【作者】 管雪梅；

【导师】 郭明辉；

【作者基本信息】 东北林业大学 ， 木材科学与技术， 2011， 博士

【摘要】 木材颜色是决定消费者印象的重要因素之一,为了提高木制品的装饰作用和产品价值,实现人工林木材的高效利用,需要通过染色技术改良劣质材。木材染色中的一个重要环节就是配色,其对染色后木材的颜色质量至关重要。在确定工艺的情况下,将计算机智能配色的方法用于木材染色配色过程中,加快染料配方生成的速度,极大地提高工作效率,节约成本。本研究以东北常见针叶材(樟子松)和阔叶材(大青杨)为研究对象,在研究了其染色工艺的基础上,探讨了利用传统计算机配色模型和利用智能决策手段对木材染色颜色配方的预测问题。具体包括,通过测定樟子松和大青杨单板解剖构造与染色效果的相关指标,对其进行多元回归分析,确定了影响木材染色效果的主要解剖因子。其二,利用计算机配色技术中的三刺激值法,对仿珍贵材贵木材染料配方的调配方法进行了研究。其三,利用RBF神经网络建立了配方预测模型,并利用仿珍贵材的数据对模型进行验证,初步说明了智能算法对于木材染色配方预测的有效性,在此基础上,提出了种基于隐层节点改进的RBF神经网络模型,并将其运用到配方预测中。其四,根据已确定的影响木材染色效果的解剖特性,建立了基于模糊神经网络预测模型,其输入增加了解剖特性,并根据研究对象的特点,提出了一种改进的隶属度函数,建立了基于改进隶属度函数的模糊神经网络预测模型。其五,建立了基于动态模糊神经网络的木材染色配方预测模型。最后,将以上方法运用C语言实现,建立了木材染色配方预测平台,对于工业普及起到至关重要的作用。研究在以下方面取得结果：(1)分析了染色工艺各参数对樟子松的表面色差的影响,确定染色工艺为：染料浓度为1%,渗透剂JFC浓度为0.1%、纯碱浓度为2％.NaCl浓度为1.5%、温度为85℃、染色时间60min、固色时间40 min、浴比17：1。(2)确定了影响樟子松木材染色效果的主要解剖因子为：管胞比量、木射线比量、树脂道比量和晚材管胞长度等因子；影响大青杨木材染色效果的主要解剖因子为：木材的早材导管直径、早材纤维长度、导管比量、木纤维比量和木射线比量。(3)利用计算机配色技术中的三刺激值法,得到樟子松单板染珍贵材的染料浓度配比分别是,仿鸡翅木(早材)：活性艳红X-3B为0.137%；活性黄X-R为0.229%；活性蓝X-R为0.042%。仿鸡翅木(晚材)：活性艳红X-3B为0.176%；活性黄X-R为0.256%；活性蓝X-R为0.165%。仿花梨木(早材)：活性艳红X-3B为0.117%；活性黄X-R为0.306%；活性蓝X-R为0.077%。仿花梨木(晚材)：活性艳红X-3B为0.162%；活性黄X-R为0.459%；活性蓝X-R为0.062%；利用大青杨单板染珍贵材的染料浓度配比分别是,仿紫檀：活性艳红X-3B为0.146%;活性黄X-R为0.184%；活性蓝X-R为0.037%。仿黑酸枝：性艳红X-3B为0.361%：活性黄X-R为0.612%；活性蓝X-R为0.179%。仿黑胡桃：性艳红X-3B为0.269%；活性黄X-R为0.203%；活性蓝X-R为0.074%；仿柚木：性艳红X-3B为0.122%；活性黄X-R为0.417%;活性蓝X-R为0.088%。(4)利用RBF神经网络建立了配方预测模型,结果表明樟子松的两种颜色空间仿珍贵材效果比较,L*a*b*空间与CMY空间的收敛速度差别不大,而仿珍贵材得到的平均误差有较大差别,分别是0.98%和1.81%,综合考虑确定L*a*b*空间作为研究樟子松的研究对象；大青杨的两种颜色空间仿珍贵材效果比较,L*a*b*空间与CMY空间的收敛速度差别较大,分别是1666步和610步,选取CMY空间为研究大青杨的颜色空间对象。从仿珍贵材珍贵材的数据来看,输出数据不是十分理想,最大误差达到8.24%。(5)基于RBF模型的缺点,提出了一种基于隐层节点改进的RBF神经网络模型,此模型的改进有效的解决了原来模型熟练速度较慢的问题,樟子松分析模型189步可以收敛,大青杨模型137步就收敛了,速度几乎可以达到在线训练的标准。从模型精度看,大青杨模型精度有所改善,但改善不多,樟子松模型没什么改变,甚至有所下降,所以此模型的改进对于精度的改善不大。从仿珍贵材的数据来看,精度有所改善,最大误差为4.36%,说明此模型的范化能力还有待进一步改善。(6)建立了基于模糊神经网络配方预测模型,并根据研究对象的特点,提出了一种改进的隶属度函数,建立了基于改进隶属度函数的模糊神经网络预测模型,结果显示樟子松和大青杨模型的误差都有所改善,分别是0.68%和0.62%,结果理想,从模型对仿珍贵材的配方预测效果来看,误差也有了明显的改善,最大误差为1.90%,基本可以接受。(7)提出了利用动态模糊神经网络建立预测模型(D-FNN),并提出一种适合木材染色配方预测的D-FNN模型及学习算法,结果显示,模型运行速度较快,且参数设置较容易,不需要过多的去设置模糊条件,大大节省了时间。从模型对仿珍贵材的配方预测效果来看,误差也有了明显的改善,最大误差也只有1.25%,说明模型的范化能力有了较大的改善。更多还原

【Abstract】 The wood color is one of the important factors deciding consumer impression. In order to improve the wood products’decoration function and product value, and realize the high efficient utilization of the plantation forest, the dyeing technique is needed to improve the plantation wood. An important link in the wood dyeing is color matching. The wood color after dyeing is the key for the product. The artificial color matching requires the personnel of high quality. It is a time-consuming task and is difficult to meet the modern industrial production requirement. In addition, it needs high cost, but it is of poor accuracy. With the increasing need of wood dyeing, after confirming the technique, the computer intelligent color measuring and matching method can be applied into wood dyeing and matching, so as to quicken the dye matching, significantly improve the work efficiency, and save costs.This research took the coniferous tree(Pinus sylvestris) and broadleaf tree(Populus ussuriensis) as the research objects. Based on the research of their dyeing technique, the research discussed that utilizing the traditional computer’s color matching model