tbohne/vehicle_diag_smach

Neuro-symbolic anomaly detection and complex fault diagnosis exemplified in the automotive domain.

vehicle_diag_smach's Issues

Suspect components are not suggested based on the priority list in the `smach`

• they should be sorted based on the priority_id

Apply improvements / bugfixes from `nesy_diag_smach`

• the vehicle_diag_smach is still not supporting many of the features, e.g., it only uses the first error code leading to some fault path, potentially following ones are ignored
• additionally, all kinds of bugs that are fixed by now in nesy_diag_smach are still present in vehicle_diag_smach

Investigate: Unexpected fault path

During a first smach-hub integration demo, the following happened:

Test case with multiple DTCs and only one fault path

• test case with multiple DTCs in which only the first one leads to a fault path, i.e., to anomalies
• additionally, case with no anomaly for first DTC, but for later ones

Extend `TestHighLevelStateMachine` with further scenarios

Bug: Not querying suspect components for next iteration DTCs

• the else branch is executed here, although this should only be the case for the next iteration of the same DTC, not others:
• 

Remove `PERFORM_DATA_MANAGEMENT` state

• already part of all the other states (each state performs its own data management)
• retrieving the latest trained classifier is no longer reasonable, we retrieve them when we know the component (this takes place in the classification states)
• PERFORM_DATA_MANAGEMENT state was born with the idea that we communicate all the data we obtain during diagnosis to some server, but actually more and more of this is either shifting to the individual states, or the info is not stored on some external server, but in the KG

Implement all aspects as part of other states - systematically:

• retrieving latest version of trained classifier from server
• reading customer complaints from session files + uploading
• reading OBD data from session files + uploading
• reading historical info from session files + uploading
• reading user data from session files + uploading
• reading XPS interview data from session files + uploading

Implement `test_hosted_knowledge_graph`

Add asserts for assumed expert knowledge (KG queries).

Replace I/O with interface implementations

• the entire I/O should be handled through the respective interfaces, not as part of the smach

Termination criterion in diagnostic loop

• the smach ends with the first detected anomaly that has been isolated, it doesn't continue with potential remaining DTCs - should it?
• it is conceivable that the remaining DTCs should also be examined, even if anomalies have already been detected

`FaultPath` generation bug

• distinguish between the following two cases:
  
• atm both lead to:
  

Communicate state transitions via interface function

• not only A -> B, but also the precise link that was used
• this is also related to the trace that is supposed to be displayed in the end

Compatibility issues

• update all the requirements.txt + setup.py in the dependencies
• works with Python 3.8 and tf-keras-vis 0.8.5
• verified with fresh Python env on Windows and Ubuntu

Implement Diagnosis Feedback

• kind of a labeling
• was the provided diagnosis helpful / correct?
• as part of PRODIVE_DIAG_AND_SHOW_TRACE

Fehlermeldung diagnostics-Service Werkstatt Hub

Edit: @T-Thurow veröffentlicht später nochmal ein Issue, das hier ist scheinbar nicht so hilfreich. :D

@tbohne Wir haben aktuell Probleme mit dem Start des diagnostics-Services in unserem Werkstatt-Hub. @sply88 hat dieses Repository hier in der dazugehörigen requirements.txt referenziert:

redis==4.5.4
celery==5.2.7
httpx==0.23.3
pydantic==1.10.5
git+https://github.com/tbohne/vehicle_diag_smach@4c2b043ef336a58471f652c8199efefa3aef84e6

Leider failed die Installation mit der folgenden Fehlermeldung:

Und das Repository wird anschließend nicht gefunden:

@T-Thurow hat gemutmaßt:

"Ich denke das Problem liegt bei dem vehicle_smach python Modul von Tim. @sply88 hat zwar das Packet auf einen bestimmen commit fest geschrieben aber das klappt nicht weil Tim das bei sich in der requirements nicht macht sondern immer den letzten commit verwendet. Dann laufen die die Versionen auseinander."

Rewire `smach` without `PERFORM_DATA_MANAGEMENT`

Transition not part of vis.: `isolated_problem_remaining_DTCs`

• in ISOLATE_PROBLEM_CHECK_EFFECTIVE_RADIUS

Provide KG URL to `smach` constructor

• important property to set when used as lib
• currently only configurable via config.py (should be default value)

Dynamic model selection based on suspect component

• replace hard-coded model with coded (with suspect component) pool of models
• select model based on considered component
• corresponding component coded in filename
• component name has to precisely match the component_name field of the corresponding SuspectComponent

Replace CLI user communication with abstract states

e.g.:

• processing
• action_required
• finised

Also include intermediate results in the communication somehow.

KG extension (diag level)

Implement KG enhancements based on information collected during diagnosis, i.e. application of the state machine:

• Vehicle with HSN, TSN, VIN, model
• DiagLog with max_num_of_parallel_rec, date
• createdFor relation between DiagLog and Vehicle
• appearsIn relation between DTC and DiagLog
• entails relation between DiagLog and FaultPath
• FaultPath with path_description
• resultedIn relation between FaultCondition and FaultPath
• ParallelRecOscillogramSet
• partOf relation between Oscillogram and ParallelRecOscillogramSet
• Oscillogram with time_series
• classifies relation between OscillogramClassification and Oscillogram
• OscillogramClassification with model_id and uncertainty and prediction
• produces relation between OscillogramClassification and Heatmap
• Heatmap with generation_method and generated_heatmap
• ManualInspection with prediction
• reasonFor between two Classification instances
• checks relation between Classification instance and SuspectComponent
• ledTo relation between DiagnosticAssociation and Classification
• diagStep relation between Classification instance and DiagLog

Log `refuted_hypothesis` case somewhere

• collect relevant information about unsuccessful diagnoses
• crucial guidance for expert knowledge acquisition
• should be stored on server as well

Extend smach vis. after refining termination criterion

• extend smach vis. after implementing #7

Incorporate hierarchical `smach` design in code

Using version tags in python requirements

The requirements.txt currently references the upstream version of obd_ontology and oscillogram_classification which are prone to change.

It would be better to reference specific version tags or commit hashes to keep older commits build clean in case there are API changes for obd_ontology oroscillogram_classification.

Adapt CAM calls after signature changes in ` Patricia dev #15 ` (`oscillogram_classification`)

• gen_heatmaps_overlay_side_by_side
• gen_heatmaps_as_overlay
• plot_heatmaps_as_overlay
• plot_heatmaps
• read_oscilloscope_recording

Refine state names etc. according to new vis.