Exploration of a Condition Monitoring System for Rolling Bearing Based on a Wireless Sensor Network

In recent years, wireless sensor networks (WSN) have attracted attention in machine condition monitoring (CM) fields for a more efficient system based on the inherent advantages of WSN, including ease of installation and relocation, lower maintenance cost and the ability to be installed in places not easily accessible. As critical components of rotating machines, bearings account for more than 40% of the various types of failures, causing considerable unpredicted breakdowns of a plant. Thus, this thesis intends to develop a cost-effective and reliable wireless measurement system for rolling bearing condition monitoring.
Based on the investigation of various wireless protocols, Zigbee has been taken as a the most promising candidate for establishing the wireless condition monitoring system as it can have an acceptable bandwidth at low power consumption. However, a comparison made between wired and wireless measurement system has found that the Zigbee based
wireless measurement system is deficient in streaming long continuous data of raw vibration signals from typical application environment with inevitable ambient interference. As a result, data loss can happen from time to time.
To solve this issue, an on-board processing scheme is proposed by implementing advanced signal processing algorithms on the sensor side and only transmitting the processed results with a much smaller data size via the wireless sensor network. On this basis, a wireless sensor node prototype based on the state-of-the-art Cortex-M4F is designed to embed customizable signal processing algorithms. As an extensively employed algorithm for bearing fault diagnosis, envelope analysis is chosen as the on-board signal processing algorithm. Therefore, the procedure of envelope analysis and considerations for implementing it on a memory limited embedded processor are discussed in detail. With the optimization, an automatic data acquisition mechanism is achieved, which combines Timer, ADC and DMA to reduce the interference of CPU and thus to improve the efficiency for intensive computation. A 2048-point envelope analysis in single floating point format is realized on the processor with only 32kB memory.
Experimental evaluation results show that the embedded envelope analysis algorithm can successfully diagnose the simulated bearing faults and the data transmission throughput can be reduced by at least 95% per frame compared with that of the raw data; this allows a large number of sensor nodes to be deployed in the network for real time monitoring.
Furthermore, a computation efficient amplitude based optimal band selection algorithm is proposed for choosing an optimal band-pass filter for envelope analysis. Requiring only a small number of arithmetical operations, it can be embedded on the wireless sensor node to yield the desired performance of bearing fault detection and diagnosis.