Condition monitoring of hardware redundant machinery subject to non-stationary operations

Abstract

The focus of this work is to develop techniques for detection of faults in hardware redundant machinery. Hardware redundancy is a design methodology that can be utilized to increase the reliability of a system. The inclusion of multiple redundant components can allow for fault tolerant design of systems, as the failure of one component typically results in increased load on other while the overall system continues to operate. However, this built in redundancy can increase the cost and complexity of the system. Therefore it is typically seen in applications where there is significant risk (cost or safety) associated with unexpected failures or downtime. These types of applications are also well suited to condition monitoring, which is another technique often employed to reduce the risk of unexpected failure and reduce the overall maintenance cost of a system. Condition monitoring (CM) is the practice of utilizing measurements or information from a system as it is operating to determine its health state in real time. This type of information can be extremely useful for planning maintenance activities and minimizing risk. CM however can be extremely difficult to apply in many industrial applications, as there are many factors that can affect a system and obscure the effects of an incipient fault. The most prevalent of which is non-stationary machinery operation. When a mechanical system undergoes changes in speed and load the measurements gathered can exhibit significant frequency and amplitude modulations. These modulations will often hide the relatively weak changes associated with a fault. This work presents five peer reviewed published works that provide solutions for, and advance our understanding of the detection of faults in hardware redundant machinery. This work demonstrates that in cases where there is hardware redundant configuration of mechanical components, it is possible to leverage a redundant data collection scheme to improve the accuracy of a fault detection system. The benefit of using redundant signals is made clear when considering non-stationary machinery. The fundamental aspect of this work is that the residual or difference between signals from hardware redundant components is less sensitive to changes in operational conditions (speed, load, temperature, etc.) than the original signals.