On the application of artificial intelligence technologies in acoustic recognition tasks

The possibilities of using artificial intelligence methods, in particular reinforcement learning, for acoustic diagnostics of autonomous systems are being investigated. Three algorithms are proposed: algorithm based on deep neural networks, multi-agent ensemble algorithm and dynamically adaptable ensemble with generative reinforcement based on generative-adversarial network. A comparative analysis of their effectiveness was carried out according to the criteria of accuracy, noise resistance, and computing resource requirements. The results show that the combined approach using reinforcement learning and a generative-adversarial network demonstrates the highest accuracy (up to 94.2 %) and adaptability to changing conditions, which makes it promising for implementation in real-time industrial systems. The methods were compared according to three criteria: diagnostic accuracy, noise tolerance, and computing resource requirements. The results showed that the multi-agent algorithm copes best with signal analysis in high-noise environments, and the dynamic ensemble with GAN provides maximum error recognition accuracy (up to 94.2 %) and the ability to adapt to previously unknown situations. At the same time, the RL-based method is characterized by low resource costs and minimal response time, which makes it convenient for systems with limited computing capabilities. The proposed solutions can be used to develop practical predictive maintenance systems that can detect malfunctions at an early stage and maintain trouble-free operation of equipment.

Authors: A. R. Salieva

Direction: Informatics, Computer Technologies And Control

Keywords: acoustic diagnostics, reinforcement learning, artificial intelligence, industrial equipment, generative adversarial networks

---

View full article