Position Estimation of In-Pipe Robot Using Artificial Neural Network and Sensor Fusion

Abstract

Abstract:Automatic position detection of water leakage in water distribution pipelines is critical tominimize the loss of labour, time, money spent on exploration and excavation in pipe inspectionprocedures. Nevertheless, the main goal of detection is to prevent water loss. In this paper,accurate position detection, crack frequency band detection, and external sphere studies of anin-pipe robot prototype were presented. During the precise position estimation, classicalExtended Kalman Filter (EKF), stationary region detection, and location estimation usingEnhanced Heuristic Drift Elimination (EHDE) were performed with two different artificialneural networks (ANNs). In this way, online precise position estimation can be done onhardware with no sufficient computational power for indoor robotic studies. In addition, thesound characteristics resulting from the crack at different hole size and water pressure intensitylevels were investigated. Finally, a new sealing sphere design was devised. Three differenthydrophone sensor data were recorded on the SD card simultaneously. The results show thatthe proposed ANN method can work online and make a similar position estimation with theclassical IMU position estimation method by 99%.