Predicting moisture content in kiln dried timbers using machine learning

The uniformity of final moisture content within a drying timber batch is crucial. Lack of such uniformity leads to undesirable moisture ranges, thus producing large percentages of over-dried and under-dried timber sub-populations, resulting in substantial quality degradation and value downgrade. Because it is an unmanageable task to collectively correlate various pre-dry timber factors to the moisture of the post-dry timber population, predicting the value and variability of final moisture is still a great challenge. In this study, for a dataset of 2304 timbers, three artificial neural networks (ANNs), including multilayer perceptron (MLP), radial basis function (RBF), and group method of data handling (GMDH) are employed to create a predictive model that connects selected initial wood attributes (basic density, initial weight, initial moisture, and target moisture) to the final moisture. Moreover, Monte Carlo simulation was applied to improve the performance of the predictive models. For each ANN approach, seven network configurations are constructed with different factors in their input layers. As a result, GMDH showed the best performance in predicting final moisture mainly due to its self-tuning capability, especially when target moisture, initial moisture, and weight were included in the model (MSE = 5.2%). Moreover, while MLP showed satisfactory performance in different configurations, RBF failed to predict final moisture in some configurations. The greatest improvement occurred when Monte Carlo simulation was combined with the RBF model, and only target moisture and initial moisture were included in the predictive model (MSE = 70.5%). Overall, initial moisture and basic density were the most and least significant parameters in predicting final moisture, respectively.