For over a decade, manufacturers have turned to automated solutions to improve their bottom line. Automation and machine vision are now being augmented and even replaced by AI. Here is the value of AI-based visual inspection in 2020.
Being replaced by AI is especially true when it comes to visual inspection. The use of AI-based visual inspection technology is transforming manufacturing’s ability to improve business operations.
AI-based visual inspection relies on two of AI’s main strengths: computer vision and deep learning. Every AI system is built with the core capacity to perceive its environment (computer vision) and act on those perceptions (deep-learning).
As a result of deep-learning, AI adapts to a range of environments, making it useful across a multitude of industries. It has unlimited potential and can be developed rapidly to meet a manufacturer’s needs.
Well-trained human eyes can detect defects. A well-trained AI-based vision system can do the same — but with greater efficiency. Like a human eye, AI-based vision systems capture an image and send it to a central “brain” for processing.
AI-based vision systems are made of two integrated components. A sensing device acts as an “eye,” while a deep learning algorithm acts as a “brain.” The integrated system successfully mimics the human eye-brain ability to interpret images.
AI-based vision systems are more efficient than human eyes because the AI “brain” stores greater amounts of information.
Robust computational power can parse through available data at rapid speeds. The system can classify objects in both photos and videos and perform complex visual perception tasks.
AI-based vision systems can search images and captions, detect objects, and classify multi-media.
Thanks to deep learning-based visual processing, AI-based visual inspection systems can perceive cosmetic flaws and detect defects across general or conceptual surfaces (mobidev dot biz).
Decades-old automated systems depend on defect libraries, lists of exceptions and complicated filters. The time it takes to accrue this information, clean it for accuracy, and re-implement it decreases its efficacy. It also wastes labor.
AI and deep learning do not require prolonged programming or tediously lengthy algorithms. AI-based visual inspection systems might be constructed by several quality engineers and a dataset of training images. The system learns rapidly and is integrated over several weeks.
Manufacturers can use AI to document inspection results and to assess product quality. Some overall process improvement initiative metrics that can be successfully tracked and correlated with concrete vision data include:
In addition, inspection images and results can also be tracked and documented. These initiatives prevent future failure, which saves time and additional production costs. Applying deep learning-based machine vision across all initiatives and inspections helps manufacturers recognize and address defects early.
AI solutions have higher rates of consistency than most expert human inspectors. Human inspectors must be trained and are only able to maintain a high degree of focus for 15-20 minutes at a time. Labor costs are incurred yearly and staff turn-over is an issue. For these reasons, AI-based vision inspections are more cost-effective than manual labor.
AI is increasing the competitiveness of manufacturers across every industry. Here are recent use cases from the aviation industry, semi-conductor manufacturing sector, and bio-science.
Alibaba has risen to meet healthcare challenges created by the coronavirus. Alibaba’s deep-learning-based visual recognition system is capable of detecting the coronavirus in chest CT scans at a 96% accuracy rate. The system accessed 5,000 COVID-19 cases and can provide a diagnosis within 20 seconds. Moreover, the system can differentiate between images of viral pneumonia and images of coronavirus.
Fujitsu Laboratories implemented an Image Recognition System at Fujitsu’s Oyama factory. The system ensures that parts are produced at optimal quality levels by supervising the assembly process. The system was so successful that Fujitsu implemented it across the entirety of the company’s production sites.
Airbus introduced an automated, drone-based aircraft inspection system in 2018. The system has improved the quality of inspections and reduced aircraft downtime.
GlobalFoundries is a leader in semiconductor manufacturing. The company designed a visual inspection system that detects defects in a scanning electron microscope (SEM) images. The system detects defects in a wafer map which then helps to determine the semiconductor device’s performance.
The use cases listed above reveal the extent to which AI is capable of automating many aspects of our lives. Although AI vision will never replicate human vision, the technology continues to classify information and advance in ways human eyes and brains cannot. And only humans might consider how to use this technology to get advantages.
I am a Data Science engineer at MobiDev (USA/Ukraine). My career started with iOS applications development but growing interest in various fields of AI. I’ve been working on different Data Science projects, from time series forecasting to face recognition. I am a speaker at the Machine Learning conference and author of articles in online Data Science communities.