Remaining useful life estimation via transformer encoder enhanced by a gated convolutional unit

Fusing physics-based and deep learning models for prognostics

“…The network has 95k trainable parameters (). This final architecture is the result of conducting a grid search wherein the search space over the hyperparameters includes: number of hidden layers [1][2][3][4], number of neurons at each hidden layer [50,100,200], and activation function type [tanh, relu].…”

“…The network has 24k trainable parameters (). As with the FNN, the final architecture is the result of conducting a grid search over the following hyperparameters: number of hidden layers [1][2][3][4], number of channels [10,20,30] at each convolutional layer, filter size [10,20], number of neurons at the fully connected layer [50,100], activation function type [tanh, relu], and window size of the sliding window [20,50,200].…”

“…The resulting network has around 4.3k trainable parameters. Motivated by the main study, this final architecture is the result of conducting a grid search with the following search space: number of convolutional hidden layers [5][6], number of channels at each hidden layer [5,10], filter size [3,5] and number of neurons at the fully connected layer [50,100].…”

A Time-Series Data Generation Method to Predict Remaining Useful Life

“…where I(condition) is an indicator function, which returns 1 if condition is satisfied, and 0, otherwise, and α ≥ 0 is a Laplace smoothing parameter to prevent Pr M 1 are selected with probabilities in ( 8) and ( 9), respectively, these probabilities should be normalized by dividing each by their sum as presented in Equations ( 10) and (11):…”

“…These models learn the degradation patterns and relationships between the pattern and the RUL from the data collected until the end of the life of the components and predict the RUL of target components. In many recent studies, ML and DL models have shown superior performance for RUL problems [8][9][10][11][12][13][14][15].…”