基于加速退化数据和现场实测退化数据的电机绝缘剩余寿命预测模型

Motor Insulation Remaining Useful Life Prediction Method Based on Accelerating Degradation Data and Field Degradation Data

该文针对机器学习、随机过程、贝叶斯滤波算法等剩余寿命(RUL)预测模型存在的不足,融合热应力作用下电机绝缘加速寿命数据和现场监测数据,结合随机过程和支持向量机模型,提出了基于拓展卡尔曼滤波的电机绝缘寿命预测模型。以剩余击穿电压为状态变量,基于Wiener过程建立了卡尔曼滤波模型的状态方程;以最大局部放电量的加速退化数据及现场监测数据为依据构建卡尔曼滤波模型的观测方程;为解决卡尔曼滤波模型由于无法获取新的监测信息而导致的预测精度不足问题,采用支持向量机建立了最大局部放电量预测模型。最后,针对电机主绝缘用6650聚酰亚胺,基于290℃、300℃、310℃、320℃下的加速退化数据构建状态方程,结合240℃下试样的局部放电数据构建观测方程,并以240℃下60h的试样实测老化数据为基准对模型进行了验证,证明了所提出模型在提高剩余寿命预测精度方面的有效性。

Insulation system is the weakest part of motor reliability. Monitoring its condition and realizing accurate remaining life prediction is an effective means to ensure the reliability and safety of motor operation.Aiming at the disadvantages of mainstream remaining useful life(RUL) prediction models, including machining learning model, stochastic process model and stochastic filtering model, a motor insulation RUL prediction model based on accelerating degradation data and field state monitoring data under thermal stress, which combines extended Kalman filtering(EKF) with support vector machine(SVM) model and stochastic process model, is proposed. This model is mainly oriented to the RUL prediction problem of motor main insulation with thermal aging as the main failure mode. First, the Arrhenius model is used as the acceleration model, and the mapping relationship between the thermal stress level and the Wiener model drift coefficient and diffusion coefficient is constructed based on the Wiener process. Taking the residual breakdown voltage as the state variable, a prediction model of motor insulation life based on accelerated degradation data under actual working conditions is established, and it is used as the state equation of the Kalman filter model. Secondly, the expression of the maximum partial discharge is deduced by the breakdown voltage estimation equation, and the observation equation of the Kalman filter model is constructed based on the accelerated degradation data of the maximum partial discharge and on-site monitoring data;Then, in order to solve the problem of insufficient prediction accuracy caused by the inability to obtain new monitoring information and the inability to update the covariance matrix of the EKF model in life prediction, this paper takes time as the input variable and the maximum partial discharge as the input variable. Based on the support vector machine, a prediction model of the maximum partial discharge is established to realize the continuous update of the covariance matrix. Finally, for the 6650 polyimide film commonly used in motors, an accelerated degradation test is designed, and the test data of insulation resistance, insulation capacitance, dielectric loss tangent, maximum partial discharge and residual breakdown voltage with aging time are recorded. Based on the accelerated degradation data at 290℃, 300℃, 310℃, and 320℃, the maximum likelihood estimation method is used to construct the state equation. The observation equation was constructed by fusing the accelerated degradation data and the partial discharge data of the material sample at 240℃. The model was verified based on the measured aging data of the sample at 240 °C for 60 hours. The results showed that the model prediction error was within 4%. Then, the prediction accuracy of the EKF model and the stochastic process model were compared, and the comparison results showed that the prediction accuracy of the Kalman filter model was higher, which verifies the effectiveness and engineering application value of the proposed model in improving the remaining life prediction accuracy.