摘要：【中文】本发明公开了一种基于故障知识库和贝叶斯网络的设备故障诊断方法，包括以下步骤：通过收集专家经验对BOG压缩机的故障模式进行故障识别，依据故障识别的结果建立BOG压缩机的故障知识库；依据故障知识库中的故障表现建立故障模式与故障表现之间互相推理的贝叶斯网络模型；依据所建立的贝叶斯网络模型，对设备的在线参数进行推理分析，得到设备每种故障模式的可能发生的概率并输出显示，对故障概率超过指定阈值的故障模式给出相应的报警信号。 【EN】The invention discloses a fault diagnosis method based on a fault knowledge base and a Bayesian network, which comprises the following steps: fault recognition is carried out on the fault mode of the BOG compressor by collecting expert experience, and a fault knowledge base of the BOG compressor is established according to the fault recognition result; establishing a Bayesian network model for mutual inference between the fault mode and the fault expression according to the fault expression in the fault knowledge base; and carrying out reasoning analysis on the online parameters of the equipment according to the established Bayesian network model to obtain the probability of each fault mode of the equipment, outputting and displaying the probability, and giving a corresponding alarm signal to the fault mode with the fault probability exceeding a specified threshold.

摘要：【中文】本发明公开了一种基于大数据的压缩机故障诊断方法，包括以下步骤：专家经验/机理相关贝叶斯网络搭建；数据驱动网络搭建，包括将压缩机振动数据处理为可操作化数据，将所有采样点振动数据绘制相关图，将所有采样点的振动数据的90％位数‑10％位数绘制相关图，初步分析故障特征；对相关振动信号进行合理降噪，将振动信号进行降噪之后，利用三层小波包分析，将原本的每个采样时间点的时域波形分解为八个频段的振动信号，进行各个频段的能量占比分析，将故障时间段的能量占比分析与正常时间段的能量占比分析进行比对，利用三层小波包分析及能量占比分析寻找出可供预测压缩机故障的特征参数，进而构建相关网络利用逻辑回归进行故障预测。 【EN】The invention discloses a compressor fault diagnosis method based on big data, which comprises the following steps: constructing a Bayesian network related to expert experience/mechanism; the method comprises the steps of establishing a data driving network, processing vibration data of a compressor into operable data, drawing a correlation diagram of the vibration data of all sampling points, drawing the correlation diagram of 90% -10% of digits of the vibration data of all the sampling points, and preliminarily analyzing fault characteristics; reasonably denoising a related vibration signal, decomposing an original time domain waveform of each sampling time point into vibration signals of eight frequency bands by using three-layer wavelet packet analysis after denoising the vibration signal, performing energy ratio analysis of each frequency band, comparing the energy ratio analysis of a fault time period with the energy ratio analysis of a normal time period, finding out characteristic parameters for predicting the fault of the compressor by using the three-layer wavelet packet analysis and the energy ratio analysis, and further constructing a related network to predict the fault by using logistic regression.