Long-term, trouble-free operation of ventricular assist devices (VADs)
is critical to the patient. A catastrophic failure of the VAD could c
ost the patient's life, thus defeating the purpose of the device. The
targeted 90% 5 year reliability also implies that the average device l
ife would exceed the 5 year limit. Time based explantation of the devi
ce after the fifth year will replace many devices with significant add
itional life, subject the patient to unnecessary surgical risk, and in
crease costs. To preclude the need for time based replacements and pre
vent catastrophic failures, a condition monitor is proposed in this ar
ticle for early detection of faults in VADs. To develop this monitor,
the effectiveness of various sensing and monitoring methods for determ
ining the VAD condition is investigated. A Hemadyne pump was instrumen
ted with a set of eight sensors, and a series of experiments were perf
ormed to record and analyze signals from the normal and abnormal pumps
with five different faults. Statistical, spectral, envelope, and ense
mble averaging analyses were performed to characterize changes in sens
or signals due to faults. Experimental results indicate that statistic
al and frequency information from the acceleration and dynamic pressur
e signals can clearly detect and identify various VAD faults.