In today's age of industrialization, sensor devices installed on equipment generate a vast amount of data. One of the engineers' main jobs is utilizing these data to provide better solutions to industrial problems. This availability of extensive data partly led to the creation of predictive maintenance (PdM). In PdM, existing and previous conditions of devices are used to predict their future behavior for optimal maintenance. Most of these PdM approaches are typical time-series predictions. Machine learning tools like Recurrent Neural Networks (RNNs) are excellent tools for time-series predictions. However, most RNNs suffer from training issues due to the unstable gradient problem. Thus, networks such as the Echo State Network (ESN), were designed to solve them. The ESN solves the gradient problem by training only the output weights using simple linear regression. Despite this ease, the selection of ESN parameters and topology is a considerable design challenge. This problem is often formulated as a typical optimization problem. Metaheuristic algorithms are known to be excellent tools for solving optimization problems. Hence, in this work, we design an improved Grasshopper Optimization Algorithm (GOA) based ESN. The proposed technique uses a new solution representation with a simplified attraction and repulsion mechanisms to enhance performance. Our target application is to predict the Remaining Useful Life (RUL) of turbofan engines. The method outperforms the Cuckoo Search (CS), Differential Evolution (DE), Particle Swarm Optimization (PSO), Binary PSO (BPSO), the original GOA, the classical ESN, deep ESN, and LSTM. We have provided all implemented codes and data at the GitHub repository. https://github.com/bala-221/Airplane-fault-prediction. © 2013 IEEE.