为了解决机械故障诊断存在特征提取带来的信息丢失问题和多段平均丢弃数据不确定性的问题，提出了一种基于概率盒理论的机械故障信号建模方法。以滚动轴承故障信号为研究对象，分析原始信号的概率分布类型，获得概率分布类型参数的不确定性区间，提出基于确定概率分布类型的概率盒建模方法。针对故障信号概率分布类型难确定问题，提取原始信号的特征，利用特征信号的有序性，提出基于特征提取的概率盒建模方法，以歪度和峭度特征为例，对比两种特征概率盒的异同点。基于概率盒定义，将原始数据的不确定性直接映射到概率盒的上下界，提出无需验证数据概率分布类型的原始数据概率盒直接建模方法。通过滚动轴承实测数据，对比三种方法的有效性及适用性，与传统特征提取方法对比，证明了方法的有效性。%Feature extraction may lead to information loss and multi-segment-average may lead to data discarding and uncertainties in machinery fault diagnosis.In order to solve these problems,a new modeling method for mechanical fault signals based on the probability box (p-box)theory was proposed.Fault signals of rolling bearings were taken as the study object.Firstly,the original signals'probability distribution types were analyzed.The uncertainty intervals of the probability distribution's parameters were calculated.The p-box modeling method based on the normal distribution was proposed.Secondly,in order to overcome the identification difficulty of fault signal data's probability distribution type,the original signals'features were extracted.Using the ordered character of the feature signals,a p-box modeling method based on feature extraction was proposed.The similarities and differences between the skewness p-box and the kurtosis p-box were contrasted.Thirdly,based on the p-box's definition,the original data uncertainties were projected into the p-box's bounds a more effective p-box modeling method directly based on the original data was proposed,it did not need data's probability distribution identification.The effectiveness and applicability of the three methods were compared using rolling bearings'measuring data,the three methods'validity was verified compared with the conventional feature extraction method.
展开▼
