Review of research from 2002 to 2025 and future guidelines

Being aware of the intensity of cavitation, the formation and collapse of vapor bubbles in a fluid, is critical to ensuring the safe and efficient operation of industrial machinery and has a significant impact on maintenance costs. Yu Sha, Ningtao Liu, Haofeng Liu and colleagues systematically review the evolution of machine learning techniques applied to this important task, tracking developments from 2002 to 2025. Their work addresses gaps in existing research by providing a comprehensive overview of how advances have been made to automatically analyze complex data, initially relying on manually defined characteristics and more recently harnessing the power of deep learning. This review not only presents the historical trajectory of cavitation intensity recognition but also highlights promising future directions, such as transfer learning and integrating physical understanding into diagnostic models, providing valuable guidance for researchers and practitioners working on complex industrial systems.

Intelligent cavitation intensity recognition trend analysis

This study pioneered a systematic investigation on Intelligent Cavitation Intensity Recognition (ICIR), analyzing hundreds of publications from 2002 to 2025. This study carefully tracked the evolution of ICIR technology and revealed a clear progression from traditional machine learning to advanced deep learning methodologies. The researchers conducted a statistical analysis of relevant publications to identify the main technical characteristics and development trends and established a framework for dividing the research into distinct phases. Initially, research focused on traditional machine learning (TML) techniques. There, scientists relied on manually designed features extracted from signals such as acoustic emissions, vibrations, and pressure measurements.

Although there was less reliance on manual judgment at this stage, feature construction was still a labor-intensive process that relied on expert knowledge and limited the generalizability of the model. More recently, research has shifted to exploring deep learning, propelling ICIR into the era of end-to-end modeling. Scientists applied deep belief networks, convolutional neural networks, residual neural networks, dense convolutional networks, MobileNet, ShuffleNet, recurrent neural networks, long short-term memory networks, gated recurrent neural networks, and transformer architectures to automatically capture cavitation features directly from raw multi-source signals. This approach significantly improved real-time performance and stability under complex operating conditions, enabled comprehensive cross-signal type analysis, and improved diagnostic robustness.

Looking ahead, this study highlights the importance of standardized, high-quality multi-source data acquisition and the design of deep learning diagnostic models based on physical information. Scientists aim to increase model generalizability, interpretability, and engineering confidence by incorporating physical mechanisms, knowledge embedding, and constraint loss into data-driven deep learning. This approach captures the nonlinear and multiscale characteristics of the cavitation process, ensuring that the diagnostic output closely matches the actual physical mechanism, and promises to pave the way for sustainable monitoring and predictive maintenance in industrial environments.

Intelligent cavitation recognition, three stages of evolution

This study details a systematic analysis of intelligent cavitation intensity recognition (CIR) research spanning from 2002 to 2025, revealing distinct advances through three distinct phases. Initial research conducted until around 2010 relied on traditional machine learning (TML) techniques using algorithms such as support vector machines, decision trees, and artificial neural networks. These approaches were limited by the need for manually designed features, preventing generalization across different operating conditions. From 2010 to 2020, the field moved towards data-driven deep learning, which demonstrated significant improvements in real-time performance and stability.

By applying architectures such as convolutional neural networks, recurrent neural networks, and transformers, automatic feature extraction from multi-source signals such as acoustic, vibration, and pressure data is now possible. Current and future research is focused on integrating physical knowledge into deep learning models with the aim of improving model interpretability and generalizability. This includes incorporating physical mechanisms and leveraging physical constraint losses to ensure that diagnostic outputs match real physical processes. The team’s analysis predicts that this approach will become important for handling high-noise data with fewer samples and achieving reliable cross-condition monitoring. This integration is expected to provide sustainable monitoring and predictive maintenance solutions in complex industrial systems. The team’s work builds on existing reviews and provides a comprehensive, integrated framework that addresses the limitations of previous research by encompassing a variety of methods and devices for a more comprehensive comparison of feasibility and cost-effectiveness.

Deep learning advances cavitation intensity recognition

This review systematically traces the evolution of cavitation intensity recognition, a critical process for maintaining safety and efficiency in hydraulic machinery. Researchers have moved from relying on manual inspection and operator experience to employing automated and intelligent systems for cavitation detection and evaluation. Early approaches utilized traditional machine learning techniques and required careful manual feature engineering based on expert knowledge. Recently, the field has been transformed by the application of deep learning models that can automatically extract relevant features from multiple data sources and significantly improve recognition performance. Current research also considers integrating physical knowledge into these deep learning models to enhance both their interpretability and ability to generalize across different operating conditions. Future research is expected to focus on developing transfer learning, multimodal data fusion, and lightweight network architectures to facilitate the deployment of intelligent agents for real-time monitoring and diagnostics.