Extracting random forest features with improved adaptive particle swarm optimization for industrial robot fault diagnosis

Feature extraction is a vital step for the fault diagnosis of industrial robots, while large-scale measured signals produce redundant features impairing the diagnosis performance. To address this problem, an improved adaptive particle swarm optimization (IAPSO) is suggested to extract effective features for random forest (RF) diagnosis. Raw data collected under different kinds of complex conditions are first represented by statistical parameters of its wavelet coefficients. A relative permutation order based scaling method with analytic hierarchy process is then used for selecting suitable updated strategies. RF is finally used to measure classification performance of each particle. The proposed method was evaluated by experiments on an industrial robot. Feature set was reduced 52 % from the initial size by using IAPSO, still achieving a superior classification precision over 96 %. The proposed method performs better than other peer methods and exhibits an essential improvement potential for the fault diagnosis of industrial robots.