Model based diagnostics of motor and pumps

The purpose of machine fault diagnosis is to 1) detect, 2) identify, and 3) predict components which are more likely to fail. This study describes a fault diagnosis method that utilizes models, parameter estimation, and Shannon’s communication theory. A centrifugal pump driven by a squirrel cage induction motor is selected as an objective system to be monitored using the proposed diagnostic method. Separate bond graph models for a motor and a pump are combined to emulate dynamics of a motorpump system. A test-bed was built and parameters in the model are “tuned” by comparing simulations to sensor measured data, and altering parameters until simulations agree with data. Degradations in different components are induced. As a fault progresses, measured signals change, and for simulations to mimic measurements, parameters must change. Inspecting deviations of parameters from their nominal values allows detection and isolation of faults since parameters of the model have direct one to one correspondence to components in the physical system. An analogy was made between a machine and a communication channel, yielding a “machine communications channel” to estimate severity of faults. Design of vii communications systems is aided by theorems of Shannon, which establish minimum signal to noise ratios for acceptable transmission and reception. The transmitter activates a communication channel with an input signal. Noise in the channel, along with component kinematics and dynamics, alters the signal. When the channel operates properly, the signal is “received” at output within tolerances. Faults disrupt functionality, and change the response. Faults generate “noise”, the difference between an actual signal and the desired signal. Unless the signal to noise ratio is kept sufficiently high, the receiver cannot reconstruct the signal within a desired tolerance, and the channel malfunctions. In terms of a machine channel, the machine cannot perform its task within tolerance. Shannon’s theory applied to machinery can establish performance limits, and develop failure criteria to assess functionality of new or degraded machinery. In this study, signals from healthy and faulty systems, and the difference between them, noise, become important diagnostic tools to assess severity of faults. Fourier transform of these signals give spectral densities, needed to estimate channel capacities and information rates. Shannon’s theorems of communication theory assess channel capacity, the maximum amount of information a machine (channel) can successfully transmit and receive. If this is less than the rate of information needed to perform a given job, Shannon’s theorems predict machine malfunction. Faults such as a damaged stator circuit in a motor will be introduced into the test equipment built for this study and diagnosed by the proposed method. We will explain the motor-pump bond graph model, and then present results of fault diagnoses via experiments, simulations, parameter estimation, and fault severity assessment.