基于时变隐马尔科夫模型的机器人故障预测

Fault Prediction for Robots Based on Time-varying Hidden Markov Model

针对多机器人系统的应用场景,提出了Weibull分布和隐马尔科夫模型相结合的多机器人系统故障预测方法。首先,根据机器人的无故障运行时间估算出机器人可靠性的Weibull分布模型;然后对机器人运动数据采用小波包变换的方法进行特征提取,并训练好状态评估模型,将经过特征提取后的待诊断数据输入训练好的状态评估模型,实现性能评价功能;最后,使用隐马尔科夫模型中期望最大化算法(expectationmaximization,EM)结合Weibull分布进行故障预测模型的训练。通过仿真验证了该方法的可行性和有效性。

Aiming at the multi-robot system scenarios, a method which combines time-varying hidden Markov model（HMM） and Weibull distribution was proposed to achieve fault prediction. In this method, firstly, the Weibull distribution of robots＇ reliability was calculated out based on the time between failure of robots. Then by using the wave- let packet decomposition method, the feature of robots＇ can be obtained from the robot movement data. Also state judg- ment HMM can be trained. Thus state judgments can be made based on the judgment model and robots＇ movement fea- ture. And finally, the Weibull distribution and the expectation-maximization algorithm was used to form fault prediction model. After the simulation experiment, the fault prediction method turns out to be effective and viable.