To analyse the causes of pneumatic valve failures in liquefied natural gas (LNG) plants, sort out the occurrence rules of failures and accurately perform failure prediction and spare parts inventory in engineering, this paper takes the damage frequency of key components of the pneumatic control valve, valve usage and instrument wind of the Kunlun Energy Hubei LNG plant from 2014 to 2019 as data samples. The Grey relational analysis model and analytic hierarchy process model are selected to perform unsupervised mathematical experiments. The results show that the faults of the pneumatic control valve can be decomposed into explicit faults, implicit faults and indirect faults. In engineering, attention should be paid to the sufficient inventory of key components of solenoid valves and positioners, and timely treatment should be administered when faults are found. In addition, targeted inspections should be regularly performed to reduce the frequency of dominant faults. At the same time, attention should be paid to the degree of damage caused by indirect faults, and regular monitoring of the water content and cleanliness of instruments should be conducted. The data of this mathematical experiment are well-coupled, and the accuracy of the model is controllable. However, the experiment shows that the basic data remain discrete. The next step should be to strengthen the data collection work, at the same time broaden the boundary of fault performance factors, and introduce data, such as the influence degree of the pneumatic control valve fault on the subordinate equipment, local process system, overall process system and its energy consumption. Standard analysis should be conducted in the process industry.

为了分析液化天然气工厂中气动阀故障的原因，找出故障的发生规律，并在工程中准确地进行故障预测和备件库存，本文以气动控制阀关键部件的损坏频率为例数据样本为2014年至2019年昆仑能源湖北LNG厂的阀门使用情况和仪表风。选择灰色关联分析模型和层次分析模型来执行无监督的数学实验。结果表明，气动控制阀的故障可分解为显性故障，隐性故障和间接故障。在工程中，应注意电磁阀和阀门定位器关键部件的足够库存，发现故障应及时处理。此外，应定期进行有针对性的检查，以减少主要故障的发生频率。同时，应注意间接故障造成的损坏程度，并应定期监测仪器的水含量和清洁度。该数学实验的数据耦合良好，模型的准确性是可控的。但是，实验表明基本数据仍然是离散的。下一步应该是加强数据收集工作，同时拓宽故障性能因素的范围，并引入数据，例如气动控制阀故障对从属设备的影响程度，局部过程系统，整个过程系统及其能耗。