In order to solve the problem that the traditional particle swarm optimization algorithm takes a long training time with accuracy or misjudgment rate as fitness function and the fault mechanism of axial piston pump is complex,a probabilistic intelligent state discrimination soft method of axial piston pump based on fitness function improved Lévy flying quantum-behaved particle swarm optimization (QPSO) multi-classification relevance vector machine (MRVM) was proposed.In order to overcome the shortcomings of imprecision and low efficiency by use of manual search for kernel parameters,QPSO based on Lévy flight was used to find the optimal kernel parameters of MRVM.In order to shorten the optimization time,the cosine similarity between samples was taken as the fitness function of the optimization algorithm,and simulation experiments were carried out by using of UCI machine learning standard data sets to verify the effectiveness and superiority of the improved method.The data of the piston pump under different fault conditions were sampled,and the time frequency domain and time domain features were extracted and input into the optimized MRVM for training and testing.The experimental results show that the proposed method can be used to effectively improve the accuracy and efficiency of fault diagnosis; at the same time,the soft classification is realized,that is,the diagnosis results can be output in the form of probability,which provides reliable and practical fault information for equipment repair and maintenance.