and the intelligent decision making means to deal with the forecast of the wood dyeing color formula. The details included:through test the index related to the dyeing effect and anatomical structure of Pinus sylvestris and Populus ussuriensis, make analysis on the Multiple Regression and Correlation Analysis, to confirm the main anatomical factors affecting the wood dyeing effect. Second, utilize the computer color matching technique’s tristimulus values to make research on the matching method of simulated precious wood dyeing formula. Third, utilize the RBF neutral network to establish the formula’s forecast model and utilize the simulated precious wood data to verify the model, so as to indicate the effectiveness of the intelligent algorithm applied to forecast the wood dyeing formula. Based on it, an improved RBF neutral network model based on the hidden units will be put forward, and utilize it into the formula forecast. Fourth, according to the determined anatomical features affecting the wood dyeing effect, the input of a forecast model based on the fuzzy neutral network increases the anatomical features; according to the characteristics of the research object, an improved membership grade function is put forward; then, establish the fuzzy neutral network forecast model based on the improved membership grade function. Fifth, establish a wood dyeing formula forecast model based on the motion blur neutral network. At last, realize above methods by C language, and establish the wood dyeing formula forecast platform. It plays important role in the industrial popularity.The research maked following achievements:(1) Analyze the effect of dyeing technique’s parameter on the surface color difference and dye absorbing rate; based on the dye concentration, dyeing temperature, dyeing time and bath ratio to analyze the variance of orthogonal test and get the good dyeing technique:dye concentration 1%, penetrating agent’s JFC concentration 0.1%, sodium carbonate concentration 2%, NaCl concentration 1.5%, temperature 85℃, dyeing time 60min, fixation time 40 min and bath ratio 17:1.(2) Confirm the main anatomical factors affecting the dyeing effect of Pinus sylvestris: tracheid proportion, wood rays proportion, resin passage proportion, late wood tracheid length; the main anatomical factors affecting the dyeing effect of Populus ussuriensis are:early wood vessel diameter, early wood fiber length, vessel proportion, wood fiber proportion and wood rays proportion.(3) Utilize the computer color matching technique’s tristimulus values to calculate the single board dyed simulated previous wood’s dye concentration proportion of the Mongolia pine:Simulated Wenge (early wood):reactive brilliant red X-3B is 0.137%; reactive yellow X-R is 0.229%; reactive blue X-R为0.042%. Simulated Wenge (late wood):reactive brilliant redX-3B is 0.176%; reactive yellow X-R is 0.256%; reactive blue X-R is 0.165%. Simulated rose wood (early wood):reactive brilliant red X-3B is 0.117%; reactive yellow X-R is 0.306%; reactive blue X-R is 0.077%. Simulated rose wood (late wood):reactive brilliant red X-3B is 0.162%; reactive yellow X-R is 0.459%; reactive blue X-R is 0.062%; calculate the single board dyed simulated previous wood’s dye concentration proportion of the Populus ussuriensis: simulated red sandalwood:reactive brilliant red X-3B is 0.146%; reactive yellow X-R is 0.184%; reactive blue X-R is 0.037%. Simulated cocobolo dalbergia retusa.reactive brilliant red X-3B is 0.361%; reactive yellow X-R is 0.612%; reactive blue X-R is 0.179%. Simulated black walnut:reactive brilliant red X-3B is 0.269%; reactive yellow X-R is 0.203%; reactive blue X-R is 0.074%; simulated teakwood:reactive brilliant red X-3B is 0.122%; reactive yellow X-R is 0.417%; reactive blue X-R is 0.088%。