AI-Powered Surface Inspection System 🏭🧠 (S7-1200 / S7-1500 & YOLOv8)

(Note: This project serves as an industrial AI-to-PLC quality control bridge designed for high-speed automated defect detection on production lines, with built-in Siemens S7-1200/1500 integration.)

This project is an Artificial Intelligence (AI) based industrial quality inspection system that detects surface anomalies and defective products on manufacturing lines in real-time, communicating directly with Siemens PLCs (S7-1200 / S7-1500) via Ethernet using the Snap7 library.

---

🌟 Key Features

• YOLOv8 Deep Learning Vision: Real-time object and anomaly detection on industrial surfaces (MVTec Anomaly Detection dataset).
• Direct Siemens S7 PLC Integration: Direct Put/Get S7 protocol memory access (DB_AI_Communication - DB1) via python-snap7 with automatic reconnection and state-change writing.
• Safety Interlock & Latching Logic: When a defect is detected, the PLC latches the fault (Fault_Active = TRUE) and immediately stops the conveyor/motor (%Q0.0 = 0, %QW96 = 0V). The motor will NOT automatically restart until an operator triggers a manual reset button or sends an [R] Restart command from the UI.
• Defect Archiving & Traceability: Automatically captures defect snapshots to the defects/ folder and logs timestamp, defect type, and confidence score to defects/defect_log.csv.
• Heads-Up Display (HUD / OSD): Live FPS calculation, bounding box visualization, motor status, and PLC link diagnostics.
• Experimental Module - Fabric Stitching Inspection: 5-Class multi-defect fabric stitching inspection with analog motor speed control (%QW96 0-10V / 27648 reference).

---

🏛️ System Architecture

┌──────────────────────────────────────────────────────────────┐
│  [Industrial Camera / Video Stream / RTSP]                   │
│       │                                                      │
│       ▼                                                      │
│  [PC / Edge Python Server]  ──── S7 Protocol (Snap7) ────►  [PLC]│
│   YOLOv8 AI Inference                                S7-1200 │
│   Defect Snapshot & CSV Logger                       TIA V17 │
│       │                                                  │   │
│       ▼                                                  ▼   │
│  [Real-Time OSD / HUD]                           [Tower Lamp]│
│  Defect type + confidence + motor status         OK / FAULT  │
└──────────────────────────────────────────────────────────────┘

---

📁 Project Structure

AI-Powered-Surface-Inspection-System/
├── src/
│   ├── __init__.py
│   ├── ai_inspection.py             # Main Surface Defect Inspection (YOLOv8 + Snap7)
│   ├── ai_stitching_inspection.py   # 5-Class Fabric Stitching Inspection (Bonus module)
│   ├── prepare_mvtec.py             # MVTec AD to YOLO Bounding Box dataset converter
│   ├── prepare_stitchingnet.py      # StitchingNet dataset converter
│   ├── train_yolov8.py              # YOLOv8 fine-tuning pipeline
│   └── train_stitching_yolov8.py    # Stitching model training pipeline
├── plc/
│   ├── plc_surface_logic.scl        # TIA Portal SCL logic for surface inspection (DB1 + FB1)
│   ├── plc_stitching_logic.scl      # TIA Portal SCL logic with %QW96 analog output
│   └── dikis_motor/                 # TIA Portal V17 project archive (.ap17)
├── weights/
│   ├── best.pt                      # Fine-tuned MVTec Leather Defect weights
│   └── yolov8n.pt                   # Base YOLOv8 Nano fallback model
├── datasets/
│   ├── mvtec_leather_data.yaml      # MVTec Leather YOLO configuration
│   ├── stitchingnet_data.yaml       # Fabric Stitching YOLO configuration
│   └── stitchingnet/                # Sample raw dataset images
├── docs/
│   ├── User_Manual.md               # Hardware & TIA Portal configuration manual (TR)
│   └── AI-Powered-Surface-Inspection-System.md # Technical specification & proposal
├── requirements.txt                 # Python dependencies
└── README.md                        # Project documentation

