The Buzzwords Industry 4.0 and Smart Factory have been in the spotlight for the last few years and more and more manufacturers are establishing IIoT (Industrial Internet of Things) standards by connecting machinery and production plants with Sensors and Edge Devices. Companies are leveraging IoT to gain their competitive edge and transform the industry sensor by sensor.
With 5G Networks emerging all over the globe, sensors getting more reliable, and edge devices increasing computational capacities with each new release, it has never been easier for manufacturers to take the first steps into the new 4.0 domain.
Over 10 billion IoT devices have been installed by the end of 2020 (IoT Analytics) and the numbers are ever-increasing.
The purpose is clear: IoT integrated machinery pushes the boundaries for manufacturers; providing the opportunity to make data-driven decisions. If manufacturers want to gain an advantage by having a complete view from a multitude of sensors and gain cost benefits from efficient operations, analytics, or even predictive maintenance, then high-quality data is the golden ticket.
Let’s find out why and where to begin.
A stable network is the starting point
Many of today’s factory floors already have some extent of a network connection already established — whether it is wifi / LAN for the adjacent office or a proper M2M network over an industrial standard protocol (MODbus, CANbus, OPC-UA, Siemens S7, BACnet, etc.). However, IoT raises the bar when it comes to challenging your network, in terms of connectivity, latency, bandwidth, power & data.
The first step for manufacturers begins with either upgrading older equipment, exchanging constrained devices for rich nodes with more computational power, or taking it one step further; establishing an edge network directly on the premise. Whatever the scope of the project may be, the stability of the incoming data determines how much value you can get out of your machines and sensors.
One of the many show stoppers for your hard-earned project is connecting your devices to extract meaningful data.
Choosing between an “On-Premise” vs. an “In-Cloud” solution
Once you have connected your devices to the internet, then you need to determine what data you are collecting and how you are processing it.
The most common mistake that people make at this point is — “let’s just connect everything and send all the data — we’ll figure it out later!”
Yep, that’s how you break IoT.
In several circumstances, it can be far more clever to consider an on-premise data logging solution and determine the select amount of data points that need to be processed in the cloud.
During your data collection period, you want to determine the best possible set-up to ensure your data flow is as stable as possible. But for now, let’s assume your machines are equipped with sensors and your network connection is established!
Making use of your data
Congratulations, you now have all of your sensors connected and are receiving your precious data sets…but way too many of them. On top of the vast amount of data you are receiving, one set of machines has data formatted conforming to totally different standards, your time series do not match due to a small difference in clock speeds, and your key data points are buried in the rubble of raw sensor data and noise.
With many different sensors, machines, and use cases for each data set, the next challenge is to treat and clean your raw data, which can be the most complicated part.
This process can be summed up in 3 parts:
1. The right data in the right table
Collect all your sensor data securely and in a readable format. This can be a challenge if you are working with multiple sensors and different communication protocols. At this stage you would need to orchestrate and consolidate all your data from the sensors into one streamlined data pipeline.
Once you have the data accessible in a readable and logical format, it’s time to trim down.
2. I don’t need this and I don’t need that
You now have huge tables and data sets, and they are ever-increasing by the minute with sensors sending real-time data to your database. Here, the data needs a rough cut, which means removing outliers and experimental data sets to protect your statistical analysis, filtering out the data noise you really don’t care for, and making sure your time series are aligned.
Make sure you have taken care of all the inconsistent and incomplete data sets, to stay within the expected behavior of your machines.
3. Getting what you want
Now that you have cleaned up your data sets, it’s time to get to the interesting part. That is, extracting the valuable information from the data sets. Instead of feeding raw unfiltered sensor data into your analytics model, you want to look at high-level conditional indicators. This reduces the strain on your network and enables a faster computing model. Here you extract your sensor’s frequency domains and other conditional indicators, depending on your project scope.
Iterate, iterate and iterate again
It’s done! Your data is cleaned, you extracted the necessary data sets, the data engineers are overworked and now your data scientists can jump in to start working with the analytics models. And now last but least, the big question for any analytics model:
How often can you iterate the projected models and how quickly can you adjust your parameters?
This is a crucial step, and mostly dependent on your infrastructure setup, and network capabilities. To have an efficient and working analytics model, you need to be able to quickly iterate and make parameter changes as needed in order to train your model. Being able to deploy your model quickly for testing and benchmarking purposes can be a great benefit, as this bottleneck will otherwise slow down the project with every other machine affected.
Implementing IoT can be tricky, with issues usually arising where you didn’t expect them and fundamental errors arising to the surface only in the later stages. The manufacturing industry is still in a trial and error phase, testing out new hardware and infrastructure solutions, as there is no set standard or best practice. The applications can be vast and the number of possible means to get there even more intimidating.
Nonetheless, whatever use case you want to establish, from simple analytics dashboards to new revenue streams with pay-per-use and blockchain, start humbly at the beginning. And that beginning is your data, where it is coming from, it’s quality, where it is going, and who can or should receive it.
Please don’t start with the analytics model, will you?