Arauco is one of the marquee clients of Aindra Labs, a crown jewel as far as our global clients are concerned. To give a brief introduction, Celulosa Arauco y Constitución (popularly known as ARAUCO, sometimes also called CELCO) is Chilean wood pulp, engineered wood, and forestry company. It’s the largest Forestry company in Chile and is a Chilean multinational (MNC) that is present in several countries around the world like Latin America (Brazil, Argentina, etc), North America (the United States and Canada), Europe, Mexico, etc. Arauco was founded in September 1979 in Chile.

The project we are doing for Arauco is called Proyecto Siscor (Proyecto is Spanish for Project while Siscor is the project name). It all started when Aindra Labs Co-Founder Ritabrata Bhaumik (Rito) was in Chile as part of Cohort 18 of one of the World’s top Start-up Accelerators “Start-up Chile”. Start-up Chile Executive Director Rocio Fonseca was impressed with Aindra Labs and Aindra’s AI platform- ‘Drishti’ which has prebuilt recognition and prediction models which identifies a range of concepts including ‘Objects, Incidents, Emotions, Targets and Predict’ which are used in various Industrial grade deployments.

Rocio was kind enough to introduce Rito to Arauco Corporate Innovation Chief Francisco Lozano, who in turn introduced Rito to Senior Innovation Manager at Arauco Mr. Marcelo Romero Morstadt. After several discussions with Marcelo and Arauco Innovation Consultant Martin Buttazzoni and a field visit to Mariquina (near Valdivia in South Chile) plant of Arauco by Rito, Aindra Labs was awarded a Pilot project (as proof-of-concept or PoC) on AI Platform powered Computer Vision-based Color Recognition in 2018 beginning. The success of this Pilot led to Proyecto Siscor awarded to Aindra Labs in partnership with TTC Systems, the Chilean company led by Juan Felipe Vallejo who was also part of Cohort 18 of Start-up Chile.

The background of Proyecto Siscor is as follows. Every day Forestry Company Plants process thousands of wooden logs for various purposes like producing engineered wood products, panels, and lumber used in furniture, construction, and packaging, etc. One of the big problems that Forestry Companies face is a high amount of wooden log rejections (as high as ~5%) due to poor quality of wooden logs supplied by the contractor (Emsefor), causing millions of dollars of annual loss to the Forestry companies! This can be reduced by bringing in more accountability to the process by identifying the contractor (Emsefor) responsible for the rejected wooden logs and taking the appropriate action accordingly. The primary objective of Proyecto Siscor in the Mariquina Plant of Arauco is to identify the contractor (Emsefor) and the specific sub-team (of that Contractor) who supplied any rejected wooden log.

Procedure to identify contractor from wood logs

The biggest challenge is to decide a methodology by which we can identify the contractor (Emsefor) and the sub-teams responsible for the rejected wooden logs. We deliberated several ways of identifying the contractor such as having a contractor face printed and pasted on a wooden log, to pasting a paper template on the wooden log but all these ways are not feasible as these wooden logs are kept in the open space and because of the rain, sunlight and other climatic conditions these marks would be unrecognizable. Finally, we agreed on a solution where Arauco contractors (Emsefor) were instructed to use two levels of identifiers for spray painting on their supplied wooden log’s face for the purpose of this project. One is the color of the Paint itself and the second one is the Shape/Symbol of the Painting on a wooden log’s face. Color of the paint is assigned to identify the Contractor (Emsefor) and the Shape/Symbol has been used to identify the sub-teams under each Contractor.

The challenge of Colors and shapes:

Once the procedure to identify the contractors was in place, the task was to assign a combination of shape and color for five contractors, each having multiple subcontractors, such that the combination is unique and recognizable for our AI-powered Computer Vision-based algorithms. Here comes the catch.

The figure below shows the interesting play of mixing of colors in general and how we perceive the net result of the mix as opposed to the actual color. 


The painting below shows the same colorful cube in red light and green light. The squares on the cube are cyan, magenta, ochre, blue, and white. Or so they seem. What colors are those squares really, objectively? In fact, the cyan square in the bottom corner of the red-lit scene is exactly the same color mixture as the red square in the upper corner of the green-lit scene. Refer to this blog.

The variations we encountered while building the dataset were the following:

  • Wood log face has multiple shades, ranges from light brown to dark brown almost black in some cases. This could be due to dirt. In fact, wood logs have concentric rings of different shades too.
  • Several lighting variations as the plant is in open ground with sunlight falling at different angles and ambient light in general keeps changing according to different times of the day.
  • The selection of colors more than the standard Red, Green, Blue was required as we had to support more than 5 Contractors for this project. Colors and their shades are generally hard to recognize given variation in lighting and wood log faces.
The left image and right image is of light green but look completely different on different wood log faces and lighting conditions

Given such variations, a dataset of wood logs with corresponding shapes and colors were collected over the course of the project and with rigorous training and constantly fine-tuning our deep learning models, and careful selection of imaging hardware, we are able to provide a robust Contractor identification solution based on images only. 

The hardware integration of using cameras to take photos of rejected wood log faces (real-time in Descortezado machine as wood-logs get processed there) and updating the results from computer vision and machine learning algorithms of Aindra Labs into Arauco database was executed by TTC Systems, partner of Aindra Labs in Proyecto Siscor. Following is the high-level workflow for Proyecto Siscor :

Proyecto Siscor is being put in Production now and Arauco intranet featured this project recently, including a short clip from Rito (the video is in Spanish – though Rito’s part after 1st 36 seconds is in English) We feel proud that our work for Arauco has also been featured in Start-Up Chile Newsletter in September 2019. Proyecto Siscor is a significant milestone in the journey of our Start-up Aindra Labs as this underscores our technical and global delivery credentials of a large scale industrial product using Our AI platform with its prebuilt recognition and prediction models.

Aindra’s AI-powered ‘Drishti’ platform enables organisations to build specific AI predictive and visual analytics products & solutions.