Spatial MTBF/MTTR analytics dashboard for manufacturing operations — bridging Operational Excellence and Predictive Machine Learning.
Manufacturing plants generate thousands of failure events per year, but most reliability and industrial engineers analyze them in flat spreadsheets, losing spatial and temporal context entirely. When a critical machine goes down on the Beta line, you want to see it on the plant floor, understand why it failed, predict when the next failure will happen, and get a concrete action plan — not find a row buried in a CSV.
This project answers four questions in one dashboard:
- Where? — Which machines need attention today and where are they on the floor?
- Why? — Which components and failure types are driving the most downtime cost?
- When? — What will the health score look like in 1–3 months?
- What to do? — What specific action should be taken, by whom, at what cost?
Interactive Plotly plant heatmap — hover over any machine to see its full health score breakdown (MTBF / Availability / Failures contributions). Value Stream Cost Intelligence cards per VSM line (Alpha, Beta, Gamma) with current + forecast scores. Top Critical and Top Healthiest machine tables.
Date range selector (From → To) across the full historical window. VSM trend lines per metric (Health Score, MTBF, MTTR, Availability, Cost). Stacked failure bar chart by area with value labels. Health score calendar heatmap (machine × month). Machine snapshot table for any selected month.
Linear trend forecast per machine with 80% prediction intervals. VSM-level forecast summary cards. Machine Risk Ranking table (HIGH / MEDIUM / LOW). Machine Deep Dive: forecast chart + insight card with trend signal quality. Technical methodology expander explaining model choice and evolution roadmap.
- A — Component Cost Pareto (80/20 rule)
- B — Component × Machine Downtime Heatmap
- C — Avg MTTR by Failure Type & Area
- D — Prioritized Prescriptive Action Plan (automated, cost-justified)
- E — Cost to Act vs Cost if Ignored (3-month projection)
- F — Component ROI Summary (Value Stream-wide)
Each machine receives a composite health score:
normalized_mtbf = mtbf / max_mtbf_in_VS
normalized_avail = availability_pct / 100
normalized_failures = 1 - (failures_30d / max_failures_in_VS)
health_score = (normalized_mtbf × 0.50)
+ (normalized_avail × 0.30)
+ (normalized_failures × 0.20)
× 100
| Score Range | Status | Color |
|---|---|---|
| 70 – 100 | Healthy | Green |
| 40 – 69 | Monitor | Amber |
| 0 – 39 | Critical | Red |
Linear regression fitted independently on each machine's monthly health score time series. In a well-run maintenance operation, health scores do not follow a clean linear trend — every corrective action resets the trajectory. The model measures the velocity of deterioration between interventions, not a fixed long-term destiny. See the in-app Technical Methodology expander (Tab 3) for the full justification and model evolution roadmap.
Fleet-relative percentile thresholds (p90 / p75 failure frequency and cost) trigger prioritized actions. Each recommendation includes cost to act, projected cost of inaction over 3 months, and ROI. Actions self-calibrate to any dataset size.
| Metric | Description |
|---|---|
| MTBF | Mean time between failures per machine |
| MTTR | Mean repair time per machine |
| Availability | (Available hrs − Downtime) / Available hrs |
| Failures (30d) | Count of failures in the rolling 30-day window |
| Monthly Downtime Cost | Downtime hrs × cost per hr |
| Health Score | Composite 0–100 reliability score |
| Component MTTR | Repair time per component type |
| Part Lead Time | Days to receive replacement part |
| ROI of preventive action | (Cost if ignored − Cost to act) / Cost to act |
Three independent VSM production lines flowing bottom → top:
┌──────────────────────────────────────────────────────────┐
│ FINISHED GOODS WAREHOUSE │
├───────────────┬───────────────┬──────────────────────────┤
│ Assembly │ Assembly │ Assembly │
│ Alpha (8) │ Beta (5) │ Gamma (3) │
├───────────────┼───────────────┼──────────────────────────┤
│ Painting │ Painting │ Painting │
│ 2PB + CF │ 1PB + CF │ 1PB + CF │
├───────────────┼───────────────┼──────────────────────────┤
│ Machining │ Machining │ Machining │
│ 5 CNC │ 3CNC+Lathe │ 2CNC + VMC │
├───────────────┴───────────────┴──────────────────────────┤
│ RAW MATERIALS WAREHOUSE │
└──────────────────────────────────────────────────────────┘
VSM Alpha VSM Beta VSM Gamma
(High-Speed) (Semi-Auto) (Flexible)
35 machines total across 3 areas (Machining, Painting, Assembly).
Place raw files in data/raw/ before running the ETL pipeline.
