基于CNN和预处理机制的球磨机故障诊断方法

Fault diagnosis method for ball mill based on CNN and preprocessing mechanism

球磨机的安全性和稳定性在工业生产中具有重要意义,其健康情况与生产效率、能耗等多项技术指标直接相关。而诊断球磨机故障的传统机器学习方法过度依赖人工特征提取,且缺乏自适应性。为此,针对球磨机工作状态下的分段多项式声信号,提出了一种小波去噪与自回归(AR)模型功率谱相结合的预处理方法来构建本征模态函数(IMF)特征向量,剔除噪声干扰信号。基于此构建卷积神经网络(CNN),以遴选的特征向量为输入,处理后变换为抽象的深层特征,以准确诊断球磨机的健康状态。实验结果表明:相较于其他传统机器学习算法,该方法在诊断准确性与诊断效率具有显著优势。

The safety and stability of the ball mill is of great significance in industrial production, and its health has great influence on many technical indicators such as production efficiency and energy consumption.However, traditional machine learning methods for diagnosing ball mill faults are overly dependent on manual feature extraction and lack adaptability.To solve the problem, a preprocessing method combining wavelet denoising and auto-regressive(AR)model power spectrum is proposed to construct feature vectors for the segmented polynomial acoustic signals under the working state of the ball mill, constructing intrinsic mode function(IMF) feature vectors, and eliminating noise interference signals.Based on this, a convolutional neural network(CNN) is constructed, which takes the selected feature vector as input, and transforms it into abstract deep features after processing to accurately diagnose the health status of the ball mill.The experimental results show that this method has significant advantages in diagnostic accuracy and diagnostic efficiency, compared with other traditional machine learning algorithms.