Failure prediction for robot reducers by combining two machine learning methods 2019-01-0508

There are many production robots used at car manufacturing plants, and each of them uses several reducers. A breakdown of one of these reducers may cause a huge loss due to the stop of all production lines. Therefore, the condition based maintenance to predict failures by predetermined thresholds of average and standard deviation is being currently used. However, this way has many false alarms or some false negative. As the way of suppressing false alarm, there are some methods such as probability density function. However when we suppress false alarm using probability density function into the operational range, we would have some false negative which leads to a breakdown of these reducers and huge loss. These false negative is caused by overlooking an anomaly with slight changes and it’s difficult to detect it using only probability density function. Then, we developed DSM (Difference Signum Method) to detect an anomaly with slight changes by emphasize slight changes. Although DSM can reduce false negative, DSM has many false alarms. In this study, we propose a new failure prediction method using ensemble learning of probability density function and DSM in order to reduce both false alarms and false negative. By using ensemble learning of probability density function and DSM, the failure prediction method can also detects anomaly with slight changes as well as anomaly with drastically changes. Using this new failure prediction method, the number of alerts reached 6 times / week under, and it was possible to drastically reduce from the 25 times / week of the conventional methods and the number of false negative reached the target value of 0 times / year from 2 times / year using probability density function. So, its performance was applicable to the actual production lines by this new failure prediction method.