Mr Dustin Helm ( - Laurentian University), Prof Markus Timusk ( - Laurentian University)
Maintenance costs can make up a significant portion of a mine’s operating budget. These costs can include not only the cost of repair (parts labour etc.) but also the opportunity cost of lost production time. Therefore effective management of maintenance for critical machinery is crucial in industry. Traditionally, maintenance strategies were either run-to-failure or preventative, however both have significant drawbacks. Due to these drawbacks many industries have been trending towards the use of condition based maintenance (CBM) practices. CBM uses information about the current and historical condition of the machine to lend decision support on when to preform maintenance. This relies on effective machine condition monitoring, which is the timely detection and diagnosis of incipient faults in machinery. This is achieved by monitoring parameters (i.e. vibrations, temperature) that contain information about the health state of the machine. In the mining industry, machine condition monitoring is often met with significant challenges due to the nature of the mining environment. The environmental changes and the non-stationary operation that faces most mining equipment creates significant changes in monitored parameters that can obscure the results of a condition monitoring system. Due to these complications, novel solutions are needed to increase the effectiveness of condition monitoring in industries such as mining. Several techniques exist to attempt to solve this problem, including signal processing tactics, complex feature extraction methods as well as classification techniques that take into account the speed and load of the system. However, these methods are often not practical as they require large amounts of training data to accurately describe the feature space for each speed and loading condition. It is this need for representative machine data that impedes advancements in CBM development work.This work will present a method designed to be able to detect faults in identical components running in parallel. This is a new approach to condition monitoring that attempts to solve many of the problems associated with non-stationary machinery. The reason for this is that when components run in parallel it is possible to reduce the systems sensitivity to unsteady conditions as well as its complexity by considering the difference between components rather than analyzing them separately. This is because the changes that are not a result of each component’s health state will be shared (or at least related) between the two signals. Preliminary results have been promising, yielding accurate detection of faults and significant increase in performance over non-parallel techniques. In this work results will be presented for several different cases, including bearing and hydraulic pump faults. Experiments for this work were carried out on the machinery fault simulator located at Laurentian University. It is designed to test a range of different machine components in a variety of faulted and healthy states while running under a realistic non-stationary duty cycle. It currently consists of two identical drive systems that are both loaded and driven in parallel. The machine is also modular, allowing for multiple different arraignments to test different components in different conditions. Using this fault simulator it is possible to test mechanical components while they operate under duty cycles, tailored to match real duty cycles as closely as possible.