jev-fault-diagnosis-poc
Industrial demo replaying compressor telemetry where code flags anomalies and Jev picks the manual fault, severity and certainty.
Overview
Fault Diagnosis PoC performs manual-grounded fault triage for an industrial compressed-air unit. It replays real compressor telemetry through an emulated Modbus device, computes windows, trends and machine state in code, and when a rule raises a suspect event, asks a decision model three questions: which fault from the machine manual (or none), how severe, and how sure. The model never sees raw time series and never writes to the machine.
On first start, the manual PDF is turned into a fault catalogue and retrieval index, so every answer names a fault and its manual section. Ground truth about which fault is active is isolated from the diagnosis code by separate packages, credentials, database roles and lint rules. The browser dashboard shows signals, alerts, tickets and running cost. Jev is the default backend with a key; Anthropic Claude and a rules-only baseline (default without a key) are alternatives. It runs with Docker Compose and is read-only by design.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Which fault from the manual explains this suspect event (or none)? | — |
| Score | How severe is it? | — |
| Score | How sure is the diagnosis? | — |
How it works
- Replay telemetry through an emulated Modbus device
- Code computes features and raises suspect events
- Decision backend picks fault, severity and certainty from the manual catalogue
- Create alerts and tickets on the dashboard
Strengths
- Answers grounded in the machine manual
- Ground truth isolated by construction
- Runs with no key via a rules baseline
Caveats
- Proof of concept
- CI screenshots use a mock endpoint, not real accuracy
Getting started
With Docker Compose v2.24+, follow the Quick start; without a key it uses the rules-only baseline.At a glance
- Type
- Demo
- Status
- prototype
- Language
- Python
- Runtime
- Docker Compose stack with web dashboard
- Has code
- yes
- Has tests
- yes
Use cases and integrations
How curated lists describe it
Fault Diagnosis PoC replays real compressor telemetry through an emulated Modbus device, spots suspicious behaviour with plain code, and asks a decision model which fault from the machine manual explains it.
hellogumbo/awesome-jev
Frequently asked questions
What does Jev decide?
Which manual fault explains a suspect event, how severe it is and how certain.
Does the model control the machine?
No, the system is read-only.
Can I try it without a key?
Yes, a rules-only baseline answers the same questions.
Related projects
Scores OpenTelemetry logs with Jev for diagnostic value, priority and routing before any expensive LLM analysis.
ProjectbetaTypeScriptSplunk search command that asks Jev typed questions about events and adds probabilities, choices and scores as fields.
AppbetaPythonRust CLI and TUI that streams Kubernetes logs, applies offline rules and Jev analysis to flag what matters, then hands off.
ProjectbetaRustSmart-meter fleet dashboard where code pre-filters telemetry and Jev classifies anomalies for remediation across millions of endpoints.
ProjectprototypeTypeScriptOpenTelemetry Collector processor that asks Jev how useful each metric is from its metadata and applies keep/drop policy.
ServiceprototypeGoOpenTelemetry trace processor that asks Jev whether an operation is diagnostic, business-critical and worth keeping, and annotates spans.
ProjectprototypeGo