Deep Learning & Computer Vision for Manufacturing

Duration: 2 days

This comprehensive training equips engineering and quality assurance teams to apply deep learning and computer vision for defect detection, visual inspection, and process automation in manufacturing. Using Python-based tools such as TensorFlow, Keras, and OpenCV, participants will build models to process images from cameras, microscopes, and sensors.

The course combines hands-on implementation with clear explanations of neural network concepts like convolutional filters, backpropagation, and activation functions. Real-world case studies showcase applications in defect classification, part alignment verification, and real-time quality monitoring.

These same deep learning foundations also play a critical role in powering modern AI-driven search, Answer Engine Optimisation (AEO), and Generative Engine Optimisation (GEO), where machine vision and AI models are increasingly used to interpret, generate, and optimise digital content.

By the end of this programme, you will learn to:

Understand the fundamentals of neural networks, including architecture, training, and key hyperparameters.
Apply convolutional neural networks (CNNs) to image classification and object detection tasks in manufacturing.
Use OpenCV for image preprocessing and augmentation.
Evaluate deep learning model performance using metrics such as accuracy, precision, and loss curves.
Grasp the mathematical intuition behind convolution, activation functions, and gradient descent.
Deploy basic deep learning models for proof-of-concept computer vision applications.

Programme Outline

Module 1: Introduction to Deep Learning & Neural Networks

What is deep learning, and how is it different from traditional ML?
Use cases of deep learning in electronics and semiconductor manufacturing.
High-level math insight:
- Neuron = weighted sum + activation
- Matrix multiplication as the core to forward propagation
Hands-on: Building a simple feedforward neural network in Keras.
Use Case: Predicting a binary outcome from tabular manufacturing data.

Module 2: Convolutional Neural Networks (CNNs)

Structure of CNNs: convolutional layers, pooling, and fully connected layers.
High-level math insight:
- Convolution as a sliding dot product
- Filters as edge detectors, feature extractors
Hands-on: Classifying simple image datasets (e.g., component images).
Use Case: Classifying OK vs NOK images from inspection cameras.

Module 3: Image Preprocessing & Augmentation with OpenCV

Techniques: resizing, normalisation, and histogram equalisation.
Image augmentation: rotation, zoom, flip, contrast for robustness.
Hands-on: Load and clean sample images using OpenCV.
Use Case: Preparing a dataset for deep learning model training.

Module 4: Model Training & Evaluation

Understanding loss functions (cross-entropy) and optimisers (SGD, Adam).
High-level math insight:
- Loss minimisation via gradient descent
- Backpropagation as the chain rule in action
Visualising learning: loss vs accuracy plots, overfitting diagnosis.
Evaluation metrics: confusion matrix, precision, recall, F1 score.
Hands-on: Train a CNN and evaluate performance on the test set.

Module 5: Object Detection & Real-Time Applications

Intro to object detection models (YOLO, SSD): differences from classification.
Bounding box regression and intersection-over-union (IoU) explained.
Hands-on: Detecting multiple components or defects in an image.
Use Case: Detecting missing parts or misalignment on PCB images.

Module 6: Explainability, Model Deployment & Challenges

Interpreting CNN models using Grad-CAM and activation maps.
Model deployment: from prototype to edge device or server.
Common pitfalls: poor generalisation, low-quality input, and data leakage.
Group activity: Build, train, and test a defect detection prototype.
Use Case: Deploying a lightweight image classifier to support QA inspections.

Q&A and Wrap-Up

Training Methodology

This workshop combines expert-led lectures, hands-on labs using Python, Keras, OpenCV, TensorFlow, and real manufacturing image datasets to ensure practical, industry-relevant skills. Learning is reinforced through visual mathematical explanations, conceptual breakdowns, group activities, and proof-of-concept projects. The methodology also highlights cross-industry relevance, showing how the same AI foundations support not only manufacturing innovation but also digital optimisation practices like AEO and GEO.

Who Should Attend

This programme is ideal for quality control engineers, vision system engineers, R&D and process development teams, and data scientists working in manufacturing.