(4) Utilize RBF neutral network to establish the formula forecast model. The result shows that, comparing the two colors spatial simulation effect of Pinus sylvestris, there is a little difference in convergence rate between the L*a*b* space and CMY space. But the average error obtained from the simulation has big difference and is 0.98%and 1.81% respectively. Through the general consideration, it is confirmed to take L*a*b* space as the Pinus sylvestris’s research object. Comparing the spatial simulation effect of Populus ussuriensis’two colors, there is a big difference in convergence rate between the L*a*b* space and CMY space,1666 steps and 610 steps respectively. Thus, select CMY space as the spatial object to make research on the color of Populus ussuriensis. Based on the simulated precious wood date, the output data is not very ideal. The maximum error reaches 8.24%.(5) Based on the shortcomings of RBF model, an improved RBF neutral network model based on the hidden units is put forward. The model improvement effectively solves the problems in slow proficiency. The analysis model of Pinus sylvestris can be converged 189 steps and the Populus ussuriensis model can be converged 137 steps. The speed almost reaches the online training standard. Based on the model precision, the Populus ussuriensis model is improved, but not significantly. The Mongolia pine model has no change and sometimes declines. Thus, the model’s precision is not improved significantly. Based on the simulated precious wood data, the precision is improved and the maximum error is 4.36%. It shows that the normalization ability of the model needs further improvement.(6) Establish the formula forecast model based on the fuzzy neutral network, put forward an improved membership grade function according to the characteristics of the research object, and establish the formula forecast model based on fuzzy neutral network with the improved membership grade function. The result shows that the error of Mongolia pine and Populus ussuriensis model is improved,0.68%and 0.62%respectively. The result is ideal. Based on the model effect to forecast the simulated precious wood formula, the error has been improved significantly. The maximum error is 1.90%. Basically, it can be accepted.(7) Put forward a forecast model (D-FNN) established based on the motion fuzzy neutral network and put forward a D-FNN model and algorithm suitable for forecasting the wood dyeing formula. The result shows that the model operates fast and the parameter is easily to be set up. There is no need to set up too many fuzzy condition. In addition, the model saves a lot of time. Based on the model effect to forecast the simulated precious wood formula, the error has been improved significantly. The maximum error is 1.25%. It shows that the normalization ability of the model is improved significantly.更多还原