Health monitoring of a rotating machine is mainly done by investigation of the vibration patterns generated by the machine. Leveraging the fact that faults occurring in different parts of a machine generate unique fault signatures, a fault diagnosis methodology is proposed that can identify nine different healthy and faulty categories under varying load and noisy conditions. Neural network is employed for classification of faults in various categories. The robustness of features such as semivariance, kurtosis and Shannon entropy make them strong candidates to train the artificial neural network. The matching of vibration textural patterns with wave atom basis functions ensures removal of noise. As a result, the enhanced features used to train the neural network have led to high accuracy in classification. The algorithm is tested at various load conditions for both bearing and gear fault experimental data sets acquired by machinery fault simulator in laboratory. Simulation results show high degree of accuracy for both bearing and gear fault diagnosis under no load to heavy load noisy conditions.

旋转机器的运行状况监视主要是通过调查旋转机器产生的振动模式来进行的。利用在机器的不同部分中发生的故障会生成独特的故障特征这一事实，提出了一种故障诊断方法，该方法可以识别在变化的负载和嘈杂条件下的九种不同的正常和故障类别。神经网络被用于各种类别的故障分类。半方差，峰度和香农熵等特征的鲁棒性使其成为训练人工神经网络的强大候选者。振动纹理模式与波原子基础函数的匹配可确保消除噪音。结果，用于训练神经网络的增强功能导致分类的高精度。在实验室的机械故障模拟器获取的轴承和齿轮故障实验数据集的各种负载条件下，对该算法进行了测试。仿真结果表明，在空载到重载噪声条件下，轴承和齿轮故障诊断的准确性很高。