Outline • Motivation • Previous Mode Estimation Approaches • Example ACS System Improving Model-based Mode Estimation • Miniature Mode Estimation System Through Offline Compilation • Rule-based system comparison • Conclusions Seung H. Chung John M. Van Eepoel Brian C. Williams Model-based Embedded and Robotic Systems Group Space Systems Laboratory Massachusetts Institute of Technology Thursday, June 21 st , 2001 Motivation General Diagnostic Engine Apollo 13 quintuple fault • Rule-based systems are becoming overly complex as spacecraft systems grow. System Models • The model-based approach allows for a more direct fault protection system. • This model-based mode estimation system will Candidate Merge Observations Partial Diagnoses Diagnoses give: Generation Diagnoses • A set of inspectable rules used in diagnosis • Faster response over previous systems. • Achieves a diagnosis through the divide and conquer approach Distant Explorers Mercury Orbiter • Generates all diagnoses, including highly unlikely ones. Mars Polar Lander failed due to a faulty sensor. Sherlock / Livingstone Miniature Mode-Estimation System Models Offline Online System Partial Diagnosis Satisfiability Conflict-directed Observations Diagnosis Models Diagnosis Rules Engine Search SAT Engine Most-likely Candidate Observations Checked Solution Conflict-directed Search Conflicts • Generates all diagnoses offline using dissents. • Generates only most likely diagnosis that pertains to the current – Methods employed are resolution, enumeration and multi- observations through a generate and test approach. resolution. • The conflict directed search is utilized at run time to generate a • Problem : The test phase - SAT-SEARCH loop is expensive in time. diagnosis. 1
Miniature Mode Estimation Example ACS Subsystem GHe • Pressure Transducer (S) Model Mini-ME N 2 H 4 – Modes – {G, SH, SL, U} * Offline Online G(S) ⇒ ((TP = nom) ⇔ (P1 = nom)) ∧ – Partial Best-first ((TP = low) ⇔ (P1 = low)) Partial Spacecraft Compiled Dissent Dissents Diagnosis Partial Diagnosis Conflicts Kernel Most Likely Repair Pressure Diagnosis Manager SH(S) ⇒ (P1 = nom) Model Model Generator Rule Rules Diagnosis Diagnosis P – Trigger Transducer Generator Generator Pipe 1 SL(S) ⇒ (P1 = low) – U(S) ⇒ ( ) Solenoid Valve S – Discrete Continuous Pipe 2 • Solenoid Valve (V) Model Monitors Observations Observations Catalyst Bed – Modes – {O, C, U} * O(V) ⇒ ((P1 = nom) ⇔ (P2 = nom)) ∧ – Hydrazine ((P1 = low) ⇔ (P2 = low)) Thruster • Offline satisfiability generates the partial diagnoses in two C(V) ⇒ (P2 = low) – steps. Inertial Sensor U(V) ⇒ ( ) – • Online search performed by: • Observable Variables • Catalyst Bed (C) Model – Pressure in Pipe 1 (P1) – { nom, low } – Triggering the appropriate partial diagnoses – Modes – {G, B, U} * – Engine Thrust (T) – { on, off } G(C) ⇒ ((P2 = nom) ⇔ (T = on)) ∧ – – Search is directed by these conflicts ((P2 = low) ⇔ (T = off)) • Unobservable Variables B(C) ⇒ (T = off) – – Tank Pressure (TP)– { nom, low } U(C) ⇒ ( ) – – Pressure in Pipe 2 (P2) – { nom, low } * All modes have an associated probability. Offline Satisfiability Online Conflict-directed Search Online Conflict-directed Search Offline Satisfiability Best-first Offline Partial Partial Diagnosis Partial Online Kernel Conflicts M o s t L i k e l y Repair Dissents Satisfiability Rules Spacecraft Compiled Dissent Diagnosis Partial Diagnosis Most Likely Diagnosis Conflict-directed Diagnosis D i a g n o s i s Manager Model Model Generator Rule Rules Diagnosis Trigger Search Generator Generator Observations Discrete Continuous Monitors Observations Observations • The satisfiability step is moved to an offline computation and dissents • Monitors discretize continuous data in two ways. are generated. – Value : Sensor Voltage = 23 V Sensor Voltage = nominal – Dissents link infeasible component modes to observations. – Time • (P1 = low) ∧ (T = on) ⇒ ¬ ( G(S) ∧ O(V) ∧ G(C) ) • Partial diagnosis trigger (P1 = nom) ∧ (T = off) ⇒ SH(S) ∨ SL(S) • Dissents are then transformed into partial diagnosis rules by making • ∨ U(S) ∨ C(V) ∨ U(V) ∨ B(C) ∨ U(C) an equivalent statement as: P1 = nom G(S) ∨ SH(S) ∨ U(S) – (P1 = low) ∧ (T = on) ⇒ ( SH(S) ∨ SL(S) ∨ U(S) ∨ C(V) ∨ U(V) ∨ (P1 = nom) ⇒ G(S) ∨ SH(S) ∨ U(S) • B(C) ∨ U(C) ) (P1 = low) ⇒ G(S) ∨ SL(S) ∨ U(S) • • … Online Conflict-directed Search Sample Diagnosis • Observations GHe Online Conflict-directed Search • P1 = low Best-first N 2 H 4 Offline • T = on Partial Diagnosis Partial Kernel Conflicts Most Likely Repair Satisfiability Rules Diagnosis Diagnosis Diagnosis Manager Trigger Generator Pressure P Transducer Pipe 1 S Solenoid Valve Pipe 2 Discrete Continuous Monitors Observations Observations Catalyst Bed Hydrazine Thruster • Best-first Kernel Diagnosis Generator then determines the set of assignments that best resolves the conflicts. Inertial Sensor G(S) U(S) SH(S) • Repair Manager uses the diagnosis and gives the appropriate action to the 0.017 0.002 0.865 spacecraft to resolve failure. 2
Sample Diagnosis Sample Diagnosis • Observations • Observations GHe GHe • P1 = low • P1 = low N 2 H 4 N 2 H 4 • T = on • T = on • Triggered Partial Diagnoses • Triggered Partial Diagnoses Pressure Pressure P G(C) ∨ U(C) P G(C) ∨ U(C) – – Transducer Transducer Pipe 1 Pipe 1 O(V) ∨ U(V) O(V) ∨ U(V) – – Solenoid Valve G(S) ∨ SL(S) ∨ U(S) Solenoid Valve G(S) ∨ SL(S) ∨ U(S) S S – – Pipe 2 Pipe 2 SH(S) ∨ SL(S) ∨ U(S) ∨ C(V) ∨ U(V) ∨ B(C) ∨ U(C) SH(S) ∨ SL(S) ∨ U(S) ∨ C(V) ∨ U(V) ∨ B(C) ∨ U(C) – – Catalyst Bed Catalyst Bed Hydrazine Hydrazine • Most-likely Diagnosis • Most-likely Diagnosis Thruster Thruster Inertial Sensor Inertial Sensor U(C) G(C) U(C) G(C) 0.865 0.0009 0.865 0.0009 O(V) U(V) 0.0846 0.0009 Sample Diagnosis Sample Diagnosis • Observations • Observations GHe GHe • P1 = low • P1 = low N 2 H 4 N 2 H 4 • T = on • T = on • Triggered Partial Diagnoses • Triggered Partial Diagnoses Pressure Pressure P G(C) ∨ U(C) P G(C) ∨ U(C) – – Transducer Transducer Pipe 1 Pipe 1 O(V) ∨ U(V) O(V) ∨ U(V) – – S Solenoid Valve G(S) ∨ SL(S) ∨ U(S) S Solenoid Valve G(S) ∨ SL(S) ∨ U(S) – – Pipe 2 SH(S) ∨ SL(S) ∨ U(S) ∨ C(V) ∨ U(V) ∨ B(C) ∨ U(C) Pipe 2 SH(S) ∨ SL(S) ∨ U(S) ∨ C(V) ∨ U(V) ∨ B(C) ∨ U(C) – – Catalyst Bed Catalyst Bed Hydrazine Hydrazine • Most-likely Diagnosis • Most-likely Diagnosis Thruster Thruster Inertial Sensor Inertial Sensor G(C) U(C) G(C) U(C) 0.865 0.0009 0.865 0.0009 O(V) U(V) O(V) U(V) 0.0009 0.0846 0.0009 0.0846 G(S) S L ( S ) U(S) S L ( S ) G(S) U(S) 0.0846 0.0017 0.0002 0.0002 0.0846 0.0017 C ( V ) SH(S) SL(S) U(S) U(V) B(C) U(C) Sample Diagnosis Mini-ME vs. Rule-Engine • Observations GHe • P1 = low N 2 H 4 • T = on Rule Repair Mode Mini-ME Engine Estimate Manager • Triggered Partial Diagnoses Pressure P G(C) ∨ U(C) Transducer – Pipe 1 O(V) ∨ U(V) – Action S Solenoid Valve G(S) ∨ SL(S) ∨ U(S) Observations Observations – Action Pipe 2 SH(S) ∨ SL(S) ∨ U(S) ∨ C(V) ∨ U(V) ∨ B(C) ∨ U(C) – Catalyst Bed Spacecraft Hydrazine • Most-likely Diagnosis Spacecraft Thruster Inertial Sensor G(C) U(C) 0.865 0.0009 O(V) U(V) 0.0009 0.0846 • Mini-ME makes explicit the deductive step of determining S L ( S ) G(S) U(S) component modes. 0.0002 0.0846 0.0017 – Models allow mapping from symptoms to system states to recovery actions easier to specify. SH(S) SL(S) U(S) C ( V ) U(V) B(C) U(C) • Full Diagnosis : SL(S) � O(V) � G(C) 3
Application to Spacecraft Summary • Sample NEAR rule • Conclusions – Symptom : – Builds upon prior model-based mode estimation heritage. • (Charger current > 0.8 A) for – The system is guaranteed to find a diagnosis. 10 sec – Coverage of all rules in a rule-based system is possible. – Recovery : – Time performance is in line with a rule-based system • Switch to the redundant charger and disengage the • Benefits primary. – Rule set is inspectable for correctness • Equivalent Mini-ME diagnosis – More intuitive set of rules – (Charger-current = high) ⇒ – Development of models is easier for engineers (Primary-charger = broken) ∨ – Real-time performance (Primary-charger = unknown) – Recovery action then specified Mini-ME based only on the mode. Offline Online Partial Best-first Partial Spacecraft Compiled Dissent Dissents Diagnosis Partial Diagnosis Conflicts Kernel Most Likely Repair Diagnosis Model Model Generator Rule Rules Diagnosis Diagnosis Manager Trigger Generator Generator Discrete Continuous Monitors Observations Observations 4
Recommend
More recommend
