Computer Vision System Eliminates 90% of Defects in Wooden Furniture Production
Prohan is a wood furniture manufacturer using hardwoods which are known to be demanding and expensive to work with.
Because of that, any flaws on the production line mean significant financial losses.
Because of that, any flaws on the production line mean significant financial losses.
Furniture Manufacturer Needed Early Detection Solution to Identify Gaps in Wood Panels
The first step of the production process with these types of wood is to glue wooden planks together to form solid panels. As humidity makes the wood likely to expand or shrink, it is impossible to use a single board of a specific size for tabletops, panels, or other vertical elements. Besides, different buyers have varying visual preferences, an example being the pattern of wood grain. Planks are therefore optimised, glued, and compressed before being cut to the desired dimensions.
Here, a specific challenge arises; there can be no gaps, cracks, and splits in the panels. The gaps are often only tenths of a millimeter wide and the wooden panels pass through the machine at speeds of up to 30 meters per second, which makes the defects undetectable by the human eye.
đź‘‹ Hey You, What Are You Working On?
Share a few details about your challenge or idea, no matter how early or complex. Our expert team will reach out to offer our expertise in terms of feasibility, budget, and scope. Let’s shape a solution for you together!
Undetected Defects Meant the Risk of Production Inefficiencies and Customer Dissatisfaction
Undetected gaps made production planning difficult for Prohan. As one small gap can ruin an entire countertop, such defects would increase the rejection rate of finished products, introducing redundancy. Out of approximately EUR 2 million generated in sales, returns only used to account for around 2%, on top of waste errors generated during the production process. When relying on manual inspections, some defects would be missed, resulting in customers returning the furniture later on. Since the solution implementation, Prohan has not had a single return triggered by this type of flaw.
Our Team has Developed a Solution That Identifies Defects at the Earliest Possible Stage of the Production Process
To avoid large financial losses in production, Prohan needed a solution that would detect the defect as soon as possible, immediately after the gluing step. They approached DAC.digital to develop and install a computer vision-powered camera that would automate the quality inspection.
In just 4 months, DAC.digital successfully developed the desired solution and deployed it at Prohan’s facility. This was possible thanks to DAC.digital’s comprehensive approach to the solution development that involved several steps that is explored in detail in the subsequent sections. The work started with an Industrial AI workshop that you can read about it here: How an AI Workshop Helped a Furniture Manufacturer Plan Quality Control Automation for Peak Season.
DAC.digital Developed a Custom Computer Vision System for Automated Quality Inspection
First, DAC.digital experts visited Prohan’s production facility to get a thorough understanding of the processes and important considerations to keep in mind throughout the solution development. During the on-site visit, the team conducted measurements of factors such as exposure levels and dust, to inform camera selection and placement. The purpose of this assessment was to ensure that the solution would operate effectively in the challenging production environment.
Then, we supported the client in preparing an adequate dataset. The client took several photos of production processes, which were then classified as “definitely no defect”, “definitely defect”, and “borderline”. These were used to train machine learning algorithms so that the camera could compare the planks it sees against the benchmarks.
Having this foundation ready, DAC.digital chose the right hardware, necessary tools, and technologies and created a prototype in our lab. Tech stack that was used included: Python, OpenCV, PyTorch, SHAP, Numpy, DVC, DVC Studio.
Following successful prototype testing, the solution was installed at Prohan’s facility. There, it underwent a smooth calibration process to align it with real-world production conditions. As the solution features cloud connectivity, any necessary configuration changes or algorithm optimisation could be done remotely by the DAC.digital team. However, according to Prohan, the solution was designed so well that no changes or additional algorithm training was needed.
The Solution Reduced the Number of Customer Returns Due to Quality Issues to Zero
Being put into practice, the solution has proven highly effective, improving the client’s quality control processes. The camera, capable of analysing images and identifying defects even in a dusty manufacturing environment with low lighting, continuously monitors the production line, operating 24/7. This has significantly increased the rate of defect detection and Prohan reports that they now receive no returns on their furniture due to defects and manufacturing faults.
Furthermore, defect detection throughout production increased by 90%. If any fault is detected by the camera, an alert light turns on and operators at earlier stages are immediately notified. Thanks to that, any defects can be rectified before proceeding. This translates to a substantial reduction in waste, as well as cost savings. Before, for every 1000 glued panels, around 5% had to be discarded. Now this rate is less that 0,5%.
Prohan’s problem could not have been solved with solutions readily available on the market as they cover the client’s unique needs only to a certain extent. Every company producing wooden furniture might have a different classification of flaws. Ultimately, the custom made solution made it possible to make the defect identification detailed and tailored to Prohan’s specific requirements.
Future Plans Involve Equipping the System with Knot Detection Capabilities
The solution that DAC.digital delivered is scalable and can be made even more advanced, using an algorithm to analyse additional parameters, in line with the evolving needs of the client. For example, the system can be trained to identify different types of knots—both aesthetic ones that add value to the furniture and pathological ones that need to be cut out.
Currently, wooden panels pass through a machine which cuts them into thinner boards that the knots are obtained from. The operator analyses the boards and uses a fluorescent marker to mark the knots that should be either kept or removed. The laser-powered machine then cuts the defects out from the whole length of the board, optimising them to avoid waste. In the ideal scenario, however, the operator would just put the boards into the machine that would be capable of detecting which knots to keep without the need for fluorescent markings.
Knot analysis was not introduced in the initial solution due to DAC.digital focusing on offering maximum project ROI in relation to the client’s budget. With this approach, the client has been able to first test the solution’s effectiveness in response to more immediate quality control needs while having the option to introduce such an enhancement in the future.
Got a Project in Mind? Let’s Discuss It.
Contact us!
Send us an email: [email protected]