---

🚀 Installation & Quick Start

1. Install Dependencies

pip install -r requirements.txt

2. Run Surface Inspection (Inference)

Mode A: With Physical Siemens S7 PLC Connected

python src/ai_inspection.py --ip 192.168.1.20 --conf 0.25

Mode B: Simulation / Test Mode (Without Physical PLC)

python src/ai_inspection.py --mock-plc --conf 0.25

CLI Parameters:

Argument	Default	Description
--source	0	Camera device index (e.g. 0, 1) or video file path
--weights	Auto-detect	Path to custom trained .pt model weights
--conf	0.25	Confidence threshold for defect detection
--ip	192.168.1.20	Siemens PLC IP address
--rack	0	Siemens PLC Rack number
--slot	1	Siemens PLC Slot number (1 for S7-1200/1500)
--db	1	Data Block number in TIA Portal (DB1)
--mock-plc	False	Run in simulation mode without PLC connection
--save-defects	True	Automatically save defect snapshots & CSV log

---

⌨️ Live Keyboard Controls

Key	Action	Description
[Q]	Quit	Exits the inspection program safely and closes connections.
[R]	Restart Motor	Sends restart command (Restart_Cmd = 1) to PLC to clear fault lock and resume line.
[S]	Simulate Defect	Triggers a simulated defect for quick functional and PLC interlock testing without physical defects.

---

🧠 Dataset Preparation & Model Training

1. Prepare MVTec AD Dataset

1. Download any category (e.g. leather, wood, metal_nut) from the MVTec AD Dataset.
2. Extract the archive into the mvtec_anomaly_detection/ folder (e.g. mvtec_anomaly_detection/leather/train and test).
3. Convert masks to normalized YOLO bounding boxes:

python src/prepare_mvtec.py --category leather

2. Train YOLOv8

python src/train_yolov8.py --data datasets/mvtec_leather_data.yaml --epochs 50 --batch 16

Trained weights are automatically saved to runs/detect/mvtec_surface_inspection/weights/best.pt.

---

📊 Training Results & Performance

• Model: YOLOv8 Nano (yolov8n.pt)
• Dataset: MVTec AD (leather category)
• mAP50: 79.7%
• mAP50-95: 47.4%

Performance Curves

---

⚙️ Siemens PLC Memory Mapping (DB1)

The Python Snap7 client directly writes to and reads from DB_AI_Communication (DB1):

Offset	Variable	Type	Description
0.0	AI_Result_Code	Int	0: Idle, 1: OK (Normal), 2: NOK (Defect Found)
2.0	Defect_Type	Int	0: None, 1: Surface Defect (or 1..5 for Stitching)
4.0	Restart_Cmd	Int	0: Idle, 1: Pulse to clear fault latch

⚠️ Important TIA Portal Settings:

1. Enable "Permit access with PUT/GET communication from remote partner" in CPU Properties.
2. Disable "Optimized block access" on DB_AI_Communication (must be non-optimized standard DB).
3. See docs/User_Manual.md for full step-by-step setup instructions.

---

🧵 Experimental Module: Fabric Stitching Defect Inspection

For textile defect inspection, an experimental 5-class stitching quality module is included:

# Run Stitching Inspection
python src/ai_stitching_inspection.py --ip 192.168.1.20

# Run in Mock Simulation Mode
python src/ai_stitching_inspection.py --mock-plc

Stitching Defect Classes:

• 1: Atlamalı Dikiş (Skipped Stitch)
• 2: Kopuk Dikiş (Broken Stitch)
• 3: Eğri Dikiş (Crooked Seam)
• 4: Gevşek İplik (Thread Sagging)
• 5: İğne Hasarı (Needle Damage)

PLC logic for stitching control includes analog speed reference regulation (%QW96 0-10V / 27648) in plc/plc_stitching_logic.scl.

---

📄 License & Credits

Developed for industrial automated vision inspection. Designed with modularity and Siemens automation compliance.