CNN Optimization for Bearing Fault Diagnostics in Adverse Industrial Conditions

In industrial environments, the reliability of machinery is often compromised by surrounding noise which can interfere diagnostic signals and lead to machine failures. Such failures usually result in unexpected downtime and considerable repair costs, emphasizing the need for more reliable diagnostic techniques. This study utilizes a CNN architecture to enhance fault diagnosis accuracy in noisy conditions. The proposed architecture is designed to extract informative features from complex vibration data even in the presence of substantial background noise. The optimization of CNN parameters is performed to maximize overall classification accuracy over various SNRs. The effectiveness of optimized model was extensively evaluated by using the CWRU Bearing dataset. The results show the accuracy of 88.89 % at the SNR of −4 dB and up to 97.96 % at the SNR of 10 dB. It demonstrates improvements over traditional fault diagnosis methods and can significantly reduce maintenance costs and enhance operational efficiency across industrial settings.