ObjectiveThe tracheid of softwood not only has the function as a medium of nutrient transportation, but also is a strong support to the trees, and its state has close relationship to the mechanical properties of timber. The investigation of internal relationship between the distribution of tracheid and the mechanical properties of wood is of great significance for the prediction of compressive elastic modulus of wood.

MethodThis paper starts with the tracheid effect of coniferous timber and introduces a set of detection platform covering the functions of light source, spot collection, spot analyses and plate traversal to build the numeric relationship between fiber angle distribution and compressive elastic modulus. First, least square method was used to fit the ellipse contour of the spots to measure the fiber angle; second, by analyzing the measurement error of fiber angle, a filtering method with mean value of 20 was selected to improve the accuracy of fiber angle measurement, and then the collection of fiber angle was completed after a traversal sampling. Finally, taking the mean value, diving coefficient and standard deviation of fiber angle distribution on the two surfaces of the plate as input, and the compressive modulus of the sample as output, a four-layer neural network with 6 inputs and 1 output was constructed to predict the compressive elastic modulus. To testify the effect of the study, 100 samples of Larix gmelini were processed in accordance with the requirement of GB/T 15777—1995, the National Standard of Compressive Modulus of Elasticity, and divided into training and testing samples with proportion of 3:1 after collecting their fiber angle and mechanic truth value with the detection platform and testing machines.

ResultThe results of the experiment revealed that when the average frequency of filtering was 20, the measurement error of the fiber angle acquisition was less than 0.65 degrees and the same time, the precision of compression modulus of the network prediction could reach 90.80%.

ConclusionThe compressive elastic modulus of the softwood can be predicted by collecting the fiber angle distribution. The combination of the least squares and filtering method can effectively express the characteristic information and ensure the measurement precision of the fiber angle. The mean value, diving coefficient and standard deviation can effectively describe the distribution characteristics of the fiber angle. By selecting different features as input, the elastic modulus prediction accuracy is directly affected. In the experiment of this paper, with the double-sided feature as input, the elastic modulus prediction accuracy is the highest.