CONDITION MONITORING OF AUTOMATIC-CONTROL SYSTEMS USING PARAMETER AND SYSTEMS IDENTIFICATION - A NUMERICAL INVESTIGATION

The intense international competition for reducing manufacturing cost has forced many industrialists to implement total automatic control over a wide range of products and manufacturing processes. This, of course, results in substantial reduction of manpower on the factory floor and the task of condition/performance monitoring must be replaced by appropriate computer based software. Many engineers have employed vibration analysis methods developed in recent years to monitor the condition of critical components such as gearboxes, bearings, belt-drives and rotating elements which are more susceptible to failure than other mechanical parts of machines. However, these methods are suitable for identifying faults that generate substantial increase in vibration level measured at a suitably selected point. The authors argue that the parameter and system identification technique presented in this paper gives far more reliable information about imminent failure than the method mentioned above. The availability of high-quality feedback signals(s) removes the need and difficulty of selecting a suitable transducer. The method is based on the concept that the dynamic behaviour of automatic control systems can be modelled by identifiable linearised/nonlinear transfer function with constant coefficients. The deviation of each parameter from the norm value indicates the failure/wear of respective elements in the system. The paper demonstrates that these parameters can be estimated relatively simply with efficient use of computer time and memory. The method uses state variable filter (SVF) and recursive least square optimisation algorithm so that it may be implemented as an ON/OFF-line condition monitoring package. The numerical investigation from simulated data shows that the identified coefficients may be used to predict components failure. In addition, the algorithm is capable of identifying the relevant coefficients accurately if the noise level on the measured signal is less than 10% which is usually the case for most automatic control systems with high-performance feedback transducers. The procedure of designing a suitable filter capable of converging rapidly is discussed and the results show that a compromise has to be made between speed of conver-gence and the magnitude of noise in the feedback signal.