Exported from your ERP or CMMS (SAP PM, Maximo, eMaint).
| Column | Type | Description |
|---|---|---|
machine_id |
string | Unique equipment identifier |
vsm |
string | Production line (Alpha / Beta / Gamma) |
area |
string | Plant area (Machining / Painting / Assembly) |
machine_type |
string | Equipment category (CNC, Lathe, VMC, Paint Booth, etc.) |
year_installed |
integer | Year of installation |
manufacturer |
string | Equipment manufacturer |
downtime_cost_per_hr |
float | Estimated cost per hour of unplanned downtime (USD) |
Exported from your CMMS work order system (corrective maintenance orders).
| Column | Type | Description |
|---|---|---|
failure_id |
string | Unique work order ID |
machine_id |
string | Equipment that failed |
failure_date |
date | Date the failure occurred (YYYY-MM-DD) |
failure_mode |
string | What failed (e.g. "Spindle fault", "Coolant leak") |
downtime_hrs |
float | Total hours the machine was out of service |
repair_hrs |
float | Hours spent on the repair |
technician_id |
string | Technician who performed the repair |
root_cause |
string | Root cause classification |
corrective_action |
string | Action taken to resolve |
component |
string | Subsystem that failed (spindle, servo_drive, coolant_system, etc.) |
failure_type |
string | mechanical / electrical / software / hydraulic |
technician_type |
string | mechanical / electrical / automation / hydraulic |
part_replaced |
string | Part number or description replaced |
part_cost_usd |
float | Cost of the replacement part (USD) |
part_lead_time_days |
integer | Days to receive the part from supplier |
time_to_diagnose_hrs |
float | Hours spent on diagnosis before repair |
Exported from your MES or shift report system.
| Column | Type | Description |
|---|---|---|
date |
date | Production date (YYYY-MM-DD) |
vsm |
string | Production line |
shift |
string | Shift (Day / Evening / Night) |
planned_hrs |
float | Planned production hours |
actual_hrs |
float | Actual hours run |
units_produced |
integer | Units completed |
units_rejected |
integer | Units rejected / scrapped |
oee_pct |
float | Overall Equipment Effectiveness (%) |
pip install -r requirements.txtdata/raw/
├── equipment_master.csv
├── failures.csv
└── production_data.csv
python _run_etl.pyGenerates three processed files:
data/processed/
├── mtbf_metrics.csv # Current-period KPIs per machine
├── monthly_metrics.csv # Time-series (machine × month) for forecasting
└── component_metrics.csv # Component-level failure aggregations for root cause
python -m streamlit run app.pyOpens at http://localhost:8501.
VS_Health_Intelligence/
├── app.py # Streamlit dashboard — 4 tabs
├── config.yaml # Path and parameter configuration
├── requirements.txt
├── _run_etl.py # ETL runner (generates all processed CSVs)
│
├── assets/ # Dashboard screenshots
├── data/
│ ├── raw/ # Source CSVs (equipment, failures, production)
│ ├── processed/ # ETL output (3 CSVs)
│ └── layout/ # zones.json + plant_layout.png
│
├── src/
│ ├── etl/ # Extract / Transform / Load pipeline
│ │ ├── extract.py
│ │ ├── transform.py # MTBF, monthly, and component metrics
│ │ └── load.py
│ ├── metrics/ # MTBF, MTTR, OEE calculation modules
│ ├── spatial/ # Coordinate mapping + Plotly interactive heatmap
│ │ ├── coordinates.py
│ │ ├── heatmap.py # generate_heatmap_plotly() with hover tooltips
│ │ └── polygons.py
│ ├── visualization/ # Plot helpers per tab
│ │ ├── plots.py # KPI tables (VSM summary, top critical/healthy)
│ │ ├── trends.py # Monthly trend charts + machine forecast chart
│ │ └── rootcause.py # Pareto, heatmap, MTTR, urgency charts
│ ├── ml/
│ │ ├── forecasting.py # Linear regression per machine + R²adj / MAE
│ │ ├── explainer.py # Machine insight text + risk table
│ │ └── prescriptive.py # Prescriptive action engine + ROI calculation
│ └── utils/ # Config, logger, helpers
│
├── notebooks/ # Jupyter EDA + methodology walkthroughs
├── tests/ # Pytest suite
└── outputs/figures/ # Generated heatmap PNG
pytest tests/ -v- Phase 1: MTBF/MTTR spatial heatmap dashboard
- Phase 2: Monthly trends, date range analysis, health score calendar
- Phase 3: Predictive simulation — linear trend forecast with 80% CI per machine
- Phase 4: Root cause & prescriptive analytics — component-level intelligence, automated action plan
- Phase 5: Interactive Plotly heatmap with hover score breakdown
- Phase 6: Holt-Winters / ARIMA upgrade (recommended at 18+ months of data)
- Phase 7: Real-time sensor data integration (MQTT / OPC-UA)
LozanoLsa · Turning Operations into Predictive Systems