Anomaly detection on the cutter bar of a combine harvester using cyclostationary analysis

Driven by increasing food demand and the urgent need for intelligent land use, the size and complexity of agricultural machinery has increased significantly. Monitoring the operation of machines during long working days is a very challenging task for the operators. Auto-matic monitoring systems can lighten the load and also pave the way towards fully autonomous machines. This paper proposes an automatic condition monitoring system for detecting operation anomalies on agricultural machine components with reciprocating motion, and applies it to a use-case on a combine harvester header cutter bar. Cyclosta-tionary analysis techniques are employed to develop filtering algorithms to extract infor-mative features, which are monitored through the implementation of statistical process control (SPC) using control charts. Together with a comparison of several filtering and feature extraction techniques, an analysis is provided on the influence of sensor type and position on anomaly detection performance. Filtering benefit was found to be highly dependent on the considered sensor type and its location, with increases in Matthews correlation coefficient (MCC) ranging from 0 to 50%, resulting in maximal MCC values of 1. Suitable feature calculation resulted in average prediction performance improvements over 10% in MCC values for nearly all considered sensor types and their locations. These results highlight the importance of intelligent sensor selection for condition monitoring purposes on agricultural machinery and the added value of SPC involving cyclostationary analysis techniques for anomaly detection. (c) 2023 IAgrE. Published by Elsevier Ltd. All rights reserved.

Automated summary

Driven by increasing food demand and the urgent need for intelligent land use, the size and complexity of agricultural machinery has increased significantly. Automatic condition monitoring systems can lighten the load for operators and pave the way for fully autonomous machines. This paper proposes a system based on cyclostationary analysis techniques to detect operation anomalies on agricultural machine components and highlights the importance of intelligent sensor selection for condition monitoring purposes.