed
- Bandwidth relief
- improved data management
- Better data security
- Improved reliability
Besides the power issue, sending, processing and analyzing huge amounts of data generated by local sensors to and in the cloud takes a lot of time and requires a lot of bandwidth. Moving a big part of this process to the edge reduces latency (making your application faster) and cuts the costs for bandwidth and data transmission. It also greatly improves security, as your raw data is processed locally, close to your assets instead of sent over a network to a central location somewhere far away.
In industrial assets we measure things like humidity, temperature, vibrations, gasses and more. All of this combined makes up a tsunami of data. Sending all of this to the cloud all the time, means you’ll be transmitting continuously, which costs a fortune. While ninety percent of the data will be thrown out anyway, because you don’t need it. You only need the data when an anomaly occurs in order to take action. So what happens is, you define predictie model with certain situations upfront and only send data to the cloud when one or more of those situations is being observed by the local processor. When the data indicates that an interference is required, that action will eventually be taken in the cloud and not on the edge.
Because of improved processors we can now also move data processing for machine learning from the cloud to the edge. Machine learning takes AI predictive maintenance a step further. Instead of just relying on the upfront determined anomalies by human beings, a machine learning model uses artificial intelligence (AI) within the sensors to constantly adjust the algorithm based on past and continuous data. This means that the predictions will get increasingly accurate the more data is collected so there is not need for scheduling maintenance which referred as preventative maintenance. With AI based predictive maintenance you can detect equipment failure before it happens, and therefore lengthen the lifetime of your assets and also reduce maintenance costs.
In this blog you can learn more about the use of AI in predictive maintenance.
AI predictive maintenance will limit most of its work to the ‘device edge’ and ‘gateway edge’. Which one depends on the type of assets, the size of the (group of) assets (e.g. a container ship, bottling factory or steel plant) and the nature of the work being done. If you only need data from one asset, you can stick to the device edge.
An asset often has several sensors/devices to measure different things. All of this data is collected in an IoT box close to the machine, still called the device edge. That box puts all the data together and sends the necessary information (i.e. processed data) on to the cloud. A bigger plant might operate on the ‘gateway edge’. Instead of processing the data on the machine itself, all the data of all the assets is being collected and processed in a central processor on the gateway.
Interesting in knowing how you can implement predictive maintenance? In this e-book we outline a step-by-step guide to get started with predictive maintenance. Grab your copy for free!
Experts are saying that in the future the edge computing will be expanded even more. With the arrival of 5G technology an even higher speed enabled us to expand the edge one step beyond the gateway: to the radio tower, creating a telco-edge.
Edge computing has many benefits over cloud computing. It saves time, money and energy and it improves the security of your data. With ongoing technological innovations the things we can do on the edge will expand even more in the future.
When we build applications for our clients we look at the best way to shape the IoT-chain and where to compute the data, while always maintaining the best security. Want to talk more about edge computing and/or asset monitoring sensors? We’re always happy to help out with specific business cases. Get in touch with us.