基于核极限学习机自编码器的转盘轴承寿命状态识别

针对低速重载转盘轴承运行工况恶劣、故障特征微弱的特点，提出基于飞蛾扑火算法优化多层核极限学习机自编码器（MFO-MLKELM-AE）的转盘轴承寿命状态识别方法. 该方法从振动信号的时域和时频域中提取出多个能够表征转盘轴承运行状态的特征向量，并将其组成高维特征集. 采用堆叠多层核极限学习机自编码器（MLKELM-AE），从高维特征集中提取最能反映转盘轴承的寿命状态信息，输入核极限学习机（KELM）模型进行寿命状态识别. 在MLKELM-AE学习训练中，采用新的飞蛾扑火算法（MFO）优化惩罚系数和核参数，提高MLKELM-AE的特征识别能力. 转盘轴承加速寿命实验表明，MLKELM-AE比多层极限学习机自编码器（MLELM-AE）、单层极限学习机（ELM）、KELM的识别精度高，多传感器、多领域特征能够全面反映转盘轴承的寿命状态.

Life condition recognition of slewing bearing based on kernel extreme learning machine based auto-encoder

A life condition recognition method based on multi-layer kernel extreme learning machine based auto-encoder optimized by moth-flame optimization (MFO-MLKELM-AE) was proposed to solve the problem of low-speed heavy-load slewing bearing, such as poor working condition and weak fault feature. Firstly, multiple feature vectors were extracted from time domain and time-frequency domain of vibration signal to make a high dimension feature set which can characterize the operation condition of slewing bearing. Secondly, multi-layer kernel extreme learning machine based auto-encoder (MLKELM-AE) was utilized to extract the vectors which best reflect the slewing bearing life condition information from the high dimension feature set. Thirdly, the vectors were inputted into the kernel extreme learning machine (KELM) for the life condition recognition. Finally, a new moth-flame optimization (MFO) was proposed to optimize the penalty coefficient and kernel parameter for the improvement of MLKELM-AE feature recognition ability in the training process. The accelerated life test of slewing bearing shows that the recognition accuracy of MLKELM-AE is better than multi-layer extreme learning machine based auto-encoder (MLELM-AE), single layer extreme learning machine (ELM) and KELM. The multi-sensor and multi-domain features can reflect the life condition of slewing bearing more comprehensively.