4. Case Study — Diagnosis

Pipelines are complementary at the scenario level

Scenario · cell = decision latency (s)Knowledge graphTime-series, singleTime-series, multiBayesian
01 Biological Filter Saturation13
02 CDRA Failure171016
03 CDRA LiOH Canister Saturation11
04 Electrolysis System Failure6060
05 Emergency O2 System Maintenance
06 Excess CO2 in Cabin
07 Excess Gas Leak131613
08 Excess Water Vapor Pressure in Cabin
09 Fuel Cell #1 and PDU Failure302525
10 Fuel Cell #2 and PDU Failure182518
11 Fuel Cell Degrade15341619
12 Fuel Cell Failure19841928
13 Loss of Pressure721715
14 MOXIE Antenna Failure32191824
15 MOXIE ECM Failure4646
16 MOXIE Fan Failure328282
17 Main Cabin Fan Failure179
18 N2 Tank Burst8275
19 PDU 4 Failure129
20 PDU 5 Failure24
21 RWGSR Malfunction211321
22 Reduced Main Cabin Fan #1 Capacity142428
23 SPE System Maintenance101717
24 TCCS Auxiliary Fan #1 Failure9811
25 TCCS Auxiliary Fan #2 Failure141425
26 TCCS Auxiliary Fan at Reduced Capacity13813
27 TCCS Filter Clog29812
28 Trace Contaminants
29 WRS Failure212122
30 WRS Maintenance
31 WRS Off-nominal pH Level191919

Only one pipeline solves

Time-series matching, multi: Biological Filter Saturation, CDRA LiOH Canister Saturation

Bayesian: PDU 5 Failure (time-series matching sees a 50 / 50 tie with PDU 4)

No pipeline solves

Emergency O₂ Maintenance, Excess CO₂, Excess Water Vapor, Trace Contaminants — rivals stay close; gate correctly waits

WRS Maintenance — never crosses a limit

Wrong decisions (knowledge graph)

Fuel Cell #2 & PDU, Fuel Cell Degrade → “Fuel Cell Failure”; N2 Tank Burst → “Loss of Pressure”