Monitoring the status of rotating components is important in modern machinery. The goal of this study is to evaluate the feasibility of using a k-Nearest Neighbors (kNN) classifier combined with a Harmony Search (HS) algorithm, to detect the operational status of rotating components within agricultural machines. Vibration data, the source data, were acquired from four accelerometers located along the chassis of a harvester. Five operational statuses of three rotating components of the harvester were studied: engine (low/maximum speed), thresher, and chopper (on/off and balanced/unbalanced). The methodology includes vibration signal acquisition, data preprocessing, smoothing, preselection of frequencies, Brute Force (BF) and Harmony Search frequency selection, and classification with kNN. The input frequencies for the classifier were chosen with either BF search or HS. The main results of the study were: i) the preselection of frequencies reduced the training time between 92.2% and 95.6%; ii) the smoothing stage improved accuracy; iii) HS reduced the training time between 82% and 90% in comparison with BF, reaching accuracies of nearly 100% in the five operational statuses with only 2 input frequencies; iv) similar levels of accuracy were obtained when using data from the accelerometers at different locations. The results suggested that it was feasible to predict the operational status of rotating components of agricultural machines using a kNN classifier with the combination of preselection, smoothing, and the HS algorithm. This feasibility was achieved both in terms of accuracy and computational burden, building upon previously proposed methods.
