OIXIO | Lumber AI
Data & AI for Manufacturing
OIXIO | Lumber production optimization with a proprietary algorithm based on stochastic evolutionary optimization methods
Forecasting – use Machine Learning and Data Science to forecast future events from historical data.
Production Modelling – use AI to find and recommend the optimum production plan for any given set of data and constraints. Simulate production into future periods to identify gaps and shortages.
Our modelling solution uses stochastic AI optimization algorithms (a self-adaptive parallelized version of differential evolution) to simulate the production into the future with a 6 weeks of time horizon. A custom meta-algorithm is used to search among the possible solutions which satisfy the needs of order deadlines
and timber expiration. A custom cost function combines the given multiobjective problem into a single evaluation metric and allows for the model weights to be tuned in different steps of the meta-algorithm
which helps with flexibility. Weight extraction for different steps in the meta-algorithm is done with an Analytic Hierarchy Process utilizing the Saaty scale to evaluate the priority of pairwise objectives.
Forecasting timber flow
Classic time-series forecasting machine learning methods are used to learn from historical data and predict the timber flow for following 52 weeks. The methods work together with an adaptive algorithm which selects the best method per each specific timber type to predict with maximum accuracy.
Production modelling and recommending the optimum solution
The aim is to model the process and provide the client with production recommendations which prevent the potential gaps and shortages all while keeping material loss to a minimum. A single percent increase in production reduces the production cost by approx. 200K € per year. Our algorithm manages to improve the production by an average of 2.5% (saving up to 500K €).
The algorithm works through a set of roughly 2 000 000 prospective solutions in a search space with a lower bound of roughly 10^50 possible solutions for a planning horizon and finds the optimal solution roughly 80% of the time.
Contact Us
OIXIO Group