摘要
Recently, the fault diagnosis of the ball mill mostly depends on the experience of workers, which brings about a lot of uncertainty for fault diagnosis. In addition, the cost of labor is getting higher, so that the research of ball mill fault diagnosis based on machine learning has become increasingly valuable. The current fault diagnosis methods are mostly judging based on instantaneous data, which makes it difficult to reflect the ball mill indicators and the occurrence of time-related correlation(such as hysteresis effect). This paper presents a ball mill fault diagnosis method based on Gate Recursion Unit(GRU),which analyzes the fault data in the form of time series and compares with other common methods such as neural network, Autoencoder and Long Short-Term Memory(LSTM). After comparison,it is concluded that the fault diagnosis method based on GRU ball mill has the lowest error rate as 4.85%.
Recently, the fault diagnosis of the ball mill mostly depends on the experience of workers, which brings about a lot of uncertainty for fault diagnosis. In addition, the cost of labor is getting higher, so that the research of ball mill fault diagnosis based on machine learning has become increasingly valuable. The current fault diagnosis methods are mostly judging based on instantaneous data, which makes it difficult to reflect the ball mill indicators and the occurrence of time-related correlation(such as hysteresis effect). This paper presents a ball mill fault diagnosis method based on Gate Recursion Unit(GRU),which analyzes the fault data in the form of time series and compares with other common methods such as neural network, Autoencoder and Long Short-Term Memory(LSTM). After comparison,it is concluded that the fault diagnosis method based on GRU ball mill has the lowest error rate as 4.85%.
