AI Vision in Smart Manufacturing: Principles and Applications

6 November 2024

Introduction

In the era of smart manufacturing, many companies aim to enhance their competitiveness through AI technology.

We believe that AI solutions are worth adopting only if they can improve operations and increase market performance.

Market reports highlight a key point: companies have strong confidence in AI vision, and ongoing investments are driving its market growth.

The growth of AI vision comes from its success in real applications.

This article covers the fundamentals and technology of AI vision, its common uses, the advantages it brings to manufacturing operations, and why it’s essential for the growth of smart factories.

AI Vision allows computers to “See, Understand, and Interpret” image data

To make computers mimic human recognition, we need large amounts of labeled image data.

To train AI, we use labeled images.

For instance, tagging one image as “cat” and another as “dog.” With enough labeled examples, the AI’s accuracy gradually improves. With enough training, AI can accurately identify new images as either a cat or a dog, just like a human.

On a production line, AI vision helps ensure a screwdriver is placed in the correct position. By training the AI model with many images of proper screwdriver placements, it learns what a correct setup looks like. After enough training, AI vision can quickly detect if the screwdriver is out of place and identify the errors.

Factory management used to rely on manual checks and video replays to find errors. Staff watched the production process to ensure no mistakes were made. Yet, mistakes could still be missed. When something went wrong, managers had to watch very second of the video to spot the issues and analyze them from experience. This process was time-consuming and relied heavily on individual judgment, making it easy to miss potential problems.

Now, with cameras installed on production lines, AI vision can “see” the process in real time. A well-trained AI model, like a manager, has enough process knowledge to check if workers are using the correct parts, placing them accurately, and following SOPs. When errors occur, it can quickly interpret the root cause based on the “what it see”.

AI vision saves time and effort for managers by efficiently extracting essential production insights from images.

Core of AI Vision Technology: Deep Learning

Most learning algorithms improve with more data, but they often hit a limit where additional data no longer boosts accuracy.

Deep learning overcomes this barrier.

Inspired by human brain, deep learning uses neural networks with multiple “layers” to process complex data. Each layer extracts features from the previous layer, passing them along to the next.

This “layered approach” allows deep learning to handle complex and large-scale data more effectively.

AI Vision in Manufacturing

AI vision has broad applications in manufacturing. Here are a few key areas:

• Safety Monitoring:
  AI vision checks if workers are wearing required protective gear, like helmets or gloves. It alerts if the worker violate the safety guidelines. Also, It sets up virtual boundaries to protect workers from hazards, such as robotic arm movements, in human-machine collaboration environment.
• Defect Detection:
  Traditional AOI often struggles with a high false rejection rate. With AI vision, models become more flexible in image recognition. It learns product features and gaining a deeper understanding of images. This helps prevent false positives caused by lighting, angles, or other non-essential factors.
• Human behavior management:
  Human behavior on production lines is harder to track compared to machines. For instance, some workers may leave their stations early before breaking time, causing the producing process slowdown. With AI vision, complex human activities can be effectively tracked and analyzed.

Focusing on “Men” and “Method” in Manufacturing: Advanced Applications of AI Vision

In manufacturing, the 4M1E framework—Men, Machine, Material, Method, and Environment—highlights the five core elements of in-process management. AI vision can effectively reduce errors specifically related to “Men” and “Method.”

Men

󠀠On labor-intensive production lines, AI vision creates an efficient management system with its “image-based” features. Human behavior is unpredictable due to reasons like attention, physical condition, or experience. All these factors impact product line efficiency and product quality. Traditionally, engineers spend many time and efforts to collect and analyze the data from the production line to reduce these potential errors.

With AI vision, key producing behavior on the line can be traced continuously. It reduces the workload on management and provides a complete production record to support decision-making.

Read More: Why root causes analysis is hard?

Method

To optimize production efficiency, industrial engineers need a complete view of the line, where data plays a crucial role in quickly identifying the root of production issues.

However, most production data collection still relies on manual recording. Engineers use stopwatches to time tasks and record observations on paper. This method is inefficient and prone to bias.

AI vision enables continuous data collection and instant root-cause analysis. It helps engineers quickly identify bottlenecks, eliminate management blind spots, and accelerate process optimization.

