Background <p>Fault detection in rotating machinery, including bearings and gears, has emerged as a significant study domain. Creating an efficient model for the early identification of faults in such equipment continues to be a considerable problem. This challenge is further complicated by noise in real-world industrial environments, which degrades signal quality and affects diagnostic performance.</p> Purpose <p>This study proposes a novel Echo State Network (ESN) design that integrates three reservoirs to augment the network's dynamic capabilities and enhance its robustness in noisy industrial conditions.</p> Methods <p>In the proposed design, three reservoirs are integrated within the ESN framework to expand its dynamic representation power. As a result, the dimensionality of the reservoir outputs increases exponentially. To address this, random projection (RP) and late fusion techniques are employed to reduce dimensionality efficiently. The proposed model is trained using two widely used feature sets—MFCC and GTCC—and their combination. Evaluation is conducted using three publicly available datasets, where multiple levels of noise are introduced during testing to ensure better generalization and performance stability.</p> Results <p>Experimental results prove that the proposed method outperforms the current state-of-the-art methods under both clean and noisy conditions, demonstrating improved classification accuracy and generalization across datasets.</p> Conclusions <p>The enhanced multi-reservoir ESN architecture offers a robust and scalable framework for fault detection in rotating machinery. Its performance under noise confirms its potential for reliable real-world industrial fault diagnosis applications.</p>