Image in modal.

Manufacturers are focusing on low-code artificial intelligence (AI) tools to simplify application development, allowing non-experts to create and customize AI workflows.

Low-code tools for AI application development are an area of focus for manufacturers, said Ed Goffin, vice president, product marketing, Pleora Technologies.

“As manufacturers get more comfortable with the concept of adding AI to an existing or new process, their attention is turning more to deployment and longer-term cost of ownership,” Goffin said. “One of the key trends we’ve seen, and been a part of, is simplifying development so a quality manager or operator can develop, train, and deploy their own AI application or workflow.”

This typically begins with a template for common requirements, such as object detection, which can be easily customized for the end-user, Goffin explained. This offers a significant cost of ownership benefit for the manufacturer, as they no longer need to contract or hire external expertise to deploy AI in their processes.

“The AI model can also learn from operator input, so instead of programming an algorithm you can train it based on real-time input from an operator,” Goffin said. “In a visual inspection step an operator first accepts or rejects product differences through image compare, while behind the scenes their decisions and associated product images are being used to train an AI model.”

Within a few inspections, the AI model will begin suggesting a decision for the operator, Goffin said. This is especially beneficial for training a system where there can be some subjective decisions around “pass” and “fail.”

Experts also see large language models, machine learning, video analytics and AI-integrated robotics advancing quality control in the future.

“Specifically, deep learning and neural networks are key technologies we’re leveraging to improve our AI-driven inspection systems’ ability to learn from vast datasets,” said Marius Kvedaravičius, CEO, EasyODM.

These tools will help operators to detect errors more efficiently during production, with machine learning algorithms improving over time, said Thomas Su, vice president, sales, North America, Vecow. “Additionally, the collaboration of AI and IoT devices allows for real-time monitoring and predictive maintenance, foreseeing equipment failures to prevent costly downtime and streamline production processes,” he said.

With this in mind, how can organizations prepare to best manage AI-driven quality management?

“My advice is to take a staged approach to AI deployment,” Goffin said. “We often are involved in projects where the goal becomes full scale automation. These are often the projects that derail; they become too expensive, too complex, and consume too many internal resources. Where we see more successful deployments is a staged approach that often starts with automating an error-prone manual process, such as visual inspection or assembly. These are easier problems to solve, have an immediate ROI, and maybe more important a manufacturer can gather the expertise and data to help guide a migration towards full scale automation. If you add AI-based decision support to visual inspection, for example, you collect the good and bad images required to automate that process as a next stage.”

Organizations should actively explore current AI technologies and engage with experts through workshops and consultations, Kvedaravičius said. For example, EasyODM hosts workshops with C-level executives to evaluate and identify AI solutions that can enhance their existing processes.

To lead in AI-driven quality management, organizations should recognize its benefits and invest in skilled personnel and advanced computer hardware, Su said.

AI decision-support tools for repetitive tasks, such as visual inspection during electronics assembly.
AI decision-support tools for repetitive tasks, such as visual inspection during electronics assembly, helps make manual processes, consistent, reliable and traceable. Image courtesy of Pleora Technologies

Maintaining Data Security

Manufacturers’ data often contains company secrets and product details. If leaked, it could cause losses, legal trouble, and harm customer trust. Companies who use automation tools are putting measures in place to ensure the security and privacy of sensitive, quality-related data.

For example, manufacturers prioritize data security and privacy by investing in advanced computer hardware and establishing private cloud infrastructure, said Su. This helps them to reduce cybersecurity risks, and they maintain control over their data storage and processing.

Companies should use custom security measures for each factory to protect sensitive data, opting for on-edge systems within the factory network rather than cloud-based storage, experts say. They should follow strict documentation, including consent and data statements, and use encryption and General Data Protection Regulation (GDPR) compliance to create a thorough security framework.

Additionally, organizations should implement solutions that do not rely solely on cloud-based AI systems.

“For most of the projects we’re involved in the information is staying on-premises,” Goffin said. “From an end-user perspective, treat AI data just as you would any other data you gather from a machine vision or automated process. In our case, we provide the software tools and templates along with the hardware to deploy AI, but we don’t have any access to the data or models.”

New Compliance Concerns

Manufacturers using AI should also stay informed about its regulatory implications, including GDPR and sector-specific guidelines. Kvedaravičius cites the upcoming EU AI Act, which is the first ever global legal framework concerning AI, as a prime example.

Artificial intelligence tools can also help companies meet existing regulatory requirements around inspection and traceability.

Traceability compliance typically incurs a good deal of manual paperwork, especially for visually inspected parts,” Goffin said. However, AI tools can lighten the load.

“By adding automated visual inspection as part of that process, an operator is also automatically gathering all the pre- and post-production product images, inspector notes, and meta data required for reporting,” he said. “That information can be saved to a MES or ERP system, or used to populate requirements for regulatory reporting.”

EasyODM AI Quality Inspection of automobile seats.
Through deep learning algorithms, EasyODM achieved exceptional precision and speed in identifying wrinkles across diverse seat materials. Images courtesy of EasyODM

Saving Costs, Enhancing Quality

When AI systems detect and address quality issues in real-time, manufacturers will save on labor costs and reduce their risk, experts say.

With significant reductions in inspection times, lowered operational costs, and improved product quality, AI systems can help production processes remain efficient and reliable, said Kvedaravičius.

Kvedaravičius cites his experience automating the defect detection process for a top automotive seat producer. Traditionally, quality inspection conducted by employees could take approximately one minute per part, but automation reduced it to less than three seconds per part.

“Adding AI inspection, whether that’s layering AI capabilities on top of an existing machine vision process or automating a manual decision or task, will reduce defects that cost time and money; that can be rework or replacement costs,” Goffin explained. It also minimizes the significant risks that arise from poor quality, especially in regulated industries where the margin for forgiveness from an end customer is very thin, he added.

Artificial intelligence tools can also offer benefits from a labor perspective, but the issue is not as black and white as it may appear.

“There’s been a consistent fear that AI will take jobs, but in our experience, manufacturers are looking at AI as a tool to help aid and remove some of the stress around decision-making for their operators,” Goffin said. “With operator-based model training, an organization can train the system based on their best inspector. That model can then be deployed across inspection stations or shifts so all operators are making the same consistent decision. That model can also be used for training new employees so they can quickly spot the difference between good and bad.”