It has been a while since I heard “don’t try to solve factory challenges with office solutions” the first time. Several decades to be precise, but it is still a valid rule and every now and then I hear myself saying it to some of the cloud advocates who haven’t quite understood industrial environments yet. As analyst and consultant, with a strong background and network in industrial infrastructure and manufacturing, I am always on the edge so to speak when it comes to industrial cloud applications (all pun intended). Is industrial cloud feasible?
My priority for AWS re:Invent 2021 is of course to identify solutions for manufacturing and industrial environments where customers can embark on a digital transformation, and inspiration for conversations with the vast majority of industrial decision makers that is still skeptical about the cloud in their factories. All this following the golden rule “don’t try to solve factory challenges with office solutions”.
During a breakout session for the DACH region, Continental showcased their solution platform build on AWS, and I could already hear the criticasters whisper “yeah, but that is not running the factory”. Fair enough. Although it does demonstrate the reliability of cloud solutions, it still isn’t embedded in the aorta of factory automation.
The AWS DACH Team didn’t take long to satisfy my desire for a real industrial application from their ecosystem. The Digital Builders Showroom demonstrated ML based leak detection for tanks and pipes, and my Industrial Automation heart started to beat a lot faster.
Leak detection is a challenge, and many have tried to crack that nut with for example Neural Networks. Problem: it takes forever to train, and the detection models are very specific. We are not talking about very abrupt leaks when a tank or pipe would burst. We are talking about the slow creeping leaks that develop over time and stay below the radar screen.
Think about for example compressed air. Compressed air is essential in many industrial processes, and although “air is free”, the required energy to deliver compressed air isn’t. Compressors are energy hungry and generate a lot of heat. Compressed air is expensive air! Leaked compressed air is a pure waste of money and energy.
Leaks occur everywhere. Couplings and hoses suffer from wear and tear. Heat doesn’t do these much good either. The leaks are difficult to detect because they are a relatively small variation of the industrial compressed air consumption and patterns. The difference is in that very small delay in pressure availability, or in that very slow decline of pressure. Hard to detect vibrations can be a clear indicator for leaks. Even the sound of a compressed air leak is very distinct but also very difficult to detect in an industrial environment.
The right mix of sensors, data, and Machine Learning can create a leak detection model for compressed air, including those hundreds of very small leaks that add up to a lot of wasted energy and money. The benefit of the AWS ecosystem is that this be can build with available services, and the learning curves and models can be shared between facilities and locations.
A practical example for this would be that a particular brief vibration in combination with a specific slow and small decline of pressure is confirmed by a maintenance team as a leaking coupling. The cloud enabled ML platform distributes this identification across all instances. As the ML engines continue to learn and refine the detection, it will also learn from those cases where the analyzes detected a potential leak, but the responsible maintenance teams could not confirm it.
This is just one practical example of how cloud enabled technology can deliver solutions for industrial and manufacturing environments, without trying to solve factory challenges with office solutions. Industrial cloud is feasible! We just need to make sure that we select the right solutions and partners!
A good way to start Digital Transformation for Manufacturing are the online educations by Dr. ir Johannes Drooghaag: