Skip to main content

Neural Grader - Digitizing the Wood Industry

Periodic Reporting for period 1 - Neural Grader (Neural Grader - Digitizing the Wood Industry)

Reporting period: 2019-02-01 to 2019-07-31

Fordaq SA has been leading the wood products B2B global market for almost 20 years. We provide our 200.000
user companies with a specialized digital platform for buying products. We have observed how low digitization of
the wood industry poses problems for success & often survival for SMEs. There is an urgent need of an automated
solution for processing lumber. The current methods are too expensive for SMEs and in practice lead to errors,
enormous economic & environmental waste.
Our newest product - Neural Grader is the solution to this problem, having the potential of disrupting the European
and global wood industry. Its innovation lies in the industrial cameras that capture images of each board and, using
state-of-the-art Artificial Intelligence methods, accurately classifies wood quality. Neural Grader also computes the
optimal way for cutting the boards and sends commands to the cutting machine, preventing wood waste.
Competitive advantages:
- Efficiency: it detects defects and grads lumber at high speed, with accuracy that exceeds the one of existing systems;
- Affordability: the price will be 1/4 of the price of our competitors’ products, making it accessible for the SMEs, who
are currently overwhelmed by overly expensive technology and dominated by large companies;
- Full-control of production: companies can manage the production lines and inventory and focus on quality.
We have a large base of existing clients and 10 sawmills in Europe & North America sent us Letters of Support.

Environmental benefits: Studies show that such an automated solution leads to improvements in wood waste reduction
of over 20%. The wood industry plays a key part in climate change and lowering the carbon footprint.

Social benefits: Currently, the wood industry in Europe is more expensive compared to the one in Asian countries, where
labor is significantly cheaper. Neural Grader could tip the scale in favor of the European industry.
Our team has gathered information regarding the market needs and identified the potential clients for the Neural Grader by carrying out surveys with a large number of sawmills on trade shows all over the world. Moreover, we have visited facilities and production sites in order to see the real world scenarios where Neural Grader will be integrated. We have managed to segment the market and estimate the size of each target group, and, based on this segmentation, decide the direction the product should follow. Initially, we will focus on a group of clients with very similar requirements - green lumber producers - which need to automate the manual graders for efficiency purposes. We based our decision taking into account not only the size of the segment, but also their technical requirements. They need a robust system that can perform continuously, with an accuracy that is slightly lower than the one required by other groups of customers. In terms of speed, the limit is higher than we initially expected - 3 seconds / board.

Using the technical specifications, we have built a new prototype development plan, adjusting the hardware and software requirements accordingly. Also, we started building a data-gathering hardware system in order to launch the training of new specialized AI algorithms.

The IPR study has resulted in a thorough analysis of existing patents in the industry and has offered us insight regarding the next steps that need to be followed for successfully protecting our research and development work.
Having visited and met a large number of companies that operate in the industry and would potentially need a Neural Grader, we have concluded that the quantity of wasted raw material that their facilities generate could be vastly reduced through a specialized system by a minimum of 20%. Companies that use manual graders have expressed their concern regarding the fact that graders do not deliver consistent results. Hiring graders is a very difficult task due to the lack of interest for manual labor. A system such as Neural Grader would allow companies to reuse human graders for less physical and more technical jobs such as operating machines.

From a R&D point of view, our main goal by the end of the project is to have a functioning test-environment prototype for the market segment we have targeted. Early tests show promising results for the 2 quality classes, having already surpassed human accuracy. We are confident that the 90% accuracy threshold can be reached with ease.
Hardware structure of the Neural Grader system
Defect delimiters on a board which mark where it should be cut
Example of manual grader marking defects with flourescent chalk on a board
Example of prototype components installed on a production line
Neural Grader output example - splitting a board into clear patches and defect areas marked in red