Multitask Convolutional Neural Network for Rolling Element Bearing Fault Identification

Algorithm 1

RMSProp optimization algorithm.

	Require: global learning rate , Decay rate
	Require: initial parameter
	Require: small constant , usually set to
	Initialize accumulated variable
	while failure to reach the stop criterion do
	Small batches containing m samples from the training set, corresponding target is
	Calculating the gradient:
	Cumulative square gradient:
	Calculating parameter updating:
	Application update:
	end while.