Empowering the Future of Maintenance: A Conversation with Pavel Konecny, Founder of Neuron Soundware
What inspired the founders of Neuron Soundware to focus on this area, and how does your AI-driven technology offer a unique approach compared to traditional methods of maintenance?
We were brought to this topic by a real story of my friend. He was driving his car, he could hear a strange noise coming from the engine. He even stopped at the service station, however there was no error on the onboard computer. So he continued driving. And about 100 km later, the cylinder broke, it blew up the whole engine and the car suffered 10.000 EUR damage. Luckily for him, it was 2 days before the warranty period ended. The human ability to learn how machines sound, to recognize all kinds of anomalies is what we replicate in software using deep neural networks. We are using the pre-trained model on a very large database of sounds, so we can be monitoring rotary but also none-rotary machines with different operating modes. The whole setup is fully automated as we are training the AI for each machine individually.
As AI and edge computing continue to evolve rapidly, how does Neuron Soundware stay at the forefront of technological advancements and continuously improve the accuracy and capabilities of your predictive maintenance algorithms?
We have strategically decided for edge computing as an essential feature of our solution. The capability of processing large amounts of data without cloud infrastructure has many benefits. Firstly, the cost of cloud computing is quite high. Secondly, the local process allows immediate reaction and machine-to-machine integration. The data also stays within the factory network perimeter. We can also analyze every second of every minute all year around. That is TB of data every month from a single machine. In addition, there is now issue with the communication bandwidth.
As Neuron Soundware has progressed in reshaping machine diagnostics, what are some key lessons learned and challenges overcome in the process?
We have proven that neural networks are very effective in anomaly detection. We have hundreds of hours of broken machines of all types in our database. As we have also implemented the traditional vibro diagnostic heuristics like signal envelope methods based on the RPM, we can compare the new approach and traditional techniques. The neural networks can detect coming issues more than 4x likely and sooner than the current methods (89% compared to 19%). Our goal is now to work with Large Enterprises and OEMs, who want to innovate their machines monitoring system. We also started working on building the distribution network of value added resellers, so we can roll-out our solution at the global scale.
Neuron Soundware's technology has the potential to disrupt traditional maintenance practices. How do you approach working with clients to transition from reactive to proactive maintenance strategies?
The combination of IoT and AI allows for a larger number of machines to be maintained based on their actual condition. However it seems that more value is created in prevention of the machine malfunction, which leads into the product quality issues. More frequently the more important is to discover the clogging of the material in the mill rather than to save on its mill maintenance. That said, the digitalization of the customer fleet would gradually change the approach to the maintenance from time based or break based to more comprehensive machine health management.
September 2023 | Pavel Konečný for TechFounders
Recent Posts
