How IoT, Edge Computing, AI, and Blockchain Fit Together in a Smart Factory Architecture

Learning Goal

Explain the five‑step process “perception, transmission, near‑processing, centralized training, trusted sharing” and decide where IoT, edge computing, AI, and blockchain belong in an architecture based on business constraints.

Scenario

A manufacturing line equips temperature, vibration sensors and industrial cameras. When an anomaly occurs, an alarm and shutdown must happen within 100 ms; the plant must keep operating locally if the network fails; the cloud aggregates data from multiple plants to train defect‑detection models; the equipment vendor, the plant, and an insurance company must jointly verify maintenance records without exposing raw video or commercial data.

Full Solution Overview

Sensor / Camera
  ↓ Data collection
Edge gateway / Edge server
  Protocol conversion, filtering, AI inference, local control
  ↓ Upload high‑value data & results
Cloud IoT platform
  Device management, data storage, model training, global analysis
  ↓ Production monitoring, predictive maintenance, quality analysis
Factory / Vendor / Insurer
  ↓ Consortium blockchain
  Record maintenance events, responsibility status, file digests

1. IoT Three‑Layer Architecture

The IoT stack consists of:

Perception layer : sensors, cameras, RFID, QR codes, actuators that identify objects and collect raw data.

Network layer : transports and processes data (industrial bus, Ethernet, Wi‑Fi, 5G, cloud platforms) and performs protocol conversion and device management.

Application layer : industry‑specific services such as monitoring, prediction, control, logistics, smart‑home, remote metering.

Example: temperature sensors collect motor temperature, vibration sensors capture waveforms, cameras capture product surfaces, and a controller receives a stop command.

2. Edge Computing

Edge computing moves compute, storage, and application capabilities close to the data source. It is not about increasing physical bandwidth; it reduces back‑haul traffic, latency, and privacy risk.

Key characteristics : massive device connectivity, first‑point data handling (filter, cleanse, aggregate, anonymise, quick analysis), resource constraints (CPU, memory, power, space, harsh environment), and distributed deployment (registration, remote upgrade, distributed scheduling, monitoring, security).

Four main values : low latency, bandwidth savings, privacy protection, local autonomy.

Typical edge nodes: smart gateways, base‑station servers, on‑site factory servers, store‑edge devices, vehicle or terminal devices.

3. AI – Training vs. Inference

AI hierarchy: AI → Machine Learning → Deep Learning. Weak AI solves narrow tasks (defect detection, speech recognition); strong AI aims at general reasoning and is still research‑level.

Learning types:

Supervised – labelled samples (e.g., “defect / good” images).

Unsupervised – no labels, discover structure.

Semi‑supervised – few labels + many unlabelled.

Reinforcement – learn actions from rewards.

Training pipeline:

Historical data & labels
  ↓ Feature processing & training algorithm
  ↓ Heavy CPU/GPU/NPU consumption
  ↓ Deployable model

Inference pipeline:

New image / sensor data
  ↓ Trained model
  ↓ Defect class, failure probability, control suggestion

Why “cloud training, edge inference”?

Cloud provides abundant compute for model training on aggregated multi‑plant data.

Edge provides real‑time, low‑latency inference, works offline, keeps raw data local.

Costs: limited edge resources, model version drift, need for monitoring latency, accuracy, and drift.

4. Consortium Blockchain

When multiple independent organisations need a shared, tamper‑evident ledger without a single trusted centre, a consortium (permissioned) blockchain is appropriate.

Public chain – anyone can join, maintained by public nodes.
Private chain – single organisation controls.
Consortium chain – authorised organisations jointly maintain.

In the smart‑factory case, the plant, equipment vendor, and insurer each have known identities; they share maintenance records, signatures, and responsibility status while keeping raw videos and commercial documents off‑chain.

Key mechanisms:

Chain data structure : each block contains a hash of the previous block, making tampering evident.

P2P network : multiple nodes store/validate the ledger.

Consensus : nodes agree on transaction order and ledger state (does not guarantee data truthfulness).

Smart contracts : deterministic programs that automatically execute agreed‑upon rules (e.g., update warranty status after joint confirmation).

Six‑layer model (data, network, consensus, incentive, contract, application) is a reference; a consortium chain may omit the incentive layer.

Chain‑off‑chain design:

Off‑chain: raw video, full maintenance report (object storage / data lake).
On‑chain: file hash, timestamp, device ID, status, signatures.
Verification: recompute hash off‑chain and compare with on‑chain digest.

Limitations: high storage cost, lower throughput, cannot guarantee source data authenticity, and regulatory compliance for deletion.

5. Integrated Architecture for the Smart Factory

Temp / Vibration sensors / Camera
  ↓ Edge gateway & server
  Protocol conversion, filtering, cache, AI inference, local control
  ↓ (a) Immediate stop on anomaly (≤100 ms)
  ↓ (b) High‑value data & results
Cloud IoT & AI platform
  Device management, data lake, model training, global analysis
  ↓ Model version distribution & monitoring
Factory ─┐
Vendor  ├─► Consortium blockchain: maintenance event, hash, signature, responsibility
Insurer ─┘
Off‑chain storage: raw video, full report
On‑chain: hash, index, timestamps, signatures

Key design points:

IoT three layers map sensors to perception, gateways to network, cloud services to application.

Edge handles sub‑100 ms control, local autonomy, privacy‑preserving filtering.

Cloud performs global data aggregation, model training, version management.

Consortium blockchain ensures multi‑party auditability without exposing confidential files.

6. Common Pitfalls and Correct Reasoning

Wrong: “Add IoT, edge, AI, blockchain just because they are trendy.” Correct: Identify the four business problems (data acquisition, low‑latency control, offline operation, multi‑party record verification) and select the mechanism that directly solves each, then discuss benefits and trade‑offs.

7. Quick Reference Checklist

IoT layers: perception → network → application.

Edge not a “fourth layer”; it is a deployment location.

AI training vs. inference are separate stages.

Consortium chain solves trust among known parties; consensus aligns state, smart contracts enforce rules.

Store large raw files off‑chain; only hashes on‑chain.