CONDITION BASED MONITORING- CBM SYSTEM FOR A CASE STUDY

Condition Based Monitoring- CBM system for a case study

Condition Based Monitoring

1. Introduction

After several decades of development, the model-based approach has been shown to be effective in the condition monitoring of a broad array of control systems. By consulting the model prediction, this approach can be configured to be sensitive to changes due to abnormal behaviour but not to those caused by system inputs, so providing enhanced capability to distinguish between faults and acceptable fluctuations in operating conditions [.M. Frank, S.X. Ding and T. Marcu, 2000]. The approach permits detection not only of system faults, such as actuator faults and component faults, but most significantly, sensor faults. It is extensively applied in various industrial fields including power plant [R. Isermann and P. Balle, 2005], jet engine [S. Simani, C. Fantuzzi and R.J. Patton, 2003], and vehicle systems.

Published applications have given good address to the model-based approach, the modelling accuracy and the residual robustness [J. Chen and R.J. Patton, 2005]. Less attention has been paid, however, to practical issues such as real-time implementation especially when advanced software package is used. Reasons for this circumstance include difficulties from either hardware resource conflict or uncertain time delay in data transmission [R. Marcus, H. Straky, T. Weispfenning and R. Isermann, 2006]. In fact, these practical issues are sometimes decisive in selection of an appropriate monitoring method. For example, when a system is controlled by a computer, it is preferable to integrate the model-based approach with the control task on the same computer. However, the integration will be restricted under certain common conditions. Sometimes, the control program cannot be changed without the help of its supplier, or its change may be subject to burdensome safety procedural requirements. Even when the controller allows the integration, its CPU loading has to trade off the control application with other tasks [F.D. McKenzie, A.J. Gonzalez and R. Morris, 2007]. The CPU in a real-time control system must read sensor measurement and process it and then send signals to actuators at periodic intervals [H. Yoshida and S. Kumar, 1995]. If a complex fault diagnostic algorithm is used and the CPU is loaded heavily by the model prediction and the fault diagnosis, the control command may be delayed to send out and the control accuracy will be affected consequently. This is the case especially when non-linearity exists in the process so that a massive computation is required not only for prediction but also for fault diagnosis. While these objections can be addressed to some extent by coding directly into a real time operating system, this approach carries a burden of expert coding and the greatly enhanced load of error checking when designing and implementing a system at this low level.

A direct way to solve above problem is to use another machine with appreciated resources in parallel with a remote or distributed implementation. Thus the control task remains undisturbed and the condition-monitoring task is taken by a new CPU in a separate ...