Interview with Pavel Konečný for

The Czech startup Neuron Soundware develops technology for automated predictive maintenance of industry machinery. Using sound/vibration analysis of the machines, the technology enables it to detect imminent failures that would mean costly repairs and production outages.

The technology is based on the combination of artificial intelligence built on neural networks and machine learning, and IoT device nEdge, which is receiving and processing the data from sensors. The Neuron Soundware technology is being used by clients in their operations all over the world from the Prague Public Transport Company to the manufacturer of semiconductors in Malaysia.

Tell us about yourself?

After studying cybernetics, I worked in an IT consulting firm where I came up with a lot of ideas and my colleagues often asked me why I didn’t start my own company. I answered that I was waiting for computers to be fast enough to create interesting applications in the field of artificial intelligence.

Early on, we explored ideas for using neural networks for data compression and applied to the StartupYard accelerator program, where we received our first investment. Thanks to the program and a better understanding of the market needs, we moved from experimenting with music or data compression to voice and finally to machine sounds. We chose machine troubleshooting because it was new to us and we couldn’t understand why no one was doing it yet. And also thanks to a friend who had a squeaky engine.

What is the inspiration behind your business?

The first use of neural networks in audio was planned in the music industry. However, it was a friend’s life experience that gave us the idea to use it in the field of machine diagnostics. He mentioned that he had a problem with his car.

The sound of the engine had changed, so he took it to the garage. However, neither the on-board computer nor the technicians detected any problem. So he continued driving and 200 miles later a cylinder blew and destroyed the entire engine of his car. He said he was lucky it was two days before the warranty expired.

What is your magic sauce?

Our advantage is certainly the versatility of the method. The non-invasive sensors and microphones are very easy to install and there are no risks of affecting machine operation. In the factory, we only need an electrical socket or a 12V power supply. The service can be started very quickly in a matter of days without the need to build IT-intensive data storage, integrate outputs from machines of different manufacturers, etc.

We can also listen outside the audible spectrum. From the point of view of fault and wear detection, sound has the advantage that anomalies become apparent very early in the sound so that problematic conditions can be detected early and future developments can be predicted. For other analyses, such as temperature, pressure, power consumption, anomalies usually do not become apparent until later – and by then it may be too late.

Where do you see your company going in 5 years?

In the future, we want to venture into robotics. I would like to buy a company that develops robots in five years. They could then become repairmen and solve the faults that our equipment identifies. We are already planning to add various instructions on what customers should do with broken machines. In the next year, we also want to start testing augmented reality glasses that would inform technicians about the state of the machine and directly advise them on how to fix it if necessary. For now, they load information about the machine via a QR code on their mobile phone.

What has been your biggest setback so far?

The biggest challenge when introducing new technology is proper communication with customers, setting expectations and deliverables accurately at the beginning of the collaboration. We had a few misunderstandings with our customers, which fortunately we managed to straighten out over time.

Introducing AI into maintenance management and process change in general in the industry doesn’t happen right away, and the technology delivers more value the longer it is deployed. Artificial intelligence needs to learn. Some customers have high expectations right from the start that algorithms will solve the entire maintenance issue for them right away. That’s not possible today. We see this in self-driving cars, which take a very slow time to deploy.

The more data the AI collects, the better it can make decisions and evaluate equipment conditions. That’s why today we recommend customers deploy the technology on as many machines in production at once as possible from the start so that we can collect data as quickly as possible, reduce learning time, and deliver the value created by the service – accurate machine diagnostics – as soon as possible.