Preventive maintenance (PM) is an effective approach to promoting reliability. Time-based and condition-based maintenance are two major approaches for PM. No matter which approach is adopted for PM, whether a failure can be early detected or even predicted is the key point. This paper presents the experimental results of a failure prediction method for preventive maintenance by state estimation using the Kalman filter on a DC motor. The rotating speed of the motor was uninterruptedly measured and recorded every 5 min from 1 April until 20 June 2001. The measured data are used to execute Kalman prediction and to verify the prediction accuracy. The resultant prediction errors are acceptable. Futhermore, the shorter the increment time for every step used in Kalman prediction, the higher prediction accuracy it achieves. Failure can be prevented in time so as to promote reliability by state estimation for predictive maintenance using the Kalman filter.
High quality and excellent performance of a system are always goals for engineers to achieve. Reliability engineering integrates quality and performance from the beginning to the end of a system life [1]. Therefore, reliability can be treated as the time dimensional quality of a system. Reliability is affected by every stage throughout the system life, including its development, design, production, quality control, shipping, installation, operation, and maintenance. Consequently, paying attention to each of the stages can promote reliability. Specifically, in the onsite operation phase, failures are the main causes that worsen performance and degrade reliability. Accordingly, failure avoidance is the main approach to reliability assurance. There are three main types of maintenance, namely improvement maintenance (IM), preventive maintenance (PM), and corrective maintenance (CM) [2]. The efforts of IM are to reduce or eliminate entirely the need for maintenance, i.e. IM is performed at the design phase of a system emphasizing elimination of failures that require maintenance. There are many restrictions for a designer, however, such as space, budget, market requirements, etc. Usually, the reliability of a product is related to its price. By contrast, CM is the repair actions executed after failure occurrence. PM denotes all actions intended to keep equipment in good operating condition and to avoid failures [2]. As a result, PM should be able to pinpoint when a failure is about to occur, so that repair can be performed before such failure causes damage.
PM is an effective approach to promoting reliability [3]. Time-based and condition-based maintenance are two major approaches for PM. No matter which approach is adopted for PM, whether a failure can be early detected or even predicted is the key point. If a device is judged to know that it is going to fail by the predicted future state variables, the failure can be prevented in time by PM. However, future state variables should be accurately predicted at a reasonably long time ahead of failure occurrence [4] and [5]. A failure prediction study entitled ‘State estimation for predictive maintenance using Kalman filter’ has been proposed [6]. In the study, failure times were generated by Monte Carlo simulation and predicted by the Kalman filter. One-step-ahead and two-step-ahead predictions were conducted. Resultant prediction errors are sufficiently small in both predictions. Even so, the failure prediction was simulated on a computer after all. In the current study, a DC motor and a data acquisition system are set to implement the simulation. The rotating speed of the motor is chosen as the major state variable to judge whether the motor is going to fail by state estimation using the Kalman filter. The rotating speed of the motor was uninterruptedly measured and recorded every 5 min from 1 April until 20 June 2001. Instead of simulated data, the measured data are used to execute Kalman prediction and to verify the prediction accuracy in the current study.
In the next section, equations for state estimation of the Kalman filter are briefly introduced. Section 3 presents the transfer function, continuous state model, and the discrete state model of a DC motor that is employed in this paper. Section 4 presents the experiment setup with related parameters. Results and discussions are in Section 5.
An experiment of state estimation for predictive maintenance using the Kalman filter on a DC motor has been performed in this paper. The resultant prediction errors for one-step-ahead prediction are acceptable. Furthermore, the shorter the increment time for every step in Kalman prediction uses, the higher prediction accuracy it achieves. Considerations for determining the required PM lead time and the increment time for prediction contradict to each other. How to compromise them and end up with an optimal value is important. Incorporating the proposed method with fault tree analysis or Petri net model for failure, can be performed on those elements in minimum cut sets of a complicated or large system instead of on all elements of the whole system. Failure can be prevented in time so as to promote reliability by state estimation for predictive maintenance using the Kalman filter.
