🏭 Industrial Predictive Maintenance Platform

AI-driven failure prediction, automated diagnostics, and real-time maintenance alerting for Industrial IoT systems.

📌 Overview

The Industrial Predictive Maintenance Platform predicts industrial machine failures before they happen and automates the entire maintenance response — from sensor reading to technician alert — using a modern data engineering and machine learning stack.

Instead of reactive repairs or fixed maintenance schedules, this system continuously evaluates live sensor data, predicts whether a machine will fail within the next 24 hours, classifies the expected failure type, generates AI-powered maintenance recommendations, and pushes real-time alerts to maintenance teams — with results visualized in a Power BI dashboard.

---

🎯 Objectives

• ✅ Predict whether a machine will fail within 24 hours
• ✅ Identify the expected failure type
• ✅ Generate automated maintenance recommendations
• ✅ Send real-time alerts to maintenance teams
• ✅ Provide operational insights through dashboards

---

🏗️ System Architecture

 ┌────────────────────┐
 │  Industrial Sensors │
 └──────────┬──────────┘
            ▼
 ┌────────────────────┐
 │   Google Sheets     │  ← Data Source
 └──────────┬──────────┘
            ▼
 ┌────────────────────┐
 │   n8n Workflow      │  ← Orchestration
 └──────────┬──────────┘
            ▼
 ┌────────────────────┐
 │   FastAPI Service   │  ← Model Serving (via ngrok)
 │  ┌────────────────┐ │
 │  │ Model 1:       │ │  Failure Prediction (0/1)
 │  │ Random Forest  │ │
 │  └───────┬────────┘ │
 │          ▼           │
 │  ┌────────────────┐ │
 │  │ Model 2:       │ │  Failure Type Classification
 │  │ Random Forest  │ │
 │  └────────────────┘ │
 └──────────┬──────────┘
            ▼  (if failure predicted)
 ┌────────────────────┐
 │  Gemini AI Agent    │  ← Condition analysis + recommendation
 └──────────┬──────────┘
            ▼
 ┌────────────────────┐
 │   Telegram Bot      │  ← Real-time alert to maintenance team
 └──────────┬──────────┘
            ▼
 ┌────────────────────┐
 │  Power BI Dashboard │  ← Operational insights & KPIs
 └────────────────────┘

---

🧰 Tech Stack

Layer	Technology
Data Source	Google Sheets
Workflow Orchestration	n8n
Machine Learning	Python, scikit-learn (Random Forest)
API Layer	FastAPI
API Exposure	ngrok
AI Analysis	Google Gemini AI Agent
Notifications	Telegram Bot API
Visualization	Power BI

---

📊 Dataset

Industrial IoT sensor dataset — 24,042 records across four machine types.

Machine types: CNC · Pump · Compressor · Robotic Arm

Features

Feature	Description
machine_type	Type of industrial machine
vibration_rms	Root-mean-square vibration amplitude
temperature_motor	Motor temperature reading (°C)
current_phase_avg	Average electrical phase current
pressure_level	Operating pressure
rpm	Rotational speed
operating_mode	Current operational state
hours_since_maintenance	Hours elapsed since last maintenance
ambient_temp	Ambient environmental temperature

Targets

Target	Type	Values
failure_within_24h	Binary	0 = Normal, 1 = Failure
failure_type	Multi-class	bearing, electrical, hydraulic, motor_overheat, none

---

🤖 Machine Learning Pipeline

1. Data Cleaning — duplicate removal, missing value imputation, outlier handling
2. Feature Engineering — thermal delta, vibration-to-RPM ratio, maintenance urgency ratio
3. Label Encoding — categorical variables encoded for tree-based models
4. Train-Test Split — 80/20 stratified split
5. Model Training — Random Forest Classifier (two stages)
6. Model Evaluation — accuracy, precision, recall, F1-score, ROC-AUC
7. Model Deployment — served via FastAPI, exposed through ngrok

Model Architecture

• Model 1 — Failure Prediction: binary classifier → failure_within_24h
• Model 2 — Failure Type Classification: multi-class classifier → failure_type (invoked only when Model 1 predicts failure)

ℹ️ Metrics are illustrative of the evaluation methodology — replace with your actual training results.

---

🔌 API Reference (FastAPI)

Endpoint	Method	Description
/predict/failure	POST	Returns failure_within_24h prediction + probability
/predict/failure-type	POST	Returns predicted failure_type + class probabilities
/health	GET	Service health check
/model-info	GET	Model version & feature schema

---

⚙️ n8n Workflow

1. Google Sheets Trigger — polls for new sensor rows
2. HTTP Request → /predict/failure
3. IF Node — branches on failure_within_24h == 1
4. HTTP Request → /predict/failure-type (failure branch only)
5. Gemini AI Agent Node — generates condition analysis & recommendation
6. Telegram Node — sends alert to maintenance team
7. Write Node — logs results for Power BI reporting

---

📲 Sample Telegram Alert

ALERT: Predictive Maintenance System
Machine: CNC-14 | Type: CNC
Status: FAILURE PREDICTED (Probability: 0.91)
Failure Type: Bearing Wear
Recommendation: Replace spindle bearing within 12
hours. Reduce load to 60% until serviced.
Time: 2026-07-19 09:42

---

📈 Power BI Dashboard

• KPI cards — machines monitored, active alerts, failures (24h)
• Failure trend line chart by machine type
• Failure type distribution (pie chart)
• Machine health heatmap
• Recent alerts & recommendations table

---

📂 Project Structure

industrial-predictive-maintenance/
├── data/
│   └── sensor_data.csv
├── notebooks/
│   ├── 01_data_cleaning.ipynb
│   ├── 02_feature_engineering.ipynb
│   └── 03_model_training.ipynb
├── models/
│   ├── failure_prediction_rf.pkl
│   └── failure_type_rf.pkl
├── api/
│   ├── main.py
│   ├── schemas.py
│   └── preprocessing.py
├── n8n/
│   └── workflow.json
├── dashboard/
│   └── powerbi_report.pbix
├── requirements.txt
└── README.md

---

🚀 Getting Started

1. Clone the repository

git clone https://github.com/<your-username>/industrial-predictive-maintenance.git
cd industrial-predictive-maintenance

2. Install dependencies

pip install -r requirements.txt

3. Run the FastAPI service

uvicorn api.main:app --reload --port 8000

4. Expose the API with ngrok

ngrok http 8000

5. Import the n8n workflow

Import My workflow.json into your n8n instance and configure:

• Google Sheets credentials
• FastAPI (ngrok) endpoint URL
• Gemini AI API key
• Telegram Bot token & chat ID

6. Open the Power BI dashboard

Open dashboard/powerbi_report.pbix and point it to your results data source.

---

🔮 Future Improvements

• Replace Google Sheets with a real streaming source (MQTT/Kafka)
• Deploy FastAPI to a persistent cloud host (AWS/Azure/GCP)
• Add model drift detection & automatic retraining
• Support multiple prediction horizons (6h, 24h, 72h)
• Explore LSTM/TCN models for time-series degradation trends
• Add authentication & role-based access control
• Build a technician feedback loop to improve model accuracy

link dashboard https://app.powerbi.com/view?r=eyJrIjoiOWU5YTQ0NTMtYTQ4Mi00OWFiLTlhNDEtOTI1N2UzYzAxMDZkIiwidCI6IjJiYjZlNWJjLWMxMDktNDdmYi05NDMzLWMxYzZmNGZhMzNmZiIsImMiOjl9

---