---
title: Monitait Step-by-Step User Guide
subtitle: Hardware (WatcherJET 3.0) & AI Training Platform
author: Monitait.com
date: '2025-10-13'
format:
  html:
    toc: true
    toc-depth: 4
    toc-location: left
    theme: cosmo
    code-fold: true
    grid:
      sidebar-width: 300px
      body-width: 900px
      margin-width: 300px
  pdf:
    toc: true
    keep-tex: true
    documentclass: scrreprt
    classoption:
    - paper=a4
    - twoside=false
sdk: docker
emoji: 🌍
colorFrom: purple
colorTo: indigo
pinned: true
short_description: Docs
---

# Monitait Step-by-Step User Guide
**WatcherJET 3.0 Hardware + AI Training Platform**  
**Version:** 1.0 – **Date:** October 13, 2025  
**Powered by Monitait** – [monitait.com](https://monitait.com)

---

## Download Full Illustrated Manuals (PDF)

| Document                                    | Pages | Link                                                                                  |
|---------------------------------------------|-------|---------------------------------------------------------------------------------------|
| Chapter 1 – Hardware & WatcherJET 3.0       | 27    | [CHAP-1-AI-HARDWARE.pdf](CHAP-1-AI-HARDWARE.pdf)                                      |
| Chapter 2 – AI Training                     | 15    | [CHAP-2-AI-TRAINING.pdf](CHAP-2-AI-TRAINING.pdf)                                      |

All wiring diagrams, screenshots, and detailed illustrations are in these PDFs.

---

# Chapter 1 – Hardware & WatcherJET 3.0

### Quick Setup (4 Steps)
1. **Connect the sensors**  
2. **Connect the power supply** (12–24 V DC)  
3. **Provide an access point (AP)**  
4. **Register at** [console.monitait.com/factory/watchers](https://console.monitait.com/factory/watchers)

### Step 1: Connect the Sensors

#### External Machine Signal
- Production Count → Connect any 12–24 V signal to **OK inputs (3 & 4)**  
- Defect Count → Connect ejector/NG signal to **NG inputs (5 & 6)**  
  → Bidirectional, opto-isolated

#### Push Button
- Production: Button → OK (4) + GND (1); bridge OK (3) → +V (2)  
- Defects: Same wiring using NG (5 & 6)

#### Obstacle Sensor
- Black wire → OK or NG input  
- Brown → +V (2)  
- Blue → GND (1)  
- Bridge remaining OK/NG → +V (2)

#### Encoder
White → NG (6) | Black → OK (4) | Brown → +V | Blue → GND  
Bridge (3) & (5) → +V (2)

#### RS485 Protocol
A → terminal 8 | B → terminal 7

### Step 2: Power Supply
- 12–24 V DC, max 2 A  
- Connect to terminals (1) & (2) → green power LED turns on

### Step 3: Network Connection
- **Best:** Wired LAN (add firewall rule `*.monitait.com`)  
- **Temporary:** Mobile hotspot → SSID: `Monitait`, Password: `p@ssword`

### Step 4: Register the Watcher
1. Go to console.monitait.com → Watchers → Add Watcher  
2. Enter Registration ID (shown on phone when connected to hotspot)  
3. Set station, multiplication factor, timeout, etc.  
4. Test: Trigger sensor → red heart LED must blink

### High-Current Power, Emitters & Actuators
- High-current PSU ≤ 48 V → terminals (9 & 10)  
- Emitters: Positive → PSU, Negative → U (12) or B (11) depending on alignment  
- Ejector output (15), Warning output (16) → connect to PLC (opto-isolated NPN)

### Keys & Indicators
- Key-1: Buzzer on/off  
- Key-3: Enable camera  
- Key-4: Enable scanner  
- Green power, red heart (data), checkmark (OK), rejection (NG), thunder (PPS)

---

# Chapter 2 – AI Training

### Table of Contents
- Step 1: AI Training Platform  
- Step 2: Administration (Tasks & Images)  
- Step 3: Evaluation  
- Step 4: Deploy Weights

### Step 1: AI Training Platform – Annotation

#### Basic Workflow
1. Confirm correct task  
2. Choose category  
3. Draw bounding box (click → drag → click)  
4. Save → next image

#### Keyboard Shortcuts (fast annotation)
`1 2 3 4 Q W E R T Y U I` → select categories instantly

#### Tips
- Overlapping objects → draw elsewhere then drag, or hide with eye icon  
- No images visible → all annotated; use < > buttons  
- Add metadata (optional) via “+” in right panel

### Step 2: Administration – Create Task

1. Tasks → **+ NEW TASK**  
   - Enter name (e.g., “Bottle Defect Detection”)  
   - Set quantity, type (object detection, classification, etc.)  
   - Add labels/categories with colors  
2. After creation → ⋮ → **Upload Images** (.jpg, .png, .jpeg)

**Best Practices**
- Use visually distinct categories  
- Always label the main object (e.g., “bottle”)  
- Use descriptive IDs (bottle-pk-100, pen-dp-105)  
- Monitait team performs training and selects best augmentations

### Step 3: Evaluation

#### Key Concepts
- **TP** = correct detection  
- **FP** = false alarm  
- **FN** = missed defect  
- **TN** = correctly identified good item

| Metric    | Formula                     | Meaning                              |
|-----------|-----------------------------|--------------------------------------|
| Precision | TP / (TP + FP)              | Accuracy of positive predictions     |
| Recall    | TP / (TP + FN)              | How many real defects were found     |
| F1-Score  | 2 × (P × R) / (P + R)       | Balance between precision & recall   |
| mAP       | Mean Average Precision      | Overall model quality (higher = better) |

#### How to Improve Accuracy & Recall
- Diverse dataset (lighting, angles, backgrounds)  
- Tight, consistent bounding boxes  
- Multiple reviewers  
- Avoid similar/confusing categories  
- Consult Monitait team for optimal augmentation and training settings

### Step 4: Deploy Trained Model

Every successful training produces:
- `best.pt` (final weights)  
- Unique **Training ID** → record it!

**Deployment on production machine:**
```bash
cp best.pt /home/projects/inference/best.pt
# Restart inference service → new model is live