The use of machine condition monitoring and fault diagnosis is one potential approach for preventing unexpected catastrophic device faults, enhancing device operational efficiency and reducing maintenance cost. Currently, online wired monitoring systems are successfully used in most large industrial sites to monitor critical devices. However, there are many other non-critical devices that are not regularly monitored. Combining these into the wired monitoring system may significantly increase system cost because of the need for additional cabling. In addition, the installation and relocation of wired systems is inconvenient, especially for temporary or specialised tests.
Wireless sensor networks (WSNs) are one promising solution to replace traditional wired monitoring systems, because of its inherent advantages, such as relatively low cost, and convenience of installation. These merits make a low cost condition monitoring system for non-critical equipment or tasks possible. WSN technology has been successfully applied in environmental monitoring, construction health monitoring, and temperature monitoring in product distribution. Compared to these, machine condition monitoring and fault diagnosis make further demands on industrial wireless sensor networks (IWSNs), such as processing heterogeneous sensor signals, higher sampling rates, faster data transmission rates, and higher reliability. At the same time, WSN monitoring systems have constrained resources, including limited radio bandwidth, computational ability and battery energy.
Several IWSNs for industrial device monitoring have been developed and reported either by individual researchers or by commercial organizations. Most of these applications only use WSNs for data acquisition and transmission, and complete the feature extraction and fault diagnosis functions on a central computer.
This thesis proposes a novel industrial wireless sensor network (IWSN) for industrial machine condition monitoring and fault diagnosis. In this thesis, the induction motor is taken as an example of monitored industrial equipment due to its widespread use in industrial processes. Motor stator current and vibration signals are measured for further processing and analysis. On-sensor node fault feature extraction, on-sensor fault feature dimensionality reduction by principal component analysis (PCA), and on-sensor fault diagnosis are then investigated to address the tension between the higher system requirements of IWSNs and the resource constrained characteristics of sensor nodes. Two typical induction motor mechanical faults, i.e., loose feet and mass imbalance, are used to validate the fault diagnosis capability of the proposed system. Experimental results show that compared with raw data transmission on-sensor fault diagnosis could reduce payload transmission data by 99%, decrease node energy consumption for data transmission by 97%, and prolong node lifetime from 106 hours to 150 hours, an increase of 43%.
To maximize the benefits of on-sensor fault diagnosis, another system operating mode is explored, which only transmits the fault diagnosis result when a fault happens or at a fixed interval. For this mode, the node lifetime reaches 1759 hours (73 days) if sensor nodes transmit diagnosis results once per hour.
In many applications, it is difficult to capture all the required information for device fault diagnosis through a single sensor, especially for sensors working in harsh industrial settings. For IWSNs device fault diagnosis systems, the situation is even worse, because the noise and interference will impact the quality of communication in the IWSNs, and further increase the uncertainty of the diagnosis results. In recent years, many wired monitoring system with data fusion have been investigated. However, using data fusion techniques together with IWSNs for machine condition monitoring and fault diagnosis has not previously been reported.
This thesis presents a novel induction motor fault diagnosis system using IWSNs and Dempster-Shafer classifier fusion. The feasibility of the proposed system is demonstrated by two examples. Four motor operating conditions - normal without load, normal with load, loose feet, and mass imbalance - are monitored to evaluate the proposed system. Experimental results show that the final fault diagnosis accuracy can be significantly improved using the proposed approaches. The result certainty using a proposed two-step classifier fusion is at least 97.5%.