Estimation of Trajectories of Pipeline PIGs using Inertial Measurements and Non Linear Sensor Fusion Douglas Daniel Sampaio Santana Newton Maruyama Celso Massatoshi Furukawa Escola Politécnica da Universidade de São Paulo Av. Prof. Mello Moraes, 2231 – PMR 05508-900 – São Paulo, SP, Brasil. douglas.santana@poli.usp.br maruyama@usp.br cmfuruka@usp.br Abstract – This paper presents a nonlinear sensor fusion algorithm for inertial navigation designed to reconstruct trajectories of a Pipeline Inspection Gauge. Outputs of a strapdown inertial measurement unit are combined with non inertial measurements provided by an incremental odometer and a set of cartographic landmarks. The navigation system is modeled as a nonlinear dynamic system and the extended Kalman filter is used to estimate the system state. On preliminary experiments, it was possible to reconstruct a closed test trajectory with 2,800 m of extension, attaining a final error of 1.7 m. Keywords – Sensor Fusion, Inertial Navigation, Extended Kalman Filter. I. INTRODUCTION An instrumented Pipeline Inspection Gauge (PIG) is an equipment that carries sensors and embedded computers to acquire information while it travels through the pipeline, pumped with the transported fluid, as shown in Figure 1. An odometer is commonly used to provide a simple but incomplete and inaccurate position reference for the logged data, since it can only measure displaced distances and is subjected to unbounded cumulative errors. Position referencing is particularly important for finding the exact coordinates of a defect detected by the PIG along the pipeline. Further, it is possible to raise the topographic outline of the pipeline in places of difficult access. Fig. 1 – Instrumented PIG moving through a pipeline The goal of this work is to reconstruct the trajectory of an instrumented PIG. In many situations, the PIG cannot assess its position coordinates (for instance, relying on GPS signals) and inertial navigation techniques must be used to estimate its trajectory. In this case, the PIG must carry inertial sensors – gyroscopes and accelerometers fixed with other sensors on a rigid frame, and integrated with powerful digital signal processors to form a so called strapdown Inertial Measurement Unit (IMU). However, it would be necessary to use a high performance IMU, with sufficient resolution to sense the Earth’s rotation, that is very expensive and access restricted. If low performance sensors are used and their outputs are integrated in time, cumulative errors grow fast – thus reliable navigation is possible for only a few minutes [3, 11]. To minimize this problem, one possible solution is to combine inertial and non inertial sensors. This technique is known as Sensor Fusion. The dynamic model of the inertial navigation system is non linear and for this reason the Extended Kalman Filter (EKF) is employed. The EKF is the kernel of the algorithm. It estimates the heading (orientation), velocity and position of the system combining the information from inertial and non inertial sensors. In this work, two non inertial source of information are employed: an odometer (already present on PIGs) and a set of detectable landmarks (like electronic transponders). For practical reasons, the fewest possible number of landmarks must be used, so it is necessary to consider the compromise between the number of landmarks and the accuracy of the navigation. However, it is not necessary to know the position of the landmarks with absolute precision since an error model for their coordinates is provided to the Kalman filter. Two extended Kalman filters are used to estimate the actual state of the system – represented by the state vector k x , at a given discrete time k. The first one updates the states of the dynamic model with the sensor data and produces the updated state vector ˆ k x . A second filter is used to track the errors of the estimated states ( ˆ k δ x ), combining the dynamic model of the errors with the measurement errors of the sensors. Fluid Rubber supports Odometers IMU, electronics and batteries Pipeline wall