Application of fuzzy logic to detection of internal leakage fault in hydraulic systems.

Abstract

Detection of fault as early as possible before it leads to any financial loss or even catastrophic failure is very important. Like other systems, faults are inevitable part of hydraulic system. Due to its high power transmission capacity, the usage of hydraulic systems is high in today’s industry and so is the need for its fault diagnosis. Fault in a hydraulic system can arise due to numerous reasons like change in environmental conditions, change in bulk modulus or viscosity of the hydraulic fluid or drop in supply pressure. A faulty sensor or malfunction of any actual component in the whole system can also lead to failure. Using fuzzy logic a yes/no decision of fault is upgraded to a percent fault severity at the output in first part of this study. When a system deviates from its normal residuals are generated which are the measure of amount of fault present in the system. These residuals are evaluated using fuzzy logic. The performance of system is successfully evaluated and final output is fault severity which ranges from 0-100%. This approach combines fuzzy logic approach which is a knowledge based technique with an already developed model based technique. Out of these numerous faults that can potentially arise, the second part of this study focuses on detection of internal leakage fault in hydraulic actuators. An algorithm which successfully detects occurrence of internal leakage in hydraulic system is developed using fuzzy logic. This research uses the available knowledge according to which “changes in pattern are observed in the second level wavelet transform of the pressure signal measured at one end of the chamber”. Combining this knowledge with measured data, membership functions and heuristic rules are developed which mimic human mind (logical reasoning) in order to make conclusions regarding occurrence of internal leakage fault. Two different algorithms are developed and both are repeated taking into consideration data in two different time intervals. These four approaches are compared based on results obtained on different sets of data at the end of this dissertation. The method proposed here is a knowledge based method which primarily uses fuzzy logic and acts as an extension to already developed model based techniques. Thus the final diagnosis made is a combination of model based and knowledge based technique.