1 Fault Diagnosis based on Particle Filter - with applications to marine crafts Bo Zhao CeSOS / Department of Marine Technology Norwegian University of Science and Technology
2 Faults Danger and harm Pollution Property loss Safety Environmental friendly Economy Purpose
3 DP incident analysis Incident in % 45 40 35 30 25 20 15 10 5 0 Incident trends 1990 2001: Primary causes Average incidents in percentage on primary causes Electrical Operator Environment Generator Other Computer Reference Year Computer Reference Thruster Operator Electrical Generator Environment Other Thruster Data from: The Software Problem ++, Marine Cybernetics, 2004.
4 DP incident report: Vessel: Date: Wind: Sea State: Drifted: Two divers on the sea bed, were testing within a subsea structure. 22:10 Bibby Topaz 18/09/12 30 knots, 316 5 240m in 7min A series of alarms activated in the vessels in relation to the DP system. Vessel started drifting away. The Divers started to leave the structure and head back to the diving bell. Diver 2 noticed his umbilical had been snagged. 22:12 Communications and video was lost to Diver 2. The DP operator were trying to control the vessel by manual operation. 22:17 Regained control of the vessel having drifted off approx. 240m. The vessel was driven manually by the master back to the structure. 22:40 22:46 The vessel was back in a position close to the drilling template. Both Divers were on the bell stage. Diver 2 was recovered into the bell.
5 Content Principle: System with faults Particle filter Diagnosis of faults Applications: Diagnosis of DP position reference system Underwater robot navigation
6 Principle Hidden Markov model State observer Switching-mode HMM Mode/state observer Kalman filter Extended KF Unscented KF Particle filter time projectile rebounds on ground
7 Principle Hidden Markov model State observer Switching-mode HMM Mode/state observer Kalman filter Extended KF Unscented KF Particle filter time projectile rebounds on ground
8 How do we diagnose a fault? Prediction Predicted Fault free behavior Predicted Faulty behavior
9 How do we diagnose a fault? Prediction Predicted Fault free behavior Observation Take the measurement Correction Obs. Predicted Faulty behavior H0 H1 Compare
10 Application to Diagnosis of DP position reference system
11 Application to Diagnosis of DP position reference system Challenge: Complex external disturbance Wave frequency motion Wind and current force Nonlinear system behavior Model uncertainty Multiple failure modes GPS Drifting Bias Outliers HPR Random excursion Outliers GPS drifting wave
12 Application to Diagnosis of DP position reference system Results: Alarm when faults happen HPR in function Diagnosis faults Acceptable positioning during failure Pros: Avoids catastrophic consequences by giving DPO time to handle faults Assists the DPO diagnosing faults GPS drifting wave Cons: Relatively poor position estimation comparing with other DP observer Time consuming
13 Application to Robust Navigation of Underwater Robot
14 y x z
15 y 2 Vertical thrusters x Vertical: 1.2 knot z 2 Main thrusters Tunnel thruster Yaw rate: 60 /s
16 y compass Yaw rate gyro x z HPR (Hydroacoustic position reference) depth sensor DVL (Dopple Velocity Log)
17 Robust Navigation of Underwater Robot ROV Model Kinetics & Kinematics Current Propulsion reduction HPR dropout outliers DVL dropout bias
18 HPR Hydro acoustic position reference Faults: 1. Dropout when no signal received 2. Outlier Measurement has significant difference from the true position
19 DVL Doppler velocity log Faults: 1. Dropout when no signal received 2. Bias small size constant difference between the measurement and the true velocity
20 Robust Navigation of Underwater Robot ROV Model Kinetics & Kinematics Current Propulsion reduction HPR dropout outliers DVL dropout bias
21 Robust Navigation of Underwater Robot Information flow Kinetics & Kinematics Current System Model Propulsion reduction Predictions HPR dropout outliers DVL dropout bias Measurement Models Estimation from last sampling time Estimation and Diagnosis
22 Robust Navigation of Underwater Robot Experiment Kinetics & Kinematics Current Model Full scale test, ROV Minerva October 17 18, HPR2012, Propulsion Trondheimsfjord. DVL dropout reduction outliers Real disturbance. Real measurement. Prediction dropo In the real time control loop Faults were triggered. ut bias Measurement Estimation from time k 1 Estimation and Diagnosis
23 Robust Navigation of Underwater Robot Experiment Kinetics & Kinematics Current Model measurment estimation command faulty Estimation measurement from KF time estimation k 1 2m 2m Full scale test, ROV Minerva October 17 18, HPR2012, Propulsion Trondheimsfjord. DVL dropout reduction outliers Real disturbance. Real measurement. Prediction dropo In the real time control loop Faults were triggered. ut bias Measurement Estimation and Diagnosis 16m 2m
24 Robust Navigation of Underwater Robot Experiment Kinetics & Kinematics Current Model Normal measurment estimation command faulty Estimation measurement from KF time estimation k 1 2m 2m Full scale test, ROV Minerva October 17 18, HPR2012, Propulsion Trondheimsfjord. DVL dropout reduction outliers Real disturbance. Real measurement. Prediction dropo In the real time control loop Faults were triggered. ut bias Measurement Estimation and Diagnosis 16m 2m
25 Robust Navigation of Underwater Robot Experiment Kinetics & Kinematics Current Model measurment estimation command faulty Estimation measurement from KF time estimation k 1 2m 2m Full scale test, ROV Minerva October 17 18, HPR2012, Propulsion Trondheimsfjord. DVL dropout reduction outliers Real disturbance. Real measurement. HPR Outliers Prediction dropo In the real time control loop Faults were triggered. ut bias Measurement Estimation and Diagnosis 16m 2m
26 Robust Navigation of Underwater Robot Experiment Kinetics & Kinematics Current Model measurment estimation command faulty Estimation measurement from KF time estimation k 1 2m 2m Full scale test, ROV Minerva October 17 18, HPR2012, Propulsion Trondheimsfjord. DVL dropout reduction outliers Real disturbance. Real measurement. Prediction dropo In the real time control loop Faults were triggered. ut Measurement HPR drop bias Estimation and Diagnosis 16m 2m
27 Experiment Fault Free HPR Outliers HPR Dropout DVL Dropout HPR + DVL DVL Bias Thruster Loss
28 Experiment Fault Free HPR Outliers HPR Dropout DVL Dropout HPR + DVL DVL Bias Thruster Loss measurment estimation command faulty measurement KF estimation
29 Experiment Fault Free HPR Outliers HPR Dropout DVL Dropout HPR + DVL DVL Bias Thruster Loss measurment estimation command faulty measurement KF estimation
30 Principle: Summary System with faults: SM HMM Particle filter for fault diagnosis Applications: Diagnosis of DP position reference system Underwater robot robust navigation Future work: Efficiency issue for particle filter in SM HMM Formal design process for the mode transition Transient after mode switching
31 Reference: Particle filter: Zhao, B.; Skjetne, R. & Blanke, M., Particle Filter and for Fault and Diagnosis and Robust Navigation of Underwater Robot, IEEE Transactions on control systems technology (Submitted), IEEE, 2013 Fault diagnosis (regarding marine crafts): Blanke, M.; Kinnaert, M.; Lunze, J. & Staroswiecki, M., Diagnosis and Fault Tolerant Control, Springer Berlin Heidelberg, 2006 Blanke, M., Diagnosis and Fault Tolerant Control for Ship Station Keeping, Intelligent Control, 2005. Proceedings of the 2005 IEEE International Symposium on, Mediterrean Conference on Control and Automation, 2005, 1379 1384 Others The Software Problem ++, Marine Cybernetics, 2004.