Core Advantages of AI Vision in Applications: Irreplaceability

24/7 Production Process Tracking

Traditionally, production tracking relied on operators observing the line to find causes of low quality or inefficiency. Often, quality control would detect defective products first, then trace the issue back to the process. This approach is inefficient and can also miss critical production issues.

AI vision provides a 24/7 tracking and root cause analysis. It offers real-time insights for both in-process management and quality control.

In one case, a factory producing high-value servers faced frequent issues with scratches or dents on products. Managers couldn’t determine the cause of these defects.

AI vision’s image tracking revealed that workers were accidentally dropping screwdrivers, leading to scratches on the products. Only “images” can capture and resolve such human errors. Current systems lack the ability to track this crucial information.

Analysis of Collaborative Movements

Most factories today use IoT systems to support quality control. However, these systems can’t capture all data, especially at critical stations where only video can fully track human behavior.

In one case on an electric scooter production line, two operators shared a workstation to finish the tasks together. Even with MES, the system could only track individual steps, not the entire collaborative process. When the abnormality was detected, production engineers struggled to identify the root causes.

AI vision became the only viable solution.

It records and analyzes the collaborative work of both operators at the workstation, tracking if they follow SOPs accurately. When errors occur, managers can easily see the root causes analysis showed on the dashboard—whether it’s due to tools, materials, or process flow. This solution cannot be found in any other management tools or systems.

AI vision is also effective at tracking workflows that involve human-machine collaboration, helping to ensure smooth operations.

Success Cases in Smart Factories

Case 1: Digital Workstation (Server Manufacturing)

Challenge: Workers often scratch the metal surface of servers when using tools like screwdrivers, affecting product appearance. Without a good tracking system, management can’t easily identify when or where issues occur. This makes problem-solving harder and risks delays.

Solution with AI Vision: AI vision continuously records the production process. It allows managers to trace issues like scratches or dents quickly, cutting down the time needed for root cause analysis.

Case 2: Line Balancing (Electronics Component Manufacturing)

Challenge: A time study for production lines requires significant time and effort and often results in incomplete data, with “non-value-added tasks” frequently omitted.

For instance, the plant noticed low production efficiency at specific times and operators leaving stations early but lacked long-term data to fully evaluate line efficiency and resource allocation.

Solution with AI Vision: AI vision automatically collects and analyzes production data 24/7, capturing cycle time, process time, and idle time with precision. Every movement or incident on the line is recorded. Based on the reports, engineers can quickly identify and fix the issues. Finally, the production line has a 5.2% improvement in UPH and over 5x ROI.

UPH UP! Top five electronics manufacturer optimizes production with AI vision󠀠

Case 3: Real-Time Error Proofing (Electric Vehicle Assembly)

Challenge: Early warranty issues persist due to operators occasionally skipping SOP steps. The MES system monitors steps individually but can’t analyze the full collaborative process, making it hard for engineers to quickly identify if issues stem from tools, materials, or procedures.

Solution with AI Vision: AI vision combined with MES and AGVs creates a process error-proofing system. If AI vision doesn’t detect key operator actions, or the MES lacks process data, the AGV won’t release items from the workstation. This approach prevents quality issues at the source, achieving zero early-stage warranty claims.

Achieving Smart Factory Goals with PowerArena HOP

In the fast-changing tech landscape, automation, digitalization, and AI are crucial drivers of industry growth. Companies are eager to gain an edge in this wave of innovation.

PowerArena HOP uses AI vision to optimize in-process quality control, reduce management workload, and support smart manufacturing.

HOP offers three core advantages to accelerate your digital transformation:

1. Transparent Production Line with Real-Time Detection

AI vision monitors production line abnormalities, quickly identifying potential failure causes.

Reduce error response time and lowers production risks.

2. Accurate Data Support, Minimizing Errors

AI vision offers 24/7 objective production records through image capture.

Combine with unbiased data analysis for reliable strategies.

3. Pre-Trained Models for Rapid Deployment

HOP’s models recognize basic elements like hands, personnel, and PCBs, eliminating lengthy training periods.

A four-week deployment allows for early intervention in production issues.

 

Don’t miss out! Book a personalized consultation and see how AI vision can elevate your smart manufacturing game!

Talk to our Consultant

󠀠