Abstract: The impending overheating of the compressor motor in a chiller is predicted by comparing measured output values in the chiller with output values predicted by passing said output values and chiller input values through a Kalman filter. Variances which exceed thresholds are used to predict the overheating condition. The Kalman filter is derived from a low order state space model of the chiller, the matrix values of which are derived by linear regression from the outputs and states resulting from random signal input excitation of a high order model of the chiller in a computer.

Abstract: A physical signal is represented by a linear system. The linear system is transformed from the time and space domain to the frequency domain, generating a transformed system. The least significant portions in the transformed system are determined. The least significant portions are deleted from the transformed system, generating a smaller pruned system. The pruned system is solved in the frequency domain, generating a solution. The solution is inverse transformed from the frequency domain to the time and space domain, generating an approximation to the physical signal.