基于代价敏感方法的智能制造故障诊断

Intelligent manufacturing fault diagnosis based on cost-sensitive method

在设备故障诊断过程中,数据集中正负分类样本数量相差较为悬殊等数据不平衡问题会导致诊断准确率降低。为减少由于正负类样本不均衡而导致的误判,提高设备故障诊断准确率,提出一种代价敏感方法。该方法借助Boosting方法,通过多次概率采样生成多个模型,并确定每个模型的权重。其中采样的概率取决于代价调整值,所提方法在每一个迭代过程中根据上一次迭代的结果对代价调整值进行调整。通过实验,并与其他方法进行对比,结果表明与采用固定的代价敏感值及非代价敏感方法相比,提出的方法具有更好的表现。

During the diagnosis process,data imbalance such as the significant gap between positive and negative samples can lead to reduced diagnostic accuracy.To reduce the misjudgment caused by the imbalance of positive and negative samples and to improve the accuracy of the diagnostic result,a cost sensitive method was proposed.By using Boosting method,the multiple models was produced with probability sampling,and the weights of each model were also determined,in which the probability of sampling depended on the cost adjustment value.In the proposed method,the cost adjustment value was adjusted according to the result of the previous iteration.Through comparing with other methods,the proposed method had better performance than the fixed cost sensitive value and the non-cost sensitive method.