Fault Recognition Based On Associative Memory Neutral Network

Huge economic losses and even catastrophic accidents often take place because ofthe fault of some critical equipment in the industrial production, engineering machinery,aerospace, ship. The fault recognition technology can analyse the cause of the fault sothat it can nip in the bud and reduce loss. The hydraulic pump is an important part in thehydraulic system, its performance directly affects the reliability and stability of the workof the hydraulic system, so the research on the fault recognition of the pump. This paperstudies the network technology which has the function of associative memory, and itrecognize the fault of the axial plunger pump.First of all, it has studied the discrete Hopfield neural network and discussed thestructure and the stability of the convergence theory. On the other hand, this paperintroduces the function and concept of associative memory. The input of the Hopfieldnetwork must be a value of two type of malpractice, and the BP network has the strongnonlinear processing advantage, this paper has combined the structure of BP andHopfield so that it can be widely applied in practice.Secondly, according to the associative memory neural network easily falling intothe minimum, it makes use of the particle swarm algorithm to optimize the weights of thenetwork and get a network owing high convergence. Before the fault recognition in theassociative memory neural network, in testing it chooses and makes sure the parametersof network, such as number of the layer, number of the neurons in the hidden layer, thelearning rate, inertia weight, acceleration factor and so on.Finally, the associative memory neural network BP-H and BP-H-PSO in this paperis proved to be effective in the fault recognition to hydraulic pump. It has got the fact thatthe associative memory neural network with the particle swarm optimization algorithmhas the identification results with high precision, high recognition rate, more reliable. Theassociative memory is more accurate and the fault recognition rate is higher by using theEbbinghaus memory forgetting curve to arrange the learning samples with crosscirculation for reducing the wrong recognition.