Particle Filter for Fault Diagnosis: Applications to Dynamic Positioning Vessels and Underwater Robotics

Safety remains an important consideration in control system design. This is particularly true for the control systems that are employed by the offshore petroleum and maritime industries, in which authorities continually claim to have developed more and more rigorous methods and processes to ensure safety and reliability. In addition to the use of more traditional approaches to conducting redundancy design, backup system design, robust design, and failure mode and effect analysis, these methods often involve the use of fault tolerant design to enhance the safety and reliability of the control system. As a common practice, segregation and redundant design are used for dynamic positioning vessels to isolate faulty components and prevent the propagation of faults. However, many incidents still occur as a result of fault escapes from the segregation on a torpid detection. In more severe cases, false detection can actually cause faults and may result in an even more dangerous situation that has more catastrophic consequences. Hence, precise and timely fault diagnosis is necessary for the operator or the automation system to take appropriate action. This thesis presents a brief overview of the existing fault diagnose methods, with a particular focus on particle-filter-based framework for fault diagnosis. The paper commences with a brief review of the background, theory, and typical features of the particle filter before progressing to examine the relationships and differences between the particle filter and other traditional stochastic filters. Switching mode hidden Markov model were employed to model a system with potential faults and a new methodology that uses a particle filter as fault diagnosis filter was developed. This method was then applied on an underwater robotic, which worked in complex environmental disturbance and suffered from different failure modes. Experimental results from ROV sea trails verified that the new fault diagnosis design is effective and reliable