How AI is Helping Streamline Defect Detection in Sheet Metal Drawer Box Production

Welding

Converting sheet metal to other products remains a process prone to defects that have to be evaluated and addressed carefully. Small failures in the production process can add up fast, degrading the overall quality of the final product by a significant margin.

Various solutions have been employed to this end, but through the last couple of decades, we’ve seen a strong focus on experimenting with AI-based approaches due to the increased reliability they provide.

Many production facilities are now rapidly integrating AI-driven solutions for defect inspection and have been seeing significant improvements in their performance and product quality as a result.

Sheet Metal Products Are Still Prone to Defects That Can Be Difficult to Identify

Sheet metal products still present various issues on the defect front, some of which remain difficult to tackle. Defects like scratches, bumps, and various imperfections caused by contamination can be identified easily, even by a human operator without any extra equipment. However, production facilities have been facing significant problems with other types of defects, particularly orange peel defects.

Those remain a challenging issue for human operators, and one that’s still prone to high failure rates. While the impact on the final product’s quality is not as worrisome as with certain other kinds of defects, it’s still something that needs to be addressed as extensively as possible.

Some companies have been investing significant resources in addressing this particular issue in their production facilities, exploring various new approaches to the situation. Automated Optical Inspection (AOI) is evolving fast, and a large part of that is due to rapid advances in AI technology.

While the situation has improved a lot over the last couple of decades, the industry is still faced with some challenging problems that will likely take a while to resolve. This makes it important to stay at the forefront of these new developments, constantly exploring new trends on the market and working with experienced specialists.

This is especially true when working with newer solutions like artificial intelligence, which often require a lot of fundamental knowledge in order to harness their full potential.

Robot

Rising Adoption Rates of AI-based Technology

Traditional methods involving having employees manually inspecting the finished product (as well as inspecting it at different stages of the production process) were able to serve the industry’s needs in the beginning, and their performance was improved by the adoption of solutions like live camera feeds from multiple angles, allowing for rapid inspection of a single piece.

However, those methods are still far from perfect, especially when production speeds are of crucial importance and the overall chain has to be optimized as much as possible.

With the rise of AI-driven image recognition solutions and their rapid integration into various technological fields, it made perfect sense to eventually see that utilized for defect detection as well.

Much of the groundwork was already there to begin with – most of the tech needed to implement an automated defect detection solution revolves around cameras capable of rapid and accurate imaging.

The only major transition was shifting away from having human operators inspect the feeds, to running them through solutions that are either partially, or even completely, automated.

An important factor in the fast adoption of these solutions was the growing list of defects that have to be accounted for in the production of sheet metal products.

Recent developments in the field have shifted the importance of certain types of defects, putting more weight on ones that have proven more problematic in the last decade. And it turns out that AI is the perfect tool for tackling the issues presented by those particular defects.

Types of Defects in Sheet Metal Production

Sheet metal production, and the manufacturing of related products, usually deals with several major types of defects. Some of the are easy to identify at a glance, while others require a more in-depth inspection, and in some cases are impossible to locate without advanced equipment.

  • Splits, or potential splits
  • Incorrectly shaped and/or sized sheets
  • Wrinkles in the sheet material
  • Irregular thickness
  • Scratches
  • Bumps
  • Contamination (dust and other particles)
  • Orange peel

Orange peel remains one of the most problematic on this list. It’s difficult to avoid it completely during the fabrication process, and it mostly boils down to identifying reasonable boundaries for the manipulation of the material.

Determining the maximum grain size that will not cause any visible cosmetic defects can be costly without the appropriate equipment and expertise. It can also vary between production processes, further amplifying the importance of deploying a universal solution that can evaluate the problem in various different contexts.

Different production processes have different tolerance levels for this, which has been adding to the difficulty of finding a more unified solution that’s suitable for the industry as a whole, as opposed to for specific individual processes.

Why AI Is the Obvious Answer

That’s where AI comes in. A large part of the work in identifying orange peel defects and filtering them out according to their severity lies in statistical analysis and the ability to accurately predict how much the product would be affected by a certain increase of the grain size.

There was already a lot of work in the field being conducted manually, experimenting with different approaches for the manufacturing of different types of products. This information proved useful when AI-based solutions were deployed.

AI-driven solutions thrive in situations where patterns have to be extracted out of large, complex data sets – which is exactly what experts are dealing with when it comes to orange peel defects.

A lot of information about the formation of those defects was already available for different contexts and different types of products, but it was hard to consolidate it all and identify patterns that could be exploited in modern production processes.

Even AI wasn’t originally perfectly suited for the job though. A 2017 approach by Y. Zhao, Y. Yan and K. Song utilized a Simple Linear Iterative Clustering (SLIC) algorithm for the detection of defects, but the technique had various shortcomings that prevented its large-scale deployment into actual production contexts.

It did, however, inspire further research into the field, with various specialists leveraging the discoveries made in the process to improve their own solutions.

In particular, deep learning has proven a very useful subset of the field with lots of potentially promising results. Since identifying defects of this type relies on a lot of prior training of the model, deep learning-based vision solutions have emerged as the leading contender in the current market, and have been showing strong promise so far.

Training data was already available in significant amounts due to the original research conducted in the field, although researchers have continued to amass even more information on the subject afterwards.

Illustration of people pushing blocks.

How Sheet Metal Defect Inspection Can Be Streamlined Further

As with most problems addressed by deep learning-based solutions, the main obstacle in the way of defect detection right now is the availability of sufficient training data.

A lot of work has been done on this front, including leveraging older data sets from before AI was even being actively considered as a solution in that field.

Models are already evolving quite fast though, with many promising features on the horizon in the coming years, especially when it comes to the detection of trickier defects like orange peel.

At the same time, the importance of proper training of human operators remains high. The transitional period between the old, completely manual approach, and the new, automated one, will require a lot of input from qualified professionals who must contribute to the growth of the field.

With that in mind, companies must continue to work with qualified experts on their production lines, and ensure that defects are being evaluated from as many different perspectives as possible on a regular basis.

In addition, all data produced from this research should be recorded and sanitized at the source whenever possible, as this can help streamline the training of related models.

Why Simerse Is Your Best Bet for Leveraging These Solutions

Simerse has been a major player on the market for AI-driven solutions for a long time now, and the company has gathered significant experience in the field.

They are among the leading experts in the field at the moment, constantly pushing the boundaries of defect detection technology and establishing new trends in the sector.

Those looking for a reliable starting point for integrating these types of solutions in their own production facilities should look into Simerse as a reliable starting point.

There are lots of challenges in this sector, and the situation is far from stable at the moment, with lots of new solutions coming out on a regular basis, and companies constantly experimenting with different approaches to classic problems.

This makes it important to know that you’re working with someone with an established track record, which is where a partner like Simerse comes into play. Get in touch with Simerse today and see what the company can do for you.