Pavel Konecny, CEO of Neuron soundware, discusses his solution to predictive maintenance to further the efficiency of machinery as we enter Industry 4.0
Over the last two years, I have been working on analysing audio signals obtained from heavy machinery, engines, pumps, wind turbines, escalators, pointing machines, and air conditioning systems. Leading European manufacturing companies such as Siemens, Airbus, Volkswagen, and MAHLE are all exploring the potential of the innovative approaches to conditional monitoring of machinery as well as their products.
There are almost no limits for the application of these approaches; for an audio diagnostic application it is anything that has a moving part that produces sound. Of course, it makes more sense to focus on the critical pieces of machinery, expensive assets, or on assets in remote areas with problematic access, first.
No matter what they might say, manufacturing executives love production assets, and machines represent the heart of their factories. But, even the best machines break down occasionally and need costly maintenance over their entire lifetime. In today’s super competitive manufacturing environment, a manufacturer’s ability to detect a malfunction early, and thus avoid unplanned maintenance, could provide a critical competitive advantage.
Smart production assets and smart maintenance
It’s quite natural that in Industry 4.0 and digitalisation strategies, manufacturers stream their early efforts for production asset management. The new concept of the industrial Internet of Things (IoT) attracts many of them. Manufacturers can discover a whole new array of possibilities for measuring and assessing the quality of their machines on the shop floor, and to increase automation and improve quality, by using:
- Open communication standards;
- Smart sensors and controllers; and
- Connecting assets to the network and to the internet.
The reality, however, is that many European manufacturers are burdened with old, often bespoke and very expensive assets that are critical to ‘run the business’. These assets are everything but ready for IoT. Besides considerations about buying new ones (which likely is not possible, or considerably costly) or inefficient retrofitting options, manufacturers should ponder one of the oldest methods that the industry has used for decades – sound diagnostics.
In a nutshell, an experienced maintenance technician can detect unusual sounds from a machine, make a judgement as to what the issue is, and prevent a breakdown. Such skills are precious and sought-after as they can save companies millions.
How smart can we get?
Sound diagnostics. That’s what our maintenance staff has done on machines for years, you may say. But, to what extent and with what results? Experienced maintenance technicians are a scarce and expensive resource; having these technicians present at all locations, and machines, is impossible.
Think of modern digital technologies like AI which are augmenting human skills – human senses, including ears, are imperfect. Therefore, even the best maintenance technician can diagnose a mechanical issue with limited accuracy. However, if technology is used in this process, these existing skills can be enhanced to near-perfection, potentially raising mechanical issue sensing accuracy to over 99%. Manufacturers can have cheap IoT devices listening to their machines continuously, without tiring or retiring. As a result, all frequency ranges of the acoustics emissions from low vibration to ultrasonic can be analysed. Businesses can carry out this process very quickly and without security risks, as there is no IT integration required. This is a low-risk, low-cost initiative that can increase:
- Asset uptime;
- Production quality; and
- Lower maintenance costs.
Augmenting maintenance capabilities with technology like AI, together with an investment strategy in smart production assets, will bring an enormous kick in competitiveness for European industry. It’s not one or the other. Building super-human maintenance capabilities, however, is typically a process handled in a separate flow. The majority of machinery suppliers are not ready to help in the field of digitalised sound diagnostics. Embedded diagnostic systems leveraging AI solutions are very rare, which means that manufacturers will have to seek specialized providers who will be able to integrate audio technology within machines, record and process audio data with AI, and predict asset breakdowns.
Audio diagnostic technology
In terms of technology, manufacturers will need no magic. Advanced pattern recognition algorithms – deep learning – is involved. It is the vast availability of computing power that allows AI to be pushed to new limits. In addition, the trend is to shift more computing power to the edge of the networks. This will allow the running of more and more complex neural networks models directly at IoT devices.
The audio part consists of high-sensitivity microphones: acoustic piezo sensors that can sense sounds at any frequency (far beyond the limits of the human ear), and an audio card. Recordings are streamed in real-time onto a cloud platform where neural models are pre-trained. The more audio data you collect, the smarter the algorithms you can build, and the more accurate the results. In more advanced stages of AI learning, machines will form a living organism, automatically updating and continuously learning from the shared experiences.
Data analyses, including automatic detection of machine malfunction, are available to operators in just seconds. They display results on demand in a visual dashboard, or via push notifications on a mobile device. Results can also be used as triggers for the automation of certain work tasks.
Predictive maintenance: Use cases and proof of concept
Savvy manufacturers work in an evolutionary mode, enhancing their production environment through a constant flow of initiatives that carry out innovative technologies and approaches.
We tested the technology on machinery of all kinds. Proof of concept projects usually take two to four weeks to familiarise production assets and set up the technology on the shop floor. Practically speaking, the set up only needs direct current (DC) electric power for IoT devices, allowing a similar time frame to perform recording analysis and neural network training. In one to two months, companies can see real data analysis and assess the impact of mechanical issue sensing (enabled by technology) on their own machines or products in greater detail.
At this point, companies need to decide the style and scale of future deployment. The audio analytics solution provides a consistent predictive monitoring platform across machines of all kind. During the proof of concept phase, manufacturers may find other use cases that could have worked better. I urge manufacturers to try them, too. I have a very positive experience working with different factories owned by one company, each testing a different application of sound diagnostics in parallel.
The future of sound diagnostics
My vision is to create a better life for both machines and their owners. I believe that one day a complex sound diagnostic will become a standard feature of all machines with moving parts. There’s no doubt that asset maintenance is amongst the most current areas being explored by European industry. In particular, companies are demanding solutions with predictive features, real-time operation, and remote diagnostics. To fulfil this vision, we are ready to build an ecosystem of partners that will help us with:
- Development of our technology;
- Audio data collection;
- Better integration to third-party solutions; and
- Scaling up implementations.
Looking towards the near future, I’m aiming to develop a standardised analytical platform that could be used for audio analytics of any machine in any industry.