Predictive
Maintenance
AI-Powered Industrial IoT Monitoring — Anomaly Detection via Hybrid CNN-Transformer Autoencoder
Datapoints
87K+
Devices
19
Features
188
Anomalies
879
Data Points
87,813
minutes of monitoring
Anomalies
879
1.0% detection rate
IoT Sensors
188
across 19 ESP32 devices
Best Accuracy
0.7073
CNN-Transformer AE
GPU Training
2.3s
per epoch · RTX 4070 Ti
System Health
88%
all devices healthy
Technical Overview
System Architecture
End-to-end IoT → ML → Dashboard pipeline
On-Site Photos
Hardware & Deployment
Goldsapa Bakery, Turkmenistan
ESP32 module + Dala Meter
Energy meter on RS-485 panel
Production line
Factory floor overview
Technical Illustrations
Hardware Design
ESP32 + RS-485 + Raspberry Pi
Monitoring panel
MIWE control panel
Wachtel Columbus oven
Compact oven 224°C
ML Pipeline
Anomaly Detection Dashboard
CNN-Transformer Autoencoder
IoT Infrastructure
Per-Device Health Monitoring
19 ESP32 devices
Абдулхамит Назар
Data Science · Group 22-08
Diploma thesis: «Development and Evaluation of a Predictive Maintenance Model for Industrial Equipment Based on Time Series Analysis from IoT Sensors»
Period
2025
Object
Goldsapa
Model
CNN-Transformer