Research Report
Current research
project "Advanced methods for computer-aided bucking of Scots pine, 2001-2003"
in Finland has focused on four different subtargets. In the work carried
out at the University of Tampere prediction models for harvesting operation
has been developed. Moderate prediction models can be created for diameter
at dead branch height, stem height and live branch height and diameter
at live branch height. The prediction models developed can be utilized
in two main ways in harvesting. One possible application is to assess the
quality limits of a stem. This information can be particularly useful in
the optimization of crosscutting. On the other hand, since the values of
stem characteristics lie on the stem curve we may use the predicted values
of these and the known part of the stem jointly in the prediction of stem
curve measurements. Preliminary results of the use of smoothing spline
functions are very promising.
Ms. Anne Puustelli has in her doctoral thesis focused in combining Bayesian statistical approach into tree prediction problem. The predictive distributions use only small proportion of trees which is the data received while harvesting the first 'corner' of the stand. The idea is to use the data received from the trees already harvested to create more accurate models for predicting the characteristics for the rest of the trees. If better prediction of the dbh distribution or the joint distribution of dbh and height could be constructed the information could be utilized in optimizing the crosscutting of the rest of the trees to be harvested. Mr. Kivinen in has in his doctoral thesis focused on developing new control methods that generate optimum price and/or demand matrices for each stand to be harvested. In the first phase a fuzzy logic based calibration system that allocates a stand-specific price list for a given log grade has been designed. In the second phase, an adaptive search algorithm that parallelly optimizes the price lists for a large number of forest stands has been developed. The algorithm based on the genetic algorithm (GA) is a versatile tool that can be used not only to derive an optimal price list for each stand but also to divide the overall target log distribution into stand-specific subtargets. The performance tests with the developed systems show that controlling price lists prior to harvesting is beneficial in most cases provided that the optimization procedure has a sufficiently reliable estimate of the characteristics of the stands to be harvested. The extensive test sawings carried out at the University of Joensuu indicate that is rather difficult to estimate internal knot properties by aid of external branch limits or by other visual defects (Uusitalo 2002). The findings strengthens the conclusion that it is not possible to predict properties of a single log with a high degree of precision due to their natural, stochastic variability. It can then be concluded that bucking operations can be automated more than have previously been believed.
Previous page | Top of page |