A High-Accuracy Least-Time-Domain Mixture Features Machine-Fault Diagnosis Based on Wireless Sensor Network

Fault diagnosis of rolling bearing plays a vital role in identifying incipient failures and ensuring the reliable operation of the mechanical system. To improve the performance of the whole machine-fault-diagnosis system and meet the requirements of low cost, low consumption, high-reliability in industrial wireless sensor networks (IWSNs), a high-accuracy least-time-domain features fault diagnosis algorithm based on the BP neural network (BPNN) for IWSNs is proposed in this article. First, the hardware of wireless multifeatures extraction sensor node is designed, which performs local-processing features extraction of four-dimensional parameters and five dimensionless features of the vibration signal. Then, the bearing-fault classification based on mentioned characteristics is investigated in the proposed BPNN with different hidden layer nodes. Furthermore, we make the comparisons of bearing-fault classification accuracy in terms of varying number of dimensional features, dimensionless features, and the combination features, searching a least-time-domain mixture features selection strategy for ensuring high-fault classification accuracy and proving the effectiveness and feasibility of the proposed method by experiments on drivetrain diagnostics simulator system. This article is conducted to provide new insights into how to select the least time-domain features for high-accuracy fault diagnosis and further giving references to more IWSNs scenarios.