Troubleshooting Laboratory

This area is restricted to invited collaborators. Enter your credentials to access the lab.

🔬 Research Environment

Troubleshooting Laboratory

A structured research environment for exploring systematic fault diagnosis methodology, functional decomposition modeling, and knowledge graph approaches to machine troubleshooting.

⚗️ 8 Active Experiments
📁 4 GitHub Repositories
👥 Invited Collaborators Only
The Research Equation

The Path to Computational Troubleshooting Systems

Computational troubleshooting systems require parallel research across eight interconnected domains. Each stream below addresses a necessary component of the whole — only their synthesis produces a system capable of supporting real-world fault diagnosis.

01
⚙️
Machine Graph Research
Modeling physical machines as networks of components and their interconnections. Without machine structure, no diagnostic system can reason about how a fault propagates from cause to symptom.
02
🔀
Functional Graph Research
Representing what each subsystem does — not just what it is. Function-based hierarchies enable efficient half-split diagnostic strategies that physical layouts alone cannot support.
03
🛠️
Troubleshooting Skill Research
Capturing how expert technicians actually diagnose faults. Without modeling expert skill explicitly, we cannot replicate, scaffold, or transfer it to less experienced practitioners.
04
🧭
Executive Agent Research
The orchestration layer. How an AI agent coordinates evidence gathering, hypothesis testing, and human collaboration across the duration of a diagnostic session.
05
🧠
Human Capability Research
Understanding what technicians can actually do at varying skill levels. Computational systems must adapt to the human in the loop — not the other way around.
06
💬
Human Emotional Research
Diagnosis under pressure is rarely calm. Stress, frustration, and confidence all shape reasoning. A useful system must account for emotional state, not just cognitive state.
07
⚖️
Utility Modeling Research
Diagnostic decisions are tradeoffs — cost vs. risk, time vs. certainty, downtime vs. thoroughness. Without utility models, no system can recommend the right next step.
08
🔍
Evidence Reliability Research
Not all observations are equal. Sensor data, technician self-reports, customer descriptions, and historical records carry different reliability. Reasoning systems must weigh them accordingly.
=
The Synthesis

Computational Troubleshooting Systems

AI-augmented diagnostic environments that integrate machine knowledge, expert reasoning, human capability, and rigorous evidence handling — producing systems that genuinely support technicians in the field rather than replacing them.

Research Experiments

Active Laboratory Work

Each experiment below represents a structured investigation into troubleshooting methodology. Click into any experiment to access documentation, data, and findings.

Structured Troubleshooting — Equipment Application
Komatsu PC110R-1 — Structured Troubleshooting Workflow Active
Complete fault-isolation workflow built from the PC110R-1 service manual. Tests whether Schaafstal/Schraagen structured methodology produces faster, more reliable diagnosis compared to intuition-based approaches.
Excavator Manual Processing Hypothesis Tracking Half-Split Strategy
Open Docs GitHub
JLG Skytrack Telehandler — Full Diagnostic Application Active
Multi-phase structured troubleshooting experiment on the JLG Skytrack. Combines functional decomposition with subsystem interaction mapping to navigate complex cross-system fault interactions.
Telehandler Multi-System Field Validation Interaction Matrix
Open Docs GitHub
Functional Decomposition — Modeling
Komatsu PC110R-1 — Functional Hierarchy In Review
Function-based (not component-based) machine hierarchy for the PC110R-1. Organized by what each subsystem does, enabling efficient half-split diagnostic routing. Includes machine-readable JSON abstraction from manual processing.
JSON Abstraction Function Hierarchy Hydraulic Electrical
Open Docs GitHub
JLG Skytrack — Subsystem Interaction Matrix Active
Complete cross-system dependency mapping for the Skytrack. Documents how hydraulic, electrical, drivetrain, operator controls, and safety systems interact — identifying fault propagation paths and diagnostic leverage points.
Interaction Matrix Cross-System Safety Systems Drivetrain
Open Docs GitHub
Maintenance Manual Processing — Methodology Active
Protocol for transforming machine service manuals (PDF) into structured JSON abstractions suitable for functional decomposition and knowledge graph construction. Validated across Volvo, Caterpillar, and Komatsu manuals.
PDF Processing JSON Schema Multi-Brand AI-Assisted
Open Docs GitHub
Knowledge Graphs — Fault Diagnosis
Machine Fault Knowledge Graph — Prototype Draft
Graph-based representation of machine diagnostic knowledge. Nodes represent components and functions; edges represent dependencies and failure relationships. Enables non-linear diagnostic routing that reflects real machine complexity.
Graph Neural Networks Bayesian Node-Edge Model
Open Docs GitHub
GNN Literature Review — Fault Diagnosis Applications In Review
Synthesis of 40+ academic papers on graph neural networks applied to machine fault diagnosis. Identifies dominant architectural patterns, datasets, and open research questions relevant to industrial equipment troubleshooting.
Literature Review 40+ Papers Research Synthesis
Open Docs GitHub
Structured Troubleshooting — Core Methodology Reference Active
Foundational methodology documentation: Schaafstal, Schraagen & van Berlo framework; half-split fault isolation strategy; hypothesis tracking protocols; competence-adaptive coaching model. Serves as the theoretical basis for all lab experiments.
Schaafstal Framework Half-Split Hypothesis Tracking
Open Docs GitHub